Learner Guide

Course Code: TGS-2022017524 · Version 1.0

Document Version Control Record

Version	Effective Date	Summary of Changes	Author
1.0	24 Jun 2026	Initial release — full 3-day, 17-lab learner guide.	Course Development Team

Common Errors & Quick Fixes
Day 1 — Foundations & Power Automate
Day 2 — Building Business Agents with Copilot Studio
Day 3 — End-to-End Workflow Automation & Workshop

Welcome! This Learner Guide takes you click-by-click through every hands-on lab in the WSQ course Microsoft Copilot Studio & Power Automate for Business Workflow Automation (Course Code: TGS-2022017524). Over three days you go from your first Power Automate flow, to AI business agents in Microsoft Copilot Studio, to complete end-to-end automated workflows — and finish by building your own in a capstone project.

Work through the labs in order: each one builds on the skills of the lab before it. Whenever you see a Checkpoint, stop and confirm your flow or agent behaves as described before moving on. The Common Errors & Quick Fixes and per-lab Troubleshooting tables will get you unstuck fast.

Course flow at a glance — Day 1: Workflow automation concepts + Power Automate (Labs 0-5). Day 2: Business agents in Copilot Studio (Labs 6-11). Day 3: End-to-end workflows + capstone (Labs 12-16).

Common Errors & Quick Fixes

Keep this handy — these are the issues learners hit most often, with the one-line fix:

Symptom	Cause	Fix
“Unauthorized” when sending email	Outlook connection expired or the account has no mailbox	Reconnect the Office 365 Outlook connection with a mailbox-enabled account; both connections must be green ✓
Approval fails: “valid users in the organization”	Approver typed as an external email, not a tenant user	Pick the approver from the people-picker dropdown (a real user in your tenant; yourself is fine for testing)
Date logs as literal text ‘utcNow()’	Expression typed into the field as text	Enter it via the fx / Expression editor so it becomes a coloured token
Excel cell shows ########	The column is only too narrow	Auto-fit the column — the value is fine
An unwanted ‘For each’ wraps your action	You inserted a list/array value into a single-value field	Use single-value fields (Outcome, trigger inputs); delete the For each and re-add a plain action
Agent can’t see its flow	Agent and flow are in different environments	Set both Copilot Studio and Power Automate to the same environment (NUS Copilot Sandbox)

Day 1 — Foundations & Power Automate

Module 1: Introduction to Workflow Automation

Read this before the Day 1 labs. It explains the "why" behind everything you'll build over the next three days. ~20 minutes.

By the end of this reading you will be able to:

Explain what a business workflow is and which tasks are worth automating
Tell the difference between an agent that only talks and a workflow that actually does work
Use the four building-block terms — trigger, actions, outputs, steps — that recur in every single lab

1. What is business workflow automation?

A business workflow is a repeatable series of steps that gets work done. You already run dozens of them by hand every week. For example:

A customer emails an enquiry → someone records it in a spreadsheet → a salesperson is notified → they reply.

Four small steps, but each one takes attention, and any of them can be forgotten, delayed, or done inconsistently.

Automation means letting software perform those steps for you — consistently, instantly, and the same way every time — instead of doing them by hand. The best candidates for automation share three traits:

Trait	What it looks like	Everyday example
Repetitive	Done the same way many times	Logging enquiries, sending confirmations
Rule-based	Clear "if this, then that" logic	If amount > $1,000, get approval
Time-consuming	Manual copying, chasing, re-typing	Re-keying form data into a spreadsheet

What you gain: fewer mistakes, faster turnaround, a complete record of what happened, nothing falling through the cracks — and people freed up for higher-value work that actually needs human judgement.

Rule of thumb: if you find yourself doing the same clicking, copying, and emailing over and over, it is probably a workflow waiting to be automated.

In this course you automate workflows with two Microsoft tools that work as a pair:

Power Automate — builds the automated steps (the flows) that do the work.
Copilot Studio — builds AI agents that understand natural-language requests and feed clean, structured data into those flows.

You'll learn Power Automate first (Day 1), then agents (Day 2), then how to combine them (Day 3).

2. Standalone agents vs integrated workflows

It helps to be clear, from the very start, about the difference between an agent that just answers and a workflow that acts.

	Standalone agent	Integrated workflow
What it does	Chats and answers questions	Performs real actions across systems
Example	"What's your return policy?"	Logs the enquiry, emails the team, files a ticket
Powered by	Copilot Studio alone	Copilot Studio + Power Automate
Outcome	Information	Action + record + notification

A standalone agent is genuinely useful, but it only talks. Ask it a question, get an answer — the conversation ends and nothing changes in your business systems.

An integrated workflow connects that same agent to Power Automate, so the conversation actually triggers work: a record gets saved, an email gets sent, an approval gets started. The customer feels heard and the back-office task is already done.

Standalone agent:   User asks  →  Agent answers           (talk only)

Integrated:         User asks  →  Agent answers
                                   └─► triggers a flow ─►  log + email + approval
                                                          (talk AND action)

That bridge between conversation and action is the heart of this course.

3. Workflow logic: the four building blocks

Every workflow you build — whether a simple flow or a full agent-driven process — is made of these four parts. Learn these terms now; you will use them in every lab.

Trigger — what starts the workflow

The single event that kicks everything off. Examples:

An email is received
A file is uploaded to a folder
A form is submitted
A schedule is reached (e.g. every morning at 9 AM)
An agent calls the flow

Every flow has exactly one trigger, and it does nothing until that trigger fires.

Actions — what the workflow does

The steps that run after the trigger, in order. Examples:

Send an email
Add a row to an Excel table
Start an approval and wait for a decision
Post a message to Teams
Create or update a record

Outputs — the data the workflow produces or passes along

The pieces of information that move between steps. Examples:

The sender's email address from a received email
The customer name and product captured by an agent
The approval result ("Approved" / "Rejected")

Outputs from one step become inputs to the next — in Power Automate this is called dynamic content, and it's how data flows down the chain.

Steps — the ordered sequence as a whole

Trigger + actions, arranged top-to-bottom, make up the steps of the workflow. Steps can also include:

Conditions — branching with if/else logic (e.g. if amount > $1,000, route to a manager)
Loops — repeat an action for each item in a list

Putting it together

TRIGGER            ACTIONS (steps)                       OUTPUTS
─────────          ──────────────────────────           ──────────
Email received  →  1. Read sender + subject          →  sender, subject
                   2. Add a row to Excel             →  logged record
                   3. Send notification to sales     →  confirmation

Read that left to right: an event happens, the flow performs a series of steps, and data (outputs) is produced and passed along the way. Keep this Trigger → Actions → Output mental model in your head — it is the spine of everything that follows.

4. How the three days fit together

Day	You build	New skill	What it gives you
Day 1	Flows in Power Automate	Triggers, actions, outputs	The "hands" that do work — email, Excel logging, approvals
Day 2	Agents in Copilot Studio	Structured capture, tools	The "brain & mouth" that talk to people and produce clean data
Day 3	Agents + flows together	Orchestration	Complete end-to-end business workflows, then your own

Day 1 is all about Power Automate. You'll master the trigger → action → output rhythm by building real flows you can run today.

Next: Module 2: Introduction to Power Automate

Module 2: Introduction to Power Automate

Read this after Module 1 and before the Day 1 labs. ~15 minutes.

This module turns the Trigger → Actions → Output idea from Module 1 into the actual tool you'll use all day: Power Automate. By the end you'll recognise the flow types, the common triggers and actions, and the one thing that trips up every beginner — connections.

1. What is Power Automate?

Power Automate is Microsoft's automation platform. It lets you build flows — automated sequences of steps that run across Microsoft 365 and hundreds of other apps, with little or no code. You design a flow visually in a browser by stacking one trigger and one or more actions (exactly the building blocks from Module 1).

With Power Automate you can:

React to events — an email arrives, a file is uploaded, a form is submitted
Move and transform data between systems — Outlook, Excel, SharePoint, Teams, Dataverse, and more
Run approvals, send notifications, and schedule recurring jobs
Be called by a Copilot Studio agent as a tool (you'll do this on Day 2)

You build flows at https://make.powerautomate.com, inside the environment you set up in Lab 0.

Low-code, not no-thought. You don't write code, but you do think like a designer: what starts the flow, what it does, and what data moves between steps.

Types of flow you'll build in this course

There are four flavours of flow. They differ only in how they start — once running, they all do the same kind of work.

Flow type	Started by	Example	Lab
Instant	A person clicking Run / a button	Send a confirmation email on demand	Lab 1–3
Scheduled	A clock/timetable (Recurrence)	Daily reminder at 9 AM	Lab 4
Automated	An event	New form response, new email, new file	Lab 5, Day 3
Agent flow	A Copilot Studio agent	Agent logs a request and notifies the team	Day 2–3

2. Common triggers

Every flow starts with exactly one trigger — the event that kicks it off. These are the ones you'll use most:

Trigger	Connector / name	Fires when…	Used in
Email received	Office 365 Outlook — "When a new email arrives (V3)"	Mail lands in a folder	Day 3, Lab 10
File upload	OneDrive / SharePoint — "When a file is created"	A document is dropped into a folder	Day 3, Lab 11
Form submission	Microsoft Forms — "When a new response is submitted"	Someone submits your form	Lab 5
Schedule	"Recurrence"	A timetable you define is reached	Lab 4
Manual	"Manually trigger a flow"	You press Run	Labs 1–3
Agent call	"When an agent calls the flow"	A Copilot Studio agent runs the flow as a tool	Day 2–3

Tip — build with a manual trigger first. When learning a new flow, start with a manual trigger so you can press Run and perfect the actions. Once they work, swap in the real trigger (form, email, schedule). The actions stay exactly the same — only the start event changes.

3. Creating workflow actions

After the trigger, you add actions — the work the flow performs. These are the core actions you'll lean on throughout the course:

Send emails

Office 365 Outlook → Send an email (V2). Send confirmations, notifications, and digests. Use dynamic content (outputs from earlier steps) to personalise the subject and body. (Lab 1)

Create Excel entries

Excel Online (Business) → Add a row into a table (and List rows present in a table). Log every record into a spreadsheet table — this becomes your audit trail and single source of truth. (Lab 2)

Excel actions only see data that lives inside a proper Table (Insert → Table), not loose cells. This is a classic first-time gotcha.

Notifications & approvals

Approvals → Start and wait for an approval. Pause the flow until a person approves or rejects, then branch on the Outcome. (Lab 3)
Notifications to Teams or email keep the right people informed at each step.

Actions are wired together with:

Dynamic content — one step's output becomes the next step's input (Module 1's "outputs")
Conditions — if/else branching to route the process

Heads-up on approvals: the approver you pick must be a real user in your Microsoft 365 tenant. Approvals can't be sent to an outside personal email address — pick someone in your organisation (yourself is fine for testing).

A note on expressions (fx)

Sometimes a field needs a small calculation or formatting — today's date, trimmed text, a number comparison. Power Automate provides an expression editor (the fx button) for this. You'll only need a few simple ones; we'll point them out in the labs when they appear.

4. Connections — how Power Automate reaches your apps

Each connector (Outlook, Excel, Approvals, Forms…) needs a connection — a saved sign-in that authorises the flow to act on your behalf. The first time you use a connector, you'll sign in and consent.

This is the number one source of "why won't my flow run?" for beginners, so commit it to memory:

Connection state	What you'll see	What to do
Healthy	Green / connected	Good to go
Broken / expired	Red ⚠️ "Reconnect"	Reconnect before running
Wrong rights	"Unauthorized" error	The account lacks rights (e.g. no mailbox) — fix or use a different account

The golden rule for every lab: green connection = ready; red connection = reconnect first. An Unauthorized error almost always means: reconnect the connector or check that the signed-in account actually has permission for that action.

5. Anatomy of a flow (what you'll see in the designer)

[ TRIGGER ]        ← one event that starts the flow
    │
[ Action 1 ]       ← e.g. Get details / read data
    │
[ Condition ]      ← optional if/else branch
   ├── If yes → [ Action ]
   └── If no  → [ Action ]
    │
[ Action 2 ]       ← e.g. Send an email / Add a row

The build-and-verify loop is the same in every lab:

Save the flow.
Test it — manually, or by triggering the real event.
Review the run history — green = success, red = error — to confirm it worked or to debug.

Get comfortable with this loop today; you'll repeat it dozens of times across the three days.

Next: Lab 1: Automated Email Workflow

Lab 0: Environment Setup — Create Your Copilot Studio & Power Automate Accounts

Lab Title

Environment Setup — Create Your Microsoft 365, Power Automate, and Copilot Studio Accounts

Lab Objectives

By the end of this lab, you will be able to:

Obtain a Microsoft 365 work or school account that can use the Power Platform
Create a dedicated Power Platform Sandbox environment (with Dataverse) for this course
Sign in to Power Automate and Copilot Studio and point both at the same environment
Confirm Outlook and Excel (OneDrive) are available for the later labs
Run a verification checklist so you start Lab 1 with zero surprises
Understand the difference between a work/school account and a personal account

Prerequisites

A web browser (Microsoft Edge or Google Chrome recommended)
A mobile phone (used once for security verification)
A credit card (only if you create a new Business trial in Option B — it is not charged during the free month)
About 30–40 minutes

Tip — Which account should I use? - Option A — You already have a Microsoft 365 work/school account (e.g. name@company.com provided by your organization). Try this first — you may already have everything you need. Skip Option B and go straight to Step 1. - Option B — You do NOT have a work/school account (you only have a personal @outlook.com / @gmail.com). Power Automate and Copilot Studio require a work or school account, so you'll create one via a free Microsoft 365 Business trial. Do Option B first, then continue from Step 1.

⚠️ Warning — Why not the Microsoft 365 Developer Program? As of 2024, Microsoft restricted the free Developer Program E5 sandbox to people with an active Visual Studio Enterprise or Professional subscription. If you sign up without one you'll see "You don't currently qualify for a Microsoft 365 Developer Program sandbox subscription." — so we do not use that path in this course. Use Option B instead.

Scenario

You work for ACME Pte Ltd and will spend the next three days automating ACME's business workflows. Before you can build anything, you need a clean place to work: a Microsoft 365 account, the Power Automate and Copilot Studio apps, and — importantly — one shared environment called NUS Copilot Sandbox that both apps point to. Getting this right now means every later lab "just works."

Step-by-Step Guide

Option B (only if you have no work/school account): Create a free Microsoft 365 Business trial (~15 minutes)

This creates a brand-new work account such as admin@yourname.onmicrosoft.com with Microsoft 365 (Outlook, Excel, OneDrive, SharePoint) — exactly what Power Automate and Copilot Studio need. Skip this entirely if you already have a work/school account.

Open a browser and go to https://www.microsoft.com/microsoft-365/business (or search "Microsoft 365 Business Standard free trial").
Choose Microsoft 365 Business Standard and select Try free for 1 month.
Enter an email address to start. When prompted, choose Set up account / Create a new account.
Fill in your details (name, business name — you may use your own name, country, phone for verification).
Create your sign-in details: a username and a domain, giving you something like admin@yourname.onmicrosoft.com. Write these down — this is the account you'll use for the entire course.
Verify with the code sent to your phone.
Add a payment method (credit card). You are not charged during the 1-month free trial.
Wait 1–2 minutes for provisioning. You now have a Microsoft 365 work tenant.

Tip: In a classroom, your trainer may provide a ready-made account. Ask before creating a trial so you don't duplicate accounts.

⚠️ Warning — Avoid surprise charges. If you don't intend to keep the trial, set a calendar reminder and cancel before the renewal date in the Microsoft 365 admin center → Billing → Your products.

Step 1: Sign in to the Microsoft 365 portal (~5 minutes)

Go to https://www.office.com (or https://m365.cloud.microsoft).
Select Sign in and enter your work/school account (Option A) or your new Business trial account (Option B), then your password.
If this is your first sign-in, you may be asked to set up multi-factor authentication (MFA). Follow the prompts using your mobile phone.
Once signed in, you should see the Microsoft 365 home page with app tiles (Outlook, Word, Excel, etc.).
Open Outlook (click the Outlook tile) and send yourself a quick test email to confirm it works — you'll rely on Outlook in Lab 1.
Open Excel and create a blank workbook to confirm it saves to OneDrive — you'll rely on this in Lab 2. You can discard the test workbook afterwards.

⚠️ Warning — No Outlook/Excel tiles? Your account may not have a Microsoft 365 license. The Send an email action in Lab 1 fails with "Unauthorized" when the account has no mailbox. Ask your IT administrator to assign a license, or use the Business trial account from Option B (which always has a mailbox).

Step 2: Create your "NUS Copilot Sandbox" environment (~7 minutes)

An environment is a container that holds your flows, agents, and data. For this course we'll create a dedicated Sandbox environment with Dataverse turned on, so Day 2/Day 3 Copilot Studio agents have a database to use.

Open a new tab and go to the Power Platform admin center: https://admin.powerplatform.microsoft.com.
Sign in with the same account from Step 1.
In the left menu, select Manage → Environments.
Select + New (top of the page).
Fill in the New environment panel: — Name: NUS Copilot Sandbox — Type: Sandbox — Region: your nearest region (e.g. Asia, Singapore) — Add a Dataverse data store?: Yes ← important
Select Next, accept the defaults (language English, currency your local currency), then select Save.
Wait 1–3 minutes. The environment appears in the list with status Ready. Refresh if needed.

Tip: If your tenant blocks creating environments (some organizations restrict this), use the existing default environment instead — just remember to select that same environment in both Power Automate and Copilot Studio in the next steps.

Step 3: Sign in to Power Automate and select the environment (~5 minutes)

Power Automate is where you build the automated workflows (called flows).

Open a new tab and go to https://make.powerautomate.com.
Sign in with the same account.
The first time, you may be asked to select your country/region — choose the correct one and select Get started.
You'll see the Power Automate home page with a left-hand menu: Home, Create, My flows, Templates, Connectors, Learn, More.
Look at the top-right corner — there is an Environment selector. Click it and choose NUS Copilot Sandbox (the one you created in Step 2). All your flows will be built here.
Select My flows in the left menu — it will be empty for now. That's expected.

⚠️ Warning: The environment selector is the single most common source of "where did my flow go?" confusion. If you build a flow in the wrong environment, it simply won't appear when you switch. Always confirm NUS Copilot Sandbox is showing top-right before you build.

Tip — Free Power Automate: A Power Automate use-rights plan is included with most Microsoft 365 licenses, which is enough for this course. If prompted, you can also start a free 90-day trial of Power Automate Premium.

Step 4: Sign in to Copilot Studio and match the same environment (~5 minutes)

Copilot Studio is where you build the AI agents (used on Day 2 and Day 3).

Open a new tab and go to https://copilotstudio.microsoft.com.
Sign in with the same account again.
If prompted, select your country/region and select Start free trial (or Try free). This activates a 30-day Copilot Studio trial at no cost.
Wait for the workspace to load. You'll see the Copilot Studio home page with options to Create an agent.
Look at the Environment selector in the top-right corner and choose NUS Copilot Sandbox — the same environment you selected in Power Automate.
Do not create an agent yet — you'll do that in a later lab. For now, just confirm the page loads in the correct environment.

⚠️ Warning — Both tools MUST use the same environment. Your agents (Copilot Studio) and your flows (Power Automate) can only call each other when they live in the same environment. If Power Automate shows NUS Copilot Sandbox but Copilot Studio shows Default (or vice-versa), they cannot connect. Fix it now by clicking the top-right selector in each tool and choosing NUS Copilot Sandbox.

Step 5: Verify your full setup (~5 minutes)

Run this quick checklist. Each item should already be true if the steps above succeeded.

#	Check	Where
1	I can sign in and see app tiles	https://office.com
2	I can open Outlook and send myself an email	Outlook
3	I can open Excel and it saves to OneDrive	Excel / OneDrive
4	My NUS Copilot Sandbox environment shows status Ready	https://admin.powerplatform.microsoft.com
5	Power Automate home page loads and NUS Copilot Sandbox is selected	https://make.powerautomate.com
6	Copilot Studio loads, my trial is active, and NUS Copilot Sandbox is selected	https://copilotstudio.microsoft.com
7	Power Automate and Copilot Studio show the same environment	Top-right selector in both

If all seven are checked, your environment is ready.

Checkpoint

You should now have:

✅ A working Microsoft 365 work/school account (Option A or B)
✅ A Power Platform Sandbox environment named NUS Copilot Sandbox with Dataverse = Yes, status Ready
✅ Power Automate open with NUS Copilot Sandbox selected top-right
✅ Copilot Studio open (trial active) with NUS Copilot Sandbox selected top-right
✅ Outlook and Excel (OneDrive) confirmed working

Troubleshooting

Problem	Solution
"You can't sign in here with a personal account"	Power Platform needs a work/school account. Use Option B to create one via the Business trial.
"You don't currently qualify for a Microsoft 365 Developer Program sandbox subscription"	The Developer Program now requires a Visual Studio subscription. Use Option B (Business trial) instead.
+ New environment button is greyed out / missing	Your tenant restricts environment creation. Ask an admin, or use the Default environment and select it in both tools.
Environment created but stuck on Preparing	Wait 2–3 minutes and refresh the Environments list; provisioning Dataverse takes a moment.
Power Automate says "no environment"	Refresh, re-select your region, then pick NUS Copilot Sandbox in the top-right selector.
Copilot Studio "Start free trial" button missing	You may already have a license — just proceed. Otherwise sign out and back in.
Different environment shows in each tool	Click the environment selector (top-right) in both tools and choose NUS Copilot Sandbox.
No Outlook/Excel tiles	Your account lacks a Microsoft 365 license (and possibly a mailbox) — ask IT or use the Business trial account (Option B).

Key Takeaways

Power Automate and Copilot Studio both need a work/school account — personal accounts won't work.
The Microsoft 365 Developer Program is no longer a free path; use a Business trial if you need an account.
An environment is the container for your work; this course uses one named NUS Copilot Sandbox (Sandbox type, Dataverse = Yes).
The #1 setup mistake is having the two tools on different environments — always verify the top-right selector matches in both.

Duration

~30–40 minutes

Next Steps

Read Module 1: Workflow Automation Concepts and Module 2: Introduction to Power Automate, then proceed to Lab 1: Automated Email Workflow.

Lab 1: Automated Email Workflow

Lab Title

Build an Automated Email Workflow in Power Automate

Lab Objectives

By the end of this lab, you will be able to:

Create a flow from a blank canvas in Power Automate's new designer
Configure a manual trigger with a text input
Add a Send an email (V2) action and create an Office 365 Outlook connection
Use dynamic content to insert the input value into the email body
Save, then Test → Run the flow and confirm the email is delivered
Read the run history to verify each step succeeded

Prerequisites

Completed Lab 0 (accounts ready)
Signed in at https://make.powerautomate.com with NUS Copilot Sandbox selected (top-right)
Outlook working with your account (a mailbox-enabled account)

Scenario

At ACME Pte Ltd, every customer enquiry should get an instant, personalized "thank you" reply so customers know they've been heard. In this lab you build the simplest possible automation — a flow you start by hand that emails a personalized confirmation. It teaches the core pattern you'll reuse all course long: Trigger → Action → Output.

Step-by-Step Guide

Step 1: Start a new flow (~5 minutes)

Go to https://make.powerautomate.com and sign in.
Confirm the Environment selector (top-right) shows NUS Copilot Sandbox — the environment from Lab 0.
In the left menu, select Create.
Under "Start from blank", select Instant cloud flow.

Tip: An instant flow is started manually by clicking a button — perfect for learning and testing. Later labs use automatic triggers (a form submission, a new email, etc.).

In the dialog box: — Flow name: Lab 1 - Send Confirmation Email — Choose how to trigger this flow: select Manually trigger a flow
Select Create. The new designer opens with one step already on the canvas: the trigger Manually trigger a flow.

Step 2: Add an input to the trigger (~5 minutes)

We'll let whoever runs the flow type a customer name, so the email can be personalized.

Select the trigger card Manually trigger a flow to open its panel on the right.
Select + Add an input.
From the type list, choose Text.
Replace the default input name with CustomerName.

Tip: This input becomes one of the trigger's outputs — a piece of data you can drop into later steps using dynamic content.

Step 3: Add the Send an email action (~10 minutes)

Below the trigger, select the + (plus) button, then Add an action.
In the search box, type Send an email.
Select the Office 365 Outlook connector, then choose the action Send an email (V2).

⚠️ Warning: Pick Office 365 Outlook, not Gmail, Outlook.com, or SMTP. Only Office 365 Outlook uses your course work account.

If this is your first use of the connector, Power Automate creates a connection: select Sign in, choose your course account, and approve. A green ✓ next to the connection means it's ready.
Configure the email fields in the action panel: — To: type your own email address (so you receive the test). — Subject: type Thank you for your enquiry — Body: build the message with the name inserted dynamically:
Click inside the Body field and type: Hi (with a trailing space)
Select the dynamic content icon (the small lightning bolt) that appears in/next to the field.
From the list, under Manually trigger a flow, choose CustomerName. It appears as a coloured token (chip) in the field.
Click after the token and continue typing: , thank you for reaching out to ACME Pte Ltd. We have received your enquiry and a team member will respond within 1 business day.

Tip: Dynamic content is how outputs from earlier steps get reused. The coloured CustomerName token is a placeholder — it's replaced with the real value when the flow runs.

⚠️ Warning — don't paste stray characters. Type the text yourself; don't copy backticks (`) or quotation marks from this guide into the field, or they'll appear literally in the email.

Step 4: Save and test (~5 minutes)

Select Save (top-right).

Tip: There is no separate "Send" button. Running the flow is what sends the email — the Send an email action does the work. So the routine is always: Save, then Test → Run flow.

Select Test (top-right) → choose Manually → Test → Run flow.
Power Automate prompts for the input you defined: — CustomerName: type Jane Tan
Select Run flow, then Done.
Watch the run status — each step should show a green check.
Open Outlook and confirm the email arrived, personalized as "Hi Jane Tan".

Step 5: Review the run history (~5 minutes)

In the left menu, select My flows, then open Lab 1 - Send Confirmation Email.
Look at the 28-day run history — you'll see your test run with a status.
Select the run to inspect each step's inputs and outputs. This is how you debug flows: a green check = success; a red ⚠️ = error (click it to read the message).

Checkpoint

You should now have:

✅ A flow named Lab 1 - Send Confirmation Email with a CustomerName text input
✅ An Office 365 Outlook connection showing a green ✓
✅ A successful test run (all steps green)
✅ A personalized email received in Outlook reading "Hi Jane Tan, …"

Troubleshooting

Problem	Solution
No "Send an email (V2)" action listed	Make sure you selected the Office 365 Outlook connector (not Gmail / Outlook.com / SMTP).
Action fails with "Unauthorized"	The Office 365 Outlook connection is broken/expired, or the signed-in account has no mailbox. Open the action's connection, reconnect with a mailbox-enabled account (green ✓).
Connection shows a red ⚠️	The connection needs to be re-authorized — select it and reconnect / sign in again.
Email not received	Check Junk/Spam; confirm the To address; re-run the test.
Can't find dynamic content	Click directly inside the Body field first, then open the lightning-bolt menu.
`CustomerName` shows as plain text, not a token	Delete it and re-insert it from the dynamic content list so it becomes a coloured token.
Clicked "Save" but no email arrived	Saving does not send. You must Test → Run flow — running the flow performs the action.

Key Takeaways

Every flow follows the pattern Trigger → Actions.
Trigger inputs become outputs that you reuse in later steps via dynamic content tokens.
A connector needs a connection: green ✓ = ready, red ⚠️ = reconnect. "Unauthorized" on Send an email means the Outlook connection is broken or the account has no mailbox.
There is no separate Send button — Save, then Test → Run to make the actions happen.
Test + run history are your tools for verifying and debugging.

Duration

~30 minutes

Next Steps

Proceed to Lab 2: Excel Data Logging Workflow.

Lab 2: Excel Data Logging Workflow

Lab Title

Capture Form Data and Log It into Excel with Power Automate

Lab Objectives

By the end of this lab, you will be able to:

Prepare an Excel workbook with a formatted Table in OneDrive
Build a flow with a manual trigger that collects Name, Email, and Message
Add an Add a row into a table (Excel) action and connect to your workbook
Map trigger outputs into Excel columns and stamp the date using an fx expression
Save, then Test → Run and verify each submission appears as a new row
Understand how to swap the manual trigger for a real Microsoft Forms trigger

Prerequisites

Completed Lab 1
Access to Excel via OneDrive (verified in Lab 0)
Signed in at https://make.powerautomate.com with NUS Copilot Sandbox selected (top-right)

Scenario

At ACME Pte Ltd, every customer enquiry must be recorded so nothing is lost. In this lab you build a flow that takes submitted details (name, email, message) and logs each one as a new row in an Excel workbook — a running enquiry register the whole team can see. You'll also timestamp each row automatically.

Step-by-Step Guide

Step 1: Create the Enquiry Log workbook with a Table (~10 minutes)

Power Automate can only read and write Excel data that is formatted as a Table — not loose cells.

Go to https://office.com, open Excel, and create a New blank workbook.
Rename it (click the file name at the top of the screen): Enquiry Log. It saves automatically to OneDrive.
In row 1, type these five column headers, one per cell (headers must be in row 1): — A1: Date — B1: Name — C1: Email — D1: Message — E1: Status
Select the header range A1:E1.
On the ribbon, select Insert → Table.
In the Create Table dialog, tick My table has headers, then select OK.
With the table selected, open the Table Design tab and set Table Name to EnquiryTable. This makes it easy to find in Power Automate.
Close the Excel tab — changes are saved to OneDrive automatically.

⚠️ Warning: If you skip Insert → Table, your file will have data but no Table, and Power Automate's "Table" dropdown will be empty in Step 3. The Table is mandatory.

Tip: Naming the table EnquiryTable is optional but strongly recommended — otherwise it appears as Table1 and is harder to pick out later.

Step 2: Create the flow and trigger (~5 minutes)

We'll use a manual trigger with inputs to simulate a submitted form. (Microsoft Forms uses the same downstream pattern — see Step 5.)

Go to https://make.powerautomate.com, confirm NUS Copilot Sandbox is selected top-right, then select Create → Instant cloud flow.
Flow name: Lab 2 - Log Enquiry to Excel
Choose how to trigger this flow: select Manually trigger a flow, then Create.
Select the trigger card to open its panel, then select + Add an input three times, choosing Text each time, and name them: — Name — Email — Message

Step 3: Add the "Add a row into a table" action (~10 minutes)

Below the trigger, select the + (plus) button, then Add an action.
In the search box, type Add a row into a table.
Select the Excel Online (Business) connector → action Add a row into a table.
If prompted, Sign in to create the connection (green ✓ = ready).
Configure where your file lives: — Location: OneDrive for Business — Document Library: OneDrive — File: browse to and select Enquiry Log.xlsx — Table: select EnquiryTable
The action now shows one field per column. Fill them in: — Date: this must be an expression, not typed text. Click the Date field, then select the fx (Expression) editor. Type exactly:

     formatDateTime(utcNow(),'yyyy-MM-dd HH:mm')

Then select Add / OK. The value should appear in the field as a single coloured token (chip).

Name: click the field → dynamic content → choose Name
Email: click the field → dynamic content → choose Email
Message: click the field → dynamic content → choose Message
Status: type the static text New

⚠️ Warning — fx, not typing. The formatDateTime(...) expression must be entered through the fx / Expression editor so it becomes a coloured token. If you just type it into the Date field, Power Automate logs the literal text formatDateTime(utcNow(),'yyyy-MM-dd HH:mm') into the cell instead of the actual date.

Tip — UTC vs local time. utcNow() returns the time in UTC, so a 4pm Singapore enquiry logs as 08:00. If you want Singapore local time, use this expression in the fx editor instead: convertTimeZone(utcNow(),'UTC','Singapore Standard Time','yyyy-MM-dd HH:mm')

Step 4: Save and test (~10 minutes)

Select Save (top-right).

Tip: As in Lab 1, there's no separate "submit" button — running the flow is what writes the row. Always Save, then Test → Run flow.

Select Test → Manually → Test → Run flow.
Enter sample values when prompted: — Name: Ahmad Rahman — Email: ahmad@example.com — Message: Interested in a bulk order of 50 units.
Select Run flow → Done. Confirm every step shows a green check.
Open Enquiry Log.xlsx in Excel — a new row should appear with your values, a New status, and a timestamp in the Date column.
Run the test 2–3 more times with different values to confirm each submission adds another row.

Step 5: (Optional) Connect it to a real Microsoft Form (~5 minutes)

To turn this into a true form submission workflow:

Create a form at https://forms.office.com with three questions: Name, Email, Message.
Build a new flow with the trigger When a new response is submitted (Microsoft Forms connector).
Add the Get response details action, then the same Add a row into a table action, mapping the Forms answers into the columns (and keep the same formatDateTime(...) fx expression for Date).

Tip: This shows the power of triggers — swap the manual trigger for a Forms trigger and the same logging logic runs automatically on every real submission.

Checkpoint

You should now have:

✅ Enquiry Log.xlsx in OneDrive with a Table named EnquiryTable (headers Date, Name, Email, Message, Status in row 1)
✅ A flow named Lab 2 - Log Enquiry to Excel with Name/Email/Message text inputs
✅ A Date column populated by the formatDateTime(...) fx expression (a coloured token, not literal text)
✅ Several test rows logged, each with a timestamp and Status New

Troubleshooting

Problem	Solution
File or Table not listed in the action	Ensure the data is formatted as a Table (Insert → Table), saved in OneDrive, with headers in row 1.
Date cell shows the literal text `formatDateTime(utcNow(),...)`	You typed the expression instead of using fx. Delete it, click the fx / Expression editor, paste the expression, and confirm it becomes a coloured token.
Date cell shows `########`	This is not an error — the column is just too narrow. Auto-fit the column (double-click its right border) to see the value.
Times look wrong / off by hours	`utcNow()` is UTC. Use the `convertTimeZone(...,'Singapore Standard Time',...)` expression for local time.
Row added but Name/Email/Message blank	Re-map each column to the correct dynamic content token.
Stray backtick or quote appears in a cell	Don't paste characters from this guide. Type values yourself or use the fx editor.
Connection shows red ⚠️	Re-authorize the Excel Online (Business) connection (sign in again) until it shows a green ✓.
Saved but no row appeared	Saving doesn't write rows. You must Test → Run flow — running performs the action.

Key Takeaways

Power Automate reads/writes Excel Tables (Insert → Table), never loose cells.
The Date must be entered via the fx editor so formatDateTime(...) becomes a token — typing it logs the literal text.
utcNow() is UTC; use convertTimeZone(...) for Singapore local time.
######## in a cell means the column is too narrow — auto-fit it; it's not an error.
The same logging action works behind any trigger — manual, Microsoft Forms, email, or a Copilot Studio agent.

Duration

~35 minutes

Next Steps

Proceed to Lab 3: Simple Approval Workflow.

Lab 3: Simple Approval Workflow

Lab Title

Build a Simple Approval Workflow with Power Automate

Lab Objectives

By the end of this lab, you will be able to:

Create an instant cloud flow with three manual trigger inputs (RequesterName, RequesterEmail, RequestDetails)
Add a Start and wait for an approval action and assign it to a real user in your tenant
Use a Condition to branch on the approval Outcome (Approved vs Rejected)
Send a different notification email in the If yes and If no branches
Test and verify both the Approve and Reject paths end-to-end

Prerequisites

Completed Lab 1 and Lab 2
Signed in at make.powerautomate.com in the NUS Copilot Sandbox environment
Your own signed-in account (it must exist in this tenant's directory) — you will be your own approver for testing

Scenario

At ACME Pte Ltd, a staff member submits a small purchase request. A manager must approve or reject it, and the requester must be automatically notified of the decision. You will build this with two essential building blocks: a human approval step and a Condition (branching). For testing, you will play both roles — requester and approver — using your own account.

Step-by-Step Guide

Step 1: Create the flow and add inputs (~7 minutes)

Go to https://make.powerautomate.com.
Top-left, confirm the environment selector reads NUS Copilot Sandbox. If not, click it and switch.
In the left menu, click + Create.
Under "Start from blank", click Instant cloud flow.
In the dialog: — Flow name: Lab 3 - Simple Approval — Choose the trigger Manually trigger a flow. — Click Create.
The designer opens with the Manually trigger a flow card. Click the card to open it, then add three inputs. For each, click + Add an input, choose Text, and name them exactly: — RequesterName — RequesterEmail — RequestDetails

Tip: Type the input names with no spaces, exactly as shown. You will reference them by these names later, and spaces make the tokens harder to find.

Step 2: Add the approval action and assign a real user (~10 minutes)

Below the trigger, click the + then Add an action.
In the search box type approval and select Start and wait for an approval (from the Approvals connector).
If prompted to sign in / create the Approvals connection, click Continue or Sign in and finish it. The connection must show a green check.
Configure the action: — Approval type: Approve/Reject - First to respond — Title: type Approval needed: then, with your cursor still in the box, open the dynamic content (lightning bolt) and insert RequestDetails. — Assigned to: click the field. A people-picker dropdown appears. Start typing your own name or email (the same account shown under your avatar at the top-right), then click your name in the dropdown so it resolves to a person chip. — Details: type Requested by then insert the dynamic token RequesterName.

⚠️ Warning: The Assigned to field MUST be a real user that exists in THIS tenant's directory, and you must pick the person from the dropdown so it becomes a resolved chip. Do not type a free-text external address like someone@othercompany.com. If you do, the run fails with "InvalidApprovalCreateRequest … Required field … valid users in the organization." Use your own signed-in account and select it from the dropdown.

Tip: Start and wait for an approval pauses the entire flow until the approver responds. The approver can respond by email, in Teams, or in the Approvals hub (left menu) at make.powerautomate.com.

Step 3: Add a Condition on the Outcome (~8 minutes)

Below the approval action, click + → Add an action.
Search condition and select Condition (from the Control connector).
Build the condition with exactly these three parts: — Left value: click the box, open dynamic content (lightning bolt), and insert Outcome (from the Start and wait for an approval step). — Operator: is equal to — Right value: type Approve

⚠️ Warning: The right value must be exactly Approve with a capital A — this is the literal text the approval Outcome returns. approve, Approved, or APPROVE will never match, and every run will fall into the If no branch.

You now have two branches below the condition: If yes (approved) and If no (rejected).

Step 4: Add a notification email to each branch (~10 minutes)

In the "If yes" branch:

Click Add an action inside the If yes branch.
Search send an email and select Send an email (V2) (from Office 365 Outlook). Complete the connection if prompted (it must show a green check).
Configure: — To: insert the dynamic token RequesterEmail (a trigger input). — Subject: Your request has been APPROVED — Body: type Hi , insert RequesterName, type , your request ", insert RequestDetails, type " has been approved. You may proceed.

In the "If no" branch:

Click Add an action inside the If no branch.
Select Send an email (V2) again.
Configure: — To: insert RequesterEmail — Subject: Your request has been REJECTED — Body: type Hi , insert RequesterName, type , unfortunately your request ", insert RequestDetails, type " was not approved. Please contact your manager for details.

⚠️ Warning: Only use single-value dynamic fields here — the trigger inputs (RequesterName, RequesterEmail, RequestDetails) and, if you want it, the approval Outcome. Do not insert any approval Responses field. Power Automate auto-wraps an action in a For each loop the moment you insert a list/array value, which breaks this simple flow. If a For each appears around your email, delete it and re-add a plain Send an email (V2) using only single-value fields.

Tip: Add the email inside each branch box, not below the whole Condition — otherwise it runs on both outcomes.

Step 5: Save and test BOTH paths (~10 minutes)

Top-right, click Save. Before testing, confirm both connections show a green check: Approvals and Office 365 Outlook.
Click Test → Manually → Test → Run flow, and enter: — RequesterName: Siti — RequesterEmail: your own email — RequestDetails: New office chair - $120
Click Run flow → Done. The flow pauses at the approval step (this is normal — it is waiting for you).
Respond to the approval. Fastest path: left menu → Approvals → Received tab → open the request → click Approve → Submit. (The email/Teams notification also works but can be slow or land in Junk.)
The flow resumes down the If yes branch. Confirm you receive the APPROVED email.
Run the test again with the same inputs, but this time Reject the approval. Confirm you receive the REJECTED email.
Open My flows → Lab 3 - Simple Approval → Run history and confirm the correct branch ran each time.

Checkpoint

✅ Flow Lab 3 - Simple Approval with manual trigger inputs RequesterName, RequesterEmail, RequestDetails
✅ Start and wait for an approval assigned to your own user (resolved as a person chip)
✅ Condition on Outcome is equal to Approve, with a Send an email in each branch
✅ Approve path → APPROVED email received; Reject path → REJECTED email received

Troubleshooting

Problem	Solution
Run fails: InvalidApprovalCreateRequest … valid users in the organization	The Assigned to value isn't a real tenant user. Clear it, type your own name, and pick your account from the dropdown so it becomes a person chip.
Flow stays "Running" forever	Expected — it's waiting for the approval response. Go to left menu → Approvals → Received and respond.
Condition always goes to If no	The right value must be exactly `Approve` (capital A) to match the Outcome text.
A For each loop wrapped your email	You inserted a list/array value (e.g. a Responses field). Delete the For each and re-add a plain Send an email (V2) using only single-value fields.
Send an email: Unauthorized	The Outlook connection is broken or the account has no mailbox. Reconnect Office 365 Outlook with a mailbox-enabled account; both connections must show green ✓ before running.
No approval email arrives	Check Junk; or just respond in the Approvals hub instead — it's more reliable.
Email actions empty / nothing sent	Make sure each Send an email sits inside its branch (If yes / If no), not after the Condition.

Key Takeaways

Start and wait for an approval pauses a flow until a real human in your tenant decides.
Assigned to must resolve to a person chip from the directory — never a typed external email.
The approval Outcome (Approve / Reject) drives a Condition, which creates branching logic for different responses.
Inserting a list/array value silently adds a For each — keep approval emails on single-value fields to avoid it.

Duration

~45 minutes

Next Steps

Proceed to Lab 4: Scheduled Trigger Workflow.

Lab 4: Scheduled Trigger Workflow

Lab Title

Build a Scheduled (Recurring) Workflow with Power Automate

Lab Objectives

By the end of this lab, you will be able to:

Create a Scheduled cloud flow that uses the Recurrence trigger
Configure the Interval, Frequency, Time zone, hours, and specific days
Send a daily reminder email automatically on a timetable
Test/Run now without waiting for the scheduled time
(Optional stretch) Read the Excel EnquiryTable and include a live count in the email

Prerequisites

Completed Lab 1 (Send an email)
(For the optional stretch) An Excel Enquiry Log workbook with table EnquiryTable from Lab 2
Signed in at make.powerautomate.com in the NUS Copilot Sandbox environment

Scenario

At ACME Pte Ltd, some work isn't triggered by an event — it just needs to happen on a schedule. Every weekday morning the team should get a reminder to review new enquiries. You will build a flow that runs automatically using the Recurrence trigger, so no one has to start it.

Tip: Trigger types so far — manual (Labs 1–3, you press Run) and scheduled (this lab, runs by the clock). Lab 5 adds an event trigger (a form submission).

Step-by-Step Guide

Step 1: Create a scheduled flow (~6 minutes)

Go to https://make.powerautomate.com.
Top-left, confirm the environment selector reads NUS Copilot Sandbox. If not, click it and switch.
In the left menu, click + Create.
Under "Start from blank", click Scheduled cloud flow.
In the dialog: — Flow name: Lab 4 - Daily Enquiry Reminder — Starting: today's date and any time — Repeat every: 1 and Day
Click Create. The designer opens with a Recurrence trigger already added.

Step 2: Fine-tune the schedule (~10 minutes)

Click the Recurrence trigger card to open it.
Set the basics: — Interval: 1 — Frequency: Day
Click Show advanced options and set: — Time zone: (UTC+08:00) Kuala Lumpur, Singapore — At these hours: 9 — At these minutes: 0 — On these days: tick Monday, Tuesday, Wednesday, Thursday, Friday (leave Saturday and Sunday unticked)
The schedule now reads: every weekday at 9:00 AM Singapore time.

⚠️ Warning: You MUST set the Time zone. Without it, the Recurrence trigger uses UTC, so a "9" would fire at 9:00 UTC = 5:00 PM Singapore — the wrong local time. This is the same UTC-vs-local trap from Lab 2.

Step 3: Send the reminder email (~8 minutes)

Below the Recurrence trigger, click + → Add an action.
Search send an email and select Send an email (V2) (from Office 365 Outlook). Complete the connection if prompted (it must show a green check).
Configure: — To: your team's address (use your own mailbox for testing) — Subject: Daily reminder: review new enquiries — Body:

     Good morning,
     This is your daily reminder to review and follow up on new enquiries
     in the Enquiry Log. Please action any items marked "New".

Top-right, click Save.

⚠️ Warning: If you hit an Unauthorized error, the Outlook connection is broken or the account has no mailbox. Reconnect Office 365 Outlook with a mailbox-enabled account (see Lab 1). The connection must show a green ✓ before running.

Step 4: Test the flow now (~5 minutes)

You don't have to wait until 9 AM — you can run it on demand to check it works.

Top-right, click Test → Manually → Test → Run flow → Done.
Confirm every step shows a green check and the reminder email arrives in your inbox.
From now on the flow also runs automatically on its schedule.

Tip: A scheduled flow only fires on its timetable in real life — Test → Run flow lets you verify it immediately instead of waiting for 9 AM tomorrow.

Step 5 (Optional stretch): Include a live count from Excel (~15 minutes)

Make the reminder smarter by counting how many enquiries are logged in EnquiryTable.

Above the Send an email action, click + → Add an action.
Search list rows and select List rows present in a table (from Excel Online (Business)).
Configure the location: — Location: OneDrive for Business — Document Library: OneDrive — File: browse to Enquiry Log (the .xlsx workbook) — Table: EnquiryTable
Open the Send an email (V2) action again. Click into the Body where you want the count, then open the fx (expression) editor and enter exactly:

   length(outputs('List_rows_present_in_a_table')?['body/value'])

Click Add / OK so it becomes a token (highlighted chip), not plain text.
Example line: There are currently (token) enquiries logged.

Click Save, then Test → Run flow again. The email now shows the live record count.

⚠️ Warning: The name inside outputs('...') must match your action's internal name exactly, with spaces replaced by underscores. If the expression errors, open the List rows present in a table action → … menu → check the action name, and adjust List_rows_present_in_a_table to match.

Checkpoint

✅ A scheduled flow Lab 4 - Daily Enquiry Reminder using the Recurrence trigger
✅ Configured for weekdays at 9:00 AM with Time zone (UTC+08:00) Kuala Lumpur, Singapore
✅ A reminder email sent on a successful Test → Run flow
✅ (Optional) A live record count pulled from EnquiryTable

Troubleshooting

Problem	Solution
Email arrives at the wrong time	Set Time zone in the Recurrence advanced options (e.g. (UTC+08:00) Kuala Lumpur, Singapore). Without it the schedule uses UTC.
Flow runs every day including weekends	In On these days, tick only Monday–Friday.
Send email Unauthorized	Reconnect Office 365 Outlook with a mailbox-enabled account; the connection must show green ✓.
`length(...)` expression error	Match the name inside `outputs('...')` to the List rows action's actual internal name (spaces become underscores).
Don't want to wait for the schedule	Use Test → Manually → Run flow to run it immediately.
Count shows as literal text, not a number	The expression wasn't added via the fx editor as a token. Re-enter it through fx and confirm it becomes a highlighted chip.

Key Takeaways

The Recurrence trigger runs flows automatically on a timetable — no human starts them.
Always set the Time zone so schedules fire at the right local time, not UTC.
Use Test → Run flow to verify a scheduled flow without waiting for its scheduled time.
Scheduled flows are ideal for digests, reminders, and clean-up jobs.

Duration

~30 minutes (45 with the optional stretch)

Next Steps

Proceed to Lab 5: Form Submission Workflow.

Lab 5: Form Submission Workflow

Lab Title

Capture Microsoft Forms Submissions to Email and Excel (Automatic Trigger)

Lab Objectives

By the end of this lab, you will be able to:

Create a Microsoft Form with three required questions and a shareable URL
Build a flow using the When a new response is submitted trigger (Microsoft Forms)
Add Get response details to read each answer (mapping the Response Id)
Email the submitted enquiry to your team
Log the same enquiry into the Excel EnquiryTable — all automatically on submit

Prerequisites

Completed Lab 1 (Send an email) and Lab 2 (Excel table + logging)
An Excel Enquiry Log workbook with table EnquiryTable (Date, Name, Email, Message, Status) from Lab 2
Signed in at make.powerautomate.com in the NUS Copilot Sandbox environment

Scenario

At ACME Pte Ltd, customers fill in an online enquiry form. The moment they submit, the workflow should email the team the new enquiry and record it in Excel — with no manual work. This is your first automatic, event-driven workflow (Labs 1–3 were manual, Lab 4 was scheduled). You'll also get a form link you can send to anyone.

Step-by-Step Guide

Step 1: Create the form in Microsoft Forms (~10 minutes)

Open a new browser tab and go to https://forms.office.com.
Sign in with the same account you use for Power Automate in the NUS Copilot Sandbox tenant.
Click + New Form.
Click the title at the top and enter: Customer Enquiry Form.
Add a short description, e.g. Tell us about your enquiry and we'll respond within 1 business day.
Add these three questions. For each, click + Add new, choose Text, type the question, then toggle Required on: — Full Name → Required on — Email → Required on — Your Message → click the … on the question and choose Long answer for more space → Required on
The form saves automatically as you type.

Tip: Keep the question titles exactly as shown — you'll match them to Excel columns later, and clear names make the dynamic tokens easy to find.

Step 2: Get the shareable form URL (~5 minutes)

Top-right of Forms, click Collect responses (older UI: Share).
Set the audience — for testing choose Anyone can respond (or your organization).
Copy the link shown (e.g. https://forms.office.com/r/XXXXXXXX).
Save this URL somewhere — it's the link you'd send to customers, and you'll use it to test at the end.

Step 3: Create the automated flow (~5 minutes)

Go back to https://make.powerautomate.com (confirm the environment is NUS Copilot Sandbox).
Left menu → + Create → Automated cloud flow.
Flow name: Lab 5 - Form Submission to Email and Excel.
In "Choose your flow's trigger", search Forms and select When a new response is submitted (Microsoft Forms).
Click Create.
On the trigger card, set Form Id → pick Customer Enquiry Form from the dropdown.

Step 4: Get the response details (~10 minutes)

The trigger only gives you a response Id — you need another action to read the actual answers.

Below the trigger, click + → Add an action.
Search Forms → select Get response details (Microsoft Forms).
Configure: — Form Id: pick Customer Enquiry Form (the same form) — Response Id: click the field, open dynamic content (lightning bolt), and insert Response Id (from the trigger)

⚠️ Warning: Get response details is required. Without it, later steps only see an internal Id, not the real answers — and your email/Excel will be blank. Always map Response Id from the trigger, then use this action's outputs (Full Name, Email, Your Message) in every step that follows.

Step 5: Email the submission to your team (~10 minutes)

Click + → Add an action.
Select Send an email (V2) (Office 365 Outlook). Complete the connection if prompted (it must show a green check).
Configure (insert each value from Get response details dynamic content): — To: your team's address (use your own working mailbox for testing) — Subject: type New enquiry from then insert Full Name — Body:

     A new enquiry was submitted via the form:
     Name: [insert Full Name]
     Email: [insert Email]
     Message: [insert Your Message]

⚠️ Warning: If you get an Unauthorized error, the Outlook connection is broken or the account has no mailbox. Reconnect Office 365 Outlook with a mailbox-enabled account (see Lab 1); the connection must show a green ✓.

Step 6: Log the submission to Excel (~10 minutes)

Click + → Add an action.
Select Add a row into a table (Excel Online (Business)). Complete the connection if prompted.
Set the location: — Location: OneDrive for Business — Document Library: OneDrive — File: browse to Enquiry Log (the .xlsx workbook) — Table: EnquiryTable
Map the columns: — Date: click the fx (expression) icon → enter exactly formatDateTime(utcNow(),'yyyy-MM-dd HH:mm') → click Add / OK so it becomes a token (a highlighted chip), never typed text — Name: dynamic content → Full Name — Email: dynamic content → Email — Message: dynamic content → Your Message — Status: type New
Top-right, click Save.

⚠️ Warning: The Date value must be entered through the fx editor and become a token. If you type the formula as plain text, the cell stores the literal text formatDateTime(...) instead of a real date. (This is the same fix from Lab 2.)

Step 7: Test the whole workflow (~10 minutes)

In Power Automate, click Test → Manually → Test. The flow goes into a "waiting for trigger" state.
Open the form URL you saved in Step 2 (new tab or your phone).
Fill in the form — Full Name Jane Tan, Email jane@example.com, Your Message Interested in 50 units — and click Submit.
Within about a minute the flow triggers. Confirm: — Every step shows a green check in the run. — The email arrives with the submitted details. — A new row appears in Enquiry Log → EnquiryTable with a clean date, the answers, and Status New.
Submit the form a second time with different details and confirm another row and another email.

Tip: Triggers can take up to a minute. If nothing happens, confirm you clicked Test before submitting — or just submit again, since a saved flow fires automatically on every real submission.

Checkpoint

✅ A Customer Enquiry Form with three required questions and a shareable URL
✅ Automated flow triggered by When a new response is submitted
✅ Get response details mapping the Response Id from the trigger
✅ Each submission → notification email to the team and a new row in EnquiryTable (Date via fx token)

Troubleshooting

Problem	Solution
Form not listed in the trigger	Sign into Power Automate with the same account that owns the form, in the NUS Copilot Sandbox environment.
Answers show as IDs / blank	You must add Get response details, map Response Id from the trigger, and use its outputs (not the trigger's) in later steps.
Flow doesn't trigger after submit	Triggers can take up to a minute; confirm Test was started before submitting, or submit again (a saved flow fires automatically).
Date shows literal text	Enter the date through the fx editor so it becomes a token (see Lab 2).
Send email Unauthorized	Reconnect Office 365 Outlook with a mailbox-enabled account; the connection must show green ✓.
Add a row Unauthorized	Reconnect Excel Online (Business) with an account that can access the Enquiry Log workbook.

Key Takeaways

A Microsoft Forms trigger turns a public form into an automatic workflow — no buttons to press.
Get response details is required to read the individual answers behind the Response Id.
The Date column is filled by an fx token (formatDateTime(utcNow(),'yyyy-MM-dd HH:mm')), never typed text.
One submission can fan out to multiple actions (notify and log) — the core pattern for Day 3.

Duration

~50 minutes

Next Steps

You've completed Day 1 and all the core Power Automate trigger types (manual, scheduled, form, email). Proceed to Day 2 — Module 3: Business Agents Concepts.

Day 2 — Building Business Agents with Copilot Studio

Module 3: Building Business Agents with Copilot Studio

Read this before the Day 2 labs. ~15 minutes.

On Day 1 you built the hands — Power Automate flows that send email, log to Excel, and run approvals. Today you build the brain and mouth: a Copilot Studio agent that talks to people in plain language and hands clean, structured data to those flows.

By the end of this reading you'll know what an agent is made of, how to make it produce predictable data (the make-or-break skill), and how an agent calls a flow as a tool.

1. What is a business agent?

A business agent is an AI assistant you build in Copilot Studio. People chat with it in natural language ("I'd like a quote for 100 units"), and the agent:

Understands the request
Asks for any missing details
Captures the information as clean, structured data
Hands off that data to a Power Automate flow to take action

   Day 1                          Day 2
 ┌────────┐                    ┌──────────┐
 │ FLOWS  │  the "hands"       │  AGENT   │  the "brain & mouth"
 │ do work│  ◄───── calls ──── │ talks &  │
 └────────┘                    │ captures │
                               └──────────┘

The agent is the friendly front door; the flow is the engine room behind it.

2. The building blocks of a Copilot Studio agent

An agent is assembled from five kinds of building block. You'll meet each one across today's labs.

Block	What it is	Example	First seen
Instructions	Plain-language directions that shape the agent's behaviour and personality	"You are a sales assistant. Always collect name, company, product, and quantity."	Lab 6
Knowledge	Documents / sites the agent can answer from (RAG)	Product catalogue, FAQ	Lab 7
Topics	Conversation flows the agent runs when the user's message matches the topic's description (generative orchestration, the default) or its trigger phrases (classic)	A "New Sales Enquiry" topic the agent chooses when someone asks for a quote	Lab 9
Tools / Actions	Things the agent can do, including Power Automate flows	"Log enquiry to Excel" flow	Lab 8
Variables	Where captured answers are stored to pass onward	`customerName`, `product`, `quantity`	Lab 9

Knowledge = RAG. When you upload documents, the agent uses Retrieval-Augmented Generation: it retrieves the relevant passages from your files and generates an answer grounded in them — so it speaks from your content, not the open internet. You'll set this up in Lab 7.

3. Prompt design for structured outputs

This is the single most important skill for connecting agents to workflows: getting the agent to produce structured, predictable data — not just free-flowing chat.

Why it matters. A flow needs clean, named fields. Compare what the agent might hand over:

The agent hands the flow…	Can the flow use it?
`"the guy from ABC wants some units maybe 100"`	❌ No — nothing to log reliably
`name=John, company=ABC, product=Widget, quantity=100`	✅ Yes — every field drops straight into a row

When the data is clean and structured, every downstream step just works. When it's messy, the whole workflow is unreliable.

Three techniques to get structured output

Capture into named variables. Use question nodes that store each answer in its own variable (customerName, email, quantity). This is the most reliable method, and you'll use it in Labs 9 and 10.
Write precise instructions. Tell the agent exactly what to collect and in what form: — > "Collect these four items one at a time: customer name, company, product of interest, and quantity. Do not proceed until all four are provided. Confirm the details back to the user before finishing."
Ask AI to format the result. When you want a generated summary, specify the format explicitly: — > "Summarize the enquiry as: Customer: <name>; Company: <company>; Product: <product>; Qty: <quantity>; Notes: <notes>."

Good-prompt checklist

✅ State the agent's role and goal
✅ List the exact fields to collect
✅ Specify the output format
✅ Say what to do when information is missing
✅ Keep tone instructions short and clear

Structured agent output = clean flow inputs. This is the through-line of Day 2. If the agent captures tidy variables, the flow receives tidy inputs — and the handoff in section 4 works first time.

4. Connecting agents to Power Automate flows

Once the agent has captured the data, it calls a flow as a tool:

USER chats with AGENT  →  AGENT captures variables  →  AGENT calls FLOW (tool)
                                                          passing variables as inputs
                                                              │
                                                              ▼
                                            FLOW runs actions (log, email, approve)
                                                              │
                                                              ▼
                                            FLOW returns a result  →  AGENT confirms to user

The key points:

The agent passes its variables (outputs) into the flow's inputs.
The flow does the work and can return a value — say, a reference number — for the agent to show the user.
This is exactly the same flow-building you learned on Day 1. Only the trigger changes — to "When an agent calls the flow."

It's still Day-1 Power Automate underneath. Same designer, same actions, same Save → Test → run-history loop. The agent simply replaces the manual "Run" button as the thing that starts the flow.

5. What you'll build on Day 2

Lab 6: Your first agent — learn the interface, instructions, and testing.
Lab 7: Add Knowledge (RAG) so the agent answers from your own documents.
Lab 8: Add Tools & Actions so the agent can do things, not just answer.
Lab 9: A Sales Enquiry Assistant that captures enquiries as structured data.
Lab 10: A Procurement Request agent that triggers a Power Automate flow.
Lab 11: Automated Response Generation — use AI prompts to draft professional replies.

Each lab adds one capability on top of the last, so by Lab 11 you'll have an agent that understands, retrieves, captures, acts, and writes.

Next: Lab 6: Create Your First Copilot Studio Agent

Lab 6: Create Your First Copilot Studio Agent

Lab Title

Create, Configure, and Test Your First Business Agent

Lab Objectives

By the end of this lab, you will be able to:

Create a new agent in Microsoft Copilot Studio using Skip to configure
Write clear Instructions that shape how your agent behaves
Identify the main building blocks of an agent (Instructions, Knowledge, Topics, Tools)
Add one Knowledge source so your agent can answer real questions
Test and iterate on your agent in the built-in Test pane
Publish your agent (optional) and understand what Channels are

Prerequisites

Completed Lab 0 (Copilot Studio trial active)
Read Module 3
Signed in at https://copilotstudio.microsoft.com (same environment as Power Automate)

Scenario

You work in IT at ACME Pte Ltd, a small company. Staff keep emailing IT the same basic questions: opening hours, who to contact, how to reset things. You will build a friendly Company Helpdesk agent that answers these questions automatically.

This lab is your guided tour of Copilot Studio. You will learn the interface and the core idea that Instructions are the heart of the agent before you build data-capturing, action-taking agents in the labs that follow.

Tip: An "agent" (sometimes still called a "copilot") is just an AI assistant you configure. You do not write code — you describe what you want in plain English, add some reference material, and test it in a chat window.

Step-by-Step Guide

Step 1: Confirm your environment (~3 minutes)

Before you build anything, make sure you are in the correct environment. This is the single most common cause of problems later in the course.

Go to https://copilotstudio.microsoft.com and sign in with your course account.
Look at the top-right corner of the screen. You will see an environment selector (a small label showing the current environment name, often with a globe or building icon).
Click it and select NUS Copilot Sandbox (your course environment from Lab 0).

⚠️ Warning: Copilot Studio must use the same environment as Power Automate. If your agent is built in one environment and your flows live in another, they will not be able to connect to each other in Day 3. Always confirm the environment name in the top-right before you start. The environment also needs Dataverse enabled, because agents are stored there.

Step 2: Create the agent with Skip to configure (~10 minutes)

In the left menu, select Create, then select New agent.
Copilot Studio opens a conversational setup screen that says something like "Describe your agent to create it." You could chat with it to build the agent — but for this lab you will fill in the form yourself so you understand every field.
Select Skip to configure (a link near the top-right of that screen). This opens a simple form.
Fill in the form: — Name: Company Helpdesk — Description: Answers general questions about ACME Pte Ltd for staff and customers. — Instructions: copy and paste the text below into the Instructions box:

     You are the Company Helpdesk agent for ACME Pte Ltd.
     Be friendly, concise, and professional.
     Answer questions about our products, opening hours, and contact details.
     If you do not know an answer, politely say so and suggest emailing help@acme.example.
     Keep replies to 2-3 sentences.

Select Create (top-right). Wait a few seconds while the agent is built. The agent's main page then opens.

Tip: Think of the three fields like this — Name is what people see, Description is a short note for you (and helps other agents/tools recognise it later), and Instructions are the actual rules the AI follows in every conversation. Instructions are where almost all of your effort goes.

Step 3: Tour the agent workspace (~5 minutes)

Now that the agent exists, get familiar with the layout. Near the top of the page you will see several tabs. The exact labels can shift slightly between versions, but you will find these building blocks:

Overview — a summary of your agent and quick links to its parts.
Knowledge — the documents, websites, and data the agent can read to find facts (you will use this in Step 4 and much more in Lab 7).
Topics — scripted, step-by-step conversation flows you design by hand for predictable questions.
Tools (sometimes shown as Actions) — things the agent can do, such as send an email or run a Power Automate flow (covered in Lab 8).
The Test pane — a chat panel, usually on the right. If you do not see it, select Test at the top-right.

Click through each tab once so you know where things are. You do not need to change anything yet.

Tip: Remember the four building blocks: Instructions = behaviour and tone, Knowledge = facts the agent can look up, Topics = scripted conversations, Tools/Actions = things the agent can do. Almost everything in Copilot Studio is one of these four.

Step 4: Add one Knowledge source (~10 minutes)

Right now your agent only knows what is in its Instructions. Give it something real to answer from.

Open the Knowledge tab.
Select + Add knowledge.
Choose Public website (the simplest option — no files needed).
In the URL box, paste a relevant website. For practice, use the Microsoft Copilot Studio docs: — https://learn.microsoft.com/microsoft-copilot-studio/ — (Alternatively, if you have a short company FAQ as a PDF or Word file, choose the Files option and upload it instead.)
Give the source a short, clear name such as Copilot Studio docs and, if asked, a one-line description of what it contains.
Select Add.
Watch the source's status. It will show Processing (or a spinner) and then change to Ready. Wait until it says Ready before testing — this can take a minute or two.

Tip: A website source lets the agent ground its answers in real content instead of guessing. You will go much deeper on Knowledge in Lab 7.

Step 5: Test your agent (~7 minutes)

Open the Test pane on the right. If it is hidden, select Test at the top-right.
Type these questions one at a time and read the replies: — What can you help me with? — What is Copilot Studio? — How do I contact support?
Notice two things working together: — Your Instructions control the style — friendly, short (2-3 sentences), professional. — Your Knowledge controls the facts — answers drawn from the website you added.

Tip: The Test pane updates as you build. After most changes you make, come back here and re-test. This build-then-test loop is exactly how real agents are developed.

Step 6: Iterate on your Instructions (~5 minutes)

The first version of an agent is rarely perfect. Tuning the Instructions is normal and expected.

Suppose replies are too long or too formal. Open the Overview tab and find the Instructions box.
Add a line to make the change you want, for example:

   Always greet the user by name if they give it.
   Never give legal or financial advice; suggest contacting a manager instead.

Select Save.
Go back to the Test pane. If your earlier conversation is still showing, select Refresh (or the circular-arrow icon) at the top of the Test pane so it picks up the new Instructions, then ask again.

⚠️ Warning: The Test pane does not always pick up changes automatically. If your edit does not seem to take effect, click Refresh at the top of the Test pane to restart the conversation.

Step 7: Publish the agent (optional) (~5 minutes)

Publishing makes your latest version live so it can be shared. You do not have to deploy it anywhere yet.

Select Publish (top-right), then confirm by selecting Publish again in the dialog.
Once publishing finishes, open the Channels tab. Channels are the places people can use your agent — for example Microsoft Teams, a demo website, or a custom app.
Browse the list to see your options. You do not need to connect any channel now; publishing alone is enough to keep your latest version testable.

Checkpoint

You have successfully completed this lab when:

✅ You confirmed you are in the NUS Copilot Sandbox environment (matching Power Automate)
✅ An agent named Company Helpdesk exists with your custom Instructions
✅ At least one Knowledge source shows the status Ready
✅ The agent gives sensible, on-style answers in the Test pane
✅ You edited the Instructions and saw the change after refreshing the Test pane

Troubleshooting

Problem	Cause	Solution
Agent ignores your new instructions	Test pane is showing the old conversation	Select Save, then click Refresh at the top of the Test pane to restart the chat.
Knowledge source stuck on "Processing"	Large website or busy service	Wait a minute or two. If it persists, remove it and add a smaller single page or a short PDF instead.
Replies are generic or unhelpful	No grounding, or vague instructions	Make sure a Knowledge source is Ready and tighten your Instructions (be specific about tone and scope).
Can't see the Test pane	Pane is collapsed	Select Test at the top-right of the agent page.
Tools/flows won't connect later	Wrong environment	Use the environment selector (top-right) to switch to the same environment as Power Automate (NUS Copilot Sandbox).
"Skip to configure" link not visible	UI variation	Look near the top-right of the "Describe your agent" screen; if missing, briefly chat a one-line description, then edit the fields afterwards on the Overview tab.

Key Takeaways

An agent's behaviour is shaped mainly by its Instructions — clear instructions give predictable answers.
The four building blocks are Instructions (behaviour/tone), Knowledge (facts), Topics (scripted chats), and Tools/Actions (doing things).
Knowledge grounds answers in real content so the agent stops guessing.
The Test pane is your build-and-iterate loop — refresh it after each change.
Always work in the same environment as your Power Automate flows, and that environment needs Dataverse.

Duration

~45 minutes

Next Steps

Proceed to Lab 7: Add Knowledge to Your Agent.

Lab 7: Add Knowledge to Your Agent (Grounding / RAG)

Lab Title

Ground Your Agent in Your Own Documents with Knowledge (RAG)

Lab Objectives

By the end of this lab, you will be able to:

Explain grounding and RAG (Retrieval-Augmented Generation) in plain, everyday terms
Add one or more Knowledge sources to a Copilot Studio agent (files, public website, or SharePoint)
Turn general knowledge off so the agent answers only from your content
Test that the agent answers from your documents and shows citations
Confirm the agent declines to answer when the information is not in your sources (no made-up answers)
Tell the difference between Instructions (behaviour) and Knowledge (facts)

Prerequisites

Completed Lab 6 (you have an agent)
A short document to use as knowledge (a PDF/Word FAQ, or a public website URL)

Scenario

Your Company Helpdesk agent from Lab 6 can chat, but it does not actually know ACME Pte Ltd's real facts — its return policy, its opening hours, its product list. An agent that only knows general internet information is not very useful for your business.

In this lab you will give the agent a Knowledge source containing ACME's own information, then prove that the agent answers from that source and nothing else. This is the technique that makes business agents trustworthy.

Tip: Grounding means tying the agent's answers to specific source material you provide, so it cannot just invent things. RAG (Retrieval-Augmented Generation) is the technology that does this.

How RAG works (in plain terms)

You add documents to Knowledge. Copilot Studio breaks them into small passages and indexes them.
A user asks a question.
The agent retrieves the few passages most relevant to that question.
The agent generates an answer using only those passages, and shows citations pointing back to the source.

RAG in one line: Find the relevant text in your documents → hand it to the AI → it answers from that text, with citations. This keeps answers accurate and up to date without retraining any model.

Step-by-Step Guide

Step 1: Prepare a knowledge source (~5 minutes)

Pick one source to start (you can add more later). Choose whichever is easiest for you:

A file (recommended for this lab): Create a one-page document called ACME Company FAQ. Save it as a PDF or Word file. Include a few clear facts the agent will answer from, for example:

   ACME Pte Ltd - Company FAQ

   Opening hours: Monday to Friday, 9am to 6pm. Closed on public holidays.
   Return policy: Items may be returned within 30 days with a receipt.
   Contact: Email help@acme.example or call +65 6123 4567.
   Top products: ACME Widget Pro, ACME Gadget Lite, ACME Toolkit.

A public website: any URL whose content you want the agent to use (e.g. a real product page).
A SharePoint site or document library, if your tenant has one with relevant files.

Tip: Keep the document focused and tidy — use clear headings and short paragraphs. Clean, well-structured content leads to much better retrieval and more accurate answers.

Step 2: Open your agent's Knowledge tab (~3 minutes)

Go to https://copilotstudio.microsoft.com and confirm the environment selector (top-right) shows NUS Copilot Sandbox.
Open your Company Helpdesk agent from Lab 6.
Select the Knowledge tab.
Select + Add knowledge.

⚠️ Warning: Make sure you are in the same environment as your Lab 6 agent and your Power Automate flows. If the environment is wrong, you may be editing a different (or empty) agent.

Step 3: Add the knowledge source (~10 minutes)

On the Add knowledge screen, choose the type that matches your source: — Files / Upload — then select and upload your ACME Company FAQ PDF or Word file. — Public website — then paste the URL. — SharePoint — then paste the site or library URL.
Give the source a clear name, for example ACME Company FAQ.
If you are prompted for a description, write a short sentence describing what it contains, e.g. ACME opening hours, return policy, contact details, and top products. The agent uses this description to decide when this source is relevant.
Select Add.
Watch the source's status. It will show Processing and then change to Ready. Larger files and websites take a little longer.

⚠️ Warning: Do not test until the status reads Ready. While a source is still Processing, the agent cannot use it and may reply that it has no information.

Step 4: Turn off general knowledge (optional but recommended) (~4 minutes)

By default the agent may blend in general AI/web knowledge. For a business helpdesk you usually want answers only from ACME's documents.

Open the agent's Settings (gear icon, usually top-right) and look for the Generative AI section. (Depending on your version, this setting may also appear on the Overview page near the Knowledge area.)
Find the option Use general knowledge (it may read "Allow the AI to use its own general knowledge" or similar).
Turn it Off so the agent relies only on your sources.
Select Save.

Tip: Turn general knowledge on when you want the agent to also answer broad, everyday questions. Turn it off when you need tight control and want to prevent answers that did not come from your documents.

Step 5: Test that answers come from your document (~10 minutes)

Open the Test pane and select Refresh at the top so it picks up your new Knowledge.
Ask questions that can only be answered from your ACME Company FAQ, for example: — What is your return policy? — What are your opening hours? — Which products do you offer?
Confirm each answer matches the facts in your document.
Look underneath each answer for citations or references (often numbered, like [1], or a "1 reference" link). Click a citation to confirm it points back to your source.

Tip: Citations are your proof that grounding worked. If an answer has no citation, it may have come from general knowledge rather than your document.

Step 6: Confirm the agent declines when it doesn't know (~5 minutes)

This is the most important test — it proves the agent will not make things up.

In the Test pane, ask something that is not in your document, for example: — What is your CEO's home address? — Do you sell aeroplanes?
The agent should say it does not have that information (and, following your Lab 6 Instructions, suggest emailing help@acme.example).
If instead it confidently invents an answer, that is a hallucination. Re-check that general knowledge is Off (Step 4) and that your Instructions tell it to answer only from provided sources.

Tip: A trustworthy business agent saying "I don't have that information" is a success, not a failure. Declining gracefully is exactly the behaviour you want.

Step 7: Tune your sources (~5 minutes)

If an answer is weak or missing, improve the source document — add clearer headings and more detail — then re-upload or refresh the source.
Add a second Knowledge source for broader coverage (for example, a product page website). Give each source a distinct name and description so the agent can tell them apart.
Reinforce grounding in the Instructions (Overview tab). Add a line such as:

   Only answer using the provided knowledge sources, and cite them.
   If the answer is not in the sources, say you don't have that information.

Select Save, refresh the Test pane, and re-test.

Checkpoint

You have successfully completed this lab when:

✅ At least one Knowledge source shows the status Ready
✅ The agent answers business questions from your document, with visible citations
✅ General knowledge is turned Off (if you chose the recommended setup)
✅ The agent declines to answer (no hallucination) when the information is not in your sources
✅ You can explain the difference between Instructions and Knowledge

Troubleshooting

Problem	Cause	Solution
Source stuck on "Processing"	Large or busy source	Wait a minute or two; if it persists, try a smaller file or a single web page.
Agent ignores the document	Source not ready, or no description	Confirm the source is Ready; add a clear description; in Instructions, tell it to answer from knowledge.
Answers come from the web, not my doc	General knowledge is on	Turn off Use general knowledge in Settings → Generative AI, then Save and refresh the Test pane.
No citations appear	Answer was not grounded	Ensure the answer actually came from a Ready source; check that general knowledge is off.
Agent makes things up	Hallucination	Turn off general knowledge and add an Instruction to decline when info is missing.
Edits not reflected in chat	Test pane caching	Click Refresh at the top of the Test pane to restart the conversation.

Key Takeaways

Knowledge = grounding (RAG): the agent retrieves passages from your documents and answers from them, with citations.
Use Instructions for behaviour and tone; use Knowledge for facts.
Turning general knowledge off keeps answers strictly inside your own content.
Citations let you and your users verify exactly where an answer came from.
A grounded agent that declines when it doesn't know is working correctly — that is how you avoid hallucinations.

Duration

~40 minutes

Next Steps

Proceed to Lab 8: Add Tools and Actions to Your Agent.

Lab 8: Add Tools and Actions to Your Agent

Lab Title

Give Your Agent Tools — Connectors, Prebuilt Actions, and Flows

Lab Objectives

By the end of this lab, you will be able to:

Explain what Tools/Actions are and how they differ from Knowledge
Add a prebuilt connector action as a tool (e.g. Send an email V2) and create its connection
Write a clear tool name and description that tells the agent when to use the tool
Set tool inputs, either as fixed values or filled by AI from the conversation
(Alternative) Add a multi-step Power Automate agent flow as a tool (using Add a row into a table), and disable a competing tool so the agent calls the right one
Test the agent actually performing an action, and read the Activity map

Prerequisites

Completed Lab 6 and Lab 7
A working Office 365 Outlook connection (a mailbox-enabled account you can sign in with)

Scenario

So far your Company Helpdesk agent can answer questions (using Knowledge). But staff at ACME Pte Ltd also want it to do things — for example, email the support team when someone needs help escalated.

Knowledge lets an agent answer; Tools let it act. In this lab you will give your agent a tool so a chat conversation can trigger real work, such as sending an email. This is the bridge to the end-to-end agent-plus-flow workflows you will build in Day 3.

Knowledge vs Tools: - Knowledge = read and answer from your documents (Lab 7). - Tools/Actions = perform actions in other systems — send an email, create a record, run a Power Automate flow (this lab).

Step-by-Step Guide

Step 1: Open the Tools tab (~5 minutes)

Go to https://copilotstudio.microsoft.com and confirm the environment selector (top-right) shows NUS Copilot Sandbox — the same environment as your Power Automate flows.
Open your Company Helpdesk agent from Lab 6.
Select the Tools tab (in some versions this is labelled Actions).
Select + Add a tool. Microsoft groups tools into several core tool types: — Prebuilt / custom connector action — a single ready-made operation from a connector like Office 365 Outlook, Excel, or Teams (you use this in Steps 2–4). — Agent flow — a multi-step Power Automate flow with conditions and logic, used as one tool (you use this in Step 5). — Prompt — a single-turn AI prompt that returns text. — REST API / MCP tool / Computer use — connect to web services, a Model Context Protocol server, or GUI automation (advanced; out of scope here).

How the agent picks a tool: With generative orchestration, the agent chooses which tool to call at runtime based purely on each tool's name and description — there are no fixed trigger phrases. That is why a clear, intent-rich description (Step 3) is the single most important thing you write, and why two tools with overlapping descriptions confuse the agent (see Step 5a).

⚠️ Warning: Tools and the flows they call must live in the same environment as the agent. If your flow was built in a different environment, the agent will not be able to see or call it.

Step 2: Add a prebuilt connector action (~10 minutes)

You will add a "send an email" action directly as a tool.

Select + Add a tool, then choose Connector (or Add an action).
In the search box, type Send an email and select Send an email (V2) under Office 365 Outlook.
If you are prompted to sign in / create a connection, do so now using a mailbox-enabled account (an account that can actually send email). Approve any consent prompt.
The action is added to your tools list. Open it to see its inputs: To, Subject, and Body.

⚠️ Warning: If you ever see an Unauthorized error when the tool runs, the connection is broken or signed in with the wrong account. Open the tool's connection settings and reconnect with a mailbox-enabled account.

Step 3: Name and describe the tool (~5 minutes)

The description is the most important part — it is how the agent decides when to call this tool.

Open the tool and set a clear name and description, for example: — Name: Send notification email — Description: Use this to email a summary to the support team when the user asks to escalate an issue or notify someone.
Be specific. A vague description like sends email makes the agent unsure when to use it; a precise description makes it reliable.

Tip: The agent reads the tool description the way you would read a label. If the label clearly says what the tool is for, the agent picks it at the right moment.

Step 4: Configure the tool's inputs (~7 minutes)

Decide how each input (To, Subject, Body) gets its value. You have two choices per input:

Fixed value — you type the value once and it never changes. Good for To: set it to your own email so test messages go to you, e.g. you@yourcompany.com.
Filled by AI (dynamic) — the agent fills the value from the conversation. Choose Fill using AI (or Dynamically) for Subject and Body so the agent writes them based on what the user said.

Tip: For your first test, set To as a fixed value (your own inbox). That way you can confirm the email actually arrives without risking sending it to the wrong person.

Step 5: (Alternative path) Add a Power Automate flow as a tool (~15 minutes)

A single connector action (Steps 2–4) is perfect for one step. But real business work is usually multi-step — you often need to log a record and notify someone and return a result in one go. A single connector tool cannot do that; a Power Automate agent flow can. In this step you will build a flow that logs a support request to a table, then add that flow to the agent as a second tool.

Why a different action here? In Steps 2–4 your agent already learned to send an email with a single connector tool. To show what a flow adds, this step uses a different action — Add a row into a table — so the flow records the request instead of emailing it. (Do not add another Send an email action here; that would just duplicate the tool you already built and confuse the agent about which one to call.)

Step 5a — Turn OFF the connector tool from Step 4 first (do not skip)

If you leave the Send notification email connector tool (Steps 2–4) enabled while you add and test the new flow tool, your agent now has two tools that both look like "do something with the user's request." The orchestrator may keep choosing the old email tool and never call your new flow — making it look like the flow is broken when it is not. Disable the old tool so this exercise has exactly one active tool to reason about.

In your agent, open the Tools tab.
Find the Send notification email tool you added in Steps 2–4.
Select its … (More actions) menu and turn the tool Off (toggle/disable it). Do not delete it — you will switch it back on later.
Confirm the tool now shows as Off / Disabled in the list before continuing.

⚠️ Warning: This was the step most learners discovered was missing. With both tools on, the agent often fires the Step 4 email tool instead of the new flow, and the flow appears to "do nothing." Turning the connector tool Off isolates the flow so you can confirm it works.

Step 5b — Create the agent flow

In Copilot Studio's left navigation, select Tools, then + New tool → choose the Agent flow tile. (From inside the agent you can instead use the Tools tab → + Add a tool → New Agent flow.)
Power Automate opens with two steps already in place: the trigger When an agent calls the flow and a Respond to the agent action. Every agent flow follows this three-part shape: trigger → do the work → respond.

Step 5c — Define the flow's input

Select the trigger When an agent calls the flow, then + Add an input.
Choose Text, name it Request summary, and give it the hint A short summary of the support request. This is the value the agent will pass in from the conversation.
Open the Overview tab → Details → Edit, set Flow name to Log support request and Description to Logs a support request as a new row in the requests table, then Save.

Step 5d — Add the work: "Add a row into a table" (the only action here)

Back on the Designer tab, select the + between the trigger and Respond to the agent to insert an action.
Search for Excel, choose Excel Online (Business), then the action Add a row into a table (reuse the workbook and table from Lab 2, or any table with a Request column).
Map the row's Request column to the trigger's Request summary input using Dynamic content.

Reminder: Add a row into a table here — not a Send an email action. The email path was already covered by the connector tool in Steps 2–4.

Step 5e — Respond to the agent

Select the Respond to the agent node, then + Add an output.
Choose Text, name it status, and set its value to Logged (or use Dynamic content to return the new row's ID). The flow must end here, or the agent never gets a result back.

Step 5f — Publish

Select Save draft, then Publish.
Return to Tools and confirm the flow's status is Ready.

Step 5g — Add the flow to the agent and map its input

In the agent, open the Tools tab → + Add a tool, apply the Workflows filter, select Log support request, then Add and configure.
Set a clear Description: Records a support request to the requests table when the user asks to log or file an issue.
Under Additional details, review When this tool may be used, Ask the end user before running, and Credentials to use (end-user credentials, with a sign-in prompt like Please sign in to log the request).
Under Inputs, for the Request summary input set Fill using → Dynamically fill with AI so the agent writes the summary from the conversation (or Custom value to bind a variable), just like the fixed-vs-AI choice in Step 4.
Under Completion, set After running → Write the response with generative AI so the agent confirms the result in natural language. Save.

Tip: Every flow used as a tool follows the same three-part shape: When an agent calls the flow (trigger) → do the work → Respond to the agent (give a result back). You will reuse this pattern throughout Day 3.

Step 6: Test the agent performing the action (~8 minutes)

Open the Test pane and select Refresh at the top so it picks up the new flow tool.
Type a message that should trigger the flow, for example: — Please log a support request: I need help with order 123.
Watch the agent decide to call the Log support request flow. If it asks for confirmation or shows a sign-in prompt, approve it.
Confirm the action really happened — open your table and check that a new row was added with the request summary.
Open the Activity map (a diagram of the conversation, if shown) to see the tool being called and the exact data passed into it. This is the best way to understand why the agent did what it did.
Switch the email tool back on: return to the Tools tab and turn the Send notification email tool (Steps 2–4) back On. With both tools live and clearly described, the agent can now log a request and email support — choosing each by its description.

Tip: If the flow does nothing, check three things in order: is the Send notification email tool still Off during the isolation test, did the flow Publish to status Ready, and did the agent actually choose the flow (look at the Activity map)?

Step 7: Refine the tool (~5 minutes)

Agent never calls the tool? Improve the tool description to clearly state the trigger situations, and add an example trigger phrase in the agent's Instructions (e.g. "When a user asks to notify or escalate, use the Send notification email tool.").
Agent calls the tool with the wrong data? Tighten the input setup — use a fixed value where the data should never change, or add a clearer hint for the AI-filled inputs.
Save, refresh the Test pane, and test again.

Checkpoint

You have successfully completed this lab when:

✅ A tool (a connector action and/or a flow) is added to your agent
✅ The tool has a clear name and description that tells the agent when to use it
✅ The tool's connection is signed in with a mailbox-enabled account (no Unauthorized error)
✅ The agent performs the action during a test conversation (e.g. the email arrives)
✅ You found the tool call in the Activity map

Troubleshooting

Problem	Cause	Solution
Agent never calls the tool	Description too vague	Rewrite the description to name the exact situations; add a trigger example in Instructions.
Connection / Unauthorized error	Broken connection or wrong account	Open the tool's connection and reconnect with a mailbox-enabled account.
Email never arrives	Wrong recipient or tool not called	Check the To value, confirm the connection, and verify the tool ran in the Activity map.
Wrong data sent to the tool	AI filled inputs incorrectly	Use fixed values where data shouldn't change, or add clearer input hints; test again.
Tool not listed after adding	Wrong environment, or not refreshed	Confirm the agent and flow are in the same environment (NUS Copilot Sandbox); refresh the page.
Flow tool fails to return	No response step	Make sure the flow ends with Respond to the agent with an output, then re-publish it.
Agent calls the old email tool, never the flow	Two competing tools both enabled	Turn the Send notification email connector tool Off while you test the flow (Step 5a), then test again.
Flow shows up but status isn't Ready	Not published	Save draft then Publish the flow; the Tools list must show status Ready before the agent can call it.

Key Takeaways

Tools/Actions let an agent act — using connectors, prebuilt actions, or full Power Automate flows.
The tool description is what makes the agent call the right tool at the right time — write it carefully.
Inputs can be fixed or filled by AI from the conversation; use fixed values for anything that must not change.
A flow used as a tool always goes When an agent calls the flow → do the work → Respond to the agent.
Use a flow (not a single connector action) when the work is multi-step — e.g. Add a row into a table to log a record and return a result.
When two tools overlap, the agent may pick the wrong one — turn off the competing tool while you test, then re-enable it once each tool has a precise description.
A broken connection causes an Unauthorized error — reconnect with a mailbox-enabled account.
Flows-as-tools are the foundation of the end-to-end workflows you build in Day 3.

Duration

~45 minutes

Next Steps

Proceed to Lab 9: Sales Enquiry Assistant.

Lab 9: Sales Enquiry Assistant

Lab Title

Build a Sales Enquiry Assistant that Captures Structured Data

Lab Objectives

By the end of this lab, you will be able to:

Create a Copilot Studio agent from scratch in the correct environment
Build a topic and set its trigger so the agent knows when to run it — a description for generative orchestration (default), or trigger phrases for classic orchestration
Add Ask a question nodes that save answers into named variables
Choose the right Identify type so text and numbers are captured cleanly
Return a tidy structured summary that inserts every captured variable
Understand how these variables will later feed a Power Automate flow (Lab 10 and Day 3)

Prerequisites

Completed Lab 6
Read Module 3, especially Prompt design for structured outputs

Tip: Sign in to Copilot Studio at https://copilotstudio.microsoft.com and confirm the environment shown at the top-right says NUS Copilot Sandbox before you start. Every lab on Day 2 must be built in this same environment, otherwise your agent and your flow will not be able to see each other later.

Scenario

You work at ACME Pte Ltd. The sales team complains that enquiries arrive in inconsistent formats — some by email, some by phone notes — so details get lost. Your job is to build a Sales Enquiry Assistant agent that walks a customer through a short set of questions and captures name, company, product, and quantity as separate, named pieces of data. The result is a clean structured summary — exactly the kind of tidy data a Power Automate flow can log and route automatically in the next lab.

Step-by-Step Guide

Step 1: Create the agent (~5 minutes)

In the left navigation, select Create, then select New agent.
A conversational setup screen may appear. Look for the Skip to configure button (top-right of that panel) and select it so you can fill in the details directly.
Fill in the configuration fields: — Name: Sales Enquiry Assistant — Description: Captures sales enquiries (name, company, product, quantity) as structured data for ACME Pte Ltd. — Instructions:

     You are a Sales Enquiry Assistant for ACME Pte Ltd.
     Your job is to collect a customer's enquiry details: full name, company,
     product of interest, and quantity.
     Collect one item at a time and be polite and concise.
     Do not answer unrelated questions; gently steer back to capturing the enquiry.

Select the Create button (top-right). Wait a few seconds while the agent is provisioned.

Tip: If Create is greyed out, the Name field is usually empty. Every agent needs a name before it can be created.

Step 2: Create the "New Sales Enquiry" topic (~10 minutes)

⚠️ Read this first — the latest Copilot Studio works differently. New agents now use generative orchestration by default. In this mode a topic is triggered by its name and description — the agent chooses the topic when the user's message matches that description — not by a fixed list of "Phrases". The old phrase list still exists, but it's now called User says a phrase and only appears if you deliberately switch the trigger type. This lab uses the modern description-based trigger (recommended and easiest); the phrase option is shown at the end of this step.

Open your Sales Enquiry Assistant agent, then in the agent's top menu select Topics. (If you don't see it, select ⋯ More and choose Topics.)
Select + Add a topic, then choose From blank. A blank canvas opens with a single Trigger node at the top. (You may also see Create from description with Copilot — ignore it for this lab so you build the steps yourself.)
On the canvas toolbar, select Details to open the Topic details panel, then set: — Name: New Sales Enquiry (don't use a full stop/period in a topic name) — Description: Use this topic when the customer wants to make a sales enquiry, ask for a quote, or say they are interested in buying a product. It collects the customer's name, company, product, and quantity.

Close the panel. This Description is what makes the agent pick this topic, so keep it specific.

Select the Trigger node once. For a generative-orchestration agent it reads The agent chooses — that is correct, and it uses the description from step 3. You do not need to add any phrases.
Select Save (top-right of the canvas).

Tip: The agent matches on meaning, not exact wording. A clear, specific Description is what makes the topic fire reliably — write it the way you'd explain to a colleague when this topic should be used.

(Optional) Want exact trigger phrases instead? Only if your agent uses classic orchestration, or you specifically want the topic to fire on set phrases:

Hover over the Trigger node and select the Change trigger icon.
Choose User says a phrase.
Add example phrases (aim for 5–10), each on its own line: I want to make an enquiry, I'd like a quote, new sales enquiry, I'm interested in a product.
Select Save.

Step 3: Capture the customer's name and company (~10 minutes)

Under the Trigger node, select the Add node icon (+), then choose Ask a question. A Question node appears.
Configure the question node: — Ask a question (message): Sure, I can help with that! What is your full name? — Identify: open this dropdown and choose User's entire response. This captures the whole reply as plain text (good for names). — Save user response as: Copilot Studio creates a variable automatically. Select the variable name and rename it to customerName.
Select the Add node icon (+) below that node and add a second Ask a question node for the company: — Ask a question (message): Thanks! Which company are you with? — Identify: User's entire response — Save user response as: rename the variable to company

Tip: Rename variables right away. A clear name like customerName is much easier to find later than the default Var1, especially when you map it to a flow in Lab 10.

Step 4: Capture the product and quantity (~10 minutes)

Add another Ask a question node for the product: — Ask a question (message): Which product are you interested in? — Identify: User's entire response — Save user response as: product
Add a final Ask a question node for the quantity: — Ask a question (message): How many units would you like? — Identify: open the dropdown and choose Number. This forces the answer to be a real number, so it can be used in calculations and stored as a number later. — Save user response as: quantity

⚠️ Warning: Make sure quantity uses Identify = Number, not "User's entire response". If you leave it as text, quantity becomes a word like "twenty-five" instead of 25, and the Excel/flow steps in Lab 10 will store messy data.

Tip: Capturing into four separate named variables (customerName, company, product, quantity) is what makes the output structured — each piece of data is isolated and reusable. A single blob of free text could not be logged into separate spreadsheet columns.

Step 5: Confirm and summarize (~10 minutes)

Select the Add node icon (+) below the quantity question and choose Send a message. A Message node appears.
Type the summary text below. Wherever you see a curly-brace variable, do not type it as plain text — instead place your cursor there, select the {x} (Insert variable) button on the node, and pick the matching variable from the list:

   Thank you! Here is a summary of your enquiry:
   • Name: {customerName}
   • Company: {company}
   • Product: {product}
   • Quantity: {quantity}
   A sales representative from ACME will contact you within 1 business day.

Select Save.

⚠️ Warning: If you type {customerName} as literal text, the customer will literally see the words "{customerName}". You must insert it through the {x} button so it shows up as a coloured variable token.

Step 6: Test the assistant (~10 minutes)

Open the Test pane on the right (if it was already open, select the Refresh / restart icon so it picks up your latest changes).
Type a message that starts an enquiry, e.g. I'd like a quote. The agent should recognise this from your topic Description and start the New Sales Enquiry topic.
Answer each question in turn: — Name: Mei Ling — Company: BrightTech — Product: Air Fryer Pro — Quantity: 25
Confirm the agent returns the structured summary with all four values filled in correctly.
Now test the unhappy path: start the topic again, and when asked for quantity, type the word twenty-five instead of a number. Because Identify is set to Number, the agent should re-ask the question. Type 25 and confirm it then accepts it.

Tip: The re-ask behaviour is a feature, not a bug. It guarantees quantity is always a clean number before it reaches a flow.

Checkpoint

You are ready to move on when all of the following are true:

✅ An agent named Sales Enquiry Assistant exists in the NUS Copilot Sandbox environment
✅ It has a New Sales Enquiry topic with a clear trigger — a description (The agent chooses) or trigger phrases
✅ Four variables are captured: customerName, company, product, and quantity (Number)
✅ A structured summary with all four values is returned during testing
✅ Typing text for the quantity causes the agent to re-ask

Troubleshooting

Problem	Solution
I can't find a "Phrases" trigger type	New agents use generative orchestration, so the trigger is The agent chooses (description-based). Either write a clear topic Description, or hover the Trigger node → Change trigger → User says a phrase to add phrases.
Topic doesn't trigger when I type	Make the topic Description more specific about when to use it (or add more varied phrases if using User says a phrase). Type something close in meaning; re-test after Save.
Variable shows as blank in the summary	You probably typed the variable name as text. Delete it and re-insert it using the {x} button. Also confirm the Save user response as name matches.
Summary shows literal `{customerName}` text	Same cause — insert variables via {x}, do not type curly braces by hand.
Quantity answer is rejected or re-asked forever	Set the quantity question's Identify to Number and answer with digits like `25`, not spelled-out words.
Test pane shows old behaviour	Select the Refresh / restart icon at the top of the Test pane to reload the latest version of the topic.
Agent answers off-topic questions	Tighten the Instructions to say it must steer back to capturing the enquiry.

Key Takeaways

Topics are triggered by a description (The agent chooses, generative orchestration — the default) or by trigger phrases (User says a phrase, classic orchestration).
Ask a question nodes capture answers into named variables, which is the foundation of structured data.
The Identify type matters: use User's entire response for free text and Number for numeric answers.
A confirmation summary builds user trust and lets you visually verify the captured data before any automation runs.
Clean, separated variables here are what make the agent-to-flow integration in Lab 10 possible.

Duration

~45 minutes

Next Steps

Proceed to Lab 10: Procurement Request Workflow, where you'll connect a capturing agent to a Power Automate flow for the first time.

Lab 10: Procurement Request Workflow

Lab Title

Connect a Procurement Agent to a Power Automate Flow

Lab Objectives

By the end of this lab, you will be able to:

Prepare an Excel table so a flow can log structured rows into it
Build an agent topic that captures a procurement request as named variables
Create an agent flow in Power Automate triggered when an agent calls the flow
Define flow inputs and map agent variables into them
Have the flow log the request to Excel and email the procurement team in one run
Return a confirmation from the flow back to the agent with Respond to the agent

Prerequisites

Completed Lab 9 (capturing variables)
Completed Day 1 Labs 1–2 (email + Excel actions)

⚠️ Warning: Your agent (Copilot Studio) and your flow (Power Automate) must live in the same environment — NUS Copilot Sandbox. If they are in different environments, the agent will not be able to find or call its flow. Before you begin, open both Copilot Studio and Power Automate and confirm the top-right environment picker shows NUS Copilot Sandbox in each.

Scenario

At ACME Pte Ltd, staff currently email the procurement team to request supplies, and someone manually copies each request into a spreadsheet. You'll replace that with an agent. When a staff member finishes their request, the agent automatically logs it to a shared spreadsheet and emails the procurement team — no manual copying, no missed requests. This is your first true agent + flow integration: the agent gathers the data, and a Power Automate flow does the real work.

Step-by-Step Guide

Step 1: Prepare the procurement log in Excel (~5 minutes)

In Excel (saved to OneDrive for Business, not your local PC), create a new workbook and name it Procurement Log.
In row 1, type these six headers, one per cell from A1 to F1: Date, Requester, Item, Quantity, Reason, Status.
Select the range A1:F1, then go to the Insert tab and select Table. In the dialog, tick My table has headers and select OK.
With the table selected, open the Table Design tab. In the Table Name box (far left), replace the default name with ProcurementTable. Press Enter.
Save and close the workbook.

⚠️ Warning: The file must be in OneDrive for Business (or SharePoint), not on your local hard drive. A flow's Excel actions can only see files stored in the cloud. The exact table name ProcurementTable matters — the flow will look for it by name.

Step 2: Create the agent and capture the request (~10 minutes)

In Copilot Studio, select Create, then New agent, then Skip to configure: — Name: Procurement Assistant — Description: Captures staff purchase requests for ACME Pte Ltd and submits them for processing. — Instructions:

     You are a Procurement Assistant for ACME Pte Ltd.
     Collect a staff member's purchase request: requester name, item, quantity,
     and reason. Collect one item at a time, be concise, then confirm the request
     has been submitted.

Select Create.

Open the Topics tab, select + Add a topic, then From blank. On the toolbar select Details and set the Name to New Procurement Request.
Give the topic a clear trigger so the agent knows when to run it: — Latest Copilot Studio (generative orchestration — default): the Trigger node reads The agent chooses. In the Details panel, set the Description to Use this topic when a staff member wants to raise a procurement or purchase request to buy supplies. It collects their name, item, quantity, and reason. No phrases are needed. — (Optional) Classic orchestration / exact phrases: hover the Trigger node → Change trigger → User says a phrase, then add phrases such as procurement request, I need to buy something, I want to order supplies, raise a purchase request.
Add four Ask a question nodes in order, using the Add node icon (+) each time. For each, set the Identify type and rename the Save user response as variable exactly as shown: — Question What is your name? → Identify User's entire response → save as requester — Question What item do you need? → Identify User's entire response → save as item — Question How many do you need? → Identify Number → save as quantity — Question What is the reason for this request? → Identify User's entire response → save as reason
Select Save.

⚠️ Warning: Set the quantity question's Identify to Number. If it stays as text, the Excel Quantity column will store words instead of numbers.

Step 3: Create the agent flow (~15 minutes)

Still in the New Procurement Request topic, select the + node after the last question, choose Add a tool, then New Agent flow. — (Alternatively, from the agent's Tools tab, select Add a tool, then New agent flow.)
Power Automate opens in a new tab with the trigger already added: When an agent calls the flow.
Confirm the Power Automate environment picker (top-right) reads NUS Copilot Sandbox.
On the When an agent calls the flow trigger, select + Add an input and create four inputs. Choose the matching type and name each one exactly: — Type Text, name requester — Type Text, name item — Type Number, name quantity — Type Text, name reason

Tip: Keeping flow input names identical to your topic variable names (requester, item, quantity, reason) makes the mapping step later much less confusing.

Step 4: Add the flow actions — log to Excel and email (~15 minutes)

Action 1 — Add a row into a table (Excel Online Business):

Below the trigger, select + New step (or +), search for Add a row into a table, and select it (Excel Online (Business)).
Set the file location fields: — Location: OneDrive for Business — Document Library: OneDrive — File: browse to and select Procurement Log.xlsx — Table: select ProcurementTable from the dropdown
Map the columns. For the Date column, you must enter an expression — do not type the date by hand: — Click the Date field, then select the fx (Expression / function) tab in the dynamic content panel. — Type this expression into the fx editor, then select Add / OK:

     formatDateTime(utcNow(),'yyyy-MM-dd HH:mm')

The field should now show a blue/purple token, not plain text.

Map the remaining columns using the Dynamic content tab — pick the values that come from the trigger (the inputs you defined in Step 3): — Requester: requester — Item: item — Quantity: quantity — Reason: reason — Status: type the literal text Pending

⚠️ Warning: The formatDateTime(utcNow(),'yyyy-MM-dd HH:mm') expression must be entered through the fx editor so it becomes a token. If you paste it as plain text into the Date field, Excel will store the literal characters formatDateTime(...) instead of a real date.

Action 2 — Send an email (Office 365 Outlook):

Select + New step, search for Send an email (V2) (Office 365 Outlook), and select it.
Fill in the email: — To: the procurement team address (use your own email for testing) — Subject: type New Procurement Request from , then insert the requester dynamic content right after it — Body: type the text below, inserting each value from the Dynamic content panel where shown (do not type the variable names by hand):

     A new procurement request has been submitted:
     Requester: {requester}
     Item: {item}
     Quantity: {quantity}
     Reason: {reason}
     Status: Pending

Action 3 — Respond to the agent:

Select + New step, search for Respond to the agent, and select it.
Add a Text output named result with the value: Logged and procurement team notified.
Select Save at the top to save (and publish) the flow.
Return to the Copilot Studio tab (or select Go back to agent if prompted).

⚠️ Warning: If the Send an email action shows an "Unauthorized" error, the Office 365 Outlook connection is broken or signed in with an account that has no mailbox. Open the action's … menu, select My connections, and reconnect using a mailbox-enabled work account. Every connection used by the flow must show a green ✓ before the flow will run.

Step 5: Wire the flow into the topic (~10 minutes)

Back in Copilot Studio, the flow should now appear as a tool/action node inside the New Procurement Request topic. If it does not appear, refresh the page and make sure the flow was published in the same environment (NUS Copilot Sandbox).
Select the tool node. Under its inputs, map each input to the matching topic variable — this step is mandatory; the flow receives nothing if you skip it: — input requester → variable requester — input item → variable item — input quantity → variable quantity — input reason → variable reason
After the tool node, add a Send a message node that shows the value the flow returned. Insert the result output via the {x} button:

   {result} Your request has been recorded. Thank you!

Select Save to save the topic.

⚠️ Warning: Mapping is the step everyone forgets. After adding the flow as a tool, you must map each flow input to its matching topic variable. Unmapped inputs arrive empty, so your Excel row and email will be blank.

Step 6: Test end-to-end (~10 minutes)

Open the Test pane and select the Refresh / restart icon so it loads the latest topic.
Type a message that starts the request, e.g. procurement request — the agent recognises it from the topic Description (or phrases). Answer the questions: — Name: Daniel — Item: Wireless mouse — Quantity: 10 — Reason: New hires
Confirm all three results occurred: — The agent shows the confirmation message ("Logged and procurement team notified. Your request has been recorded. Thank you!"). — A new row appears in Procurement Log.xlsx (with a real date, the four values, and Status Pending). — The notification email arrives in your inbox.
If anything is missing, open the flow in Power Automate and review its run history (28-day run history) — select the latest run to inspect the inputs each step received and any error messages.

Checkpoint

You are ready to move on when all of the following are true:

✅ A Procurement Log workbook with a ProcurementTable exists in OneDrive
✅ A Procurement Assistant agent captures four variables (requester, item, quantity, reason)
✅ An agent flow triggered by When an agent calls the flow, with four matching inputs
✅ One conversation produces both a new Excel row and a notification email
✅ The agent shows the flow's returned confirmation message

Troubleshooting

Problem	Solution
Flow doesn't appear in the topic	Refresh the Copilot Studio page; confirm the flow was published and is in the same environment (NUS Copilot Sandbox).
Inputs unmapped or arriving empty	Select the tool node and map each flow input to its matching topic variable.
"Unauthorized" error on Send an email	Reconnect the Office 365 Outlook connection with a mailbox-enabled work account; every connection must show a green ✓.
Date column shows `formatDateTime(...)` as text	You typed the expression instead of entering it in the fx editor. Re-enter it via fx so it becomes a token.
Excel row is blank	Map the columns to the trigger dynamic content (the inputs), not to outputs of later steps.
No email received	Check your Junk folder, verify the To address, and review the flow run history for errors.
Flow can't find the table	Confirm the file is in OneDrive for Business and the table is named exactly `ProcurementTable`.

Key Takeaways

An agent flow uses the When an agent calls the flow trigger and ends with Respond to the agent.
Agent variables map to flow inputs; this mapping is mandatory and easy to forget.
The flow can return a value (result) that the agent shows back to the user.
Dates come from the formatDateTime(utcNow(),'yyyy-MM-dd HH:mm') expression entered via the fx editor as a token — never typed as text.
One short conversation now triggers real business actions: logging and notifying together.

Duration

~50 minutes

Next Steps

Proceed to Lab 11: Automated Response Generation.

Lab 11: Automated Response Generation

Lab Title

Use AI Prompts to Generate Professional Responses Automatically

Lab Objectives

By the end of this lab, you will be able to:

Add an AI prompt action (AI Builder "Create text with GPT" / Run a prompt) inside a flow
Feed captured enquiry data into the prompt to generate tailored text
Control the length, tone, and format of the AI output
Email the AI-generated response automatically
Compare a static template versus an AI-generated response and know when to use each

Prerequisites

Completed Lab 10
A Sales Enquiry Assistant agent (from Lab 9) you can reuse

⚠️ Warning: As in Lab 10, the agent and the flow must both be in the NUS Copilot Sandbox environment, or the agent cannot call its flow. Confirm the environment picker (top-right) in both Copilot Studio and Power Automate.

Scenario

At ACME Pte Ltd, every sales enquiry currently gets the same copy-paste reply. It works, but it feels impersonal. In this lab you'll reuse your Sales Enquiry Assistant from Lab 9 and add a flow that uses an AI prompt to draft a warm, personalised reply from the customer's captured details — then emails it automatically. This shows how generative AI turns the structured data you captured into polished, individual communication, while still keeping you in control of length and tone.

Step-by-Step Guide

Step 1: Understand where the AI prompt runs (~5 minutes)

An AI prompt can run in two equivalent places: — In a Power Automate flow — using the AI prompt action (AI Builder). — In Copilot Studio — by adding a Prompt tool to a topic.
This lab uses the flow approach because it builds directly on the agent + flow skills from Lab 10 and keeps the "generate" and "email" steps together in one place.

Tip: Both options call the same underlying AI Builder model. Choosing the flow approach simply means generation and emailing live in one flow you can inspect via run history.

Step 2: Reuse the Sales Enquiry Assistant and add a flow (~10 minutes)

Open your Sales Enquiry Assistant agent (from Lab 9) and open the New Sales Enquiry topic.
After the summary Send a message node, select the + node, choose Add a tool, then New Agent flow.
Power Automate opens with the trigger When an agent calls the flow. Confirm the environment reads NUS Copilot Sandbox.
On the trigger, select + Add an input and add four inputs (matching the variables from Lab 9): — Type Text, name customerName — Type Text, name company — Type Text, name product — Type Number, name quantity

Step 3: Add the AI prompt action (~15 minutes)

Below the trigger, select + New step, search for AI prompt (it may also appear as Run a prompt or Create text with GPT using a prompt) under AI Builder, and select it.
Choose to create a new prompt (or Build your own prompt).
Write the prompt instruction below. Wherever a value comes from the enquiry, insert it from the Dynamic content panel rather than typing it:

   Write a warm, professional email reply (maximum 120 words) to a sales enquiry
   for ACME Pte Ltd.
   Customer name: {customerName}
   Company: {company}
   Product of interest: {product}
   Quantity: {quantity}

   The email should:
   - Thank the customer by name
   - Acknowledge the product and quantity
   - Say a sales representative will follow up within 1 business day with pricing
   - End with a professional sign-off from "The ACME Sales Team"
   Output only the email body text. Do not include the prompt or these instructions.

Select Save on the prompt. The action now produces a generated text output (often called Text or Response).

Tip: Notice the prompt explicitly states length (max 120 words), tone (warm, professional), content (what to mention), and format ("output only the email body"). Controlling all four is what separates a reliable AI step from an unpredictable one — this is the structured prompt design from Module 3 applied to output generation.

Step 4: Email the generated response (~10 minutes)

Select + New step, search for Send an email (V2) (Office 365 Outlook), and select it.
Fill in the email: — To: your own email (for testing; in production you would use a captured customer email variable) — Subject: type Re: Your enquiry about , then insert the product dynamic content — Body: insert the AI prompt output (the generated email text) from the Dynamic content panel
Select + New step, add Respond to the agent, and return a Text output named result with value Reply drafted and sent.
Select Save to save (and publish) the flow, then return to the Copilot Studio tab.

⚠️ Warning: If Send an email shows an "Unauthorized" error, reconnect the Office 365 Outlook connection with a mailbox-enabled work account. Every connection used by the flow must show a green ✓ before it will run.

Step 5: Wire the inputs and test (~10 minutes)

Back in the topic, select the flow's tool node and map each input to its matching topic variable — the flow gets empty values if you skip this: — input customerName → variable customerName — input company → variable company — input product → variable product — input quantity → variable quantity
(Optional) After the tool node, add a Send a message node showing {result} so the agent confirms the email was sent.
Select Save to save the topic.
Open the Test pane, select the Refresh / restart icon, type a message that starts the enquiry (the agent recognises it from the topic Description), and run the enquiry: — Name: Mei Ling, Company: BrightTech, Product: Air Fryer Pro, Quantity: 25
Confirm: — The flow runs successfully (check run history if unsure). — You receive a uniquely worded, personalised email — not a fixed template.
Run it again with different details and compare the two emails; the AI adapts the wording each time.

Tip: If the email body contains the prompt instructions, your "Output only the email body text" line is missing or you mapped the wrong field — map only the prompt output, and keep that instruction in the prompt.

Step 6: Compare template versus AI (~5 minutes)

Recall the static template email from Day 1: it is fast and 100% predictable, but identical for every customer.
The AI-generated response is personalised and natural, but you must guide it with a precise prompt and review it for accuracy.
Note the best-practice rule: use AI for drafting the wording, and keep critical facts (prices, policies, legal terms) as controlled static text or knowledge — never let the model invent them.

Tip: A safe production pattern is "AI drafts, human (or fixed text) confirms the facts." Personalisation from AI plus controlled facts gives you the best of both.

Checkpoint

You are ready to finish when all of the following are true:

✅ A flow that calls an AI prompt with the four enquiry inputs
✅ A personalised email generated and sent automatically from the agent conversation
✅ The email contains only the body (no leaked prompt instructions)
✅ You can explain when to use a static template versus an AI-generated reply

Troubleshooting

Problem	Solution
No AI prompt action available	Search AI Builder / Create text with GPT; confirm AI Builder is available in this environment (the trial / developer environment includes it).
"Unauthorized" error on Send an email	Reconnect the Office 365 Outlook connection with a mailbox-enabled account; every connection must show a green ✓.
Output too long or wrong tone	Tighten the prompt: state the word limit, the tone, and "output only the email body."
Captured values not appearing in the email	Confirm you inserted them as dynamic content inside the prompt, and that the topic inputs are mapped to the variables.
Email body shows the prompt instructions	Add "Output only the email body text" to the prompt and map only the prompt output into the email body.
Flow doesn't appear in the topic	Refresh Copilot Studio; ensure the flow is published in the same environment (NUS Copilot Sandbox).

Key Takeaways

AI prompts turn the structured data you captured into polished, personalised text.
You control quality by specifying length, tone, content, and format — especially "output only the email body."
Combine AI drafting with controlled facts for responses that are both personal and reliable.
The same agent + flow pattern from Lab 10 now powers generative output, not just logging.

Duration

~45 minutes

Next Steps

You've completed Day 2. Proceed to Day 3 — Module 4: End-to-End Orchestration Concepts.

Day 3 — End-to-End Workflow Automation & Workshop

Module 4: End-to-End Workflow Automation

Read this before the Day 3 labs. ~15 minutes.

You now have both halves of an automated business process. Today you join them into complete, end-to-end workflows — and learn the design discipline that keeps them reliable.

1. Agents + Power Automate = end-to-end automation

By now you've built both halves:

Power Automate flows (Day 1) — the doing: send email, log to Excel, run approvals.
Copilot Studio agents (Day 2) — the understanding: capture requests as structured data and call flows.

End-to-end automation chains these into a complete business process where a single trigger flows all the way through to a finished outcome — with little or no manual work in between:

 CAPTURE  →  RECORD  →  DECIDE  →  NOTIFY  →  CLOSE
 (agent/    (Excel    (approval  (email/    (final
  email/     log)      or rule)   Teams)     status)
  form)

One event goes in the top; a finished, recorded, communicated result comes out the bottom.

2. Workflow orchestration

Orchestration is arranging multiple steps — and sometimes multiple flows and agents — so they run in the right order, pass data correctly, and handle every outcome.

A well-orchestrated workflow has five qualities:

A clear trigger — one starting event (email arrives, file uploaded, agent finishes capturing).
Sequenced actions — each step's output feeds the next step's input.
Branching — conditions route the process (approved vs rejected, high value vs low value).
Notifications — the right people are told at the right moments.
A definite end state — the record is updated to its final status (Logged / Approved / Rejected / Completed).

No dead ends. Orchestration is mostly about making sure every path — including the unhappy ones — leads somewhere defined. A rejected request still needs an owner, a notification, and a final status.

Common orchestration patterns

You'll build one of each this morning. Notice how they grow in complexity:

Pattern	Shape	What's new	Day 3 lab
Capture → Log → Notify	Linear: record it, tell someone	Pure sequence, no branch	Lab 12
Upload → Approve → Act	File trigger, then human decision	Adds an approval + branch	Lab 13
Request → Approve → Notify	Multi-party approval chain	Adds routing by who/what	Lab 14

Approver reminder (from Day 1): the approver in any approval step must be a real user in your tenant, not an outside email address. Pick a colleague — or yourself — for testing.

3. Managing workflow outputs and next steps

Every step produces outputs. Managing them well is what separates a demo from a dependable workflow:

Pass data forward with dynamic content — sender's email → notification; captured quantity → approval title. (Use the fx expression editor for any small formatting, like a date or a trimmed string.)
Use a Status column as the source of truth: New → Pending → Approved/Rejected → Done. Update it at each stage so anyone can see where a request stands at a glance.
Decide the next step for every outcome. Never leave a branch empty — an approved request and a rejected request both need a defined follow-up.
Notify deliberately. Tell the requester the result; tell the team when action is needed. Avoid noisy, redundant emails.
Make it traceable. The Excel log plus the Power Automate run history together give you a complete audit trail.

Designing a workflow — a simple method

Use this every time before you start clicking in the designer:

Write the sentence: "When ___ happens, do ___, then ___, and finally ___."
Identify the trigger (the "when").
List the actions in order (the "do / then / finally").
Mark the decision points (where it branches).
Define each end state (final status + who is notified).
Build, test the happy path, then test every branch.

Sentence first, build second. If you can't say the workflow in one plain sentence, you're not ready to build it yet. The sentence reveals your trigger, your steps, and your end states before you touch the tool.

You'll apply exactly this method in Lab 16 (Capstone Workshop).

4. What you'll build on Day 3

Lab 12: Email enquiry → log to Excel → notify the team (Capture → Log → Notify).
Lab 13: Invoice file uploaded → approval workflow (Upload → Approve → Act).
Lab 14: Purchase request → manager approval → notification (Request → Approve → Notify).
Lab 15: Order Processing — an agent captures an order, then the flow confirms, logs, and raises a restock alert.
Lab 16: Capstone — design and build your own end-to-end workflow for Sales, Finance, Procurement, or Order Processing.

Labs 12–14 drill the three core patterns; Lab 15 brings the agent back into the loop; Lab 16 is where you put it all together yourself.

Next: Lab 12: Email Enquiry → Excel Logging → Notification

Module 5: Business Workflow Workshop

Read this before the Capstone (Lab 16). ~10 minutes.

This is your briefing for the capstone. Everything you've learned across three days now comes together in one workflow that you design, build, and explain.

1. Purpose of the workshop

You've now learned every building block:

Power Automate flows — email, Excel logging, approvals, scheduled and event triggers (Day 1)
Copilot Studio agents — instructions, knowledge/RAG, tools, structured capture (Day 2)
End-to-end orchestration — combining agents + flows across business domains (Day 3 morning)

The workshop is where you put it together yourself: design and build a complete, working workflow for a real business process — and explain its value to the business.

Think of this as moving from following recipes to cooking your own dish with the techniques you've practised.

2. The four business domains

Your capstone targets one (or more) of these domains. Each maps directly to labs you've already built — so you're never starting from a blank page; you're adapting and combining work you understand.

Domain	Typical workflow	Built on
Sales	Enquiry/lead capture → log → acknowledge → notify rep	Labs 9, 12
Finance	Invoice/expense submitted → approve by amount → notify	Labs 3, 13
Procurement	Purchase request → manager approval → notify + log	Labs 10, 14
Order Processing	Order captured → confirm → log → restock alert	Lab 15

Pick the domain you understand best. A workflow you can explain confidently is worth more than an ambitious one you can't finish.

3. Design method (recap from Module 4)

Run through this before you build — it's the same discipline from Module 4:

One sentence: "When ___ happens, do ___, then ___, and finally ___."
Trigger: manual / scheduled / form / email / file / agent call
Capture: the fields/variables you need
Actions in order: log, generate, approve, notify…
Decision points: the condition(s) and each branch
End states: final status + who is notified for each outcome

If you can write step 1 cleanly, the rest of the design tends to fall into place.

4. Quality bar

A strong capstone workflow:

✅ Has a clear trigger and runs end to end
✅ Captures or receives data and logs it (Excel)
✅ Includes at least one condition (branching)
✅ Includes an approval or an AI-generated response
✅ Notifies the right people for every outcome
✅ Is tested on the happy path and every branch

Treat this as your self-check list. Before you call the capstone "done," walk down it and tick each item.

5. Tips for success

Start small, then add. Get a basic happy-path flow working end to end first, then layer on conditions, approvals, and notifications.
Test after each step — don't build everything and then test once. Small tests catch problems while they're still easy to find.
Reuse your earlier labs as templates — see the Capstone's reuse list. Adapting working flows is faster and safer than rebuilding from scratch.
Check connections are green before running — a red connector or an Unauthorized error means: reconnect first.
Remember the approver rule — approvals must go to a real user in your tenant (yourself is fine for testing), never an outside personal email.
Use a Status column in Excel as the single source of truth, updated at each stage.
Keep agent output structured — clean variables in means clean flow inputs out.

One sentence, built carefully, tested fully. That's a winning capstone. Good luck.

Next: Lab 16: Capstone Workshop

Lab 12: Email Enquiry → Excel Logging → Notification

Lab Title

End-to-End Workflow: Capture an Email Enquiry, Log It, and Notify the Team

Lab Objectives

By the end of this lab, you will be able to:

Create an automated cloud flow that starts on its own when an email arrives
Configure the When a new email arrives (V3) trigger with an optional subject filter
Log each enquiry as a new row in an Excel table, with a clean timestamp
Notify the sales team automatically with the enquiry details
Orchestrate the classic Capture → Log → Notify pattern into one hands-off workflow

Prerequisites

Completed Day 1 Labs 1–2 (email + Excel actions)
An Excel Enquiry Log with table EnquiryTable (from Lab 2) — or create one (Date, Name, Email, Message, Status)

Scenario

ACME Pte Ltd receives customer enquiries by email to a shared inbox. Right now, staff copy each enquiry into a spreadsheet by hand and forget to tell the sales team. In this lab you will build an automated flow in the NUS Copilot Sandbox environment that, the moment an enquiry email arrives, logs it to Excel and emails the sales team — completely hands-off. This is the classic Capture → Log → Notify pattern.

Step-by-Step Guide

Step 1: Create an automated cloud flow (~5 minutes)

In your browser, go to https://make.powerautomate.com and sign in.
Top-right, confirm the environment selector shows NUS Copilot Sandbox. If it shows a different environment, click it and switch — every tool in this course uses the same environment.
In the left menu, select Create.
Under Start from blank, select Automated cloud flow.
In the Build an automated cloud flow dialog: — Flow name: Lab 12 - Email Enquiry to Excel and Notify — In Choose your flow's trigger, type when a new email arrives in the search box. — Select When a new email arrives (V3) under Office 365 Outlook.
Select Create.

Tip: If this is your first Office 365 / Excel action, Power Automate may ask you to sign in to create a connection. Always sign in with your own course account. A working connection shows a green ✓ — you will rely on this later.

Step 2: Configure the email trigger (~10 minutes)

The flow opens in the new designer with the trigger card already placed. Select the When a new email arrives (V3) card to open its panel on the right.
Set: — Folder: Inbox (use the folder picker — do not type it) — Importance: Any — Only with Attachments: No — Include Attachments: No
Select Show advanced options (or Show all) to reveal more fields, then set: — Subject Filter: Enquiry — This means only emails whose subject contains the word Enquiry will start the flow — which keeps your log clean while testing.

Tip: Without a subject filter, every incoming email starts the flow. During training, set a filter so unrelated mail (newsletters, replies) does not flood your Excel log.

Step 3: Log the enquiry to Excel (~12 minutes)

Under the trigger card, select the + (plus) → Add an action.
Search add a row into a table and select Add a row into a table under Excel Online (Business).
In the action panel set: — Location: OneDrive for Business — Document Library: OneDrive — File: browse to and select Enquiry Log.xlsx — Table: select EnquiryTable from the dropdown
The five table columns (Date, Name, Email, Message, Status) now appear. Map them: — Name: click the field, then from Dynamic content insert From (or From display name if you prefer) — Email: insert Dynamic content → From — Message: insert Dynamic content → Body preview — Status: type New
For the Date column, enter an expression so the timestamp is clean: — Click the Date field, then open the fx (Expression) editor. — Type exactly: formatDateTime(utcNow(),'yyyy-MM-dd HH:mm') — Select Add (or OK). The field now shows a blue fx token, not the raw text.

⚠️ Warning: Enter the date through the fx editor so it becomes a token. If you type formatDateTime(...) directly into the box as plain text, Excel stores the literal text and you get a meaningless string instead of a date.

Step 4: Notify the sales team (~10 minutes)

Under the Excel action, select + → Add an action.
Search send an email and select Send an email (V2) under Office 365 Outlook.
Configure: — To: the sales team address — for testing, use your own email — Subject: type New enquiry received: then insert Dynamic content → Subject — Body: type the message below, inserting the matching Dynamic content tokens where shown in braces:

     A new enquiry has arrived and been logged.

     From: [From]
     Subject: [Subject]
     Message: [Body preview]

     Status: New — please follow up.

Select Save (top right).

⚠️ Warning: If saving or testing later shows "Unauthorized" on the Send an email action, your Office 365 Outlook connection has expired or used the wrong account. Open the action's … menu → My connections (or the connection link at the bottom of the panel) and reconnect with a mailbox-enabled account. Every connection must show a green ✓ before the flow will run.

Step 5: Test the end-to-end workflow (~10 minutes)

Top right, select Test → Manually → Test.
From another account (or the same one), send an email to your inbox with: — Subject: Enquiry - bulk order — Body: a short sentence, e.g. Hi, I'd like a quote for 200 units.
Within about a minute the flow should trigger. Confirm: — Every step shows a green ✓ in the run view. — A new row appears in Enquiry Log.xlsx with the sender, message, and a readable date/time. — The notification email arrives with the enquiry details.
Now send a second test email without the word "Enquiry" in the subject (e.g. subject Hello). Confirm the flow does not run — proving the subject filter works.

Tip: If the new Excel row shows ######## in the Date column, the column is just too narrow. Double-click the column border in Excel to auto-fit — the value is correct.

Step 6: Review and harden (~3 minutes)

In the left menu, open My flows → Lab 12 - Email Enquiry to Excel and Notify → the latest run to inspect inputs and outputs of each step.
Discuss optional improvements: — Add a Condition to flag high-priority enquiries (e.g. subject contains "urgent"). — Capture the sender's display name in a separate column.

Checkpoint

✅ Automated flow triggered by an incoming email
✅ Each matching email → new Excel row with a clean timestamp
✅ Each matching email → team notification email
✅ Non-matching emails are correctly ignored by the subject filter

Troubleshooting

Problem	Solution
Flow doesn't trigger	Confirm the email reached the Inbox and the subject contains `Enquiry`; triggers can take up to a minute.
"Unauthorized" on Send an email	Reconnect the Office 365 Outlook connection with a mailbox-enabled account; every connection must be green ✓.
Date shows raw text like `formatDateTime(...)`	You typed it as text. Re-enter it via the fx editor so it becomes a blue token.
Date column shows `########`	The column is too narrow — double-click the column border in Excel to auto-fit.
Empty Excel fields	Map to the trigger's dynamic content (From / Subject / Body preview), not static text.
Too many rows logged	Add or tighten the Subject Filter on the trigger.
No notification email	Check Junk; verify the To address; check run history for errors.

Key Takeaways

Automated triggers (an email arrives) make workflows run with zero clicks.
One trigger can fan out into multiple actions — log and notify in a single flow.
Use the fx editor for dates so Excel stores a real, readable timestamp.
Subject filters keep automation focused and your data clean.

Duration

~40 minutes

Next Steps

Proceed to Lab 13: Invoice Upload → Approval Workflow.

Lab 13: Invoice Upload → Approval Workflow

Lab Title

End-to-End Workflow: Trigger an Approval When an Invoice File Is Uploaded

Lab Objectives

By the end of this lab, you will be able to:

Use a file upload trigger — When a file is created in OneDrive
Pass file details into a Start and wait for an approval action
Assign the approval to a real tenant user so the run succeeds
Branch on the approval outcome with a Condition
Move the file and notify by outcome, completing the Upload → Approve → Act pattern

Prerequisites

Completed Day 1 Lab 3 (approvals + conditions)
Access to OneDrive (to create folders)

Scenario

At ACME Pte Ltd, staff drop supplier invoices into a shared OneDrive folder, but they sit unseen until someone remembers to check. In this lab you will build an automated flow in the NUS Copilot Sandbox environment so that when an invoice lands in Invoices/Incoming, finance is asked to approve or reject it. Approved invoices move to an Approved folder and rejected ones to Rejected, with an email sent either way. This is the Upload → Approve → Act pattern.

Step-by-Step Guide

Step 1: Prepare the OneDrive folders (~5 minutes)

Go to https://www.office.com, sign in, and open OneDrive.
Select + Add new → Folder and create a folder named Invoices.
Open the Invoices folder and create three subfolders inside it: — Incoming — Approved — Rejected
Leave Incoming empty for now — you will upload a test file later.

Tip: Create all four folders before building the flow. The flow's triggers and Move file actions let you browse to these folders, which is far safer than typing paths by hand.

Step 2: Create the automated flow with a file trigger (~8 minutes)

Go to https://make.powerautomate.com and confirm the environment selector reads NUS Copilot Sandbox.
In the left menu, select Create → Automated cloud flow.
In the dialog: — Flow name: Lab 13 - Invoice Upload Approval — Search when a file is created and select When a file is created under OneDrive for Business.
Select Create.
Select the trigger card to open its panel, then set: — Folder: use the folder picker (the small folder icon) and browse to Invoices/Incoming. Do not type the path.

Tip: The trigger fires whenever a new file appears in that exact folder — that is your "upload" event. It does not fire for files already there before the flow was turned on.

Step 3: Start the approval (~10 minutes)

Under the trigger, select + → Add an action.
Search start and wait for an approval and select Start and wait for an approval under Approvals.
Configure: — Approval type: Approve/Reject - First to respond — Title: type Invoice approval needed: then insert Dynamic content → File name — Assigned to: start typing your own name or course email, then pick the matching person from the people-picker dropdown so it resolves to a person chip (a rounded tag with the name). — Details: type Please review the uploaded invoice: then insert Dynamic content → File name.

⚠️ Warning: The Assigned to approver MUST be a user that already exists in this tenant. Pick the person from the dropdown so it becomes a person chip — do not type a free-text external email. If you paste an outside address, the run fails with "The approvers/assigned to must contain valid users in the organization."

Step 4: Branch on the outcome (~12 minutes)

Under the approval action, select + → Add an action, search condition, and select Condition (Control).
Configure the Condition: — Left value: insert Dynamic content → Outcome — Operator: is equal to — Right value: type Approve exactly (capital A, no quotes)

In the True branch (Approved):

Select + → Add an action → Move file (OneDrive for Business): — File to Move: insert Dynamic content → Id (from the trigger) — Destination Folder: use the picker to select Invoices/Approved — If another file is already there: Replace
Select + → Add an action → Send an email (V2) (Office 365 Outlook): — To: your own email (for testing) — Subject: type Invoice APPROVED: then insert Dynamic content → File name — Body: The invoice has been approved and moved to the Approved folder for payment processing.

In the False branch (Rejected):

Select + → Add an action → Move file: — File to Move: Dynamic content → Id — Destination Folder: Invoices/Rejected — If another file is already there: Replace
Select + → Add an action → Send an email (V2): — To: your own email — Subject: type Invoice REJECTED: then insert Dynamic content → File name — Body: The invoice was rejected. Please review and resubmit if necessary.
Select Save.

⚠️ Warning: Insert File name and Id as single-value dynamic content. If the designer ever wraps an action in an automatic "Apply to each" loop, you accidentally selected an array/list output. Delete the For each, then re-add the plain Move file / Send an email action and pick the single-value tokens (File name, Id). The Outcome and trigger file fields are all single-value.

⚠️ Warning: If Send an email shows "Unauthorized", reconnect the Office 365 Outlook connection with a mailbox-enabled account. All connections must be green ✓ before the flow can run.

Step 5: Test both paths (~10 minutes)

Top right, select Test → Manually → Test.
In OneDrive, upload a sample PDF (any file) into Invoices/Incoming.
The flow triggers and pauses at the approval. Respond in one of these ways: — Open the Approvals hub from the left menu of make.powerautomate.com (or Power Automate), open the request, and choose Approve. — Or respond from the approval email / Teams card.
Confirm the True path: — The file moved from Incoming to Invoices/Approved. — You received the APPROVED email.
Repeat: upload another file, let the flow pause, and this time choose Reject. Confirm it moves to Invoices/Rejected and you get the REJECTED email.

Tip: Approval emails can be slow or land in Junk. The most reliable place to respond is the Approvals hub in the left menu — it always shows pending requests assigned to you.

Step 6: Review (~5 minutes)

Open My flows → Lab 13 - Invoice Upload Approval → the latest runs and confirm the correct branch executed each time.
Optional enhancement: at the start of the flow, add a row to an Invoice Register Excel table with Status Pending, then update it after the decision for a full audit trail.

Checkpoint

✅ Flow triggered by a file upload to Invoices/Incoming
✅ Approval requested, assigned to a real tenant user, and awaited
✅ Approved → file in Approved folder + approval email
✅ Rejected → file in Rejected folder + rejection email

Troubleshooting

Problem	Solution
Run fails: "valid users in the organization"	Assigned to must be a real tenant user picked from the people-picker dropdown (a person chip) — not a typed external email.
Trigger doesn't fire	Confirm the file went into the exact `Invoices/Incoming` folder; allow up to a minute; the file must be new.
An "Apply to each" wrapped my action	You selected an array output. Delete the For each and re-add the plain action using single-value tokens (Id, File name, Outcome).
Move file fails	Use the trigger's Id as File to Move; make sure the destination folders already exist.
Wrong branch runs	The Condition right value must equal exactly `Approve`.
"Unauthorized" on Send an email	Reconnect the Office 365 Outlook connection with a mailbox-enabled account; connections must be green ✓.
No approval email	Respond in the Approvals hub; check Junk; emails can be slow.

Key Takeaways

A file-created trigger turns a simple upload into a full workflow.
Approvals only run when Assigned to is a real tenant user chosen from the dropdown.
Conditions route the file and notifications by the Outcome.
Moving files between status folders gives a visible, auditable process.

Duration

~45 minutes

Next Steps

Proceed to Lab 14: Purchase Request → Manager Approval → Notification.

Lab 14: Purchase Request → Manager Approval → Notification

Lab Title

End-to-End Workflow: Agent-Captured Purchase Request with Manager Approval and Notification

Lab Objectives

By the end of this lab, you will be able to:

Combine an agent (capture) with an agent flow (log + approve + notify) in one process
Log a request to Excel as Pending with a clean timestamp
Use a Condition (cost threshold) to decide auto-approve vs manager approval
Assign approvals to a real tenant user and notify the requester of every outcome
Build a complete Capture → Log → Approve → Notify business workflow

Prerequisites

Completed Day 2 Lab 8 (agent calling a flow) and Day 1 Lab 3 (approvals)
An Excel Procurement Log with table ProcurementTable (from Lab 10) — add a Status column if missing

Scenario

At ACME Pte Ltd, staff raise purchase requests through the Procurement Assistant agent built in Day 2 Lab 10. In this lab you will extend it so each request is logged to Excel, routed by cost threshold (small requests auto-approve; larger ones go to a manager), and the requester is automatically notified of the result. Everything runs in the NUS Copilot Sandbox environment — the full procurement loop, end to end.

Step-by-Step Guide

Step 1: Reuse the Procurement Assistant agent (~5 minutes)

Go to https://copilotstudio.microsoft.com, confirm the environment is NUS Copilot Sandbox, and open your Procurement Assistant agent (from Day 2 Lab 10).
Open its New Procurement Request topic and confirm it captures these variables: requester, requesterEmail, item, quantity, reason. — If requesterEmail is missing, add an Ask a question node: What is your email? → save the answer to a variable named requesterEmail. — Add an estCost variable: add an Ask a question node What is the estimated total cost (SGD)?, set the Identify type to Number, and save to estCost. — Confirm quantity is also a Number variable.

Tip: quantity and estCost must be Number variables, not Text. The Condition in Step 4 compares estCost to 500 numerically — a Text value can throw a comparison error.

Step 2: Build the agent flow (~8 minutes)

In the topic, after the questions, select + → Add a tool → New Agent flow.
The flow opens with a When an agent calls the flow trigger. Select it and add these inputs (matching name and type exactly): — Text requester — Text requesterEmail — Text item — Number quantity — Text reason — Number estCost

Step 3: Log the request as Pending (~7 minutes)

Under the trigger, select + → Add an action → Add a row into a table (Excel Online (Business)).
Set Location OneDrive for Business, Document Library OneDrive, File Procurement Log.xlsx, Table ProcurementTable.
Map the columns from the flow's inputs (under Dynamic content): — Requester: requester — Item: item — Quantity: quantity — Reason: reason — Status: type Pending — (If your table has a Cost column, map estCost.)
For the Date column, click the field, open the fx (Expression) editor, type exactly formatDateTime(utcNow(),'yyyy-MM-dd HH:mm'), and select Add. The field shows a blue fx token.

⚠️ Warning: Enter the date via the fx editor so it becomes a token. Typing the expression as plain text stores literal characters in Excel instead of a real timestamp. If Excel later shows ######## in the Date column, the column is simply too narrow — auto-fit it.

Step 4: Decide the approval path with a Condition (~12 minutes)

A common rule: small requests auto-approve; larger ones need a manager.

Under the Excel action, select + → Add an action → Condition (Control): — Left value: Dynamic content → estCost — Operator: is greater than — Right value: type 500

In the True branch (needs manager approval):

Select + → Add an action → Start and wait for an approval (Approvals): — Approval type: Approve/Reject - First to respond — Title: type Purchase approval: then insert item, type x, then insert quantity — Assigned to: type your own name or course email and pick the matching person from the people-picker dropdown so it resolves to a person chip. — Details: Requested by + requester + . Reason: + reason + . Est cost: SGD + estCost + .
Under the approval, add a nested Condition: — Left value: Dynamic content → Outcome — Operator: is equal to — Right value: type Approve — Nested True: add Send an email (V2) → To: requesterEmail, Subject: Purchase APPROVED: + item, Body: confirm the request is approved. — Nested False: add Send an email (V2) → To: requesterEmail, Subject: Purchase REJECTED: + item, Body: explain the request was rejected.

In the False branch (auto-approved, ≤ 500):

Select + → Add an action → Send an email (V2) → To: requesterEmail, Subject: Purchase AUTO-APPROVED: + item, Body: confirm the request is approved automatically.

⚠️ Warning: The Assigned to approver MUST be a real user in this tenant — pick from the people-picker dropdown so it becomes a person chip. A typed external email makes the run fail with "valid users in the organization."

⚠️ Warning: Insert requesterEmail, item, Outcome etc. as single-value dynamic content. If the designer wraps a Send an email in an "Apply to each" loop, you picked a list/array output — delete the For each and re-add the plain action using the single-value tokens (Outcome and the trigger inputs are all single-value).

Tip on status updates: to flip the row to Approved/Rejected later, use Excel Update a row with a key column (e.g. add an ID column set to a unique value when the row is created, then update by that key). Status updates are optional for this lab — focus on the approve + notify flow.

Step 5: Return a result and wire inputs (~10 minutes)

Under the branches, select + → Add an action → Respond to the agent. Add a Text output named result with value Your request has been submitted for processing.
Select Save, then Publish the flow, then Go back to agent.
In the topic's tool node, map every flow input to the matching agent variable: requester, requesterEmail, item, quantity, reason, estCost.
After the tool node, add a Send a message node showing {result} so the user sees the confirmation.
Select Save on the topic, then Publish the agent.

⚠️ Warning: If Send an email returns "Unauthorized", reconnect the Office 365 Outlook connection with a mailbox-enabled account. All connections must show a green ✓ before the flow can run.

Step 6: Test all paths (~8 minutes)

Open the Test pane in Copilot Studio. Run a request under the threshold: — item Notebook, quantity 5, reason Stationery, estCost 120 — Expect: a Pending row logged, an AUTO-APPROVED email, and the confirmation message in chat.
Run a request over the threshold and approve it: — item Laptop, quantity 2, reason New hires, estCost 3000 — Respond in the Approvals hub (left menu) → Approve. Expect an APPROVED email.
Run another over-threshold request and Reject it → expect a REJECTED email.
Open My flows → run history and the Procurement Log to confirm each path behaved correctly.

Tip: Approval emails can be slow or land in Junk. The Approvals hub in the left menu is the reliable place to approve or reject pending requests.

Checkpoint

✅ Agent captures the request and calls the flow with all inputs mapped
✅ Request logged to Excel as Pending with a clean timestamp
✅ Threshold Condition routes auto-approve vs manager approval
✅ Approvals assigned to a real tenant user; requester notified of every outcome

Troubleshooting

Problem	Solution
Run fails: "valid users in the organization"	Assigned to must be a real tenant user picked from the people-picker dropdown (a person chip) — not a typed external email.
Condition errors on estCost	Ensure `estCost` (and `quantity`) are Number variables/inputs and you enter digits only.
An "Apply to each" wrapped my email action	You picked a list/array output. Delete the For each and re-add the plain action using single-value tokens (Outcome, requesterEmail, etc.).
"Unauthorized" on Send an email	Reconnect the Office 365 Outlook connection with a mailbox-enabled account; connections must be green ✓.
Date shows raw text / `########`	Re-enter the date via the fx editor so it is a token; widen the Excel column to fix `########`.
Approval never resumes	Respond in the Approvals hub; check Junk; emails can be slow.
Inputs empty in the flow	Re-map each agent variable to the matching flow input in the tool node.

Key Takeaways

This is the full loop: capture (agent) → log → approve → notify.
Approvals only run when Assigned to is a real tenant user picked from the dropdown.
Conditions implement real business rules such as value thresholds.
Keep single-value tokens out of accidental For each loops, and use the fx editor for dates.

Duration

~50 minutes

Next Steps

Proceed to Lab 15: Order Processing Workflow.

Lab 15: Order Processing Workflow (Agent + Flow)

Lab Title

End-to-End Order Processing — Agent Captures the Order, Flow Logs It, Confirms It, and Alerts the Warehouse

Lab Objectives

By the end of this lab, you will be able to:

Build a Copilot Studio Order Assistant agent that captures a customer order as structured data (name, email, product, quantity, address)
Connect that agent to a Power Automate agent flow using When an agent calls the flow and Respond to the agent
Log each order into an Excel table (Order Log workbook → OrderTable) with an automatic timestamp and a Received status
Use an AI prompt to draft a friendly order-confirmation email, then Send an email to the customer
Use a Condition so that large orders (quantity over 100) trigger a restock alert email to the warehouse — and small orders do not

Prerequisites

Completed Day 2 Labs 6–11 (create an agent, capture variables, connect an agent to a flow, AI prompts)
Completed Day 1 Labs 1–2 (email + Excel actions)
Environment NUS Copilot Sandbox with a working Outlook connection

Tip: This lab reuses the exact same skills as the Procurement workflow in Lab 14 — you are simply applying the "capture → log → generate → notify → branch" pattern to a brand-new business domain: Order Processing.

Scenario

You work for ACME Pte Ltd. A customer places an order by chatting with an agent. The moment the order is captured, the workflow should do four things automatically:

Record the order in a shared Excel order register.
Confirm the order to the customer by email.
Check the quantity — if it is a large order, alert the warehouse to restock.
Tell the agent it is done, so the agent can thank the customer.

You will build this as a Copilot Studio agent (the part the customer talks to) plus a Power Automate agent flow (the part that does the back-office work).

Step-by-Step Guide

Step 1: Prepare the Order Log Excel table (~5 min)

The flow needs somewhere to write each order. We use an Excel Table (not just a sheet) because Power Automate can only add rows to a named table.

Go to OneDrive (office.com → OneDrive) and create a new Excel workbook named exactly Order Log.
In row 1, type these seven column headers, one per cell, starting in cell A1:

Select the header cells A1:G1, then go to the Insert ribbon → click Table. In the dialog, make sure My table has headers is ticked, then click OK.
With the table selected, open the Table Design ribbon (top of the screen). In the Table Name box on the left, type OrderTable and press Enter.
Close Excel. Your changes save automatically to OneDrive.

Tip: Later, if a date cell shows ########, the column is just too narrow — widen it by double-clicking the border between the column letters. The data is fine.

Step 2: Create the Order Assistant agent (~10 min)

Open Copilot Studio (copilotstudio.microsoft.com) and confirm the environment picker (top-right) shows NUS Copilot Sandbox. The agent and the flow must live in the same environment.
Click Create (left menu) → New agent → at the top of the setup screen click Skip to configure.
Fill in: — Name: Order Assistant — Instructions:

     You are an Order Assistant for ACME Pte Ltd.
     Collect a customer's order: customer name, email, product, quantity, and delivery address.
     Collect one item at a time, be friendly and concise, then confirm the order has been placed.

Click Create.
In the left menu open Topics → + Add a topic → From blank. On the toolbar select Details and set the Name to New Order.
Set the topic trigger (latest Copilot Studio): — Generative orchestration (default): the Trigger node reads The agent chooses. In Details, set the Description to Use this topic when a customer wants to place an order or buy a product. It collects the customer's name, email, product, quantity, and delivery address. No phrases needed. — (Optional) exact phrases: hover the Trigger node → Change trigger → User says a phrase, then add I want to place an order, new order, I'd like to buy, order products (one per line).
Below the trigger, select the Add node icon (+) → Ask a question five times to add five questions. For each one, type the message, set what to Identify, and save the answer to a variable:

Question message	Identify	Save answer as variable
`What is your name?`	User's entire response	`customerName`
`What is your email?`	User's entire response	`email`
`Which product would you like to order?`	User's entire response	`product`
`How many units?`	Number	`quantity`
`What is the delivery address?`	User's entire response	`address`

⚠️ Warning: For the quantity question you must change Identify to Number. If it stays as text, the "greater than 100" comparison in Step 6 will not work correctly.

Click Save (top-right).

Step 3: Start the agent flow (~10 min)

Still inside the New Order topic, click the + below your last question → Add a tool → New Agent flow. This opens the Power Automate flow designer in a new tab.
The first (trigger) node is already When an agent calls the flow. Click it to open it.
Click + Add an input once for each field below. Choose the type, then type the exact name: — Text → customerName — Text → email — Text → product — Number → quantity — Text → address

Tip: These input names are how the agent will hand its variables to the flow. Spell them exactly as shown — you will match them up again in Step 7.

Step 4: Log the order to Excel (~5 min)

Click + New step → search Excel Online (Business) → choose the action Add a row into a table.
Fill in the action's first three boxes from the dropdowns: — Location: OneDrive for Business — Document Library: OneDrive — File: browse to Order Log.xlsx — Table: OrderTable
The seven table columns now appear as fields. Map them: — Date: click inside the box → click the fx (function) icon to open the formula editor → type exactly:

     formatDateTime(utcNow(),'yyyy-MM-dd HH:mm')

then click Add / OK. The formula collapses into a single token. Never type this expression directly into the field — always enter it through the fx editor, or it will be saved as plain text.

Customer: dynamic content customerName
Email: dynamic content email
Product: dynamic content product
Quantity: dynamic content quantity
Address: dynamic content address
Status: type the literal word Received

Step 5: Draft and send the order confirmation (~10 min)

(Recommended) Add an AI step: + New step → search AI prompt → choose Create text with GPT using a prompt (sometimes shown as Run a prompt). In the prompt box, type:

   Write a short, friendly order confirmation email (max 100 words).
   Customer: {customerName}; Product: {product}; Quantity: {quantity}; Delivery to: {address}.
   Confirm the order is received and will be processed. Sign off as "The ACME Order Team".
   Output only the email body.

(Insert each {...} value by picking it from the dynamic content list rather than typing it.)

+ New step → search Office 365 Outlook → choose Send an email (V2). Fill in: — To: dynamic content email — Subject: type Order confirmation: then add dynamic content product — Body: add the AI prompt output (the generated text from step 1). If you skipped the AI step, type your own short message using the dynamic content tokens.

⚠️ Warning: If the Send an email action shows Unauthorized (a red error), the Outlook connection is signed in with the wrong account. Click the action's … menu → + Add new connection, then sign in with a real mailbox-enabled tenant account. Every connection on the action should show a green check mark ✓ before you continue. (Same fix as Lab 1.)

Step 6: Add the restock-alert Condition (~10 min)

Large orders should warn the warehouse; small orders should not.

+ New step → search Condition → choose Condition (under Control).
Build the test: — Left box: dynamic content quantity — Operator: is greater than — Right box: type 100
In the If yes branch (this runs only for large orders): — Click Add an action → Send an email (V2). — To: the warehouse address — for testing, use your own email so you can see it arrive. — Subject: type RESTOCK ALERT: then add dynamic content product. — Body: type a message and drop in the tokens, e.g. — Large order received: quantity units of product for customerName . Please check stock and restock if needed.
Leave the If no branch empty — small orders need no alert.

Tip: If Power Automate ever wraps an action in an automatic For each loop, it means you selected a field that can return multiple values. Here every field (quantity, product, etc.) is a single value, so no loop should appear. If one does, delete the action and re-add the dynamic content carefully.

Step 7: Return a result and wire the inputs (~10 min)

+ New step → search Respond to the agent → add it. Add a Text output named result with the value:

   Your order has been placed and confirmed by email.

Click Save (and Publish if prompted), then close the flow tab and return to the agent (Go back to agent / the Copilot Studio tab).
In the New Order topic, click your flow's tool node. You will see the five flow inputs. Map each one to the matching agent variable from the dropdown: — customerName → customerName — email → email — product → product — quantity → quantity — address → address
Below the tool node, click + → Send a message and type:

   {result} Thank you for your order!

(Insert result from the flow's outputs.)

Click Save.

Step 8: Test end-to-end (~10 min)

Open the Test pane (top-right). Click the refresh icon so it picks up your latest changes.
Type new order and answer the questions for a large order: — Name: Mei Ling — Email: your own email address — Product: Air Fryer Pro — Quantity: 150 — Address: 123 Orchard Rd
Confirm all four results: — The agent shows your "Thank you for your order!" message. — A new row appears in Order Log.xlsx with Status = Received and a timestamp in the Date column. — The order confirmation email arrives in your inbox. — Because 150 is greater than 100, a RESTOCK ALERT email also arrives.
Run it once more with a small quantity (e.g. 5) and confirm: — The order is still logged and confirmed. — No restock alert email is sent.

Checkpoint

You are done when:

✅ The Order Assistant captures all five order fields (and quantity is a Number)
✅ The agent flow logs the order, confirms by email, and conditionally alerts the warehouse
✅ A large order (150) gets a restock alert; a small order (5) does not
✅ A new row with Status = Received and a real timestamp appears in OrderTable

Troubleshooting

Problem	Solution
Flow does not appear in the topic	Refresh the page; make sure the flow is Published and in the same environment (NUS Copilot Sandbox).
Flow inputs arrive empty	In the tool node, re-map each agent variable to the matching flow input (Step 7).
Restock alert never fires (or always fires)	The Condition must compare `quantity` is greater than `100`, and `quantity` must be identified as a Number in Step 2.
Send an email shows Unauthorized	Reconnect Outlook with a mailbox-enabled account; confirm every connection shows a green ✓.
Date column shows the literal text `formatDateTime(...)`	You typed the expression instead of entering it through the fx editor. Re-enter it via fx.
Date cell shows `########`	The column is too narrow. Widen it — the data is correct.
An action got wrapped in For each	You selected a multi-value field. Remove the action and re-add single-value dynamic content.

Key Takeaways

Order Processing is a complete capture (agent) → log → confirm → alert loop, end to end.
A Condition on quantity turns a business rule (the restock threshold) into automatic behaviour.
The timestamp comes from an fx expression (formatDateTime(utcNow(),'yyyy-MM-dd HH:mm')) entered through the formula editor, never typed as text.
This is the same pattern as Procurement (Lab 14) applied to a new domain — proof that one set of skills covers many business workflows.

Duration

~55 minutes

Next Steps

Proceed to the Lab 16: Capstone Workshop to design and build your own end-to-end workflow.

Lab 16: Capstone Workshop — Build Your Own End-to-End Workflow

Lab Title

Business Workflow Workshop: Design and Build a Complete Automation

Lab Objectives

By the end of this workshop, you will be able to:

Design an end-to-end workflow for a real business scenario on paper before you build
Build it by combining a Copilot Studio agent with one or more Power Automate flows
Orchestrate triggers, actions, conditions, approvals, and notifications into one working process
Test the happy path and every branch using a structured test log
Present your workflow and explain its business value

Prerequisites

Completed all Day 1–3 labs (you will reuse skills from across the whole course)
Your environment NUS Copilot Sandbox set up, with working Outlook and Excel connections (all green ✓)

Tip: This is the last-afternoon capstone. You are not learning a new technique here — you are combining the pieces you already practised into one workflow that is yours. Pick a scenario you genuinely care about and the rest gets easier.

Scenario

You work for ACME Pte Ltd. Choose one real business process from your day-to-day work (or one of the four tracks below), then design and build a complete automation for it: something triggers it, data is captured and logged, a business rule decides what happens, the right people are notified, and the process ends in a clear final state.

You will work individually or in a small group and finish with a working, tested workflow plus a short demo.

Step-by-Step Guide

Step 1: Choose a track (~10 min)

Pick one scenario below (or adapt one to your own job). Each maps directly to skills you have already practised.

Track A — Sales

Lead-to-CRM workflow. A customer submits an enquiry (via the Sales Enquiry Assistant or an email). The workflow logs the lead, generates a personalised acknowledgement with an AI prompt, and notifies the assigned salesperson. High-value leads (quantity or deal size over a threshold) also raise a manager alert.

Track B — Finance

Expense / invoice approval workflow. An expense or invoice is submitted (agent form or file upload). It is logged, routed for approval based on amount, and the requester is notified. Approved items move to a "to-pay" status; rejected items are returned with a reason.

Track C — Procurement

Purchase-request workflow. Staff raise a purchase request via an agent. It is logged, routed for manager approval when it exceeds a cost threshold, and the requester is notified. The procurement team gets a consolidated notification for approved items.

Track D — Order Processing

Order intake workflow. A customer places an order (an agent captures product, quantity, address). The order is logged, an order-confirmation email is generated with an AI prompt, large orders trigger a restock alert to the warehouse, and the customer is notified of the order status.

Step 2: Design on paper first (~20 min)

Use the orchestration method from Module 4. Complete this design sheet before you build anything — a clear plan makes the build fast and the testing obvious.

Design item	Your answer
One-sentence summary ("When ___, do ___, then ___, finally ___")
Trigger (email / form / file upload / agent call / schedule)
Data to capture (the fields / variables, and their types)
Actions in order (log, generate, approve, notify…)
Decision point(s) (the condition and the two paths)
End states (final status + who is notified for each outcome)
Tools used (agent? which flows? Excel? Outlook? Approvals?)

⚠️ Warning: Get this design sheet checked by a facilitator or peer before building. Fixing a plan on paper takes a minute; fixing a half-built flow takes much longer.

Step 3: Build it incrementally (~60–90 min)

Build one piece at a time and test after each piece — never build everything and test once at the end.

Suggested build order:

Data store — create the Excel log as a named Table (include any status / key columns your end states need).
Capture — build the agent topic (Ask a question nodes → variables) or configure the automated trigger (email / file). Remember to set numeric fields to Number.
Log — add Add a row into a table; test that one record is written correctly. Use the fx editor for any timestamp: formatDateTime(utcNow(),'yyyy-MM-dd HH:mm') (enter it via fx, never type the expression into the field).
Generate (if used) — add an AI prompt action; test that the generated text reads well.
Decide — add Condition(s) for your business rule; test both branches (just above and just below the threshold).
Approve (if used) — add Start and wait for an approval; test approve and reject.
Notify — add Send an email actions for each outcome; test that each one is delivered.
Connect — if using an agent, add When an agent calls the flow as the trigger and Respond to the agent at the end, then wire the agent variables → flow inputs and show the returned confirmation message.

⚠️ Warning — green connections: Before testing any Send an email or Excel action, confirm every connection shows a green check mark ✓. An Unauthorized error on Send an email means the Outlook connection is the wrong account — reconnect with a real mailbox-enabled tenant account.

⚠️ Warning — approvals: In Start and wait for an approval, set Assigned to by picking a real tenant user from the dropdown. Do not type an external email address — external approvers will never receive the request and the flow will appear to hang.

Tip: If an action gets wrapped in an automatic For each loop, you picked a multi-value field. Most of your fields are single values — remove the action and re-add the dynamic content carefully.

Reuse your earlier labs as templates:

Email sending → Lab 1
Excel logging → Lab 2
Approval + condition → Lab 3
Scheduled / recurring trigger → Lab 4
Form submission trigger → Lab 5
Agent fundamentals & instructions → Lab 6
Knowledge / RAG grounding → Lab 7
Tools & actions → Lab 8
Agent capture (structured data) → Lab 9
Agent → flow integration → Lab 10
AI-generated response → Lab 11
Email/file trigger end-to-end → Labs 12–13
Full approval loop → Lab 14
Agent-driven order processing → Lab 15

Step 4: Test thoroughly (~20 min)

Complete this test log. Your workflow is not done until every relevant row passes. (Skip the approval rows only if your track has no approval step.)

Test case	Expected result	Pass?
Happy path (typical input)	Logged + notified + correct end state
Approval → Approved	Approved branch runs + correct email sent
Approval → Rejected	Rejected branch runs + correct email sent
Threshold / edge case (just above & just below)	Correct branch chosen each time
Bad / missing input	Agent re-asks; flow does not crash
Run history	Every step green; logged data correct

Tip: Open the flow's Run history (Power Automate → your flow → 28-day run history) to see exactly which step failed and read the error. Green = success, red = failure.

Step 5: Present your workflow (~5 min each)

Demo to the group. Cover:

The business problem and who it helps.
The trigger and the steps — walk through your one-sentence design summary.
A live run of the happy path and at least one branch.
The business value — time saved, errors avoided, consistency gained.
What you would add next with more time.

Assessment checklist

Your capstone is complete when:

✅ It has a clear trigger and runs end to end
✅ It captures or receives data and logs it to an Excel Table
✅ It includes at least one Condition (branching on a business rule)
✅ It includes an approval or an AI-generated response
✅ It notifies the right people for every outcome
✅ Every relevant test case in Step 4 passes
✅ All connections are green ✓ and any approver is a real tenant user
✅ You can explain its business value in plain language

Stretch goals (optional)

Add a scheduled flow that emails a daily summary of new records.
Use Update a row to keep the Excel Status column current at each stage (e.g. Received → Approved → Completed).
Deploy the agent to Microsoft Teams so colleagues can use it where they already work.
Add a second approver or an escalation if no response within a time limit.
Post outcomes to a Teams channel instead of (or as well as) email.

Duration

~2–3 hours

Congratulations 🎉

You have completed the Microsoft Copilot Studio & Power Automate for Business Workflow Automation course. You can now design and build end-to-end automations that capture requests, apply business rules, route approvals, and notify stakeholders — across Sales, Finance, Procurement, and Order Processing.

Where to go next:

Explore the Power Automate Templates Gallery for ready-made patterns.
Review Copilot Studio documentation for advanced agent features.
Identify one manual process at your workplace and automate it this week.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Learner Guide

Document Version Control Record

Table of Contents

Common Errors & Quick Fixes

Day 1 — Foundations & Power Automate

Module 1: Introduction to Workflow Automation

Module 2: Introduction to Power Automate

Lab 0: Environment Setup — Create Your Copilot Studio & Power Automate Accounts

Lab 1: Automated Email Workflow

Lab 2: Excel Data Logging Workflow

Lab 3: Simple Approval Workflow

Lab 4: Scheduled Trigger Workflow

Lab 5: Form Submission Workflow

Day 2 — Building Business Agents with Copilot Studio

Module 3: Building Business Agents with Copilot Studio

Lab 6: Create Your First Copilot Studio Agent

Lab 7: Add Knowledge to Your Agent (Grounding / RAG)

Lab 8: Add Tools and Actions to Your Agent

Lab 9: Sales Enquiry Assistant

Lab 10: Procurement Request Workflow

Lab 11: Automated Response Generation

Day 3 — End-to-End Workflow Automation & Workshop

Module 4: End-to-End Workflow Automation

Module 5: Business Workflow Workshop

Lab 12: Email Enquiry → Excel Logging → Notification

Lab 13: Invoice Upload → Approval Workflow

Lab 14: Purchase Request → Manager Approval → Notification

Lab 15: Order Processing Workflow (Agent + Flow)

Lab 16: Capstone Workshop — Build Your Own End-to-End Workflow

Uh oh!

FilesExpand file tree

LEARNER-GUIDE.md

Latest commit

History

LEARNER-GUIDE.md

File metadata and controls

Learner Guide

Document Version Control Record

Table of Contents

Common Errors & Quick Fixes

Day 1 — Foundations & Power Automate

Module 1: Introduction to Workflow Automation

Module 2: Introduction to Power Automate

Lab 0: Environment Setup — Create Your Copilot Studio & Power Automate Accounts

Lab 1: Automated Email Workflow

Lab 2: Excel Data Logging Workflow

Lab 3: Simple Approval Workflow

Lab 4: Scheduled Trigger Workflow

Lab 5: Form Submission Workflow

Day 2 — Building Business Agents with Copilot Studio

Module 3: Building Business Agents with Copilot Studio

Lab 6: Create Your First Copilot Studio Agent

Lab 7: Add Knowledge to Your Agent (Grounding / RAG)

Lab 8: Add Tools and Actions to Your Agent

Lab 9: Sales Enquiry Assistant

Lab 10: Procurement Request Workflow

Lab 11: Automated Response Generation

Day 3 — End-to-End Workflow Automation & Workshop

Module 4: End-to-End Workflow Automation

Module 5: Business Workflow Workshop

Lab 12: Email Enquiry → Excel Logging → Notification

Lab 13: Invoice Upload → Approval Workflow

Lab 14: Purchase Request → Manager Approval → Notification

Lab 15: Order Processing Workflow (Agent + Flow)

Lab 16: Capstone Workshop — Build Your Own End-to-End Workflow