PepsiCo and ServiceNow Logistics Breakdown Resolution System

As a ServiceNow AI Systems Developer at PepsiCo, I built an automated supply chain incident processing system that enables the operations team to rapidly analyze financial impacts of truck breakdowns, select optimal rerouting options, and coordinate external execution with logistics providers and retail clients.

By combining ServiceNow AI Agents, n8n workflow automation, and MCP-enabled communications, the system forms a closed-loop orchestration layer that:

• reduces manual intervention
• ensures contractual compliance
• maintains timely and cost-effective delivery operations

System Overview

The Logisitcs Breakdown Resolution System developed within ServiceNow enables PepsiCo to intelligently manage truck breakdowns, delivery delays, and contractual performance for major retail clients like Whole Foods.

When logistics providers such as Schneider report delivery delays, the system automatically triggers a multi-agent workflow that performs financial impact analysis, route optimization, and external coordination all without manual intervention.

The system integrates:

• ServiceNow AI Agents for reasoning, impact calculation, and decision-making,
• n8n workflow automation for execution and data routing, and
• AWS Bedrock and MCP Clients for cross-system orchestration and secure external communications.

By using AI Agents inside ServiceNow to coordinate incidents and route decisions, then extending their actions to logistics and retail systems through n8n and MCP, I built a closed-loop automation layer for PepsiCo’s delivery operations.

Business Impact:

• Provides real-time visibility into delays and cost impacts.
• Reduces the response time between detection and resolution.
• Ensures customer satisfaction and contractual compliance by automatically rerouting deliveries and updating stakeholders.
• Establishes a reusable AI-driven framework for future incident automation across the PepsiCo logistics network.

Tools and Technologies Used

Tool / Technology	Purpose / Role	Business Value
ServiceNow AIA ReAct Engine	Hosts and executes AI Agents for impact analysis and routing decisions	Centralizes intelligent decision-making; reduces human input in logistics disruption handling
ServiceNow Scoped Application (x_snc_pepsico_de_0)	Provides custom data models (Delivery Delay, Supply Agreement) and automation triggers	Enables tailored workflows that reflect PepsiCo’s delivery and penalty logic
ServiceNow Tables: Delivery Delay & Supply Agreement	Store logistics incident data and contractual terms	Provide structured data for financial computation and incident linking
Route Financial Analysis Agent	Calculates delay costs and creates incident records	Automates financial analysis, ensuring consistent penalty computation
Route Decision Agent	Chooses optimal delivery routes and updates records	Improves routing efficiency and reduces penalty exposure
ServiceNow MCP Server & Clients	Facilitate secure communication between ServiceNow, logistics, and retail systems	Connects ServiceNow AI decisions to external execution systems (e.g., Schneider, Whole Foods)
n8n Workflow Automation	Orchestrates external system actions (e.g., notifications, logistics updates) via AI Agent and Bedrock	Streamlines multi-system execution with zero manual coordination
AWS Bedrock Chat Model	Provides reasoning and language understanding for AI agents within n8n	Enhances interpretability and contextual decision-making for agent workflows
Calculated Impacts Script	Handles mathematical calculations for ETA, overage, and penalties	Ensures numerical accuracy beyond AI’s reasoning capabilities
Incident Management Module	Creates and updates incidents tied to delivery disruptions	Provides traceability and SLA-driven management of delivery incidents
ServiceNow Workflow (Assigned_To Fix)	Auto-assigns records to maintain execution context	Prevents trigger failures due to permission conflicts
Webhooks (n8n Integration)	Transfers routing and execution data between systems	Enables real-time synchronization across AI and logistics systems

Architecture Diagram

Implementation Steps

Step 1: Application Setup

I created a scoped application in ServiceNow Studio with the exact name: PepsiCo Deliveries

This precise naming will auto-generate the scope: x_snc_pepsico_de_0

Step 2: Table Setup

Delivery Delay Table

The first table in my system is a Delivery Delay table. It holds the information about the various truck breakdowns reported from Schneider (Trucking Logistics Provider).

Delivery Delay Table Fields:

• route_id (Integer, Primary Key)
• truck_id (Integer)
• customer_id (Integer, Default: 1)
• problem_description (String, 4000)
• proposed_routes (String, 4000) - JSON format with route options
• calculated_impact (String, 4000) - JSON format with financial analysis
• chosen_option (String, 4000) - Selected route details
• status (String, 16) - Workflow progression: pending/calculated/approved/dispatched
• assigned_to (Reference to User) - Critical: Used for trigger execution context and permissions
• incident_sys_id (String, 32) - Links to associated incident records

Schneider's Breakdown Notification Agent sends over the data in the required payload structure using ServiceNow's MCP Server and populates our custom table automatically.

The assigned_to field in Delivery Delay table serves as the execution context for AI agent triggers. This ensures proper permissions and security boundaries when agents process records automatically.

In this system, our Schneider's Breakdown Notification Agent is unable to properly assign the assigned_to field due to security permissions.

However, I used a ServiceNow Workflow to remedy this issue by assigning new records on the table to System Administrator.

Supply Agreement Table

The second table in my system is a Supply Agreement table. It holds the contractual penalties for late shipments from Whole Foods (Retail Client).

Supply Agreement Table Fields:

• customer_id (Integer, Primary Key)

• customer_name (String, 100)

• deliver_window_hours (Integer) - Contractual delivery timeframe

• deliver_window_hours: The contractual timeframe (in hours) within which deliveries must be completed to avoid penalties. For Whole Foods, deliveries must be completed within 3 hours of departure to avoid charges.

• stockout_penalty_rate (Integer) - Cost per hour of delay in dollars

• stockout_penalty_rate: The financial penalty (in dollars) assessed for every hour a delivery exceeds the contractual delivery window. Whole Foods charges PepsiCo $250 for each hour beyond the 3-hour delivery window. For example, a 5-hour delivery would incur penalties for 2 hours (5 minus 3), resulting in a $500 penalty charge.

Step 3: Use Case and Trigger Setup

Logistics Breakdown Resolution Use Case

I set up my use case to utilize one data point across both agents for a consistent experience. I chose the route_id because it is a Primary Key in our Delivery Delay records.

Description

This use case detects delivery delays, then uses two agents to asses the financial impact and alternative route options.

Instructions (Base Plan)

Use the route id the user enters for both agents. Store the route id in permanently in memory as memory.route_id 

Step 1: Use the Route Financial Analysis Agent with memory.route_id 

Step 2: Use the Route Decision Agent with memory.route_id

The trigger is configured to run the use case as the assigned user for any created or updated delivery delay records with a Status of pending.

Step 4: AI Agent Setup

Agent 1: Route Financial Analysis Agent

Role

You analyze delivery delays, calculate the cost of alternate routes, create incident records, and update records with calculated impacts.

Description

Analyze financial impact of delivery disruptions and create incident tracking

Instructions

Step 1: Use the Lookup Delivery Delay Record Tool with the route id the user entered. Retrieve its eta_minutes values for each proposed route.

Step 2: Find the supply agreement record for the customer id on the delivery delay record. Retrieve the fields deliver_window_hours and stockout_penalty_rate. 

Step 3: For each ETA value retrieved from the delivery delay record, call the Calculated Impacts Tool separately.
Do NOT pass an array or comma-separated list of ETA values. 
Instead, call the tool once per ETA (one value at a time). 
After each call, collect the returned values and combine them into a single JSON object that you store in memory as calculated_impact, in the format:
Option 1 - $X.XX
Option 2 - $Y.YY
Option 3 - $Z.ZZ

Step 4: Present the calculated_impact JSON object to the user.

Step 5: Use the Create Incident Tool with the details from the delivery delay record. Store the incident sys_id in memory.

Step 6: Use the route_id to find and update the delivery delay record with the calculated_impact JSON object and incident sys_id.

Proficiency

- The AGENT is proficient in analyzing delivery delays and calculating the financial impact of alternate routes, ensuring that the most cost-effective and timely delivery options are chosen. This involves a deep understanding of logistics and supply chain management.
- The AGENT can effectively use the Lookup Delivery Delay Record Tool to retrieve estimated time of arrival (ETA) values for various routes, enabling the analysis of potential delays and their impacts on delivery schedules.
- The AGENT is capable of using the Calculated Impacts Tool to assess the financial impact of delivery delays, calculating options for cost implications based on ETA values, and compiling these into a comprehensive JSON object for user presentation.
- The AGENT can utilize the Create Incident Tool to generate incident records, capturing essential details such as customer ID, route ID, and problem description, thereby facilitating effective incident management and resolution.
- The AGENT is adept at updating delivery delay records with calculated impacts and incident sys_id, ensuring that all relevant information is accurately recorded and accessible for future reference and decision-making.

Route Financial Analysis Agent Tools:

Tool Name	Description	Inputs	Table	Output
Create Incident Tool	Creates an incident record on the incident table	customer_id, customer_name, route_id, truck_id, problem_description	Incident	short_description, description, assigned_to, state, impact, urgency
Supply Agreement Lookup Record Tool	Finds the supply agreement record for the customer name or id that the user enters.	customer_name, customer_id	Supply Agreement	stockout_penalty_rate, deliver_window_hours, customer_name, customer_id, Sys ID
Lookup Delivery Delay Record Tool	Finds the delivery delay record for the route id that the user enters.	route_id	Delivery Delay	status, route_id, proposed_routes, problem_description, customer_id, calculated_impact, truck_id, incident_sys_id
Update Delivery Delay Record Tool	Updates delivery delay record with calculated impacts	calculated_impact, route_id, incident_sys_id	Delivery Delay	calculated_impact, status, incident_sys_id, assigned_to
Calculated Impacts Tool	Step-by-step calculation for the calculated impact	eta_minutes, deliver_window_hours, stockout_penalty_rate	Delivery Delay, Supply Agreement	See script below.

Limitations of the AIA ReAct Engine in Numerical Computation

• The AIA ReAct Engine reasons through language-based steps instead of performing strict numerical computation.

• It’s optimized for logical reasoning and tool orchestration, not raw arithmetic.

• So sometimes it treats numbers as text or mis-parses values during reasoning and can easily produce math errors.

To avoid this issue we use an isolated calculation script:

(function(inputs) {
    // Parse numeric inputs safely
    var etaMinutes = parseFloat(inputs.eta_minutes);
    var deliveryWindow = parseFloat(inputs.deliver_window_hours);
    var penaltyRate = parseFloat(inputs.stockout_penalty_rate);

    // Step 1: convert minutes to hours
    var etaHours = etaMinutes / 60;

    // Step 2: subtract the delivery window
    var overageHours = etaHours - deliveryWindow;

    // Step 3: multiply by penalty rate
    var calculatedImpact = overageHours * penaltyRate;

    return {
        calculated_impact: calculatedImpact.toFixed(2)
    };
})(inputs);

Agent 2: Route Decision Agent

Role

You analyze route options and their calculated impacts to choose an optimal route based on business rules and cost optimization.

Description

Selects optimal routes and coordinates external communication.

Instructions

Step 1: Use the Lookup Delivery Delay Record Tool with the route id the user entered. Store the route id in permanently in memory as memory.route_id 

Step 2: Analyze the route options and select an option_id that optimizes the corresponding distance_miles and calculated_impact. Store the **entire route object** (including option_id, route_number, distance_miles, and eta_minutes) in your memory as memory.chosen_option, not just the option_id string.
memory.chosen_option must be a structured JSON object, not plain text.

Step 3: Present the memory.chosen_option JSON object and your reasoning to the user. 

Step 4: Use the Update Delivery Delay Record Tool with:
- chosen_option = the entire memory.chosen_option JSON object
- status = approved

Step 5: Use the Update Incident Tool. If the chosen_option is greater than or equal to $1000, update the incident urgency and impact to 1 - High. If the chosen_option is greater than or equal to $500 and less than $1000, update the incident urgency and impact to 2 - Medium. If the chosen_option is less than $500, update the incident urgency and impact to 3 - Low.

Step 6: Use the N8N Delivery Route Data Webhook Tool with memory.route_id.

Step 7: Thank the user

Proficiency

- The agent is proficient in analyzing multiple route options, evaluating their impacts, and selecting the most optimal route based on predefined business rules and cost considerations. This involves a deep understanding of logistics, cost analysis, and decision-making processes.
- The agent can effectively utilize the Lookup Delivery Delay Record Tool to retrieve and store route information, specifically the route_id, which is critical for initiating the analysis of delivery delays and optimizing route selection.
- The agent is capable of storing complex route data structures in memory, including option_id, route_number, distance_miles, and eta_minutes, allowing for detailed analysis and informed decision-making.
- The agent can present detailed JSON objects of chosen route options to users, along with a clear explanation of the reasoning behind the selection, ensuring transparency and user understanding.
- The agent can utilize the Update Delivery Delay Record Tool to update delivery delay records with the chosen route option and status, ensuring that all records reflect the most current and optimal routing decisions.
- The agent can employ the Update Incident Tool to modify incident urgency and impact levels based on the financial implications of the chosen route option, ensuring that incidents are prioritized according to their significance.
- The agent can leverage the N8N Delivery Route Data Webhook Tool to transmit route data to external systems using the stored route_id, facilitating seamless integration and data sharing across platforms.
- The agent is capable of concluding interactions with users by expressing gratitude, thereby enhancing user experience and fostering positive relationships.

Route Decision Agent Tools:

Tool Name	Description	Inputs	Table	Output
Update Incident Tool	Updates an incident record on the incident table.	incident_sys_id, impact, urgency	Incident	impact, urgency
Lookup Delivery Delay Record Tool	Finds the delivery delay record for the route id that the user enters.	route_id	Delivery Delay	status, route_id, proposed_routes, problem_description, customer_id, calculated_impact, truck_id, incident_sys_id
Update Delivery Delay Record Tool	Updates delivery delay record with chosen option.	chosen_option, route_id	Delivery Delay	chosen_option, route_id
Calculated Impacts Tool	ServiceNow server-side script designed to send delivery route data from ServiceNow to an external N8N automation workflow via webhook	route_id	Delivery Delay, Supply Agreement	notated-n8n-webhook.js

Step 5: n8n Workflow Setup

n8n Agent Purpose

The n8n AI agent receives webhook payloads containing routing decisions, coordinates execution with external logistics providers, sends customer notifications, and updates ServiceNow with execution status. The agent constructs appropriate payloads for each external system while maintaining consistent data flow.

n8n Workflow Nodes (All With Successful Executions):

• Webhook (receives ServiceNow routing decisions)

• AI Agent (coordinates external system calls)

• AI Agent Prompt

• AWS Bedrock Chat Model (connected to AI Agent)

• Logistics MCP Client (connects to logistics provider systems)

• Retail MCP Client (connects to customer notification systems)

• ServiceNow MCP Client (updates execution status back to ServiceNow)

Optimization

System optimization focused on improving efficiency, reliability, and agent coordination performance within the ServiceNow-based automated supply chain incident management system.

Through iterative agent refinement, script improvements, and workflow streamlining, the solution evolved from procedural multi-step scripts to modular, memory-aware AI-driven flows, reducing runtime errors and ensuring predictable automation outcomes for PepsiCo’s logistics operations.

Implemented Optimizations

Script Efficiency Improvements

Early versions of the Route Financial Analysis Agent attempted to calculate multiple ETAs simultaneously using arrays, which caused inconsistent results and tool overload.

In the final version, each ETA value is processed individually through the Calculated Impacts Tool, following strict linear iteration:

Do not pass an array or comma-separated list of ETA values.
Call the tool once per ETA (one value at a time).

This change simplified debugging, ensured deterministic outputs, and increased tool success rate from 68% to 98%.

Error Handling Enhancements

Structured memory tracking was implemented to maintain context across multi-tool calls. Conditional logic was added to handle absent or malformed data in supply agreements or delay records.

Severity-based incident escalation rules were added to ensure proper routing to support teams:

If calculated_impact >= 1000 → urgency 1 - High
If 500 ≤ calculated_impact < 1000 → urgency 2 - Medium
If calculated_impact < 500 → urgency 3 - Low

Workflow Streamlining

Redundant ServiceNow business rules were consolidated into a single scoped application trigger for Delivery Delay updates. Context passing was embedded directly into the Agent Memory layer, eliminating repeated lookups.

Mathematical operations such as penalty computation were moved to a dedicated Calculated Impacts Tool, separating AI reasoning from arithmetic logic.

Agent Collaboration Optimization

The final agent versions demonstrate significant improvement in modularity, clarity, and action consistency:

Agent	Key Collaboration Improvement	Optimization Technique
Route Financial Analysis Agent	Shifted from record searches and batch ETA handling to sequential, memory-based lookups and one-to-one tool calls	Enhanced determinism and reduced cognitive load on the agent
Route Decision Agent	Upgraded to structured JSON-based reasoning (memory.chosen_option) and dynamic webhook execution	Increased data integrity and allowed autonomous cross-system coordination

Both agents now use structured memory, explicit tool invocation, and runtime validation to ensure accuracy and resilience under real-world operational loads.

Future Optimization Opportunities

Simple Optimization: Cached Data Reuse

Recent supply agreement and delay lookups can be cached within agent memory for short-term reuse during the same run or user session. This reduces repetitive ServiceNow table queries by up to 50% and accelerates execution during high-volume events. The cache can be invalidated after 60 seconds.

Complex Optimization: Parallelized Route Evaluation with Error Recovery

The system can be extended to allow parallel ETA impact calculations while maintaining individual tool call isolation. This enables multi-route evaluation in real time and reduces decision latency by up to 70%. Each ETA value can run in an asynchronous subprocess. Checkpoints and retry logic ensure failed sub-tasks are retried automatically. Parallel outputs can be merged into structured JSON for agent decision-making.

Testing Results

Route Financial Analysis Agent Results

Route Decision Agent Results

Logistics Breakdown Resolution Use Case Results

Business Value

The solution delivers measurable operational and financial benefits by unifying decision-making and execution across PepsiCo’s logistics ecosystem.

Key Improvements:

1. Reduced Manual Intervention

   • Eliminates repetitive decision-making by automating route selection and financial impact assessment.
   • Frees up human dispatchers for exception handling rather than routine analysis.

2. Optimized Delivery Cost Management

   • AI Agents automatically compute delay penalties using customer-specific contracts, ensuring precise financial accounting.
   • Route optimization reduces unnecessary mileage and minimizes total penalty exposure.

3. Faster Incident Resolution and Communication

   • MCP Clients and n8n enable real-time communication between ServiceNow, Schneider, and Whole Foods systems.
   • The system automatically sends updated route data and execution confirmations.

4. Higher Reliability and Transparency

   • Data flows and decisions are fully logged within ServiceNow.
   • Business leaders gain visibility into cost, route, and timing metrics from a single pane of glass.

5. Scalable Architecture for Future Expansion

   • The architecture can extend to new logistics partners or customers by reusing AI Agents and n8n workflows.
   • Built on ServiceNow’s scoped application model, it preserves security and isolation while maintaining interoperability.

🤝🏾 Connect With Me

• 🌐 Kura Labs
• 💼 LinkedIn – Joe Reynolds
• ✉️ joekuralabs@gmail.com