Skip to content

PRD.md

Latest commit

 

History

History
147 lines (90 loc) · 6.38 KB

PRD.md

File metadata and controls

147 lines (90 loc) · 6.38 KB

Foresight — Business PRD

Scope: Web app for a single hotel (initially), with real‑time AI pricing recommendations and one manager user role. Tech: Next.js (frontend) + FastAPI (backend) + MongoDB (potential data store).

1) Vision & Problem

Hotels routinely leave 10–20% revenue on the table due to static or manual pricing. Revenue managers juggle spreadsheets, instincts, and delayed signals (events, pace, seasonality). Foresight delivers revenue intelligence: timely, transparent, and trustworthy AI price recommendations that are easy to act on.

Primary outcome: Increase RevPAR by surfacing the right price at the right time with minimal effort from the manager.

2) Target User & Roles

  • Hotel Manager (single role) — Reviews/adjusts prices, inputs special dates/events, applies recommendations to PMS or exports.

3) Core Value Proposition

  • MVP 0: Prove the end‑to‑end loop from historical data → AI model → daily recommendations in a clean UI.
  • MVP 1: Make it live: integrate one major PMS, keep recommendations fresh, and add lightweight event/seasonality awareness.

4) Success Metrics (north‑star & supporting)

  • NSM: Increase in expected RevPAR vs. baseline for the next 30 days (simulation/what‑if analysis).
  • Adoption: % days with recommendations viewed; time‑to-first-recommendation (< 10 minutes from login).
  • Quality: Manager‑rated usefulness of recs (≥ 4/5); % recs accepted/applied.
  • Reliability: Data freshness (e.g., last sync < 60 min in MVP 1), uptime ≥ 99%.

5) MVP 0 — Requirements (3–4 months)

Goal: Demonstrate value with the simplest path to recommendations.

In‑Scope Features

  1. Historical data ingestion

    • Manager uploads CSV/XLSX following a simple template (dates, available rooms, ADR, occupancy, bookings, room type optional).
    • Basic validation and schema hints.

  2. AI price recommendations (batch)

    • Train a lightweight model on historical data (day‑of‑week, seasonality, pace proxies).
    • Output daily recommended rate for the next 30–60 days (per hotel; per room type optional if data allows).
    • Show confidence/strength signal (e.g., low/med/high) and a simple reason snippet where possible.

  3. Manager UI (minimal)

    • Secure login (single user).
    • Upload page, "Generate Recommendations" action.
    • Results: table (Date | Recommended Price | Notes). CSV export.

  4. Ops & guardrails

    • Error handling for malformed files.
    • Basic audit log (upload time, model run time, result set ID).

Out‑of‑Scope (MVP 0)

  • Real‑time sync, automatic PMS write‑back, competitor scraping, multi‑property, multi‑user, rich analytics.

Acceptance Criteria

  • Given valid historical data, the system returns a non‑flat recommended price curve for the next 30+ days.
  • End‑to‑end time (upload → results) under ~2 minutes on sample data.
  • Exportable recommendations; UI understandable by a non‑technical manager.

6) MVP 1 — Requirements (follow‑on ~4–5 months)

Goal: "Always‑fresh" recommendations with one PMS integration and light event awareness.

In‑Scope Features

  1. PMS integration (one provider)

    • Read reservations/availability/rates via provider API (polling or webhook) on a schedule (e.g., every 15–60 min).
    • Normalize into our datastore; surface last‑sync status in UI.

  2. Real‑time(ish) recommendations

    • Automatic recompute cadence (e.g., hourly) and/or trigger on significant pickup deltas.
    • Optional one‑click export (CSV) or push back to PMS for selected dates (push may be a stretch goal; export is in‑scope).

  3. Event & seasonality awareness (lightweight)

    • Manager can add special events/blackout periods with expected impact (low/med/high).
    • Model or post‑processing applies uplift/guardrails for those dates.

  4. Manager UI (usable daily)

    • Dashboard (today’s occ %, current vs. recommended ADR, 7‑day view).
    • Calendar view of recommendations; event badges; quick filters.
    • Simple explanations (e.g., "High weekend demand" / "Low midweek pickup").

  5. Ops & reliability

    • Status page: last model run, last PMS sync, data health.
    • Basic alerting if sync fails or recommendations are stale.

Out‑of‑Scope (MVP 1)

  • Multi‑property tenants; competitor rates; channel manager/CRS; native mobile apps; advanced RL pricing.

Acceptance Criteria

  • PMS data ingests on schedule; UI shows fresh timestamp.
  • Recommendations change in response to synthetic booking shocks (QA scenario).
  • Manager can add an event and see uplift reflected on those dates.
  • "npm run build" (frontend) and test suite (backend) pass in CI for release.

7) Constraints & Assumptions

  • Single hotel and single role until post‑MVP 1.
  • One major PMS integration only (depth over breadth).
  • Data science remains explainable; no opaque black‑box requirements.
  • Hosting on reliable cloud (separate staging/prod). Secrets managed securely.

8) Risks & Mitigations

  • PMS API access delays → start integration early; use mock/test data; keep export workflow as fallback.
  • Data sparsity/quality → include validators and templates; allow manual event input to improve signal.
  • Model trust → ship simple explanations + guardrails (min/max rate bands), and CSV export for human review.

9) Go‑to‑Market for Pilot

  • Target one committed pilot hotel (the dataset we have). Weekly check‑ins; gather acceptance signals (recs applied %, qualitative feedback). Create simple ROI case studies for fundraising and future sales.

10) Release Plan (indicative)

  • Month 1–2: Data model, upload/validation, baseline model, MVP 0 UI.
  • Month 3–4: MVP 0 polish, export, basic logging/metrics; pilot demo.
  • Month 5–6: PMS integration scaffolding, data adapter, scheduler, dashboard/calendar.
  • Month 7–8: Event inputs, explanations, auto refresh, reliability/alerts, pilot hardening.

11) Non‑Functional Requirements

  • Security: HTTPS, least‑privilege credentials, audit of data imports/exports.
  • Performance: Recommendation generation < 2s per date range after model is warm; sync jobs complete reliably within cadence.
  • Observability: Logs, request IDs, job status, basic dashboards.
  • Documentation: Runbooks for onboarding, model retrain, and integration keys.