Product Requirements Document (PRD)

AI Dynamic Pricing for India Coffee Shops

Status: Side Project → Production-Ready Concept
Target Market: Indian Independent Coffee Shops (500-5,000 transactions/month)

Executive Summary

India coffee shops lose an estimated ₹2-4k monthly in revenue by using static pricing that ignores demand signals. This AI-powered dynamic pricing system adjusts prices in real-time based on weather, time-of-day, and product type to capture willingness-to-pay during high-demand periods while maintaining customer trust through explainability.

Key Metric: 8-12% revenue lift without customer churn (validated via backtest simulation)

1. Problem Statement

1.1 Business Problem

Independent India coffee shops face margin pressure from rising costs (rent, labor, ingredients) but hesitate to raise prices due to customer sensitivity. Static menus leave money on the table during peak demand (cold rainy mornings, 8-9 AM commute) while potentially overcharging during slow periods.

1.2 User Pain Points

Coffee Shop Owner:

"I know customers will pay more on cold mornings, but I can't change menus daily"
"My POS system doesn't support time-based pricing"
"I'm worried dynamic pricing will upset loyal customers"

Operations Manager:

"How do I explain price changes to baristas?"
"What if customers complain about different prices?"

Finance Director:

"How do I justify dynamic pricing to franchisees?"
"What's the ROI vs. risk of customer backlash?"

1.3 Market Opportunity

Target Market: 25,000 independent coffee shops in India
Addressable Segment: 5,000 shops with digital POS systems
Revenue Potential: ₹2k-₹4k additional monthly revenue per shop
Total Opportunity: ₹10M-₹20M annual revenue lift (at 5% penetration)

2. Solution Overview

2.1 Product Vision

An AI pricing engine that runs in the background, recommending optimal prices based on contextual signals, with full explainability for every recommendation to build stakeholder trust.

2.2 How It Works

For Coffee Shop Owners:

Integrate via API with existing POS system
System learns from 6-12 weeks of historical transaction data
Receives hourly price recommendations via dashboard
Can accept, reject, or modify recommendations
Views impact reports (revenue lift, customer satisfaction)

Technical Architecture:

XGBoost model trained on historical transactions + weather data
SHAP explainability layer for every price recommendation
Confidence scoring (High/Medium/Low) guides manual override decisions
Backtest engine validates recommendations before deployment

2.3 Key Differentiators

Explainability-First: Every price change comes with a "because..." explanation
Human-in-the-Loop: Owners control final pricing decisions
India-Specific: Weather integration via Open-Meteo API (Bengaluru focus)
Ethical Design: No surge pricing on essentials; caps at +15% above base price

3. User Personas

Persona 1: Priya (Independent Coffee Shop Owner)

Age: 34, runs 2 coffee shops in Bengaluru
Pain: Thin margins, can't afford to lose regular customers
Goal: Increase revenue without alienating loyal base
Tech Comfort: Medium (uses Square POS, Instagram for marketing)
Decision Criteria: "Will this make me more money without costing me customers?"

Persona 2: James (Operations Manager, Chain of 10 Shops)

Age: 42, manages operations for mid-size chain
Pain: Staff confusion, inconsistent pricing across locations
Goal: Standardize pricing strategy while capturing local demand patterns
Tech Comfort: High (manages POS, inventory, scheduling software)
Decision Criteria: "Can my team explain this to customers? Is it scalable?"

Persona 3: Ananya (Finance Director, Franchise Network)

Age: 48, oversees 30 franchisee locations
Pain: Needs to justify pricing changes to franchise owners
Goal: Prove ROI before rollout; minimize franchisee pushback
Tech Comfort: Medium (Excel power user, reads basic SQL reports)
Decision Criteria: "Show me the numbers. What's the risk?"

4. Success Metrics & OKRs

4.1 North Star Metric

Revenue per Transaction (RPT) - Must increase 8-12% without impacting transaction volume

4.2 Primary Metrics

Metric	Baseline	Target (Month 3)	Measurement
Revenue Lift	0%	+10%	Daily revenue vs. static pricing baseline
Customer Retention	85%	85% (maintain)	Monthly repeat customer rate
Price Acceptance Rate	N/A	80%	% of AI recommendations accepted by owner
Complaint Rate	2%	<3%	Customer complaints re: pricing per 1000 transactions

4.3 Secondary Metrics

Model Performance: RMSE < ₹0.20 on test set
Explainability Score: 90%+ of owners understand SHAP explanations (survey)
System Uptime: 99.5% availability
Recommendation Latency: <500ms per prediction

4.4 Counter-Metrics (Guard Rails)

Customer satisfaction score (CSAT) must stay ≥4.2/5
Average basket size must not decrease >5%
No single product can exceed +20% from base price

5. User Stories & Acceptance Criteria

Epic 1: Core Pricing Engine

Story 1.1: As an owner, I want to see hourly price recommendations so that I can decide whether to accept them

AC1: Dashboard shows next 4 hours of recommendations
AC2: Each recommendation shows: predicted price, confidence (H/M/L), explanation
AC3: One-click accept or manual override available

Story 1.2: As an operations manager, I want price explanations in plain English so that baristas can answer customer questions

AC1: Each price change has a printable "staff guide" (e.g., "₹0.30 higher due to cold morning + high demand")
AC2: Explanation uses <20 words, no technical jargon
AC3: Includes suggested customer response script

Epic 2: Explainability & Trust

Story 2.1: As a finance director, I want to see SHAP waterfall charts so that I can audit pricing logic

AC1: Clicking any recommendation opens detailed SHAP breakdown
AC2: Shows: base price → feature contributions → final price
AC3: Export to PDF for board presentations

Story 2.2: As an owner, I want to set maximum price caps so that I maintain brand positioning

AC1: Can set global cap (e.g., no price >₹5.00 for any coffee)
AC2: Can set product-specific caps (e.g., flat white max ₹4.50)
AC3: System respects caps; shows "capped" label on recommendations

Epic 3: Monitoring & Learning

Story 3.1: As an owner, I want to see daily impact reports so that I know if dynamic pricing is working

AC1: Daily email/dashboard showing: revenue vs. baseline, top revenue-generating time slots, customer feedback summary
AC2: Weekly report comparing projected vs. actual revenue lift
AC3: Monthly A/B test summary (dynamic vs. static pricing performance)

6. Technical Requirements

6.1 Functional Requirements

FR1: Ingest transaction data via CSV upload or API integration (Square, Shopify POS, Lightspeed)
FR2: Fetch real-time weather data (Open-Meteo API)
FR3: Generate price recommendations every hour (batch) or on-demand (API call)
FR4: Store model explainability artifacts (SHAP values) for audit trail
FR5: Support manual override with reason logging

6.2 Non-Functional Requirements

NFR1: Model retraining: Weekly (automatic) with human validation
NFR2: API response time: <500ms for single prediction
NFR3: Dashboard load time: <2 seconds
NFR4: GDPR compliance: All personal data anonymized; 30-day audit log retention
NFR5: Scalability: Support 10,000 requests/hour per shop

6.3 Data Requirements

Minimum Viable Data: 3 months of historical transactions (recommended: 12 months)
Required Fields: Date, Time, Product, Price, Quantity
Optional Fields: Customer ID (anonymized), Payment method, Location
External Data: Weather (temperature, precipitation, cloud cover)

7. Go-to-Market Strategy

7.1 Launch Plan (Phased Rollout)

Phase 0: Portfolio Demo (Current)

Streamlit app with simulated data
Target: Showcase to recruiters, interview portfolio

Phase 1: Pilot (3 months)

Partner with 5 independent Bengaluru coffee shops
Free service in exchange for data + testimonials
Weekly check-ins, manual overrides allowed
Success criteria: 8%+ revenue lift, <5% customer complaints

Phase 2: Beta (6 months)

20-50 paying customers (₹99/month subscription)
Self-service onboarding via web app
Email support + monthly business reviews
Build case studies for marketing

Phase 3: Scale (12 months)

Target 500 customers across India
Tiered pricing: Starter (₹99), Pro (₹199), Enterprise (custom)
Partner with POS providers (Square, Shopify)
Hire customer success team

7.2 Pricing Model (Future State)

Starter: ₹99/month - Single location, email support
Pro: ₹199/month - Up to 5 locations, priority support, custom caps
Enterprise: Custom - 10+ locations, API access, dedicated CSM

7.3 Key Partnerships

POS Systems: Square, Shopify POS, Lightspeed (API integrations)
Trade Associations: India Coffee Association (credibility, distribution)
Data Providers: Open-Meteo (weather), Google Places API (foot traffic)

8. Risks & Mitigations

Risk	Likelihood	Impact	Mitigation
Customer backlash against dynamic pricing	Medium	High	Lead with explainability; cap price increases at +15%; A/B test messaging
Model drift (seasonal patterns change)	High	Medium	Weekly retraining; confidence scoring; human-in-loop overrides
Competitor copies model	Medium	Low	Focus on explainability + India localization as moat; build brand trust
POS integration complexity	High	High	Start with CSV upload MVP; partner with 1-2 POS providers for deep integration
GDPR compliance issues	Low	High	Anonymize all data; EU hosting; legal review before launch

9. Ethical Considerations

9.1 Fairness & Bias

Concern: Algorithm could discriminate based on proxy variables (e.g., location as proxy for demographics)
Mitigation: SHAP audit for protected attributes; monthly bias testing; exclude location-based features in v1

9.2 Transparency

Concern: "Black box" pricing erodes customer trust
Mitigation: SHAP explanations for every price; public FAQ on website; "Why this price?" button in app

9.3 Consumer Protection

Concern: Surge pricing on essentials could harm vulnerable customers
Mitigation: Price caps (+15% max); no dynamic pricing on basic items (filter coffee stays fixed); low-income discount program

9.4 Responsible AI Checklist (Completed)

☑ Model explainability implemented (SHAP)
☑ Bias audit conducted (no demographic proxy features)
☑ Human oversight required (owner approval loop)
☑ Data privacy compliant (India DPDP Act, 2023)
☑ Rollback plan (can revert to static pricing in <1 hour)

10. Open Questions & Future Roadmap

10.1 Open Questions (Needs Research)

What's the optimal price change frequency? (Hourly? Daily? Dynamic throughout day?)
How do customers perceive "fair" price variation? (Survey needed)
Which product categories have highest price elasticity? (A/B test needed)
Can we predict foot traffic using Google Maps API? (Prototype needed)

10.2 Future Features (Post-MVP)

V2: Personalized pricing (loyalty members get stable prices)
V3: Competitor price monitoring (adjust based on nearby cafes)
V4: Inventory-aware pricing (discount items expiring soon)
V5: Predictive ordering (suggest inventory purchases based on forecasted demand)

11. Appendix

11.1 References

Prediction Machines (Agrawal, Gans, Goldfarb) - Decision-making under uncertainty
The AI Product Manager's Handbook - Explainability as product requirement
Applied Artificial Intelligence (Leong Chan) - Adoption gap framework

11.2 Related Documents

METRICS.md - Detailed metrics framework
ETHICS.md - Responsible AI audit
BACKTEST_METHODOLOGY.md - How we validated 8-12% revenue lift
SHAP_EXPLAINABILITY_GUIDE.md - For technical stakeholders