PRD: Demand Forecasting for EWM Inbound/Wave Planning (MVP)

Project Priority: 🥉 THIRD MOST IMPORTANT
Version: 1.0 MVP
Status: Ready for Development

1. Product Vision (MVP)

Create a practical demand forecasting service that predicts SKU-level daily demand and exposes a simple output that could be used by an EWM wave-planning process: “what volume is coming in/out per day for key SKUs.” This demonstrates understanding of how forecasting connects to warehouse operations.

Why This Matters: Demand forecasting directly feeds wave management and inbound planning—core EWM functions. Supply/demand reconciliation logic demonstrates understanding of business operations, not just data science. This shows you can translate analytics into operational decisions.

2. Users & Jobs-to-be-Done

Primary User: Supply Chain Planner / EWM Wave Planner

• Need: Reliable short-horizon demand signal for operational planning
• Job-to-be-Done: “I need a reliable short-horizon demand signal to decide when to schedule inbound receipts and when to trigger outbound picking waves.”

Secondary User: SAP EWM Consultant

• Need: Demonstrate forecasting capabilities that complement EWM wave planning
• Job-to-be-Done: “I want a lightweight forecast that I can conceptually plug into wave creation rules.”

3. MVP Scope (Must-Haves)

3.1 Data Requirements (Simplified)

Single Warehouse, Top 20 SKUs:

Required Fields: - date - Date of transaction - sku - SKU identifier - demand_qty - Daily demand quantity (orders or picks) - inbound_qty - Daily inbound receipts (optional) - on_hand_qty - Inventory on hand (optional)

Data Requirements: - Minimum 6 months of daily history - Top 20 SKUs by volume (can start with fewer for MVP) - Include realistic patterns: - Weekly seasonality - Monthly trends - Holiday spikes - Promotional periods

Optional: - Calendar of holidays/promotions (CSV file with dates and event types) - External factors (weather, economic indicators) - nice to have but not required

Data Sources: - Synthetic data generation based on realistic demand patterns - Or aggregated from logistics/retail datasets (Kaggle) - Include seasonality, trends, and noise

3.2 Modeling Approach (MVP)

Per SKU Forecasting:

Model Options: - Option A: Prophet (Recommended for MVP) - Handles seasonality and trends well - Built-in holiday support - Provides confidence intervals - Easy to interpret and explain

• Option B: SARIMA (via statsmodels)
  • More traditional time series approach
  • Requires more parameter tuning
  • Good for understanding time series fundamentals

Model Features: - Trend component (capturing long-term growth/decline) - Weekly seasonality (day-of-week patterns) - Monthly seasonality (if data supports) - Holiday effects (using holiday calendar if provided) - Uncertainty quantification (confidence intervals)

Forecast Output: - 4-week daily forecast for each SKU - Lower and upper confidence bounds (80% and 95% intervals) - Forecast components breakdown (trend, seasonality, holidays)

Aggregation: - Daily total forecast volume (sum across all SKUs) - Category-level aggregations (if SKU categorization available) - Warehouse-level load forecast

3.3 Simple Wave Planning Hints (Logic, not full optimizer)

On top of forecast, derive 3 simple planning metrics per day:

1. expected_lines_to_pick
   • Derived from forecasted demand
   • Assumes average lines per order (configurable)
   • Example: forecast_qty * avg_lines_per_order
2. expected_inbound_volume
   • If inbound data available, forecast inbound receipts
   • Otherwise, use historical inbound patterns
   • Helps plan receiving capacity
3. expected_peak_days
   • Flag days exceeding threshold (e.g., 1.5x average daily volume)
   • Helps identify when extra resources needed

Wave Planning Suggestions:

Based on volume distribution, suggest 3 wave windows per day: - Morning Wave (e.g., 8 AM - 12 PM) - Afternoon Wave (e.g., 12 PM - 4 PM) - Evening Wave (e.g., 4 PM - 8 PM)

Simple Heuristic: - If daily volume < threshold → “2 waves enough” - If daily volume > threshold → “3 waves needed” - If daily volume >> threshold → “Consider 4 waves or overtime”

Output Format: - Daily summary table with: - Date - Total forecast quantity - Expected pick lines - Expected inbound volume - Peak day flag - Wave suggestion (e.g., “3 waves needed”)

3.4 UI / Deliverables (MVP)

Notebook or Streamlit App:

Dashboard Components:

1. SKU-Level Forecast Visualization:
   • For chosen SKU: historical + forecast plot
   • Show confidence intervals
   • Highlight forecast period vs historical
   • Decompose forecast (trend, seasonality, holidays)
2. Warehouse-Level Summary:
   • Daily total forecast with threshold line indicating “high load”
   • Peak days highlighted
   • Weekly/monthly aggregation views
3. Wave Planning Table:
   • Next 14 days with columns:
     • date
     • total_forecast_qty
     • expected_pick_lines
     • expected_inbound_volume
     • peak_day_flag (Yes/No)
     • wave_suggestion (e.g., “3 waves needed” vs “2 waves enough”)
4. Model Performance:
   • Accuracy metrics (MAPE, RMSE) on holdout period
   • Forecast vs actual comparison chart
   • Per-SKU accuracy breakdown
5. SKU Comparison:
   • Top 10 SKUs forecast comparison
   • Identify fastest-growing SKUs
   • Identify declining SKUs

Optional Features: - Interactive SKU selector - Date range selector - Export to CSV functionality - What-if scenarios (adjust forecast manually)

4. Out-of-Scope for MVP

• Full wave creation algorithm (only suggestions)
• Integration to actual SAP APIs
• Multi-echelon or multi-DC network forecasting
• Real-time demand updates
• Advanced optimization (OR-Tools for wave optimization)
• Inventory optimization (reorder points, safety stock)
• Multi-SKU correlation modeling

5. Tech Stack (MVP)

Core: - Python 3.9+ - Pandas / NumPy (data manipulation) - Prophet (time series forecasting) OR statsmodels (for SARIMA) - Matplotlib / Plotly (visualization)

SAP-EWM Alignment: - SKU-level forecasting aligned with SAP EWM material master data - Wave planning logic that mirrors SAP EWM wave creation rules - Understanding of how forecasts feed into EWM inbound/outbound processes - SQL queries for extracting demand data from EWM transaction tables - Potential OData service exposure for integration with SAP systems - CDS Views understanding for semantic data modeling of forecast outputs

UI: - Streamlit (recommended) OR - Jupyter Notebook with interactive widgets

Data: - CSV/Parquet file storage - Synthetic data generator (numpy/pandas) - Or aggregated from existing datasets

Validation: - Time series cross-validation - Holdout period testing - MAPE, RMSE, MAE metrics - Forecast accuracy by SKU

6. Success Criteria (MVP)

Functional Requirements:

• ✅ For at least 10–20 SKUs:
  • Train simple models
  • Generate 4-week forecasts
  • Summarize in a single planner-friendly table
• ✅ Produces wave planning suggestions
• ✅ Generates visualizations and reports

Model Performance:

• ✅ MAPE < 25% on holdout period (aggregate level)
• ✅ Captures weekly seasonality patterns
• ✅ Provides reasonable confidence intervals
• ✅ Handles missing data gracefully

Interview Readiness:

• ✅ You can say:
  • “This is not just a Kaggle forecast; it’s aligned with wave planning and inbound scheduling logic”
  • “Demand forecasting directly feeds EWM wave management and inbound planning”
  • “Uses SQL skills that transfer to SAP Open SQL for extracting EWM transaction data”
  • “Forecast outputs can be exposed via OData services for SAP integration”
  • “Demonstrates understanding of CDS Views for semantic data modeling”
  • “Shows how Python analytics skills complement ABAP technical implementation”
  • “Shows how analytics translates into operational decisions”

7. MVP Deliverables

1. Source Code:
   • demand_forecast.py - Main forecasting script
   • data_preparation.py - Data loading and preprocessing
   • wave_planning.py - Wave planning logic
   • app.py - Streamlit dashboard (if using Streamlit)
   • requirements.txt - Dependencies
2. Data:
   • data/demand_history.csv - Sample historical demand data
   • data/holidays.csv - Holiday calendar (optional)
   • data/synthetic_data_generator.py - Script to generate synthetic data
3. Documentation:
   • README.md with setup instructions
   • Model explanation document
   • Wave planning logic explanation
   • Sample outputs and visualizations
4. Outputs:
   • Forecast CSV file (per SKU, per day)
   • Wave planning recommendations CSV
   • Visualization charts (PNG/HTML)
   • Model performance report

8. Development Timeline (MVP)

Week 1: Data & Foundation - Create synthetic data generator - Implement data loading and preprocessing - Basic exploratory data analysis - Identify top SKUs

Week 2: Forecasting Model - Implement Prophet model (or SARIMA) - Train per-SKU models - Generate forecasts with confidence intervals - Validate model performance

Week 3: Wave Planning Logic - Implement wave planning heuristics - Create planning metrics - Build summary tables - Add peak day detection

Week 4: Visualization & Polish - Build Streamlit dashboard (or notebook visualizations) - Add SKU comparison views - Create documentation - Prepare demo

9. SAP-EWM Technology Complementarity

How Python Forecasting Skills Complement SAP Technologies:

While ABAP is core for SAP EWM wave creation logic and BRFplus handles rule-based wave planning, this Python-based forecasting demonstrates complementary skills:

1. CDS Views & Data Modeling:
   • Understanding of CDS Views for semantic data modeling of forecast outputs
   • Ability to expose forecast data via SAP’s modern data access layer
   • Demonstrates S/4HANA cloud-ready thinking
2. OData Integration:
   • Forecast outputs can be exposed via OData services for SAP integration
   • Understanding of RESTful API patterns used in SAP Fiori apps
   • Enables integration with EWM wave planning processes
3. SQL & Data Extraction:
   • SQL skills for extracting demand data from EWM transaction tables
   • Understanding of EWM material master and delivery data structures
   • Ability to analyze historical demand patterns
4. Operational Understanding:
   • Demonstrates understanding of how forecasting connects to warehouse operations
   • Wave planning logic that mirrors SAP EWM wave creation rules
   • Shows ability to translate analytics into operational decisions

Key Differentiator: Most SAP consultants understand EWM configuration; fewer understand how analytics translates into operational decisions. This project demonstrates the bridge between data science and warehouse operations.

10. Key Interview Talking Points

1. Operational Relevance:
   • “Demand forecasting directly feeds wave management and inbound planning—core EWM functions”
   • “This shows how analytics translates into operational decisions”
   • “Wave planning logic mirrors SAP EWM wave creation rules”
2. Technical Approach:
   • “Uses Prophet for time series forecasting with seasonality and holiday effects”
   • “Provides confidence intervals for risk-aware planning”
   • “SQL skills transfer to SAP Open SQL for extracting EWM transaction data”
3. SAP Technology Alignment:
   • “Forecast outputs can be exposed via OData services for SAP integration”
   • “Understanding of CDS Views for semantic data modeling”
   • “Demonstrates S/4HANA cloud-ready architecture thinking”
4. Business Impact:
   • “Enables proactive wave planning vs reactive order processing”
   • “Helps optimize warehouse capacity and labor allocation”
5. Complementary Skills:
   • “Shows how Python analytics skills complement ABAP technical implementation”
   • “Understanding of EWM material master and delivery data structures”

11. Future Enhancements (Post-MVP)

• Multi-SKU correlation modeling
• Integration with SAP EWM APIs
• Real-time demand updates
• Advanced wave optimization (OR-Tools)
• Inventory optimization (reorder points, safety stock)
• Multi-echelon forecasting
• Promotional impact modeling
• External factor integration (weather, economic indicators)
• Automated wave creation recommendations
• Integration with WMS/ERP systems