2025 / UC Berkeley Capstone 121
Operations-Driven Analytics in E-Commerce
Decision-focused forecasting for fulfillment and inventory
Overview
Capstone project exploring how demand forecasting can directly drive inventory allocation and routing decisions in large-scale e-commerce.
Problem
E-commerce operations need forecasts that translate into concrete decisions across inventory, fulfillment, and last-mile routing.
Role
Capstone project (UC Berkeley, advisor: Prof. Paul Grigas)
Timeline
Sep 2025 - Present
Tools
Python / pandas / optimization / Power BI / Excel
Tags
Forecasting / Optimization / Analytics / Supply Chain
Data
- - Course-provided e-commerce and last-mile datasets
- - Demand, fulfillment, and routing signals aligned to decision stages
Approach
- - Defined prediction-to-optimization pipeline with nested validation
- - Evaluated decision-focused loss formulations against forecasting baselines
- - Mapped outputs to inventory allocation and routing scenarios
Evaluation
- - Nested cross-validation across demand segments
- - Scenario comparisons vs. baseline heuristic policies
Results
- - Scenario analysis indicated improved order-fill rate (+12%)
- - Routing cost reductions of ~9% in simulated policy comparisons
Deployment & Monitoring
- - Dashboards for scenario review and sensitivity checks
- - Clear handoff between forecasting and optimization stages
Limitations
- - Anonymized datasets and simulated policies; results are scenario-based
- - Live deployment pending access to real-time operational data
Gallery
E-commerce analytics context
Fulfillment network focus
Last-mile delivery framing
Dashboard-style insights
Repro Steps
- - Slides and scripts available in the Capstone folder
- - Rebuild visual summaries using the provided assets
Next Steps
- - Add real-time data feeds for rolling forecasts
- - Integrate uncertainty bounds into allocation decisions
- - Pilot the pipeline on a live SKU subset