AI-Driven Process Automation for an E-Commerce Operator
Predictive ML replacing gut-feel decisions across the supply chain
Key outcomes
- Forecasting accuracy improvement
- 25%
- Reduction in manual ops tasks
- 60%
- Time to first production model
- 3 weeks
- Reduction in overstock incidents
- 18%
A multi-category e-commerce operator was losing margin to manual demand forecasting and fragmented operational processes. We designed and deployed ML models for demand prediction, price optimisation and inventory reordering — cutting manual workload and improving margin.
Stack
Demand planners were spending 80% of their time on spreadsheets and still getting forecasts wrong. Overstock and stockouts were both common. Leadership had no visibility into which product categories were actually profitable after fulfilment costs.
How we tackled it, step by step.
Ran a 2-week discovery to map all operational decision points, data sources and failure modes
Built a demand forecasting model using LightGBM with category, seasonality and promo features
Deployed the model on GCP Vertex AI with weekly automated retraining against fresh sales data
Built an inventory reordering engine that triggers purchase orders based on predicted lead time and demand
Created a profitability attribution dashboard showing true margin per SKU after fulfilment costs
Ran a parallel-run validation period before full cutover — measured against actuals for 6 weeks
Outcomes that speak for themselves.
Forecasting accuracy improvement
Reduction in manual ops tasks
Time to first production model
Reduction in overstock incidents
Professional execution and long-term scalability thinking. The models were built for growth, not short-term fixes.