FMCG Demand Forecasting & Root Cause Analysis
A leading beverage distributor needed to move from reactive stockout discovery to proactive prediction. We deployed a demand forecasting engine with root cause classification and alerting on GCP in 23 weeks.
The challenge
Stockouts were discovered in-store, too late to act. Root causes were unknown: delayed restocking, incorrect stock levels, or lack of shelf rotation? Promotional activities created unpredictable demand spikes. There was no integration between sell-out data, inventory data, and external factors. Field teams operated reactively, visiting stores only to find empty shelves.
What we built
We built a demand forecasting engine at SKU/store level with a 14-day horizon. A root cause classification model (multiclass) identified primary stockout drivers. A proactive alerting system flagged stores at high stockout probability before it happened. A performance dashboard tracked prediction accuracy and model health. Everything was deployed on GCP using BigQuery and Vertex AI.
Outcomes
Stockout detection by individual store and SKU combination
Automated classification across multiple stockout cause types
Proactive alerts to field teams before stockouts materialize
End-to-end from requirements to production on GCP
“We went from discovering stockouts in-store to predicting them days in advance and understanding why they happen.”