Demand forecasting that survived festival season
28%
MAPE reduction vs incumbent forecasts
34%
reduction in markdown exposure
9wk
from kickoff to first store live
The challenge
Where it started.
The client's statistical forecasts collapsed around festivals, wedding season, and regional events — exactly when the margin was made or lost. Buyers over-ordered as insurance, and markdowns ate the difference.
The approach
How we built it.
Built hierarchical demand models on five years of sales history, enriched with regional calendar, weather, and pricing features via the ontology layer.
Deployed Vibalitics to stream store-level sell-through signals — demand shifts surface in days, not month-end reviews.
Forecasts flow into replenishment as ranked recommendations with confidence bands; buyers keep the final call.
AIDLC drift operations retrain models on trigger, not calendar — festival patterns update themselves.