A demand forecast that cut overstock by 23%
A direct-to-consumer brand in India. SKU demand predicted 30 days ahead, inside Excel.
23%less overstock
30 daysforecast horizon per SKU
Zero new toolsthe team kept working in Excel
The challenge
What wasn’t working
Purchase orders were based on gut feel and last month’s sales. Some SKUs piled up in the warehouse while others ran out.
What we built
The solution
- 1
We cleaned and joined two years of order history.
- 2
An ML model forecasts demand for every SKU 30 days ahead.
- 3
Forecasts are written straight into the planning sheet the team already used.
- 4
Accuracy is tracked and the model is retrained as patterns change.
Built with: Python, scikit-learn, pandas, Excel
“The forecasting model paid for itself in the first quarter through reduced overstock alone.”
Next case study
200+ invoices a day entered without typing