Case Study

The Client

A retail chain with several points of sale and a shared warehouse. Replenishment was ordered from each store manager’s experience, so some locations ran a surplus while others ran short on the very same SKUs.

Goals

One shared, measurable demand forecast per store and SKU; early warning on items heading for a stock-out or an overstock; and a view management can read without asking anyone for a report.

Challenges

Historical sales data was inconsistent between stores, with gaps from periods when the system had changed. Demand was strongly seasonal and affected by promotions that were never recorded systematically. Data had to be cleaned before any model could be trusted.

Our Solution

Cleaning and unifying the data: We pulled sales history from every location into one format, filled the gaps and flagged promotional periods so they would not contaminate the forecast.

Models per SKU and per store: We trained models that account for seasonality, day of week and promotion history, producing a separate forecast for each store/SKU combination.

Alerts instead of reports: The dashboard does not wait to be opened: it flags the SKUs heading for a stock-out or an overstock over the coming weeks.

Replenishment suggestions: For each SKU it proposes an order quantity based on the forecast, the supplier lead time and current stock on hand.

Implementation

We mapped the data sources and agreed what counts as a "clean" sale in each store.

We backtested the models against historical data before production, so the improvement was measurable rather than theoretical.

We piloted on one product category and expanded once the accuracy was confirmed.

We trained store managers to read the alerts and use them in the weekly order.

Results

Next Steps

Connecting to the ERP so suggested orders are created automatically as drafts for approval.

Feeding external factors (public holidays, local events) into the forecasting models.