A 22-location specialty retailer was managing inventory across 3,400 SKUs using manual buying decisions informed by last year's sales. Seasonal peaks created chronic stockouts on bestsellers while end-of-season markdowns on slow movers eroded margins. The buying team had no visibility into demand signals beyond sales history.
We consolidated sales, POS, weather, and promotional data into a Snowflake warehouse, then built an ML forecasting model using LightGBM that incorporated seasonal patterns, promotional uplift, and weather sensitivity by product category. The model outputs a weekly buying recommendation per SKU that buyers review and override with one click.
The system went live at three pilot locations before rolling out company-wide. Buyers see a recommendation dashboard in Tableau, with confidence intervals and the key drivers behind each forecast. An alerting system flags SKUs tracking above or below forecast mid-week so buyers can adjust orders before the situation becomes a problem.
Gross margin improved by 18 points in the first full quarter. End-of-season markdowns dropped by 32%. Stockout incidents on top-100 SKUs fell from 14% to under 4%. The buying team reports spending 60% less time on routine replenishment decisions and focusing more on new product selection.
Book a free consultation to explore what's possible for your organization.
Book free consultation
