Cutting dead stock at a 14-store outdoor retailer
Ridgeline Supply was ordering on last season instinct. We built a per-store demand forecast that reorders on actual sell-through.
Buying decisions made twice a year, for a market that moves weekly
Ridgeline placed seasonal orders from a spreadsheet built on the previous year total, with no view of variation between stores. Coastal branches ran out of waterproofs by October while inland stores marked the same stock down in January. Nobody could say which was which until the year closed.
A forecast per store, per category, retrained weekly
We consolidated four years of till data, weather records and promotion calendars into one warehouse, then trained a gradient-boosted model per store-category pair. Backtesting against the last eight seasons set the confidence bands. The output is not a dashboard nobody opens: it writes suggested reorder quantities directly into the purchasing system, with a one-line explanation of what drove each number.
Six weeks to one store, eleven to all fourteen
We launched with the two highest-turnover branches so the buying team could sanity-check the numbers against their own instincts. Where the model disagreed with a buyer, we logged it and reviewed weekly. After four weeks the disagreement rate had dropped enough that the team asked to roll out the rest early.
Markdowns down, availability up
Across the first two seasons, end-of-season markdown volume fell 31% while in-stock rates on core lines improved. The freed working capital went into extending the rental programme, which the forecast now covers as well.
The first month I checked every number against my own gut. By the second month I stopped, and that is the highest compliment I can give a piece of software.
RSHead of Buying, Ridgeline Supply