.png)
How a reseller replaced gut feel with 86% forecast accuracy
.png)
+86%
Forecast accuracy, and climbing
Millions
In annual purchasing now sized by the system
5 mo
From raw data to a live system
Ever bought Nike or Adidas gear on Amazon?
Chances are, you've purchased from one of the platform's top 100 Nike and Adidas resellers. We partnered with them to overhaul their inventory forecasting, helping them predict demand more accurately, reduce stockouts, and avoid tying up cash in excess inventory.
Here's what we learned from applying AI to inventory planning at scale, and how the same principles can help businesses of any size make smarter operational decisions.
What we built
The system sits between JWC's sales history and its purchase orders. We built it on PostgreSQL, pulling live marketplace data through the Amazon SP API and forecasting in Darts. Five parts: a pipeline that unifies internal and external data, two applications that size the orders, a clustering layer so products are compared against the right peers, and a governance loop that watches its own accuracy and retrains before it drifts. We left the order itself with the buyer.
Unifies the data before anything models it
Finds the creators who fit the brief
Surfaces the new products worth stocking
Compares products against the right peers
Watches its own accuracy and retrains
The largest recurring decision in the business now runs on a number instead of an instinct. Sales history and market data arrive clean, every product is forecast to size and colour against the right peers, new candidates surface monthly, and the system retrains before it drifts. Accuracy sits at 86% and is still climbing.
Want a forecasting system built like JWC's? Get in touch.
.svg)
.svg)
.svg)
Suggested stories
.png)
.png)
.png)
You’re all set!
.svg)
