The brief
Northwind’s category managers were drowning in dashboards but starving for direction. Sell-through, basket-mix, and dwell-time signals lived in three different systems, refreshed on three different cadences, and arrived in three different units. The result: every store made its own call, and most calls came too late.
We were asked the question every retailer eventually asks — “can we make the data tell us what to do, instead of just what happened?” The answer needed to fit on a tablet at the back of a shop, not a slide at a head-office meeting.
What we built
A single signal layer that fuses point-of-sale, footfall, weather, and shelf-camera data into one ranked feed of actions. No charts. No dashboards. Just a prioritised list of “do this next” suggestions a store manager can act on in under thirty seconds.
Three principles guided every decision
- One signal per decision — never show two charts when one number will do.
- Action over insight — every screen ends with a button, not a graph.
- Reversible by default — managers can undo any AI suggestion without paperwork.
The first morning the new screens went live, I watched a store lead spot a category drift in eight minutes that would have taken our regional team a week to find. That was the moment we knew.
Sarah Mehta, Director of Retail Operations, Northwind
How it played out
We picked twelve pilot stores across three regions, instrumented them in week one, and shipped the first usable build by the end of week two. The remaining four weeks were spent doing what we always do — sitting next to the people who’d use the thing, watching them try, and shortening the loop between “I noticed something” and “I did something about it.”
By week six, all 312 stores were live. By week ten, the pilot cohort had a clear, measurable edge over the rest of the estate. We rolled the playbook nationally in the following quarter.
The result
A 38% lift in conversion on featured SKUs is the headline, but the number we care about is quieter: out-of-stock incidents fell by 27%. That’s not just margin — that’s trust. Customers who find what they came for, twice in a row, come back a third time.
The system now runs as a permanent part of Northwind’s stack. Every Monday, the model proposes; every Friday, the floor disposes. The conversation between data and decision finally happens in the same room.
