A large US retail chain was losing money at both ends: fresh stock thrown out by the crateload every month, and popular basics missing from the shelf exactly when shoppers wanted them. We replaced gut-feel ordering with demand forecasting a buyer still signs off.
Inventory is where a retail chain quietly loses fortunes. This one was losing at both extremes at once — over-ordered fresh stock rotting in the bin, and best-sellers out of stock at peak hours while customers walked to a competitor.
The root was that buying ran on gut feel. Buyers and inventory managers worked off old spreadsheets and ordered by what "felt right" or by what they'd ordered last week, never really analysing the demand in front of them. Double-digit percentages of the most perishable lines — dairy, bakery, fruit and veg — went straight to the bin because too much was ordered to sell in time. And nothing connected the tills in the branches (what's selling right now) to the logistics centre and the suppliers in real time.
And kept a human hand on the order button, so nobody was ordering blind.
Instead of every branch and department working alone, we connected the live till data straight into the chain's existing logistics system — one real-time picture of what's actually selling, where.
We fed the model the external variables that really move a shopping basket — peak days, holidays and weather — so a forecast reflects the week that's coming, not just the week that passed.
Rather than a buyer guessing how many crates to order, the model reads the real-time rate of sale and produces a precise, tailored order recommendation for every product in every branch.
Special weight on the short-life lines — dairy, bakery, produce — where over-ordering means the bin and under-ordering means an empty shelf, and the right quantity is a moving target.
We did not let the AI order blind. Every morning the purchasing manager gets a tidy recommendation list on their phone, reviews it, and approves the order to the supplier in a single tap.
All of it runs against the systems the chain already uses — tills, logistics, suppliers — so it became part of the daily routine rather than another tool bolted on the side.
Within the first quarter, both ends of the loss started to close.
Expired stock going to the bin fell by around 30% in the first quarter alone — a large, recurring monthly loss cut out. Empty-shelf events on popular products dropped by 60%, which lifted turnover directly and left customers happier in-store.
The buyers' daily grind of inventory maths and ordering shrank from about four hours a day to roughly fifteen minutes — a quick review and a tap to approve the system's recommendations, with the rest of the day back for the work that needs a person.
The reason it stuck is that it never took the human out. The AI does the analysis a spreadsheet can't and hands over a clear recommendation; the buyer, who knows the floor, still decides and approves. That balance — sharp forecasting with a person on the order — is what makes it safe to run every day, and it's the same discipline behind our legacy-integration work.
Reef TRH