Every Monday, the operations team at a 14-store outdoor retailer spent the morning building the same report. Three systems, a dozen exports, one enormous spreadsheet — and by the time it reached store managers, it was already a day old.

Six hours every Monday

Each store manager also built their own version for their store, adding roughly six hours a week per person. Worse, everyone was looking at slightly different numbers, so meetings started with arguments about which spreadsheet was right.

The problem wasn’t effort or skill. It was that the definitions lived in people’s heads. “Sell-through” meant one thing to merchandising and another to finance, and every export quietly baked in a different assumption. More automation on top of that would only have produced wrong numbers faster.

One model, ten numbers

We started not with AI but with agreement. Over two workshops, leadership and managers agreed the ten numbers that actually drive decisions, including:

• Sales against the same week last year, by store and category

• Sell-through rate on the top 50 products

• Days of stock remaining against supplier lead time

• Products selling more than 30% faster than forecast

Then we connected point-of-sale, inventory and supplier data into one model, so those numbers are calculated once, the same way, every night.

A brief that writes itself

On top of the model, an AI step writes a one-page brief for each store every morning: what changed, what to reorder today and what deserves a closer look. It only describes numbers the model has already calculated — it never invents its own.

Managers can reply to the brief with a question — “why is footwear down at Riverside?” — and get an answer drawn from the same data, with the underlying figures linked so they can check it themselves.

A dashboard shows you everything. A good brief tells you the three things that matter before your first coffee.

What changed

Within a quarter, stock-outs fell 34% and each manager got about six hours a week back. Monday meetings now start with decisions instead of reconciliation.

The less visible win was trust. Because every number now comes from one definition, leadership stopped asking where a figure came from — and started asking what to do about it. The whole project took four weeks, and the first brief went out at the end of week three.

Try this first: Before automating any report, ask who reads it and what they decide from it. If nobody can answer, the report probably needs deleting, not automating.