How long until SKU and store forecasting stops fighting the buyer

Implementing SKU and store forecasting for a 50 to 80 store chain, with a defined scope and data in reasonable shape, takes 14 to 22 weeks. Without those two conditions, it takes however long it takes.

The fight with the buyer does not end when the model ships. It ends when the buyer looks at the week’s number and recognizes the operation in it. That is the criterion, not accuracy.

The realistic range

14 to 22 weeks, with a squad of 3 to 5 people, starting on one category, not the whole chain. The range assumes nobody discovers midway that the sales history does not separate sales from shelf-out.

When they discover it, and they almost always do, add 4 to 8 weeks. Not for the model. For fixing the data.

What “data in reasonable shape” means

This is where most projects live or die. Reasonable means concrete things.

• History of 18 to 24 months, at SKU by store by day granularity.
• Shelf-out flagged in the history. Without it, the model learns censored sales as if it were demand, and under-orders forever.
• A SKU registry with no duplicates and no phantom items.
• A promo calendar on record, not in the buyer’s memory.

If three of the four exist, the range holds. If only two, double the first phase.

The phases, in weeks

Diagnóstico: 2 weeks. It ends with the criterion written and a pilot category chosen.

Implementação: 12 to 20 weeks. The model serves daily forecasts per SKU and store, the buyer validates them against their own instinct, and shelf-out on the pilot category enters the weekly review.

Operação: quarterly, renewable. One improvement per quarter, the internal team running the day to day. The internal team keeps the operation; that is the model’s signature.

What moves the timeline by 50% each way

Four decisions, and you can see all of them coming:

• How many categories at once. One speeds it up; five stall the project.
• Quality of the shelf-out history. It is the single biggest factor.
• Number of integrations with legacy systems. Each one is a negotiation.
• Whether the buyer joins the criterion in week 1, or gets told in week 12.

The week-one criterion

The number that tells you it is working is not aggregate MAPE. It is shelf-out plus shrink on the pilot category, measured on peak days, compared against the week-1 baseline.

Take an AI project on your roadmap right now. In ten seconds, name the three indicators that tell you whether it is working. If you stalled, the project has no criterion yet, it has a hope.

Before choosing between building, buying, or implementing, it is worth reading where each path wins. But the criterion comes before the three.

Send me the three candidate indicators your forecasting project will measure. In one hour I will send back a diagnosis: which predict success, which predict silent failure, and what review trigger each needs. If the project shape confirms, the two-week Diagnóstico ends with a one-page document, three indicators, three ranges, three triggers, that the squad holds for the rest of the Implementação.