How StockAI calculates what to order.

No black boxes: models compete on your own history, the one that costs you least to be wrong wins, your supplier’s lead time is learned from your receptions, and the signal rule can be done by hand. This page explains it in plain language.

From your file to the suggested quantity

1. Your file is checked
Before training, every product is examined: missing dates, outliers, insufficient history. If the file would give a wrong forecast, it does not run until you choose how to correct it.
2. Each product is classified by how it sells
Stable, seasonal, intermittent or volatile. That decides which models compete: none you have not enabled is trained, and none that does not fit the pattern.
3. Models compete on your history
Up to nine models: LightGBM, XGBoost, ARIMA, SARIMAX, Prophet, ETS, Croston, an LSTM neural network and a global model that learns from your whole catalogue. Each has the final stretch of your history hidden and is asked to forecast it, in several cuts.
4. The one that costs least to be wrong wins
The winner is chosen with a cost where running out weighs three times more than having extra. And it has to beat two naive forecasts (repeat the last value, repeat the previous season).
5. The winner is retrained on everything
It is retrained on the full history, your latest weeks included, and projects demand with a probable range around it.
6. It is crossed with your stock and your supplier
Forecast demand, stock on hand, what is on its way and the supplier’s lead time give the cover, the reorder point, the signal and the suggested quantity.

How each supplier’s lead time is learned

At first you use the lead time you type on the supplier’s card. Each time you log a reception, full or partial, StockAI stores how many days really passed between the order and the delivery.

From that supplier’s third reception, the reorder point uses the real average instead of the declared one, and the screen shows which of the two it is using. The safety cushion also grows when the supplier is irregular in its deliveries.

How accuracy is shown

Error is measured on dates the model did not see
What you see on the forecast screen is how far off each product was on the hidden stretches, not a number computed on what it had already learned.
Each product, separately
There is no single percentage for the whole catalogue: each product carries its own error and its own range, so you know how far to trust its suggested quantity.
Against what actually sold
When you upload new sales you can compare the forecast you had published with what sold afterwards.
No marketing accuracy percentage
We do not publish a hit-rate figure on this site: it would depend on your data, and the one you see is yours.

The limits, said up front

  • A product with fewer than 20 periods of history is left out of the forecast and shows as SIN DATOS: no number is invented for it.
  • The forecast range is not a guarantee: it is the band where sales will probably fall.
  • It does not guess what never happened: a new promotion or a new large customer is tested in the scenario simulator.
  • What decides the signal’s colour is a published rule, not the model: AI forecasts, the rule decides.
  • It does not buy for you: it suggests what, how much and from whom; you review, adjust and send the order.

For the full technical detail (metrics, validation, routing rules) there is a technical manual and a user manual in PDF.

See the full ruleDownload the manual (PDF)

Keep reading

Put your own AI to work forecasting your sales. Today.

Three ways to start, depending on how much you want to commit right now. All three lead to the same product.

Create your account

With the whole engine. Upload your sales file and see your first forecast, and with it your first list of what to order.

Create my account

Look before giving your email

An instant trial account with sample data. It lasts 24 hours and is then erased.

Try it without signing up

Talk to us

For the corporate plan, the source code, or to see it with your data and someone beside you. We answer within 24 hours.

Message us on WhatsAppEmail us

Sales and contact: contacto@stockai.es. Phone: +506 7186 2820.

Sales forecasting made in Costa Rica, for companies across Latin America.