AI forecasts your buyers will actually trust.
Every night Stokk runs five forecasting models for every product at every store and keeps the one that has been most accurate, then Stokk AI reads the numbers against last year’s season and your notes. Seasonal lines, patchy sellers and items that ran out are all handled, and every forecast shows which model it chose.
| Store | Sells a week now | In 8 weeks | On the shelf |
|---|---|---|---|
| Harbour St | 4.1 | 9.8 | 0 |
| Old Town | 3.2 | 7.6 | 2 |
| Riverside | 2.4 | 5.7 | 0 |
A forecast you set up once and never trust again.
Most planners make you pick a method up front and leave it there. Seasonal items overshoot in summer and run out in December. A week out of stock reads as a week nobody wanted it. So the buyer goes back to gut feel.
- One method chosen up front, foreverFive models scored every night, the best kept
- A week out of stock reads as zero demandDays off the shelf don’t count against the item
- One national average for every storeA forecast per product, per store
- Christmas lines spotted in DecemberLast year’s season read in time
- A one-off bulk order skews next monthOdd days left out of the history
- No idea whether the forecast is any goodEvery forecast scored against what sold
Forecasts that earn their keep, item by item.
Proven models chosen automatically for every item, and AI that knows when the history isn’t the whole story.
- 01
Five models, the best kept
Smoothed average, trend, Croston for patchy sellers, an 84-day average and same-week-last-year, scored every night. The most accurate wins, per product per store. - 02
Stockout-aware
Sales are counted over the days an item could actually be bought, so a week on an empty shelf doesn’t drag its forecast down. - 03
Seasons, learned
Each item learns its month-by-month pattern from its own history, or from its category when it sells too little to tell. - 04
Demand patterns
Every item is read as smooth, erratic, intermittent or lumpy, so a now-and-then seller isn’t forecast like a daily one. - 05
ABC classes
A items carry a higher service level and can be reviewed every week. C items carry smaller buffers and don’t soak up cash. - 06
Scored every night
Every forecast is saved and later scored against real sales, so you can see per model and per item how close it got. - 07
New items from day one
A new product gets a forecast the first night it sells. Link it to the item it replaces and it inherits that item’s history. - 08
Clean history
Write-offs, samples, house accounts and one-off bulk sales can be kept out of demand, so the forecast learns from real customers.
Judged on what sold, and told what the history can’t show.
The accuracy report shows which model is winning where. The item file is where your team adds what the numbers don’t know.
Five models run for every item at every store. The most accurate over the last 120 days is kept.
- Smoothed average · Steady sellers41% of items
- Same week last year · Seasonal lines23% of items
- Trend · Growing or fading items14% of items
- Croston · Patchy, now-and-then sellers12% of items
- 84-day average · Everything else10% of items
Orders built on what will sell, not on what sold last month.
- Fewer
- stockouts on the lines that matter
- A items get the highest service level and the closest watch.
- Less
- cash in slow stock
- Patchy sellers are forecast as patchy, so they’re not topped up like daily ones.
- In time
- for the season
- Seasonal lines are read against last year, before the shelf goes empty.
What the way of working changes, stated directionally rather than as audited figures. Results vary with category mix, lead times and the quality of the data in your ERP.
A forecast that picks its own model, with Stokk AI on top.
The engine learns which model fits each item from how accurate it has been, and scores them all again every night. Stokk AI adds the judgement: when you open an order it reads each forecast against thirteen months of sales, last year’s season and your item and supplier files, and changes or questions the numbers a good buyer would.
A stockout, a bad batch, a one-off bulk buyer or an item that replaces another goes in the item’s file once. The forecast drops those days or inherits the old item’s sales, and Stokk AI reads the file on every check.
What people ask about Forecasting.
Do we have to choose the model?
How much history do you need?
What about brand new products?
What if a number looks wrong?
Same stock, same sales, same login. The next module is a switch, not a project.
Order for the season ahead, not the one behind.
Half an hour on a working sample: a seasonal item, the five models and what they mean for this week’s order. If Stokk isn’t right for your shop, we’ll say so on the call.