Rolling performance
How it’s doing over time
Forecast accuracy, sell-outs, and waste — against how things ran before the AI.
Forecast accuracy
AI forecastOld way (last week)
Higher is better · % of demand predicted correctly
Waste
Waste rate
7.2%
Thrown away
199
Donated
1435
Share of delivered stock written off each week · baseline ~10%, target ~9.3–9.5%
Counted in pieces (baht values not yet available). Donations are counted separately from thrown-away stock.
Ran out (sold out)
Lower is better · before the AI, ~43% of items sold out
Just right (1–3 left at close)
Higher is better · before the AI, only ~25% landed here
Ordering style
3-month simulationHow much to order trades off running out vs having waste. Every AI style below orders fewer or similar total pieces than today — yet runs out less, by placing stock where demand actually is.
| AI style | Pieces ordered | Sold out |
|---|---|---|
| Leanorder less | 17,004 | 40.1% |
| Careful | 19,559 | 36.3% |
| Balancedrecommended | 22,066 | 29.3% |
| Generousrarely runs out | 25,010 | 20.9% |
For comparison — today, without AI: 26,251 pieces ordered, 43.4%sold out. That’s more pieces than every AI style yet the most sell-outs — because orders aren’t matched to each store & item’s real demand. The opportunity is better placement, not just volume.