09/02/2026
What they say… and what they don’t 🤫
(A reality check about AI forecasting in retail)
In retail tech, many claims the same thing:
“Our platform uses AI to forecast at SKU–store level with price, promotions, weather, seasonality, cannibalisation and more.”
Sounds perfect. But here’s what they don’t tell you:
Most platforms do not actually learn these effects at SKU–store level.
Instead, they learn one global or region-level coefficient (price elasticity, promo uplift, weather sensitivity, etc.) and then simply:
👉 apply it top-down,
👉 weight it by store data,
👉 and call the result “SKU–store forecasting.”
The forecast number becomes store-specific —
but the behavior behind the number is not. And that’s a big problem!
Price sensitivity differs dramatically by store type and neighborhood
Promo uplift depends on local habits & competitor intensity
Weather impact is extremely location-specific
Cannibalisation is always local — driven by assortment, shelf proximity, and shopper missions
You can’t solve these with a global elasticity and some multipliers.
You can’t “allocate down” true customer behavior.
At MySales Labs, we chose the harder but correct path:
👉 True bottom-up causal learning — per SKU, per store, per factor.
That means the model actually understands how customers in that store react to: price changes, promotions, weather, assortment shifts, cannibalisation, competition, local events
No shortcuts. No global averages. No fake granularity.
If you want to instantly separate real forecasting platforms from marketing fiction, ask this:
“Do you learn causal impacts per SKU–store, or do you apply top-level coefficients and allocate them down?
Can you show how?”
This question ends the storytelling very quickly. 😎
