wavexplain
Pick a product, then hide a driver to see what it was worth.
A sales forecast estimates how much a product will sell next, so a store can order the right amount. This shows why the forecast is what it is.
What if we hide a driver and forecast again?
Unchecking replaces those days with the product's typical baseline and shows the model's real forecast without them. The drop is what that driver was worth.
Opens a Colab notebook. No install, no signup.
Not technical? Let me run it on your dataEmail a short sample (daily sales for a product or two, and which days had a promotion) and I'll send back the explanation card. No install, no code. I only need a small slice and won't share it.
How it works
No forecasting knowledge needed. Here is the whole idea.
Take a product's recent daily sales.
The model predicts the coming days.
Hide the promotion days, or the recent days, and forecast again. The change shows what each part was worth.
Typical pattern + promotion + trend, adding up exactly to the forecast.
wavexplain is a research tool. The toggles above switch between real model forecasts computed ahead of time, not numbers made up in your browser. They show how the model responds to hiding part of the input, not a claim about real-world cause.