Research
Can AI predict what shoppers will buy?
We tested skuel on 36 million real shopping events. Using browsing behavior alone, it predicted which products would sell as accurately as models trained on years of data.
The test
We gave skuel anonymized browsing sessions from a major European e-commerce platform and asked it to predict how often each product would be bought. It saw no product names, descriptions, or images. Only how people browsed.
As accurate as trained models. Far ahead of a raw LLM.
skuel matched the deep neural network and came close to gradient boosting, the best trained model. Used on their own, GPT-5.5 and Claude Fable 5 made roughly four times the error.
Prediction error
Average gap between predicted and actual purchase rate. Shorter is better.
Ranking products by sales
How well each method ordered products from best to worst seller. Longer is better.
Why this matters for pricing. A trained model can only learn from offers you've already run. skuel reasons from behavior, so it can judge a price or promotion your store has never tried.
Whitepaper · 2026
Test pricing before customers see it
- How skuel turns browsing behavior into synthetic experiments
- The full case study, method, and results
- How skuel plans promotions under your goals and limits
- Data, privacy, and integration