+38%
Checkout conversion
12M
Daily predictions
−22%
Wasted impressions
4 mo
To production
The challenge
Rich data, generic experience.
A retail media network sat on years of behavioural data but served every shopper the same shelf. Merchandising rules had grown into thousands of hand-tuned conditions nobody dared touch.
Our approach
Models where rules used to be.
We consolidated event data into a single feature store, then shipped ranking and recommendation models behind a service the storefront could call in under 40ms. Every model change ran as an experiment with margin, not clicks, as the success metric.
The results
Relevance that pays for itself.
Checkout conversion rose 38% and wasted impressions fell by a fifth. The system now serves twelve million predictions a day and merchandisers steer it with guardrails instead of rules.
“
Sharp strategy, beautiful execution, and numbers that held up in the boardroom.
Lena Fischer
CMO, Lumen Health