Aurelia Retail
/ THE NUMBERS
Recommendations that actually convert.
/ THE PROBLEM
A homepage that looked the same for everyone.
Aurelia's previous personalization stack scored every product against every shopper at request time — and at 8M weekly visitors the cost was prohibitive. They'd quietly degraded to a single "trending now" rail that the merchandising team manually curated each Monday.
The team wanted true per-shopper recommendations back, but only if the unit economics worked at scale.
/ THE SOLUTION
Vector recall, then a cheap ranker.
NeuralPress shipped a two-stage system: a vector recall step pulls a candidate set of ~200 products in under 8ms, then a small ranker scores those for the specific shopper. The total inference budget is 14ms — a fraction of what the old system cost, and personalized.
CTR more than doubled in the first four weeks. Average order value on recommended items lifted 18%. Most importantly, merchandising got their week back.
/ YOUR ORG NEXT?
See what one platform could ship for you.
Talk to a founder, see Aurora live on your data, walk out with a 90-day rollout plan.