Proof
Measured against a live holdout. Not against a hunch.
The only number we publish is one a brand can check: pooled experiment traffic against a holdout group of real shoppers, over one measured campaign window.
Apparel brand, paid in-app ad campaign
From July 8 to 16, 2026, Lutiq ran 22 generated page variants against an apparel brand's existing page on live paid traffic. A randomized half of the people stayed on the original page as the holdout.
The pooled 22-page prediction system produced 64% more buyers per person and 55% more revenue per person than the original-page holdout (p = 0.005): 19,606 people, 9,607 on Lutiq pages and 9,999 held out. People are counted with the ad network's own person ID, assigned before the split, so both halves are counted the same way. The comparison is pooled experiment vs holdout, not best single page vs holdout.
No one page owns that lift. The measured result belongs to the pages working together across different consumer journeys, with Lutiq deciding which approved page gets each click.
Method: Holdout Protocol · Methodology · What A/B tests cannot see
Lutiq home · Pricing · Product