AI product recommendations rollout: +21% homepage conversion
Commentary byLiftArchiveOur analysis, not theirsReplacing manually managed product recommendations with AI-driven ones was followed by a 21% rise in homepage conversion rate and a 21% rise in average basket value.
AB Tasty rolled this out for Groupe Équipement de la Maison, the parent of Bricomarché, and it was published as a VWO case study.
This was a product rollout compared before and after, not a split test.
- Homepage conversion rate: +21%.
- Average basket on product pages: +21%.
- Engagement with recommendations: 10% of shoppers interacted with them. No baseline was given, so this is a usage figure, not a lift.
- The team also reported time saved on manual merchandising; no figure was shared.
The change
Manually managed product picks are replaced by AI-personalized recommendations.
Simplified mockup drawn by LiftArchive from the shared screenshots.
Control
Product recommendations were picked and managed by hand across a catalog of more than 250,000 products.
Variant
AI-driven, personalized recommendations appeared on the homepage, product pages and even 404 error pages. The source describes the setup as cookie-free.
Why it likely won
Our reading of the result, not the practitioner's claim.
- With a very large catalog, hand-picked recommendations can't stay relevant for every shopper.
- Showing related items on product pages gives shoppers easy ways to add to the basket.
- Recommendations on 404 pages give lost visitors somewhere to go instead of leaving.
Caveats: This was a vendor product rollout measured before and after, not an A/B test, so seasonality or other changes could explain part of the lift. No sample size, duration or test setup was shared, and the figures are vendor-reported.
Who ran it
AB Tasty ran this test for Bricomarché (Groupe Équipement de la Maison) and published the results (VWO case study). The numbers are theirs; the summary and commentary on this page are ours.
Read the original case study on AB Tasty →
How to run this test on your store
- Hypothesis
- If we replace hand-picked recommendations with personalized ones on the homepage and product pages, then conversion and basket value will rise because shoppers see items more relevant to them.
- Primary metric
- Revenue per visitor
- Guardrails
- Homepage conversion rate
- Page load time
- Average order value
- Who should test it
- Retailers with large catalogs where manual merchandising can't keep up. Small stores with a few dozen products will see less benefit. Run it as a split test rather than a full rollout so you can measure the real effect.
Questions
Do AI product recommendations increase sales?
They can, especially on large catalogs. In this case the gains came from a before-and-after rollout, so a split test is the best way to confirm the effect on your own site.
Where should product recommendations go?
This rollout used the homepage, product pages and 404 pages.
Product pages tend to drive basket size, while homepage and 404 placements help visitors find something to click.
Why measure a rollout with an A/B test?
Before-and-after numbers mix the change with seasonality, promotions and traffic shifts.
A split test isolates the effect of the recommendations.
Similar tests
@oliverkenyon
OptiMonkOptiMonk
@kasper_dolk@kasper_dolk
@joshtCRO