At a glance
- Client
- An online fashion and lifestyle retailerName withheld for confidentiality
- Industry
- Retail
- Service
- Artificial Intelligence
- Scope
- Recommendation engine, A/B testing and storefront integration
The challenge
Every visitor saw the same 'trending' and 'bestseller' carousels, whatever they had browsed or bought before. Recommendations rarely matched what a shopper was actually looking for, and they were mostly ignored.
The marketing team knew personalisation could help but had no way to act on the browsing and purchase data sitting in their systems.
Our approach
Unify shopper data
We combined browsing sessions, purchases, returns and catalogue attributes into one clean dataset.
Train recommendation models
We built deep learning models that learn which products go together and which a given shopper is likely to want next.
Test before rollout
New recommendations ran as an A/B test against the old widgets so the business could see the impact in real revenue.
Place them where they help
Recommendations were added to product pages, the cart and post-purchase emails.
The solution
Each shopper now sees products chosen for them: items that complete an outfit on product pages, add-ons in the cart, and relevant new arrivals in follow-up emails.
The models update as stock and behaviour change, so out-of-stock or returned-heavy items drop out of recommendations automatically.
Results
- Recommendations based on each shopper's behaviour, not site-wide bestsellers
- Impact proven through A/B testing before full rollout
- Cross-sell placements on product pages, cart and email
- Out-of-stock items filtered out automatically
Technology used
- Python
- TensorFlow
- Vector search
- A/B testing
- Headless commerce APIs
Facing a similar challenge?
Tell us what you are working on. We will walk you through how we would approach it, with no obligation.