AI commerce platform
An AI-native shopping experience — personalised discovery, search and merchandising — across web and mobile, serving millions of users.
The brief
The brand had a capable but generic storefront: the same catalogue for everyone, manual merchandising, and search that returned lists rather than relevance. Conversion plateaued and the team couldn't personalise at the scale of millions of shoppers by hand.
They wanted an AI-native commerce experience — discovery, search and merchandising driven by intelligence — that lifted conversion and held up fast and consistent for millions of users across web and mobile.
- Deliver a personalised, AI-native shopping experience across web and mobile.
- Lift conversion through genuinely relevant discovery and search.
- Scale to millions of users with fast, consistent performance.
- Merchandise intelligently without growing the operations team.
- Unify the experience across channels.
- Measure and improve continuously.
Our AI-native approach
We rebuilt the experience around a recommendation and ranking core: discovery, search and merchandising all read from one product graph and a shared event stream, so every surface is personalised to the shopper's intent. A console lets merchandisers steer the AI with full transparency.
We validated with experiments throughout — every model change measured against conversion before it became the default.
What we built
Personalised discovery
Each shopper sees a storefront ranked to their intent, not a fixed catalogue.
Intelligent search
Relevance-ranked results that understand intent, not just keywords.
Smart merchandising
Automated, signal-driven merchandising with human guardrails.
Cross-channel experience
One consistent, personalised experience across web and mobile.
Merchandiser console
The team steers the AI — pin, set rules, review decisions transparently.
Experimentation
Every change is A/B measured against conversion before rollout.
How we built it
We shipped a unified product graph and event stream first, because personalisation, search and merchandising all depended on clean, shared data. From there a senior pod built the recommendation, ranking and search models in increments.
Every model was placed behind an experiment — measured against conversion on live traffic — so the experience improved on evidence, and performance was engineered to hold sub-second at the scale of millions of users.
How it works
Ingest
Behaviour, catalogue and content stream into a shared event log.
Enrich
Products are embedded and resolved against the product graph.
Rank
Recommendation and search models personalise every surface.
Serve
Personalised storefronts render through a fast, cached layer.
Experiment
Outcomes feed A/B tests; winners become the default.
The recommendation and ranking layer blends behavioural signals with content and image embeddings, so even new products find the right audience. Search understands intent rather than matching keywords, which is what moves conversion.
Everything is experiment-gated: models earn their place against live conversion, not offline metrics alone, and the architecture holds sub-second response at millions of users.
The impact
Conversion rose materially against the previous generic storefront.
The experience served millions of users with sub-second response.
Merchandising scaled through AI without scaling the team.
A consistent, personalised journey ran across web and mobile.
Technology stack
Building something similar?
If this maps to a problem you're facing, tell us what you're building. We'll show you how we'd engineer it — and come back within one business day.