Cross-category recommendations: +36% add-to-cart on item pages, +15% on cart pages
Recommending general merchandise to grocery shoppers — a milk frother with the milk — has long been hard; LLM retrieval plus real-time cart context moved production A/B metrics.
Key numbers
- Setting: recommending general merchandise to grocery-focused shoppers (pairing milk with a milk frother) is a long under-explored cross-category problem requiring product associations and real-time cart context.
- Two-stage framework: candidate generation combines co-purchase market basket analysis with an LLM-based approach to find novel item-item associations, followed by a transformer ranker using real-time sequential cart context and optimising engagement signals such as add-to-carts.
- Online result one: LLM-based retrieval on the item page lifted add-to-cart rate by 36%.
- Online result two: the cart-context-based ranker lifted add-to-cart rate by 15% on the cart page; results come from offline analysis plus online A/B tests.
Method and sample
arXiv paper “Grocery to General Merchandise: A Cross-Pollination Recommender using LLMs and Real-Time Cart Context” (2025-09-02): a cross-pollination framework combining basket analysis and LLM retrieval for candidate generation with a transformer ranker over real-time cart context, validated through offline analysis and online A/B tests.
What it means for brands
Retail recommendation competition has narrowed to the next step within a session: being added to the cart on the right page with the right context matters more than static ranking. Product-to-product relationships — bought together, complementary in use — are machine-readable assets.
How SourceSight handles it
For retail and e-commerce projects SourceSight uses purchase and pairing prompts (which to choose, what to buy with it, pricing) to see how AI describes you, while the content factory structures category, specifications and use cases so pairing models have fields to work with.
Limits
Both percentages come from one platform's online A/B tests and depend on category mix and traffic; no cross-platform replication is reported, so they do not transfer to every retailer.
References
Run it on your own project
What the same prompts return across 10 engines is only knowable by running them.
Related research
Home services: 22% of homeowners already use AI to find a provider, and 80% of businesses don't know how to be found
The local services funnel is moving upstream: before a homeowner opens a website or calls, AI has already synthesised reviews, reputation and service descriptions into one answer.
Industries · 2026-10-07Handing appliance scheduling to AI: it captures 96.7–98% of the savings — and still needs a validator
Five LLMs acting as home energy managers: cheap is not the same as safe or feasible, models both invent schedules and explain a constraint correctly while committing an invalid one.