Retail AI needs better product discovery, not simply more choice

The problem is not catalog size

Retailers have spent years expanding assortments, but a larger catalog can make it harder for shoppers to find a suitable product. In an October 16 article, search-platform vendor Vespa argues that the competitive task is shifting from offering more options to making relevant options easier to discover. The company connects that argument to a Kearney consumer study that, according to Vespa, surveyed more than 21,500 people in 20 countries.

The study figures are useful context, but they arrive through Vespa's interpretation. Vespa says 86% of respondents cared more about whether a product worked for them than about its brand. In grocery and meal planning, it reports that 26% would let technology suggest options, while 7% would let it make the decision and purchase. Those results suggest an appetite for assistance without showing that most consumers want autonomous shopping.

What a shopping assistant has to retrieve

A fluent chat interface cannot answer a practical request from language alone. If a shopper asks for a lightweight waterproof jacket below £150 that can arrive by Friday, the system has to translate that intent into product attributes, price, stock and delivery constraints. It then has to rank the eligible products for that particular shopper and moment. A plausible answer therefore depends on search infrastructure and current retail data as much as on the language model producing the response.

Vespa says its platform combines keyword search, vector retrieval—which finds items by semantic similarity—and structured filters with ranking and continuously updated signals. The company contrasts that approach with personalization based mainly on purchase history, broad customer segments or frequently bought-together rules. Historical behavior can still help, but it may not reflect what the shopper needs today.

Customer results are vendor-reported evidence

Vespa points to two customers. It says German marketplace Kleinanzeigen serves more than 32 million monthly users across 58 million listings and uses Vespa for search, recommendations and real-time personalization. It also reports that Vinted moved search for roughly one billion items from Elasticsearch to Vespa and subsequently recorded a 1.1% rise in transactions, a 0.6% increase in gross merchandise value and more than €3.5 million in uplift. These are customer-case figures presented by the vendor, not independent measurements, and the article does not disclose enough detail to generalize them to another retailer.

The practical claim is narrower than “add an AI assistant.” Retailers first need reliable product descriptions, live availability and delivery data, retrieval that can combine text with hard constraints, and ranking that can respond to the current session. A language model can interpret a request and explain results, but it cannot compensate for missing inventory facts or irrelevant candidates. The open questions for any deployment remain its cost, latency and measured effect on customer outcomes.