Tesco's App Assistant Plans the Meal, Then the Basket

Tesco's AI assistant, in the trial announced on 9 April 2026, plans a meal and then builds a shopping basket in the Tesco app. It is a customer feature. Colleagues are the beta testers.

The Tesco PLC release calls it a large-scale colleague trial. About 280,000 colleagues get early access to a beta in the app, so they can shape the assistant before a wider customer launch. The release said that customer launch would follow later in 2026. This article does not confirm whether that has happened.

What the release says the assistant does:

• Meal planning: a two-way conversation, recipe ideas, and the shopper's dietary preferences. One stated use is cooking from ingredients already in the house.
• The basket: once a recipe is chosen, the assistant finds products in the app and uses previous shopping history and preferences.
• The stated aim: less time, less stress, less money spent and less food waste. Tesco attaches no figure to any of the four.

Ken Murphy, Tesco's CEO, said colleagues are the right people to shape the assistant, and he tied the project to saving customers time and money. Colleagues will also help name it. Tesco's in-house app, data science and engineering teams built it with Tomoro AI, a UK consultancy founded in 2023 and working with OpenAI. The release says that work ran under secrecy protocols from autumn 2025. Tesco also cites a three-year partnership with Mistral, and says its technology team has doubled in five years. Clubcard is the example Tesco gives of AI it already uses for personalisation.

None of that is a PIM announcement. The product-data consequence sits in the two steps Tesco did describe. A recipe that respects a diet only works when the diet tag on the product is filled in and means the same thing as the preference the shopper stated. A basket only works when the ingredient maps to a SKU the customer can buy. Tesco has not published the attribute model, the failure rate, or what the assistant does when the tag is empty. The trial names the job. It does not show that the data passed.

The colleague beta is a large panel, 280,000 people, and it will surface bad matches before customers see them. That kind of feedback names features. It does not fill a missing diet attribute. Someone still owns the tag.

AI-ready product data is what a shopping assistant has to find in the record before it plans or recommends. Product data quality and governance is who owns the diet and ingredient attributes when the assistant and the shelf disagree. The commerce stack separates the app the customer sees from the product record that has to answer it. Where ERP stops and PIM starts is the same split for attributes a transaction system was never meant to hold.