Kroger's chief digital officer, Yael Cossett, went on stage at the Groceryshop conference in Las Vegas last week and summed up the company's entire AI approach in two words: control and flexibility. That sounds simple, but it is actually a smart answer to a question every traditional company is wrestling with right now: how much of your AI should you build yourself, and how much should you rent from outside companies.
Kroger's answer is both. The company has spent more than 20 years building a data science operation, originally meant to personalize its online store, and later expanded to run its supply chain and merchandising. That data is the foundation. On top of it, Kroger plugs in AI models from outside partners, so it can swap in better tools as the technology improves without starting over.
This is playing out in two products already live. At the start of the year, Kroger gave employees an AI assistant for their daily work. Months later, it launched a shopping assistant for customers that plans weekly meals, reads a photo of a handwritten grocery list or a recipe card, and builds a cart around a budget or dietary need. Kroger also recently expanded its work with Instacart to bring a similar AI cart tool to its mobile app, and it is not alone. Every major grocery chain, including Walmart and Albertsons, is racing to ship a version of the same idea.
Two findings from the rollout are worth paying attention to. First, Cossett expected the shopping assistant to mainly attract younger customers. Instead, it appealed equally across every age group, which suggests the appetite for AI shopping tools has less to do with generation and more to do with whether the tool actually removes a real hassle. Second, making shopping faster did not make people browse less. It did the opposite. Once the boring, repetitive parts of the task were handled, shoppers used the extra time and confidence to explore new products.
But Cossett also drew a clear line. Kroger is choosing not to let AI handle automatic refills or recurring orders on customers' behalf, even though the technology could do it today. His reasoning: meal planning is still an emotional task tied to family and identity, and pushing full automation risks breaking the trust that makes shoppers comfortable using AI in the first place.
That restraint is the real lesson here. Any company sitting on years of customer data has a genuine advantage in the AI era, because that data is what makes rented AI models actually useful for your specific customers. But the businesses that win will be the ones that use AI to remove friction, not the ones that use it to remove the customer's sense of control over their own decisions.