Over the past year we’ve been thinking a lot about AI at MBS. In fact, we wrote a column back in February that explored how impactful agentic AI would be on the consumer sectors, and now, six months later, we can see that is coming true around the world. However, an area where we’ve been seeing this really explode is grocery.
In the US, Walmart has built several grocery assistants over the past eighteen months, and of those, Sparky has seen the most success. The way the tool works is by replacing the search bar with a conversation, which then allows the shopper to describe what they are trying to do. From that conversation, Sparky assembles the products rather than returning a list of results. This way of interaction with the customer has seen real success, as those who use it put 35 per cent more in their baskets than those who do not, plus, the assistant’s weekly users have more than doubled over the past six months.
These are significant developments in a part of retail whose process has stayed remarkably consistent, however, there have been changes which have revolutionised the physical process of the supermarket shop.
The same trajectory towards agentic AI in grocery is happening across the globe. In July, Pick n Pay, one of South Africa’s largest grocery chains launched Penny. Rather than a conversation, the user photographs a handwritten list or asks for a recipe, and then Penny finds the products and fills the basket. Also in South Africa, Checkers in-house built Pixie works the other way round, drawing on Xtra Savings loyalty data to predict what a household is due to reorder proactively.
These are significant developments in a part of retail whose process has stayed remarkably consistent, however, there have been changes which have revolutionised the physical process of the supermarket shop. After the second world war, self-service arrived in British supermarkets through the trolley, whilst the barcode, which was first scanned in an Ohio supermarket in 1974, made the checkout quick and gave grocers their first real view of what was selling. More recently, online ordering, and then click and collect, moved the shop out of physical retail. However, the one thing that remained through all of these developments was that the customer was always making the decisions.
Until now, building an assistant such as Olive – Woolworth Australia’s own agent which plans a week of meals, reads photographed recipes and suggests cheaper swaps to fit into a budget – has needed Walmart-scale engineering, but that is about to become more accessible. Earlier this month Anthropic open-sourced a commerce agent blueprint, developed with Visa, Mastercard, Shopify and Priceline, which is working code a retailer can take and adapt rather than a system it has to design from nothing. This development turns a multi-year programme into something a competent technology team can build in months.
Yet, what customers say they want pulls in two directions. Recent research from PSE Consulting found that some 74 per cent wanted an assistant able to compare across retailers, against 10 per cent who wanted one tied to a single shop. Work by dunnhumby found that 69 per cent trusted recommendations drawn from their own past purchases, well ahead of the 30 per cent who trusted suggestions based on what similar shoppers buy.
But despite this fragile consumer trust, Leah Weckert, Chief Executive Officer of Coles, told an AFR AI Summit that she expects agents to place up to 30 per cent of the grocer’s online orders within five years, which is significant, but not unimaginable given the developments we are already seeing across the sector.
Interestingly though, despite British grocers holding more purchase history than almost any of the retailers already offering these tools, the UK market in this space is only just beginning, with the Institute of Grocery Distribution putting the share of UK shoppers using AI for their grocery shop at just three per cent. For the British customer, the question is one of habit. Close to nine in ten belong to at least one loyalty scheme and the average household carries more than three cards, yet Kantar found households still visited just under five different grocers in February 2025. The question here is what will an AI assistant mean for shopping at multiple supermarkets if one has more accurate data, and is more helpful than the others?
Work by dunnhumby found that 69 per cent trusted recommendations drawn from their own past purchases, well ahead of the 30 per cent who trusted suggestions based on what similar shoppers buy.
In a clear step towards these sorts of developments, this month Tesco appointed Lucy Cooper – who was previously at THG Ingenuity – as its first Group Customer Experience and Agentic Director. Announcing her move on LinkedIn, she wrote that commerce now turns on “what happens when AI starts shopping for them”. This appointment followed a beta trial in April that put a meal-planning assistant – built with Tomoro AI on Clubcard data – in front of 280,000 Tesco staff, and a three-year agreement with Mistral AI signed in December.
Roles of this kind are appearing elsewhere. American retail corporation, Kroger created a Chief Data and AI Officer position in March, with a brief to build agentic capability across the company and a reporting line to the Chief Digital and Technology Officer. These sorts of appointments have the same ring as the genesis of the Chief Digital Officer who began as a specialist hire brought in to handle something the rest of the business did not yet understand. Now, they are a regular fixture on any executive team.
There is a lot currently happening with AI across all sectors, but this development in grocery is one that feels extra pertinent. The introduction of Chief AI Officers feels inevitable as the technology continues to improve and customers adapt their ways of shopping with it. AI agents could be revolutionary for the grocery landscape, and it’s a shift that will only continue to grow and grow.
