CIOs agree: Agents have left the lab and are doing the shopping
'Nobody wants to buy a drill, they want to put up a shelf’
Agentic AI is arriving at a perfect storm moment in terms of costs, but how are companies actually using them in production?
Costs are in the spotlight because agentic AI is no longer limited to PoCs and experiments – Google’s customers, speaking on a panel at Cloud Summit London, are already putting systems into production, with an overarching focus on outcomes.
Rather than making it easier to buy a single product, for example, agents are focusing on customers’ end goal and helping them to reach it.
For ecommerce firm THG Ingenuity, that’s in the form of the AI Shopping Assistant, which it’s piloting with MyProtein. This is a chatbot/shopping adviser that helps customers navigate large, overwhelming product catalogues to find the right one for them via hyper personalisation.
CTO Jo Drake said, “If you go to MyProtein as a first-time customer, everyone's telling me you've got to shove protein into every orifice possible, you've got to take electrolytes, I've got to do all of these things, but it's so many different types [of product] and I don't know what to do. So, it’s really guiding people through the experience.”
THG will scale the Assistant by offering it for download on Google Marketplace within the next 2-3 weeks. Anyone with a retail website – including non-THG customers – will be able to download it and apply it to their platform quickly and easily.
Matalan, we learned at the event, will be going live with the Assistant this week. Other major retailers like Debenhams were also showing interest when we visited THG’s stand on the show floor – and the staff spun up a demo for the company’s website while we watched.
Mohsen Ghasempour, chief artificial intelligence officer (CAIO) of Kingfisher PLC, which owns brands like B&Q and Screwfix, said his company has replaced keyword search with agents that understand the outcome a customer wants.
“Nobody wakes up in the morning and says, ‘I want to buy a drill.’ They want to put up a shelf.”
If that’s the case for DIY, it’s even more true when trying to buy the place to do that DIY in: a new house.
Property search company Rightmove has launched a conversational AI assistant to help users find homes, using data pulled from both images and listings to create hundreds of new filters – so now people can search for “houses with a porch,” or “sash windows” - as well as an online valuation tool for estate agents.
But the people browsing estate agent listings are thinking about their new home, not the (many) steps to take before walking through the front door.
Rightmove CDO Steve Pimblett named just a few: “You’ve got the valuation, you've got viewings, you've got affordability of mortgages, you've got the transaction itself; you've got to sort your utility bills, your energy, when you move in, [and] you want to understand everything about the local area, commute times and cost of ownership.”
By leveraging its previously locked and siloed data estate, Steve said, “we think we can turn [Rightmove] into a home move assistant..., enable partners on the platform to get the right outcomes for them, and we also think we could just make the whole property industry more efficient.”
Data still does the heavy lifting
Rightmove launched its conversational assistant in six weeks. Steve says the way the company has built its data estate was key to that.
“We leverage hub and spoke as an operating model. Hubs provide centralised capabilities and services like BigQuery, Booker, Vertex: they manage the estate, they manage the scalability...and that means [spokes like customer service] can innovate really quickly within their domain and their business use case...
“The hubs develop the guardrails, the platforms, and the access to the data in a self-serve way, and the business division can communicate really quickly.”
Kingfisher has also centralised data in its lakehouse, Nucleus, and is using that to speed time to market:
Mohsen said, “We started our journey a few years ago to build this data platform. The idea was, if you bring all the data to central location, it will help us to launch new products quicker and faster. Since then, we have deployed 50-plus AI services live into production.”
It’s encouraging to see agentic AI rolled out in production at scale, with customers already using the technology to collapse complex customer journeys into outcome-led conversations. But, that puts pressure on data foundations, governance and legacy operating models.
As an IT leader, you should be thinking about how your approach will change in a world where your customer can simply ask for the outcome they want.