Morrisons taps Google’s Gemini AI to solve the supermarket scavenger hunt
Morrisons has rolled out a generative AI-powered tool, backed by Google Cloud and Gemini, that helps customers find products in store, part of its ongoing digital revamp and data infrastructure overhaul, City AM can reveal.
The feature allows users to type in queries for items, ranging from ‘tahini’ to ‘that tomato puree in a tube’, and returns real-time product locations down to aisle level.
It has already processed more than 50,000 daily searches during peak periods.
“We’ve used Google’s foundational models, but we’ve built the application and the way we use the models completely internally”, Peter Laflin, director of data at Morrisons, told City AM.
The system uses Google Cloud’s Vertex AI platform and Gemini large language models (LLMs), but was developed by Morrisons’ in-house data science team.
Laflin claimed that was a deliberate choice following increasing concerns around third-party vulnerabilities.
“We’re in control of it, and that has huge advantages. We’re able to build for scale, so if it’s Christmas or Easter and usage jumps, the system responds”, he said.
Sector shifts and cyber threats
The launch comes as UK grocers accelerate their use of AI, both to streamline operations and drive media revenue.
Sainsbury’s is preparing to roll out its in-house media platform ‘Pollen’, which uses AI to optimise campaigns using Nectar data.
Meanwhile, US retailer Walmart has also invested heavily in similar platforms.
But that innovation comes amid rising cyber risk. In June, Marks & Spencer confirmed that data from some customer accounts had been compromised following a breach at a third-party service provider.
Morrisons itself was affected by last year’s ransomware attack on its logistics partner, Blue Yonder.
Laflin said building internally and hosting the tool securely in-house was ‘non-negotiable’: “We can be confident that our data is secure, it’s always at the forefront of what we do”.
Where can I find…
Laflin told City AM that the idea for the product search tool came from a simple, yet common customer problem.
“The most common question that our colleagues in store get is: ‘Where can I find X?’ So if we can help customers answer that themselves through the app, that’s a win for them — and for our colleagues too”, he claimed.
He said the team focused on everyday pain points, like finding niche items that aren’t always familiar or well signposted in store.
“Tahini is a good example”, he said. “It’s something specific, that’s small, and can be difficult to find”.
The model has been trained to handle typos, brand references, and ambiguous phrases, attempting to return the closest matcges based on each store’s specific layout and inventory.
Laflin also said the system was designed to scale on demand during seasonal trading spikes.
