Case study: Kingfisher Group takes DIY approach to AI roll-out across e-commerce sites

AI at the beginning

However, when Ghasempour first joined the company three years ago, Kingfisher knew it wanted to use AI to help achieve its strategic goals, but was still figuring out what role the technology would play in its business.

“When we started, there was no plan in terms of ‘This is how we’re going to use AI’,” he said. “So, the question became ‘How are we going to use it?’”

The answer to that came through trying to address what Ghasempour describes as one of the businesses’ biggest problems: a customer wanting to buy a product online that is no longer in stock.

“It wasn’t an AI problem, it was a product availability issue [that needed solving] that was affecting customer experience,” he said. “At that time, the challenge was ‘How are we going to solve it?’, but we did not necessarily think the answer was in using AI.”

While addressing this challenge, the idea of creating an “alternative product” recommendation algorithm emerged, which Ghasempour said gave way to an exploration of what role AI could play in the process.

“We started investigating how we can use AI when customers are at the point of buying a product that is not available, and how you can recommend a product which is very similar to the product that they’re looking for as an alternative,” he said. “That was the first recommendation service we developed, it went live in early 2023 on [B&Q’s online site] diy.com.”

This service has now been rolled out, in one form or another, across all of Kingfisher’s brands, and since B&Q became the early adopter of the technology, the brand has seen more than 10% of its e-commerce sales originate from product recommendations, according to the company’s own stats.

“From the basic algorithm to solve one problem, today we have 10 different recommendation algorithms that try to help the customer journey in different ways by offering [serving customers information about] frequently bought together products and personalised recommendations,” said Ghasempour.

And the early success achieved from its first forays into building AI-powered recommendation engines allowed the company to take the concept of Powered by Kingfisher even further by providing it with the proof points needed to ditch some of its legacy tech providers, he added.

“We had some legacy recommendation providers on [our]  e-commerce platform, and we started running tests A-B tests against those providers to demonstrate that we can achieve better performance, which justified building [out] this in-house [data and AI] capability even more,” he said.

“We are completely replacing all the third-party providers we used for recommendation engines, so all of that, across all of our e-commerce platforms, is now powered by internal capabilities.”

These capabilities have also been created using Google Cloud’s portfolio of AI tools, with Ghasempour revealing that Kingfisher has partnerships in place with Microsoft and Amazon Web Services (AWS) too.

“Anybody wanting to build any kind of AI capability needs some infrastructure and at Kingfisher we have a partnership with all three cloud providers, but when it comes to AI and data science capability, Google has a bit more of a mature platform, from our point of view,” he said. “It was more intuitive and easier to use, so we started building that capability in Google’s infrastructure.”

Attuned to AI with Athena

Google Cloud’s fully managed development platform, Vertex AI, is playing a foundational role in the delivery of Kingfisher’s AI and data strategy, as it forms the basis of the company’s AI orchestration framework Athena.

Before the introduction of Athena, Kingfisher was effectively setting about addressing individual customer pain points, such as lack of product availability, by creating the AI microservices needed to address these problems from scratch each time.

In Kingfisher’s own words, this way of working resulted in lengthy development times for each microservice, which in turn slowed down the release time for them and caused scalability issues.

What Athena does is allow the Kingfisher team to automatically select the correct, ready-made microservice needed to answer a specific user issue or query, which it claims has cut the development time for new AI services from months to weeks.

“This is a fairly new technology for us, and is probably about a year old,” said Ghasempour. “And the idea behind Athena was, ‘How can we actually build a framework that means we can start to utilise the services in a in a safe and secure way, but also move fast because whoever is using this technology fastest is going to get the competitive advantage?’”

Athena acts as a “wrapper” around existing large language models, such as Google Gemini and Chat GPT, that allows Kingfisher to tap into the respective capabilities of these competing tools at once.   

“Athena can wrap around all of those large language models, and provide a stronger and more powerful service because it can utilise all of those language models at the same time, plus build the security model around them. So, we can we can track all the conversation and we can make sure there is nothing inappropriate happening,” said Ghasempour.

This means Kingfisher can essentially take a “build once, apply everywhere” approach to rolling out AI services across its retail brands.

“You can just do the development once but you can scale it up to more banners [brands] while you’re still secure in the safe environment,” said Ghasempour.

Presently, Kingfisher is using Athena to create services that will make it even easier for the company’s customers to find products using AI-based conversational, image and text searches.

For instance, if a customer does not know the name of the piece of equipment they need to replace on a household item or what the name of a certain tool is, Athena makes it possible for the customer to search the product catalogue for what they need using an image and get a result in seconds.

“All they have to do is upload a photo of the part and we’ll show them exactly what they need,” said Ghasempour.

It is also experimenting with using Athena to moderate the content of the listings published on the marketplace section of diy.com, which allows third-party sellers to sell their home improvement wares online through its website.

“Athena assesses the description of the product to check for any racism or sexism, for example, and offers visual moderation of all the product images,” said Ghasempour.  

Furthermore, the technology is being put to use internally at Kingfisher, to assist its 82,000-strong workforce with finding information about the group’s employment policies and guidelines that are contained within hundreds of internal staff documents.

“In any organisation you have a lot of documentation, from the legal team or HR, that tell staff what the rules of working there are, but people don’t go read the documents. So, at the moment, we’re putting [Athena] on top of those documents, so staff can ask an [internal chatbot] about the maternity leave policy, for example, and get the information they need,” said Ghasempour.

“Over the next couple of months, we’ve got a few more services going live internally to empower our colleagues using this technology to do their day-to-day jobs more efficiently.”