Two years ago, DIY giant Kingfisher made the decision to build its own in-house AI capabilities to power the business.
Leading the venture is group AI director Mohsen Ghasempour and his rapidly growing team of 30 people, who “can make sure we deliver the end-to-end solution”.
The department is in growth mode and is working on “at least 20-plus” new AI initiatives this year alone, with Ghasempour explaining that a lot of the projects involve using “generative AI internally to improve our efficiency”.
“We are working on a lot of things, and I don’t see that slowing down anytime soon. Next year, especially, there is going to be a lot of new AI initiatives that we’re going to work on,” he shares.
Ghasempour explains his team’s efforts have been focused on creating initiatives that improve customer experience and internal operations – many of which have helped to boost sales under each of the group’s DIY banners.
Mohsen Ghasempour, Group AI director at Kingfisher
Creating the framework
Kingfisher has opted to do all the work in-house, which enables it to “really fine tune” the technology to the DIY industry and make the sector more accessible to attract new customers.
Most of its AI developments are built on the company’s AI orchestration framework called Athena, which is named after the Greek goddess who is known for her ability to strategise and problem-solve.
To put it into perspective, Ghasempour explains that Athena brings all the company’s AI solutions together in one place – as if all the initiatives are different pieces of Lego, and Athena is the bigger, more complex structure.
The framework manages prompting and interaction across a range of large language models which, as well as the AI tools, are developed in-house.
This avoids the risk that ChatGPT-type applications hold when put in front of customers, as they can provide incorrect information or can be bias or inappropriate.
Kingfisher’s framework makes it “more secure” for customers as it is based on the product data or articles the DIY retailer already , says Ghasempour.
“That’s one of the big advantage of using Athena and the second one is it significantly reduces go-to-market time,” he adds.
Ghasempour says his team spent around three months building Athena but now has the general framework to deploy any application based on a large language model, which means that 80% of the job is already handled for new projects.
AI assistants and product recommendations
So how has Kingfisher used this AI?
One of the Ghasempour’s proudest projects is Hello Casto, the first AI-powered assistant in the home improvement sector that uses generative AI to help customers with their DIY projects.
Customers can use the online tool, which is currently available on its Castorama France banner, in their “natural language” to ask DIY related questions that they might typically ask a sales assistant in store.
“When you talk to Hello Casto, you explain your problem, ‘I want to remove my old wallpaper’, and then our agent will tell you ‘you need wallpaper remover, you need this part of product, you need this and you need that to do your project’.”
“It’s quite new and we have to educate both internally and people to use this technology, but it’s one that’s potentially going to have a very big impact on the younger generation, especially those getting into DIY because we’re providing that level of advice digitally to our consumer”, says Ghasempour.
He notes that thanks to Athena, “it was only a couple of weeks to get the first version [of Hello Casto] live on the website and that timeline is going to be even shorter when we want to talk about diy.com [B&Q’s website]”.
Ghasempour says the team is already working on launching an English version of the assistant for its B&Q and Screwfix banners, which should happen this year.
Product recommendations are also a key area as Ghasempour notes “personalisation is definitely one of our big pillars”.
Last year, Kingfisher launched a suite of ‘best product’ solutions for its app and online, as well as in multiple formats such as ‘frequently bought together’ carousels, ‘substitute products’ or direct personalised offers based on customer shopping trends and preferences.
For B&Q, where the solutions have been in place the longest, more than 10% of its ecommerce sales had come from product recommendations.
On top of that, its own recommendation engine had driven a more than 100% increase in online sales from product recommendations compared to its previous third-party solution.
Ghasempour explains it use machine learning technology, which has been trained from the group’s product data and sales history, to deliver the right recommendations.
The algorithm can also spotlight a very similar product if the one it would typically recommend is out of stock – something its previous third-party solution was unable to do.
Powering the Kingfisher business
Much of Ghasempour’s job is launching new initiatives that “both internally impacts our operations, but also brings a better service to our customer”, he explains.
“At the moment, we are using a lot of different types of AI technology to improve the efficiency of our colleagues through different ways of identifying what they do day-to-day,” he says.
Ghasempour notes the potential to use AI in the supply chain is huge, as there is “so much manual effort going into that process”.
At the moment, Kingfisher uses AI for demand forecasting – predicting the quantity of products consumers will want – as well as working out how to clear stock in a more beneficial way for the business.
“About 12 to 18 months ago, we started a big project around mark down and promotions for B&Q [looking at] using technology to identify the best route to mark down a product,” he says.
“When you want to mark down or get rid of a product, it’s very important for you to understand discount you can put on the product and what period you have to leave it for.”
The pilot tested in the first half of last year delivered encouraging gross margin improvements, while also increasing the sell-through of seasonal stock and improving the efficiency of range changes.
“Similar technology is also being used for promotions, which combines demand forecasting and optimisation to “provide the best price for our customers”, says Ghasempour.
The AI tool, which was first tested at B&Q, will be introduced to Castorama France and Castorama Poland this year.
Much of the projects Ghasempour and his team are working on are adapting the AI tools to be introduced across the rest of the Kingfisher banners.
Kingfisher is not only a DIY giant, it is quickly turning itself into an AI powerhouse too.
Click here to sign up to Retail Gazette‘s free daily email newsletter
