As an international home improvement company with over 2,000 stores across Europe, Kingfisher plc helps make better homes and better lives for everyone. Under our banner brands, B&Q, Screwfix, Castorama, Brico Dépôt, and Koçtaş, we provide a wide range of products, services, and customer support to help professionals and consumers complete any home improvement project.
To do this, we use leading technology to provide a balance between speed and choice. Professionals know what products they want, so our goal for those customers is to deliver those items as quickly as possible. For DIY consumers working on their own homes, it’s more about helping them make informed decisions.
For years, we had been building various AI microservices to solve individual problems, such as product recommendations, churn prediction, demand forecasting, and more. But the development times for each of these microservices slowed our time to market and made it difficult to scale up the number and complexity of these services.
We wanted a way to bring these AI solutions together easily, in a secure way, to build more sophisticated solutions, without having to build every service from scratch. That’s why we built our orchestration framework, Athena. Built on Google Cloud, Athena enables us to quickly build sophisticated generative AI solutions by automatically selecting the right microservice to answer a specific user query. And because security is built into Athena’s processes, the platform improves the safety of the AI services we build.
With Athena, we have significantly reduced our development time. Services that would have taken months to build from scratch now take just a few weeks, reducing our time to market and allowing us to develop secure AI solutions at scale.
Laying the foundation for scalable AI solutions
With its strong history of generative AI development, Google Cloud was the natural choice to meet our needs around scalability, cost efficiency, and performance. And because it offers a mix of bare metal and packaged solutions, it gives our developers the flexibility to build solutions from scratch or use an off-the-shelf solution to build services more quickly.
When large language models (LLMs) emerged, we recognized their power to elevate our service to customers but we didn’t want to use it in isolation. Designed with this in mind, Athena uses Vertex AI to solve two core problems. First, it creates a secure, compliant environment in which to use LLMs. Second, by enabling integration between LLM technology and our AI services, we can provide even more relevant product recommendations and a better customer journey.
We opted for a diversified approach to give us flexibility and adaptability in our solutions so we gave Athena the ability to choose the right AI service and LLM for the particular use case, such as the wide range of models in Model Garden. This agility means we can iterate the backend solutions for each of our AI services as needed, with no negative impact on the output.
For example, the way we describe a product in French could be very different from how we describe it in English. If we want to address a user query in both languages, Athena can evaluate multiple LLMs and pick the right one for each. The end result is more relevant, appropriate content for each product page and a much smoother process for creating it.
Accelerating development with Google Cloud
In March 2022 , we launched our B&Q marketplace, taking the number of different products available from around 40,000 to more than 1 millions in less than two years. While customers benefit from a wider selection of products, it’s important that we make it easy to find the right product for their needs. We introduced our in-house recommendation engine to guide customers towards the right products, and saw our recommendation conversion rate double, with the revenue making up over 10% of B&Q’s total online sales. That recommendation engine is now live at B&Q, Castorama, Screwfix and Brico Depot.
Now we are using Athena to make it even easier for our customers to find products. By combining our recommendation engine and Vertex AI Search with enterprise data and conversational AI, Athena can use text, voice, images, and video to help customers and employees quickly get the information they need, no matter what they’re looking for.
So, if a customer needs to replace a broken piece of their sink but they don’t know what the part is called, Athena can answer their query and perform a visual search of our entire product catalog within seconds. All they have to do is upload a photo of the part and we’ll show them exactly what they need.
Athena helps us internally, too. For example, by combining an LLM with Vertex AI Search, Athena can search hundreds of internal documents and guidelines to help our employees find answers to important questions, such as how many vacation days they are entitled to or how to initiate parental leave.
We used Athena to build Hello Casto, an AI assistant that provides DIY help for Castorama customers. Having built Athena, we are now able to build new production-ready services like this in just two weeks and can use that template to replicate AI assistants across all our banner brands. We’re using that foundation to run a proof-of-concept for FAQ assistants. These chatbots pull information from user manuals and other manufacturer’s documents to answer questions related to specific products, such as the battery life of a drill.
To automate, monitor, and govern all our ML systems, we use Vertex AI Pipelines. This service runs in a serverless manner by using ML pipelines to orchestrate workflows, allowing us to create repeatable processes to continuously retrain our models on the latest production data. We now have 48 pipelines for our recommendation engine, which wouldn’t be possible if we had to build them manually.
Underpinning all of these capabilities is automated compliance and security built directly into the Athena platform. Since everything is in place, we don’t need to add security layers as we build additional AI services or replicate existing ones.
Partnering on sustainable AI growth
Our use of Google Cloud also helps us to use AI in a sustainable way. With sustainability monitoring built into the platform, we can make informed decisions about how we architect our solutions. This will allow us to reduce our carbon footprint while achieving the same latency requirements.
Our Athena framework, built on Google Cloud, has accelerated the use of generative AI, discovery, visual search, and much more across our five brands, allowing us to develop AI solutions faster, more securely, and at scale. We’re excited to see where it will take us next.
