AXA UK provides competitive car insurance quotes for customers via online aggregator sites, and the company needs the agility to respond to market conditions and quickly change its pricing algorithms. So, it’s moving to a Microsoft Azure platform as a service (PaaS) cloud environment that uses managed endpoints in Azure Machine Learning to automate deployment of new pricing models. AXA looks forward to staying ahead of the market, innovating, and delivering more value to its customers.
“We make it our mission to try new ideas and go beyond to differentiate AXA UK from other insurers. We see managed endpoints in Azure Machine Learning as a key enabler for our digital ambition.”
Nic Bourven, Chief Information Officer, AXA Insurance UK
The online quote race
Online aggregator sites—like Expedia, Shopify, and specialized aggregators for mortgages, insurance, and other services—entice us with the promise that we can compare and save. In just seconds, we can find the best deals on vacation travel, appliances, a home loan, and everything in between.
It certainly makes it easy to buy car insurance. But insurance companies have to work hard behind the scenes to pull it off. Turning around millions of quotes that are competitive and accurate and deliver real insurance value—in less than half a second each—takes significant effort.
As one of the largest car insurance companies in the United Kingdom, AXA provides 30 million car insurance quotes to aggregator and broker sites every day. But the digital marketplace doesn’t stand still for long. The pricing group at AXA realized they need to streamline the way they manage online quotes. They need a solution that can help them speed up deployment of new pricing models—one that won’t sacrifice application performance or resilience.
Ticking the boxes
In 2020, the AXA data science team was working hard to figure out how to build a more flexible framework and shorten the time it took to get new models into production. The team heard about managed endpoints, a new capability in Microsoft Azure Machine Learning that AXA could use to automate the deployment of its models, and was eager to give it a try.
“Managed endpoints in Azure Machine Learning just ticked all the boxes,” recalls Tom Snowdon, Head of Data Science at AXA Insurance UK. Snowdon and his team know that if AXA can automate its deployments, it can dedicate more resources to building and testing new models. That, in turn, means AXA can deliver more precise online quotes and better value to its customers.
“We make it our mission to try new ideas and go beyond to differentiate AXA UK from other insurers,” adds Nic Bourven, Chief Information Officer at AXA Insurance UK. “We see managed endpoints in Azure Machine Learning as a key enabler for our digital ambition.”
Agility, performance, reliability
To use managed endpoints in Azure Machine Learning, the company needs to switch from its current infrastructure to using Azure Machine Learning in an Azure platform as a service (PaaS) environment in the cloud. The data science team adopted managed endpoints during private preview and worked with Microsoft to test and validate the models. The system must be able to handle 400 quotes per second and deliver quotes within a lightning-fast response time of 200 milliseconds. Snowdon and his team had to be certain that the quote generator could meet or beat its current performance benchmarks using Azure PaaS resources and deploying its models via managed endpoints.
The AXA engineers tested models built and deployed directly in managed endpoints. They also tested ONNX open-source models deployed through endpoints. Overall, they optimized performance and shaved response times down to as low as 130 milliseconds—a reduction of around 60 percent.
“We’re looking to do as much as we can in open source and push the innovation boundaries to build better, more competitive models,” notes Snowdon. “Azure Machine Learning gives us that access to open source, with a really fluid way to deploy.”
Focus on innovation, deliver more value
AXA is not yet running these new models in production, but with validation and testing complete, the company is set to go live with its first use case in early 2022. Bourven and Snowdon foresee a future where the company’s engineers will have more time to focus their efforts on building new, innovative models that bring more value to customers.
AXA is already exploring other Azure Machine Learning resources, such as machine learning operations, along with responsible AI. The company intends to use those technologies to better understand its customers, offer them customized products, and streamline customer journeys.
“Insurance quotes are not just about price,” comments Bourven. “We can use Azure Machine Learning to help us deliver value, relevance, and personalization to the customer.”
The road ahead
By using managed endpoints in Azure Machine Learning, the pricing team hopes to establish a more efficient and agile process. The company sees this as a strategic driver for sustainable growth and plans to scale the process across the business. “With managed endpoints, our data science team will be able to experiment a lot more and see results materializing much faster than they did in the past,” says Bourven. ”That’s a massive plus for our customers.”
“Good systems are key to a stable, innovative platform,” Snowdon adds, “and we think that Azure Machine Learning will become a big part of our ability to deliver value to our customers.”
As for the road ahead? Bourven sums it up, ”Once we have managed endpoints and Azure Machine Learning working for us, we’ll have fewer constraints. This is really the beginning of the next phase of our digital journey.”
“Managed endpoints in Azure Machine Learning just ticked all the boxes.”
Tom Snowdon, Head of Data Science, AXA Insurance UK
Follow Microsoft