'Insurers are the original big data companies,' says AXA UK data chief Paul Hollands

GenAI has created a step change in data perception

Paul Hollands, Chief Data & Analytics Officer for AXA UK explains how the company perspective on data has shifted, talks about some GenAI case studies and explains how to avoid data perfection becoming the enemy of progress.

Unless you happen to work for AXA, the chances are that it's a considerably larger organisation than you you may have realised. AXA is present in more than 60 countries and has approximately 108 million customers. The group employs around 145,000 people around the world.

Paul Hollands is Chief Data & Analytics Officer, AXA UK. Speaking to Dael Williamson, Field CTO at Databricks during the London leg of the Databricks Data + AI World tour, Hollands emphasised the extent to which data and AI are absolutely core to AXA strategy in the next five years. Many businesses in that market would say the same thing. As Hollands says:

"Insurance based businesses, along with banks are probably the original big data companies."

How is that translating operationally? Specifically, what impact has generative AI (genAI) had on AXA customers and those providing customer service?

"GenAI has created a step change in how business stakeholders think about data and AI," says Hollands.

"I think a lot of organisations have gone through this phase in the last few years of talking about how to build the foundation, putting technology in place, but we're now seeing the pivot very quickly to working out how we deliver value. The value story is front and centre."

Hollands points out that, whilst there is a cost involved in building models, the step change created by genAI also means that there's a cost to not building them. That opportunity cost balance has shifted very quickly - which is a challenge in an industry not renowned for being at the bleeding edge of technology.

"The insurance industry has been slower paced at adopting change so how do we best accelerate that?" asks Hollands. "We have a moral obligation to our customers to provide a great service and great outcomes. We've also got a fantastic set of colleagues who we need to support with data and AI tooling as well. So what we're trying to think about is use cases. What's the value story that we can drive within AXA which allows us to make that fundamental step change in terms of how we serve customers, support colleagues and ultimately drive value for the business."

AXA began working with Databricks in 2017. The company wanted to leverage their data and machine learning but had a disparate set of platforms which didn't scale. Databricks provided the desired unified platform which enabled AXA to go from data engineering to data science in a full stack solution.

"That one stop shop has given us the ability to go from feature to model really effectively," says Hollands.

Data democratisation and governance in a federated model

AXA UK & I is comprised of four individual units - retail insurance, commercial insurance, healthcare and a combination of all of those in Ireland. Each of those businesses is used to considerable autonomy. Autonomy and consistency aren't always easy to reconcile. Hollands explains:

"We have a very federated delivery model. My role is to bring together the narrative over the top around the role of data, the use cases and the value we generate. The execution sits within the four business units. It's important for us to give each unit autonomy, but also consistency and the support in terms of how they execute. Within the data ingestion team, we've adopted that domain model of how we ingest and how we allow work and prioritisation."

It's proven a successful model.

"In some business areas productivity has gone up seven times this year in terms of throughput and pace because of the adoption of a domain model and the individual prioritisation that each of the business units can engage with."

Given the nature of AXAs business, data governance is extremely important. However, Hollands is wary of letting perfection be the enemy of progress.

"Data governance is vital because it allows you to understand the condition and the state of the data that you're using at any point in time. You probably aren't at a perfect starting point but if you understand the context of what you're using, you can still do something with it. But people have to understand the conditions and implications of that.

"So, we've taken the first step change in data ownership and business owners taking that data seriously and really trying to lean into that. We're now working on the next stage and determining the role of business owners in access to data and how we start moving to greater democratisation data using that ownership model."

The second big piece is AI. We want to put more data products and models to live. So the question is what's the role of model governance within that? What does model governance actually mean? The idea of model risk in financial services isn't that new but as I look at some of the regulation and emerging frameworks that are coming out, what I see is real rigour around how to validate a model, how to peer review it and drive that forward. It is absolutely central to our thinking about model risk and AI framework and AI governance piece for us too"

GenAI early wins

AXA have scored some early wins with genAI, building on the machine learning that's been central to its pricing for some time. Hollands explains:

"The first use case is in our retail business. We've built a genAI model over our policy information, which allows customer service advisors to put a question into a window and it returns all the policy information. It's about augmenting their capabilities not replacing them and providing better outcomes for customers.

"Another use case involves the risk assessment of RAAC. [This is the ‘bubbly' concrete which was used extensively in the construction of public buildings from the 1950s until the late 1980s which has turned out to have a much shorter life span than first thought.] Commercial customers have survey reports and risk reports and we were in a matter of hours able to review 70,000 documents and hone that down to a small shortlist and then a number of interactions. We had to help clients understand where there was a risk and that we could help them. It's a great story about how you can be responsive, proactive and really help customers understand the extent of a risk and how we can help. It took about a week from having the idea to actual execution and contacting businesses."

The RAAC risk assessment sounds like a textbook case of using genAI to solve a problem rather than seeking problems to fit a solution - "bright shiny thingitus" as Hollands calls it. However, speaking with Computing, Hollands admitted that that these use cases provided an enticing opportunity to give genAI a spin.

"On that first use case we did want to try genAI out and see if it could provide a solution. We probably found a problem for the solution as much as finding a solution for the problem. But as important as whether we could technically leverage it, was finding out it's about how colleagues would respond to it, connect with it and use it day-to-day. They're the co-pilots to these solutions and it's really about understanding how people interact and how humans respond to these technologies."

The importance of people

Hollands is emphatic about the interaction of humans with AI because it's the people who work for a company that are the differentiator.

"Everyone's got access to the same tools and technology he says. The differentiator is your people. How do you help stakeholders become great stakeholders? How can you help them understand this new dimensionality of the world we're moving into? What does that mean for them? How do they run a business in that world?

"The second piece is in the people executing this work. How do you help them come up with that capability, sophistication and scale? How do you help those scientists, data engineers and ML engineers, prompt engineers. How do we keep investing in those roles? How do we help enrich their careers, build their capabilities and help them succeed?"