Why culture is at the core of AI strategy at AXA UK

Cali Wood, Head of Data & AI Strategy and Culture at AXA UK, on scaling AI through culture and community

AXA UK’s Head of Data & AI Strategy and Culture talks through some AI use cases and explains how a highly risk averse organisation of 8000 people is inculcating a positive AI culture and using it to implement and scale generative AI to potentially transformative effect.

CIOs will affirm that organisational culture is a crucial factor in the successful implementation and integration of technology. This is especially true for AI and generative AI technologies, where the success or failure of initiatives depends entirely on not just whether the technology is adopted, but how effectively and thoughtfully it is used.

At AXA UK, culture is embedded in the DNA of its data and AI strategy, sitting alongside the creation of value and the connection of data and tooling. Cali Wood, Head of Data & AI Strategy and Culture, makes it clear: culture isn’t an add-on. It’s foundational.

“Culture isn’t something that sits outside of strategy, it’s a core component of the strategy,” she says. “Data culture means different things to different people. At AXA, it’s all about the people.

“You can have the most modern data stack and tooling but if you don’t have the people who can operate it or the people in the business that can run with the output then you’ll always be limited. That’s why culture is a core imperative.”

Use cases

AXA UK isn’t new to AI. One of its most mature use cases, Repair or Replace Intelligence (RoRI), has been guiding motor claims teams for nearly five years—helping the business save millions in indemnity costs by improving decision-making on vehicle repairs.

Another tool triages complex, high value claims early in the process, ensuring they’re routed to the right teams. This not only saves time and money, it improves the customer journey.

“They’re a small number but a big proportion of our claims spend, “Wood explains. “It helps identify them early which means they go to the right teams to manage them which improves the customer experience and settlement speed.

“So far this has been built predominantly on AI but it’s moving to LLMs. Everything we’re doing in the data and AI space links up.”

AXA UK has also used AI to automate a large part of complaints final response letters, using GenAI to produce consistent, case-specific responses—reducing writing time by 40%.

But the real differentiator? Domain expertise is non-negotiable.

“It’s not just about keeping a human in loop. It’s business domain expert in the loop. We can’t create a generative AI product without a domain expert being a part of it.”

This approach was pivotal in developing AI-powered call summarisation. Instead of building in isolation, data scientists and engineers sat shoulder-to-shoulder with call handlers—refining prompts, tuning output, and building trust.

“What we don't want is the data team saying, ‘here's a model, off you go and use it.’ The call handlers were pivotal in refining, fine tuning that model so that they can trust that what we've created has been listening to their feedback. “

Derisking LLMs

Of course, LLMs can engender productivity and efficiency gains, but they also make mistakes. Hallucinations could find their way through the system to underwriters when pricing policies. AXA UK has access to healthcare and financial data and ensuring the privacy and integrity of that data is critically important.

For the sake of compliance, decisions must be transparent – a task rendered more difficult by the ‘black box’ nature of LLMs. Even the companies that built them seem to struggle to explain how they work.

AXA UK has customised its GenAI use to try to remove as much risk as possible. All GenAI usage is routed through Secure GPT—an internal platform designed to keep data safe within AXA’s infrastructure.

What happens when Secure GPT makes a mistake? The answer Wood explains lies in ensuring that users are able to critically evaluate output.

“We have a Secure GPT learning portal,” explains Wood. “It contains content created in house on what GenAI is, prompt engineering 101, what GenAI should and shouldn’t be used for etc. It’s all vetted by our AI governance team and data protection team. The key one is to never assume that what you’ve got is gospel. It’s there to supplement your own knowledge and expertise, and perhaps help with a creative block, but you should never assume that what you have is perfect.”

“We are very risk averse. My job is to get people to use this to improve their day-to-day roles, but we need to make sure that’s balanced with good governance.”

Scaling culture through community

Implementing AI at scale is a cultural challenge—especially when working across multiple data teams in AXA UK’s retail, health, and commercial divisions, alongside a central data hub.

To foster alignment and energy, Wood launched Ignite – AXA UK’s internal data and AI conference.

“In the morning, we did a refresh of strategy, thinking about what we’ve achieved and where we’re going. We hosted a panel of leaders from across the organisation so they could demonstrate how they are embedding data & AI into their strategy.

“In the afternoon we hosted a promptathon. It was brilliant, we created synthetic data for three real-life use cases so people could experiment within a Secure GPT environment with what those solutions might look like. We then did presentations and awards. It was a terrific opportunity to get people together and collaborate.”

In addition to one off events, AXA also launched a Data and AI Academy just over a year ago.

“We’ve had over 37% of the organisation engage with it and 62% of leaders which we’re proud of. The academy is the mechanism with which we share the development and learning we want to get out to the organisation.

“We’re building strong relationships with the business to deliver at pace, and then the collaboration across the teams to operate at scale. We've done this through a few ways, and we've got 2500 people across four communities. We’ve got a Secure GPT community, a Power BI one where we have monthly drop-in clinics. We've got a data community, which is all 300 or so data professionals, and then we've got a data ambassadors community, which is over 1200 people who have actively signed up because they want to be involved. “

AXA UK has also recently announced a GenAI Champion apprenticeship which should embed generative AI expertise more deeply into the business.

By viewing data and AI as strategic imperatives, AXA UK is succeeding at weaving them through every part of the organisation, scaling them through communities of ambassadors and power users.

Culture isn’t the only organisational trait necessary for transformative change – but it is a prerequisite.