AI transformation in insurance – Part 2

Insights

Romain Caillet, Strategy Lead for Insurance in EMEA at Accenture, outlined how insurers are rapidly expanding their use of AI and moving beyond isolated use cases toward broader process reinvention. Drawing on a recent Accenture study of 263 insurers worldwide, he highlighted three major trends.

1. Insurers are increasing the pace of their AI adoption but few firms have attained mature implementations. Most are scaling AI across functions rather than confining it to pilots or isolated use cases. About 23% report enterprise-wide integration, up from roughly 10% a year earlier, yet only 1% see themselves as leaders. 
   
   
2. Improving customer experience and retention is now the biggest driver of AI investment, followed by refining underwriting and pricing accuracy, and enhancing claims efficiency and outcomes.
   
   “Insurers are turning more and more to customer-facing AI improvement whereas in the past two years they were looking more at the back office.” — Romain Caillet, Accenture.
   
   
3. AI investment is still focused mainly on improving existing insurance operations rather than creating new business models. Accenture research shows that insurers are achieving measurable gains from end-to-end process reinvention. These included gains of 30% in underwriting, 25% in customer service, 40% in claims handling, and up to 50% in procurement and other support functions.

Caillet added that impact at scale comes from fully reinventing end-to-end processes, not from layering AI onto existing workflows. Insurers need to reimagine the work, reshape the workforce, and redesign the workbench. That involves rethinking processes around business outcomes, using digital twins and data-driven decision-making to redesign workflows, equipping teams with new skills and more iterative ways of working, and building user-friendly platforms with the knowledge, models, agents, and backbone needed to scale AI securely and without adding complexity.

Participants’ polls

Two polls conducted during the roundtable discussion showed that participants are taking a pragmatic approach to adopting AI. Most participants, 43%, are keeping AI development in-house, while 29% are partnering with Big Tech firms. Collaborations with smaller specialists and undecided positions each account for 14%. AI implementation is concentrated most heavily in customer service and engagement functions. All participants are using AI in contact center support, 80% are also applying the technology in customer interaction, while 40% are using it to improve operational efficiency and information technology development. Only 20% of participants are using AI in their marketing and sales.

1. Partnerships and buy versus build

Hybrid sourcing models are becoming the norm. Most insurers favor a mixed approach to AI sourcing rather than a pure buy or pure build model. They want to combine external technology, internal platforms, and in-house teams in ways that preserve flexibility while allowing them to move forward with implementation.

“I see more and more of build on top of buy rather than one or the other.” — Yann Bry, AXA.

Control, compliance, and lock-in shape partnership decisions. Buy-versus-build decisions are heavily influenced by concerns about regulation, data protection, vendor lock-in, and knowledge transfer. Insurers want to ensure that external partnerships do not create dependency, weaken control over sensitive data, or leave internal teams without expertise.

“We build it and we control it ourselves. It’s safer for us in that environment.” — Luka Posedi, Generali Hrvatska.

Insurance-specific capabilities are usually built in-house. Insurers are more likely to build AI applications themselves when use cases are closely tied to core operations, local market conditions, or insurance-specific processes and when they clearly support business objectives. 

“We put a team of data scientists in the business line, on the floor where they learn about the processes of that business, how it works, and what they can add. They then propose a model that we develop. It could be for claims or agentic commerce, but it must have a business case.” — Harry van der Zwan, NN.

Startups and large vendors each have limitations. Startups may offer speed and innovation, but many insurers view them as unsuitable for tightly regulated environments. Large technology vendors offer stronger platforms and resources, but they often lack sufficient understanding of insurance processes and operational complexity. 

“Going with startups is unacceptable due to the GDPR and AI Act. Startups usually don’t address them. On the other hand, we also found that Big Tech companies have great technology and know how their technology works but unfortunately they are not aware of the specifics in our industry.”
— Michal Kozub, UNIQA.

Legacy integration remains a big problem. Connecting new AI tools to legacy systems, fragmented data, and existing workflows remains a major challenge. 

“To unlock value, you need to tap into legacy systems, old data, and data coming from very different sources. Using AI to try to solve that middleware problem is something we’re considering.” — Yann Bry, AXA.

2. Going beyond automation with AI

AI implementations are moving from automation to end-to-end process redesign. Insurers are no longer using AI just to automate individual tasks. They are starting to redesign full processes, especially in claims and customer service.

“We are not just embedding AI large language model capability into our business process. We are now thinking from the customer’s perspective and trying to redefine the end-to-end process.” — David Wei, Ping An.

Decision support promises further value. The next stage of AI adoption is likely to be improved decision-making. In claims, this could include better fraud recovery, faster and more consistent handling, while in customer service AI could help insurers better interpret client needs, recommend solutions, and guide policyholders through complex interactions.

“We are leaping from automation to decision-making and delivering customer results.” — David Wei, Ping An.

AI is beginning to create feedback loops across the insurance value chain. Better claims data and analysis, for example, can feed back into underwriting and pricing, while customer-facing tools can connect multiple services in a single interaction. 

“If you have a data layer that is feeding the workbench for your claims handlers, then you can have much more knowledge about the cause of your claims and that can feed back into your underwriting and pricing models.”— Yann Bry, AXA.

3. Measuring the value of AI

Measuring AI value requires a broader framework than simple efficiency metrics. AI creates value through customer experience, efficiency, growth, loss reduction, and technical performance. Insurers need frameworks that connect adoption and user satisfaction with business outcomes.

“I still believe that it’s important to measure automation, efficiency, and things like that because they unlock opportunities to create value elsewhere.” — Yann Bry, AXA.

Measurement works better when AI is used to solve existing business problems. Value is easier to track when insurers use AI to address an existing problem in functions such as claims handling, fraud detection, or underwriting. By contrast, generic AI initiatives are harder to evaluate because the link between adoption and measurable business value is often weak or delayed.

“It’s super hard to measure exactly what is changing. Often when I talk to my clients about how to see some of these investments, I start with personal productivity.” — Romain Caillet, Accenture.

Small-scale and local AI investments may need to be treated differently from broad transformation programs. Some limited AI initiatives, especially those that support personal productivity or are experiments, are hard to measure and may require insurers to take a “leap of faith” rather than enforce strict return on investment discipline. By contrast, larger end-to-end transformations are expensive and need clearer value targets and stronger business ownership. Similarly, insurers may need to set different metrics for their various local operations as well as their group-level functions. 

“After long discussions at the holding company level we set up a north star target that includes efficiency gains and bottom-line gains. We set the number top down because we couldn't really do the exercise bottom up.” — Nida Gorkem, Ageas.