From AI adoption to AI control: how can insurers scale with confidence?

A recent Insurance Post roundtable explored how insurers can avoid AI fragmentation, moving from pilots to controlled, enterprise-wide AI that actually delivers value? Vicki Summerhayes reports.

AI adoption is accelerating across the market, but many insurers are struggling with the shift from experimentation to deployment. Progress is being hampered by growing complexity, as firms grapple with integration and oversight.

A recent Insurance Post roundtable, held in partnership with GitLab, explored how insurers can transition from fragmented AI initiatives to a more structured, enterprise-wide approach, scaling AI effectively without losing control.

AI adoption challenges

AI brings unique pressures that sets it apart from previous waves of technology innovation, not just in terms of the speed of change but also the very nature of the transformation. “When GenAI first emerged, the focus was on productivity. Agentic AI brings far greater complexity, because it’s integrating systems and reimagining workflows. It’s an end-to-end system redesign,” said Mansoor Reehana, head of AI at Allianz.

Along with internal efforts, insurers are now faced with a multitude of existing software providers embedding AI into their solutions and a continual wave of emerging tools, platforms and point solutions providers. “The more tooling you have, the more challenges you have with integration,” observed Ania Collins, AI transformation lead at Zurich.

The more tooling you have, the more challenges you have with integration.

Ania Collins, Zurich

Pardeep Bassi, global proposition leader, data science at WTW, also outlined the increasing challenges of cutting through this endless parade of options and plotting a clear path forward. “Leadership is struggling to separate hype from real value. They want hard evidence to justify where to invest, but they also want to know what not to do,” he said.

Business-led, technology-guided

There was unanimous agreement among attendees that extracting AI’s operational value requires technology teams to work closely with business functions, bridging the gap between technical skills and insurance expertise.

“AI transformation has to be driven by domain experts who know the processes and understand the problems,” explained Dominika Kampa, group head of genAI value creation at Admiral.

However, Reehana highlighted another common disconnect, noting that while business functions may be eager to deploy agentic AI, they can often struggle to define what they want, or even understand what it can achieve. “This requires collaboration with technology, data science, and senior leadership. It should not be technology-led, but technology-guided to uncover what is possible,” he said.

AI transformation has to be driven by domain experts who know the processes and understand the problems.

Dominika Kampa, Admiral

Creating the right structure

While early GenAI adoption was characterised by bottom-up experimentation and pilots, enterprise scale requires a different approach.

Jemima Pitceathly, senior product manager at Flock, described their culture of experimentation: “We have open forums and workshops for employees to share what they are working on. It empowers people to think about where they can bring in LLMs and agentic workflows. But obviously, that is much easier to execute at a small organisation.”

At larger firms, unguided experimentation risks rapidly dissolving into fragmentation, with thousands of use cases and no clear direction. Even within structured frameworks, there is a risk that cross-departmental silos can result in different business units trying to solve the same problem with different tools.

We have open forums and workshops. It empowers people to think about where they can bring in LLMs and agentic workflows.

Jemima Pitceathly, Flock

Kampa describes Admiral’s highly structured, top-down approach. “We don’t believe in small-scale experimentation right and left. Instead, we provide big, strategic themes in a top-down manner, backed by C-level executive sponsors.”

Similarly, Allianz relies on robust group guidelines and prescriptions for responsible AI adoption. “This covers the end-to-end life cycle of a use case from ideation to production and even decommissioning,” said Reehana.

C-suite sponsorship was described as critical for success, particularly in overcoming any resistance from middle management. Attendees reported high levels of involvement and engagement from C-level executives and board members, with leaders firmly embedded in decision-making and driving change. This includes involvement in internal showcases and all-in conferences, solution demos at exco meetings and hackathons.

Effective governance and guardrails

With such rapid and pervasive change, the roundtable discussion shifted focus to how governance and regulation can keep pace with technology without choking innovation.

Allianz has adopted a streamlined, cross-functional governance approach. “We have a global, centralised tool and repository for all use cases. Every function – from IT and InfoSec, to risk, compliance and legal – has visibility and can raise issues,” Reehana explained. He acknowledged that case-by-case governance will soon hit a ceiling due to the ever multiplying numbers of use-cases, adding: “We are already looking at platform-based governance, building an environment where teams work seamlessly on pre-allowed use cases within a safe perimeter.”

The firm has also introduced strict guardrails and assessments for vendors, with contractual clauses and obligations covering subcontracting, data transparency, shared clusters, and prompt protection. Zurich has adopted a similar approach: “We have very robust operational assessment processes for both AI enhancements to existing tools and any new AI tools, in order to determine whether to proceed or pivot,” said Collins.

Whatever technical systems are put in place, the panel agreed that human accountability remains the anchor and will be expected by regulators. Yet Kampa cautioned on the importance of not stifling innovation. “Regulation needs to stay outcome-based with a human in the loop and solid underlying principles that allow you to safely scale as the technology evolves,” she said.

We have a global, centralised tool and repository for all use cases. Every function – from IT and InfoSec, to risk, compliance and legal – has visibility and can raise issues

Mansoor Reehana, Allianz

Building AI capability

Even with governance in place, scaling ultimately depends on people, with talent described as one of the primary hurdles.

Insurers need both deep domain experts to enhance existing processes and visionaries focused on future operating models. However, cultivating that vision internally and changing mindsets can be difficult. “Transformation comes from proper re-imagination and most of the big legacy players are still struggling with this,” said Kampa. “A domain expert who has been doing something the same way for 15 years is not going to suddenly change their view on how a process should work. You need to bring in a bit of external expertise, to inspire these challenging conversations.”

However, intense competition is compounding the insurance talent challenge, with tech giants and other FS firms creating thousands of AI jobs in the UK. Significant efforts are under way to equip existing teams and attract new hires, but as Collins explained: “Insurers have more people to reskill and upskill than ever before. It needs to happen across the entire business, not just within technology.”

AI-ready data

Attendees also debated the importance of data transformation for successful AI deployment. Pitceathly explained how rebuilding Flock’s underlying data infrastructure has unlocked both customer and business value. “Our data lake has become a shared asset across pricing, underwriting and customer success. Rather than needing a data scientist or data engineer, everyone in our organisation can query our data using natural language,” she said.

However, several executives questioned the benefits of wholesale data transformation for large established firms, asserting that the focus should be on data accessibility and usability, rather than waiting for data perfection before acting.

“Unless you are a greenfield build, your data will never be completely ready,” said Kampa. “Absolutely, there are specialised parts of the data stack that need investment, but I question investing millions of pounds in building organisation-wide ontologies. AI is developing so quickly that soon a good LLM will understand data connections and relationships without perfectly structured data.”

Speed of execution is an advantage...but expect that edge to be eroded six months later.

Pardep Bassi, WTW

Reehana agreed, noting that tabular data, once a challenge for LLMs, is now simple to query using modern AI integrations. “It’s not about traditional data architecture and data transformation, it’s about AI-ready data. This has a different path and different semantic layers need to be created,” he said.

Future-proofing business models

As the roundtable drew to a close, the discussion turned to whether scaling AI offers competitive advantage or is simply the baseline required to survive. “Speed of execution is an advantage,” noted Bassi. “But expect that edge to be eroded six months later.”

This relentless pace demands an entirely new operational mindset, adapting to a vastly accelerated way of working. While the initial hurdle of moving past pilots remains a challenge, participants highlighted that the hardest part is getting off the starting block. “Once you’ve completed an initial implementation, you can scale much more easily,” said Collins. Bassi agreed, noting the compounding impacts, with speed increasing for each new initiative as insurers lay down the foundational capabilities in their data and evaluation frameworks.

Ultimately, the roundtable concluded that the true threat is not existing market rivals, but agile new players unburdened by legacy constraints. “It is no longer about beating established firms,” Kampa summarised. “Instead, it is a race to future-proof the entire business model against an entirely new breed of start-up.”

Sponsored content