Agentic AI: from pilot to purpose – key takeaways from InsTech’s 'The Rise of the Agents'

Last week, the InsTech event ‘The Rise of the Agents’ brought together a mix of insurance experts to explore how agentic AI is beginning to shape operations across the sector.

Speakers included Dr. Thomas Kuhnt, Member of the Executive Board at HDI, Vincent de Ponthaud, Group Head of Software & AI Engineering at Axa, and Ed Ackerman, Chief Operating Officer at Qover, who offered a perspective from a technology-led scale-up.

The discussion ranged from future potential to real-world friction. While many larger organisations are working out how to integrate AI into long-established systems, the conversation also touched on what it looks like to use these tools now, in more agile settings.
‍

Why scale matters when adopting agentic AI in insurance and how it shapes the approach

The session started with a broad question to the panel: ‘what is it like leading AI transformation within organisations of different sizes and maturities?’ 

Our Chief Operating Officer, Ed, noted that with around 150 people, Qover operates under very different conditions compared to large insurers. This size enables a more flexible and iterative approach to testing and implementing new technologies, including agentic AI.

Rather than relying on extended planning cycles, experiments with agentic AI have been integrated incrementally into live workflows. This approach gives teams the opportunity to learn from experience and adjust course based on real outcomes. While a smaller size doesn’t automatically make things simpler, it does make it easier to adapt quickly when needed.
‍

What agentic AI in insurance looks like in practice

Ed also explained that Qover’s AI journey began with customer support and has since evolved into a focused drive toward achieving full straight-through processing in claims. 

The vision is to build an end-to-end system where hyper-personalised agents handle the entire process, with human input focused solely on defining requirements and ensuring quality.

This would allow Qover to hyper-scale its operations effectively without increasing headcount, while continuing to deliver a high standard of customer experience with a consistent, human tone.

The progression has followed a phased structure that reflects our ambition and discipline:

• Crawl: we started by enhancing our customer care experience. Generative AI was introduced to help agents draft replies to common questions more efficiently – the beginning of AI-human collaboration in live operations.
• Walk: building on early gains, we moved quickly into automation. AI began handling common customer queries directly, across both email and voice channels. This wasn’t just about faster responses: it marked the start of always-on, multilingual support at scale.
• Run: now, we’re applying that same logic to claims, a more complex and high-stakes part of the journey. Our focus is on refining and scaling a network of around 20 highly specialised internal agents that work together to handle claims from start to finish. The goal is full automation, without compromising accuracy or empathy.

The decision to build or buy AI tooling depends on where Qover sees the greatest advantage. When a use case is highly specific, like claims processing trained on our proprietary data, and not readily available on the market, we build it ourselves.

For more general needs where off-the-shelf tools are sufficient and ownership offers no added value, we prefer to buy. This approach lets us focus on building what’s uniquely impactful for Qover, while relying on external solutions where speed and simplicity matter more.
‍

Why culture often moves slower than technology

One of the clearest takeaways from the panel was that introducing AI is rarely just about the tech. Especially in a regulated space like insurance, concerns around accountability, compliance and operational risk shape what’s possible.

Ed pointed out that the biggest challenge isn't the technology itself, but rather the inertia in working with traditional insurance partners. 

‘Delegation of authority is a big deal, and people rarely lose their jobs for displaying an abundance of caution,’ he said. That tension, between the need to innovate and the instinct to avoid risk, often means decisions take longer and progress is slower.

At HDI, Thomas Kuhnt explained that adopting AI at enterprise scale involves difficult trade-offs. While the technology holds clear promise, committing too early can carry reputational and financial risk if the outcomes don’t match expectations. His role on the executive board includes pushing for meaningful investment in AI but always with a measured, realistic view of how fast and how far to go.

Vincent de Ponteves from Axa described how enthusiasm for AI is increasingly coming from the business side. Rather than engineers pushing new technologies internally, it’s now business teams driving demand and asking technical teams to support them. This shift is creating a new kind of collaboration, one where business leads the charge and tech enables the transformation.

Even when the systems are ready, internal governance and external validation processes often dictate the pace. Caution is appropriate in an industry like insurance, where the stakes are high and trust is essential. But the potential rewards for executing effectively, without unnecessary delay, could be enormous.

Customers don’t dislike bots, they dislike bad bots

There’s sometimes an assumption that customers resist AI-led interactions. In reality, most are simply looking for fast, accurate answers and smooth experiences.

Qover’s data suggests:

• The vast majority of claimants (99.7%) are happy for AI to handle their case.
• Satisfaction scores for AI-led interactions match those of human agents, averaging around 90%.
• Customers have positively referenced digital agents in feedback.

The lesson is not that AI is always preferred, but that it’s accepted when it works well. Maintaining that standard requires continuous attention and iteration.
‍

Customer expectations are high and growing – and agentic AI in insurance can help

For Ed, the concern isn’t about whether AI works. It’s about whether we’re doing enough to stay ahead of the risks that come with scaling it.

As agents take on more critical responsibilities – like cancelling policies, approving claims, or updating coverage – the challenge becomes ensuring that the platform behind them can be trusted to act with the same care and accuracy as a human would.

This requires more than just good performance metrics. It means building stronger guardrails, investing in oversight and making sure our teams are asking the right questions. It’s not just about reviewing what the AI did, but establishing whether the outcomes meet company and customer expectations.

In practical terms, this includes:

• Robust monitoring and audit capabilities
• Hybrid workflows that blend automation with human review
• Ongoing training to help teams work confidently alongside AI

The risks of complacency are real but so are the opportunities. If done right, agentic AI can help scale operations without losing the human qualities that make good service feel personal.

Whether someone is interacting with a bot or a person, standards should remain consistently high.

Closing thoughts from the panel on agentic AI in insurance

The panel made it clear that the industry is in an active learning phase. Different organisations are moving at different speeds, shaped by their size, structure and appetite for change.

Qover’s contribution reflected a pragmatic approach offering a transparent look at what’s been tried, what’s worked and what’s still being figured out.

Much remains to be developed in the sector, and there’s still significant learning ahead. However, with the appropriate safeguards and strong collaboration, agentic AI has the potential to simplify insurance for all stakeholders.