Agentic AI is gaining traction in insurance, but most firms remain unsure how to use it meaningfully. A recent InsTech webinar shed light on where the technology is working today, and what it takes to move from experiments to enterprise-scale adoption.
In this article, we’ll explain…
- What is Agentic AI, and how does it differ from generative AI?
- Early deployments in underwriting and claims
- Implementation realities and scaling challenges
- Cultural shifts and selecting the right partners
- How the upcoming Rise of the Agents event will move the conversation forward
Introduction
Agentic AI has become the hottest word of the year. It refers to systems that can perform tasks or make decisions proactively, not just respond to prompts. “There’s more excitement about Agentic AI than anything I’ve seen for a long time,” said Robin Merttens, Chairman, InsTech.
Despite the buzz, deployment remains limited. Most applications are isolated pilots or narrow-function tools. In October, InsTech hosted a webinar, Agentic AI – friend not foe, with speakers from Google, HDI Global and OIP Insurtech to examine what’s real and what is not. The discussion focused on tangible benefits, organisational enablers and what to expect next.
What makes Agentic AI different?
“Insurance is text-heavy, process-heavy and ripe for this,” said Mladen Subasic, Chief Product Officer, OIP Insurtech. Traditional software could store information and guide step-by-step processes, but “the true work” happened in between, handled by humans. “Agents in AI are expanding that area heavily,” he explained. Unlike traditional AI, which predicts outcomes, agentic systems can now collect, collate and act on information – executing multi-step workflows without necessarily scaling costs and labour.
Christina Lucas, Global Director & Market Leader, Google, pointed to the move from automation to autonomy. “Claims was once a linear, adjuster-led process. With Agentic AI, the system can ingest a first notice of loss, then trigger and manage the next five actions without waiting.”
“Agentic AI lets systems decide and act – not just suggest.”
Where insurers are deploying it now
Real deployments are starting to appear across insurance lines. Christina cited fraud detection in US medical claims as an early win: “Voice, video, document and image inputs are all being ingested to flag fraud within minutes, not days.”
Mladen described real mid-office automation gains at OIP Insurtech: “Ingesting submissions, firing emails, fetching quotes and matching markets – these are now real tasks agents are doing.” He noted that some implementations are already running in production, replacing workflows that once required multiple manual touchpoints.
Ulf König, Head of Innovation & Strategy, HDI, noted their team is focusing on underwriting: “We picked front-line enthusiasts and gave them tooling. Results matter, but it’s the enthusiasm that’s powering progress.” Their approach has been deliberately grassroots. Rather than imposing technology top-down, HDI identified enthusiastic underwriters already pushing boundaries and gave them tools to experiment. “We started with the innovators who were already thinking this way,” Ulf explained. This strategy helps build proof points and internal champions before wider rollout.
Inevitably, firms are experimenting most in data-heavy, high-volume tasks like document ingestion and quote comparison. Success depends as much on finding willing adopters as it does on the technology itself.
How to scale responsibly
Every panellist agreed: pilots are easy, but scaling means enterprise transformation.
Christina highlighted the need for “grounded data and secure sandboxes” creating trusted environments where AI operates only on verified information. Insurers must combine first-party data with carefully vetted third-party sources to limit hallucinations. “This is about curating your playground – not just plugging into the web,” she added. The alternative, letting AI access unrestricted internet data, introduces too much risk for regulated environments.
Governance, too, is crucial. Agentic AI requires new control models. “Decide where the human remains in the loop – and when,” said Christina. The answer varies by use-case: high-stakes decisions like claims payouts above certain thresholds might always require human sign-off, while routine data entry can run autonomously. “This changes depending on line, geography and regulation.”
“Agentic AI needs grown-up governance, not just good tech.”
Therefore firms must embed governance, security and human-in-the-loop checkpoints into their architecture from the start. Retrofitting controls later creates risk and slows adoption.
Choosing the right partners
No insurer plans to build these systems alone. “We look to both startups and hyperscalers,” said Ulf. “Startups bring speed and innovation, but hyperscalers bring stability.” The choice often depends on the use case: startups may be ideal for experimental, niche applications, while hyperscalers provide the infrastructure and compliance frameworks needed for enterprise-wide deployments. Many insurers are taking a hybrid approach.
Mladen reinforced this: “Be careful who you bring to the party. True Agentic AI needs understanding of your business, workflows and architecture.”
Three things to remember
- Agentic AI is moving from prediction to action: systems now execute multi-step workflows autonomously
- Early wins are in high-volume, data-heavy processes: fraud detection, submission ingestion, quote comparison
- Scaling requires enterprise transformation: governance frameworks, data curation, and strategic vendor partnerships are non-negotiable
What’s next
InsTech is hosting The Rise of the Agents on 25 November in London. The half-day event gathers insurers, CIOs and technology leaders to assess where Agentic AI stands today.
Sessions will include:
- Results from the first Insurance AI Index
- Enterprise architecture for agentic systems
- Real-world use cases across underwriting and claims
- Panels featuring HDI, AXA, Qover, Zurich, NatWest and more
- Live multi-agent demos and curated networking
“This is the most significant AI shift we’ve seen – and our moment to shape it,” said Robin.
The Rise of the Agents
Learn more about The Rise of the Agents and register for the event on 25 November.
