Unlocking value for underwriting and claims with agentic AI

Insurance leaders see agentic AI as a shift from reactive systems to proactive tools that improve risk, claims and customer outcomes.

“Everybody wants to go to heaven, but nobody wants to die,” said Yasir Andrabi (pictured) when asked about the challenges of innovation in insurance. Genpact’s global head of agentic AI solutions for insurance believes building a culture where experimentation and failure are acceptable is the only way for insurers to unlock the transformative potential of agentic AI.

Yasir Andrabi will join the line up of speakers to debate the opportunities and challenges of Agentic and Generative AI for insurance at Intelligent Insurer’s Agentic and Generative AI for Insurance event on November 18, 2025.

“Innovation, by definition, means you’re going to have failures. As leaders, you have to build a culture where failure is OK, as long as you're rewarding experimentation,” Andrabi said.

Insurance is entering a new phase of AI adoption; one that moves beyond rule-based systems to autonomous decision-making. For Andrabi, agentic AI “not only follows a path, it actually decides the path it needs to take”, and that could fundamentally reshape underwriting, claims and customer service across the industry.

Explaining what distinguishes this technology, Andrabi used the leap from mathematical calculations to self-navigating tools as an analogy. “Think about it like a calculator that used to do rule-based X plus Y equals this much. Agentic AI decides the path it needs to take and which one is best to accomplish the objective.”

That ability, he argued, can be applied across the entire insurance ecosystem: carriers can evaluate risks and profitability, brokers can find the best fit for clients and customers can access personalised products in near-real time.

One of the most compelling shifts for Andrabi is the move from reactive to proactive decision-making and he described scenarios ranging from policyholder retention to catastrophe response.

“If you have agentic AI which is listening for clues in a conversation, it’s able to make recommendations.”

Anrabi describes a customer call where the conversation sounds perfectly standard. On the surface, nothing signals that this customer is about to churn and shop for another policy.

“But with agentic AI listening in, picking up on subtle cues, context and intent, it can proactively flag that this customer is likely exploring other options and recommend an immediate retention offer. “

Another powerful example that Anrabi gives concerns catastrophe response: agentic AI can fuse multiple data streams. such as real-time weather models, projected hurricane landfall, expected severity by zip code, and map that against a carrier’s exposure.

“It can then monitor impact and enable the insurer to prepare a claim in advance, even before the policyholder makes contact.”

“The experience changes completely,” Andrabi emphasised. “The conversation would be more: ‘Are you safe? I know this is where your house was. I know this is what the damage in that area was. We’ve set up your claim. Where do you want me to send the money?’”

“Most companies run governance of AI like any other project, and this approach absolutely does not work.”

Beyond underwriting, agentic AI is being applied in claims, fraud detection and customer service. “Agents can ingest any kind of information – images, voice calls, documents – within seconds, validate it against third-party sources and orchestrate the entire workflow,” Andrabi stated. He cited fraud prevention in workers’ compensation, where discrepancies between injury photos, reports and social media could be spotted quickly, as well as call-centre augmentation where AI would prompt staff with the best responses in real time.

Yet adoption at scale remains difficult. Genpact’s recent research found “51% of executives plan to invest in better data quality in the next couple of years”; a recognition that ageing systems and fragmented data pose barriers. Governance is another.

“Most companies run governance of AI like any other project, and this approach absolutely does not work,” Andrabi stated, pointing to the difficulty of moving from successful proofs of concept to enterprise deployment.

Talent is the third hurdle, and Andrabi said their research had found that “only 2% of insurance executives said all their team members are AI fluent”, leaving a critical knowledge gap.

So what’s the way forward? Andrabi highlighted investment in modern data architectures, APIs and continuous model retraining. He urges insurers to adopt explainable, auditable AI methods to satisfy regulators and maintain trust.

On workforce readiness, he believes in “AI champions” from within teams who can demystify the technology. “When my own peer group is able to demystify it and stand behind it as a champion, my willingness to test it out is much higher.”

“A black hole doesn’t create trust. Fear does not help us build trust. Take away the fear, provide the tools and watch the confidence grow.”

Looking a decade ahead, Andrabi sees hyper-personalisation as inevitable. “It could be very specific to, for example, the time of day I’m driving and the route I’m taking. Hyper-personalisation is definitely on the horizon.”

He also expects lower fraud levels and potentially lower premiums, particularly for small businesses and personal lines.

His final advice for insurers is to take a holistic approach. “Don’t think that one agent can do everything. You have to think about it as a symphony of agents.” Combined with strong governance and investment in data quality, this orchestration could deliver the true promise of agentic AI.

Did you get value from this story?  Sign up to our free daily newsletters and get stories like this sent straight to your inbox.