For years, artificial intelligence has quietly tiptoed in the background of insurance claims, surfacing information, flagging risks and helping humans make decisions, but rarely taking action itself. That is beginning to change. Those stealthy steps are now becoming bashful plods, as AI’s impact on the decision-making process surges.
“Agentic AI looks to represent a fundamental shift: AI that actively participates in the work,” said Rajeev Gupta, Co-Founder and Chief Product Officer at Cowbell.
In practical terms, that shift is already visible inside claims workflows. Systems that once answered questions are now being asked to move work forward.
“It analyses the First Notice of Loss, identifies the exposure, and actively proposes concrete next steps,” Gupta said. “It is technology that actually moves the work forward, rather than just waiting to be queried.”
Others describe the change less as an upgrade and more like a fundamental redefinition of how claims operate.
“Agentic AI represents a significant step forward,” said Sudhir Upadhyay, Senior Consultant at Capco. “It shifts the focus from automating tasks to orchestrating outcomes.”
“Instead of simply assisting a claims handler with individual activities, it coordinates the sequence of decisions and actions required to move a claim from intake toward resolution.”
Where agentic AI is already making an impact
That orchestration is most visible at the very start of the claims journey. First notification of loss and triage, long considered some of the most manual and fragmented parts of the process, are becoming testing grounds for more autonomous systems.
“Agentic AI can analyse documents, images and written descriptions simultaneously,” Upadhyay said, “enabling claims to be assessed more efficiently and routed to the appropriate workflow with greater accuracy.”
The effect is not just speed, but momentum. Improvements at intake tend to carry through the rest of the process, shaping how quickly and accurately claims are resolved.
In high-pressure situations, that speed can have immediate consequences.
“The most critical window in any cyber incident is the ‘Golden Hour’ of response,” Gupta said. “When an SME is facing a ransomware event, speed is a security feature, not just a customer service metric.”
“A purpose-built claims agent could instantly triage an incoming claim, assess the urgency, and draft the initial response,” he added. “The measurable value here isn’t just operational efficiency; it is the tangible reduction of financial devastation for the policyholder because the response was immediate.”
For some, the implications extend beyond faster claims altogether. At Quensus, a leading provider of leak management systems, agentic AI is being used not to process claims, but to prevent them.
“Agentic AI represents the leap from detection to prevention,” said Dan Simmons, Managing Director at Quensus. “Our systems identify abnormal water behaviour and autonomously trigger a shut-off valve to isolate the supply, performing the task of protecting the building without waiting for human intervention.”
“The measurable value is found in the drastic reduction of cycle time and business interruption,” he added. “By the time a traditional automated system would have alerted a human to file a claim, our autonomous agents have already prevented the damage.”
The risks beneath the surface
As the technology moves closer to decision-making, however, the conversation becomes more complex.
Much of the industry’s focus has been on efficiency, but Paulo Ferreira, CTO at KYND, argues that the more important question lies elsewhere.
“Most of the conversation around agentic AI in insurance focuses on operational efficiency,” he said. “But the more consequential question is a different one entirely: how will the widespread adoption of agentic AI generate claims that insurers haven’t priced for?”
Agentic systems, he argues, introduce a different kind of exposure.
“They don’t just suggest actions; they take them. When an AI agent autonomously negotiates with a supplier, authorises a payment or interacts with customer data, the liability surface looks nothing like a human operator using an AI-assisted tool.”
That shift from tool to actor changes not just how claims are handled, but how risks emerge.
“The autonomy that makes agentic systems useful is the same thing that makes errors propagate further before anyone catches them,” Ferreira said. “Feed it incomplete or inconsistent data, and it accelerates bad decisions, with greater confidence and at greater speed than a human ever could.”
Others see similar risks from a technical perspective.
Kit Ruparel, Chief Technology Officer at TCC Group, said that organisations are often underestimating how quickly reliability can degrade when multiple AI systems are combined.
“Even if each model is highly accurate, the combined reliability falls sharply as you add more steps,” he said, describing how chains of AI-driven decisions can compound small errors into larger failures.
Trust, control and the human role
That tension, between capability and control, is shaping how insurers approach deployment. Across the industry, there is broad agreement that autonomy cannot come at the expense of oversight.
“We must ensure that the ‘human-in-the-loop’ is non-negotiable,” Gupta said. “Agentic AI should propose actions, but it must leave judgment, accountability and control exactly where they belong: with people.”
Ferreira points to a similar principle at a systemic level. “The principle of controlled delegation rather than blind automation is well understood,” he said, though he notes that implementation still varies widely across the market.
The challenge, increasingly, is not just managing internal systems, but understanding how agentic AI is being used externally. “We are seeing early signs of what might be called ‘silent AI’,” Ferreira said.
“Organisations are deploying agentic tools that interact with customer data and operate across critical infrastructure, often without explicit coverage in existing policies. When one of those agents makes a consequential error, the claim will land somewhere — and many policies have not been written with that scenario in mind.”
He emphasises the need for insurers to know what they can and cannot detect. “Some categories of AI exposure and risk are detectable from outside an organisation,” Ferreira explained.
“Others are largely invisible: employee use of hosted models, unsanctioned agents on personal devices. The nuance is knowing what your signals can and cannot prove, and being conservative about it.”
A different kind of claims function
Even where the opportunity is clear, adoption remains uneven. “Agentic AI delivers its greatest value when it can access reliable, well-structured information and interact seamlessly across systems,” Upadhyay suggested.
Without that, even advanced systems struggle to perform consistently. There are also practical limitations. Orchestration tools are still maturing, explainability remains imperfect, and integration into existing workflows can be slower than expected.
“Much of the C-suite language around AI adoption has been reported in terms of cost-savings,” said Paul Donnelly, Global Head of Insurance at Version 1, noting that this has contributed to apprehension among claims teams.
In practice, however, the shift is less about replacement and more about redistribution.
“Agentic AI can guide claims handlers through processes, retrieve documents and autofill documentation,” explained Graham Gordon, Product & Strategy Director at Sapiens. “It ensures ‘claims knowledge’ is shared more evenly across the workforce.”
Over time, that redistribution is expected to reshape the structure of claims operations.
“We will see fully autonomous straight-through processing of relatively lower complexity claims,” Donnelly added, alongside more personalised and data-driven customer interactions.
For some, the longer-term shift goes further still.
“The claims function will evolve into a resilience department,” Quensus’ Simmons said. “The industry’s promise to pay is being replaced by a promise to protect, where autonomous agents manage risks 24/7.”
It is a view that reframes claims not as a moment of response, but as part of a continuous system, one that increasingly sits upstream of loss itself.
And while the path to that future remains uneven, shaped by data, governance and emerging risks, the direction of travel is becoming harder to ignore.
Keep up with all the latest FinTech news here
Copyright © 2026 FinTech Global
