Insurers have spent the last two years experimenting with AI at the edges of their business. Diego Devalle, Chief Product Development Officer at Guidewire, believes the foundations are now strong enough to put AI agents inside the workflows that matter most: claims and underwriting.
What has changed, Devalle argues, is not the appetite but the infrastructure. “Carriers have spent the last two years proving that AI has a role to play,” he said. “What’s changed is the infrastructure around it: AI can now be connected more securely to the data, rules, and controls that core insurance workflows depend on.”
Claims and underwriting rely on policy data, claims history, business rules and regulatory requirements, and without secure, real-time access to that information “there will always be a limit to what an AI agent can do”. What is changing, he said, is the ability to give AI that access within the right controls, so it can be embedded more deeply into core workflows and support multi-step processes rather than isolated tasks. “This doesn’t mean insurers are becoming less cautious. It’s more about the foundations becoming strong enough to put AI to work in core workflows with greater confidence.”
Which workflows go first
Guidewire has chosen claims summarisation, policy change and first notice of loss as the workflows to automate first. A good candidate for an AI agent, Devalle said, is “a repeatable, data-rich workflow with a clear outcome, but that still requires significant manual effort”. The three starting points are well-understood processes where an agent can bring information together, complete defined steps and remove administrative work.
Not every process qualifies. “I would be much more cautious about using agents for decisions that are highly subjective or carry significant financial or regulatory consequences,” he said. “AI can still gather information or make recommendations, but that does not mean it should make the final decision. The goal is to use agents to improve speed and consistency, with the right data, governance and oversight in place.”
Control without free rein
Once an agent has real-time access to policy, claims and billing data and can complete multi-step processes on its own, the question becomes how an insurer keeps meaningful control. “Giving an agent direct access does not mean giving it free rein,” Devalle said. “The control comes from how the agent is built into the platform and how the workflow around it is designed.”
With Guidewire’s Agentic Framework, carriers can choose the right AI model for each task and give agents secure access to the policy, billing and claims information they need, when they need it, with the agent working inside the core environment rather than in a separate AI layer disconnected from the systems and controls insurers already rely on. “That makes it possible to automate multi-step processes without removing human oversight,” he said. “In insurance, there will always be points where regulation, complexity or judgement require a person to be involved. The value is in automating the work around those decisions while keeping the insurer in control.”
Accountability before regulators ask
Accountability is the hard question in regulated processes, and Devalle is unambiguous about where it sits. “Accountability can’t sit with the AI agent,” he said. Before a carrier puts an agent into production, it needs to be clear who owns the workflow, what the agent is allowed to do and where human review or approval is required. That is particularly important in claims and policy administration, where decisions can have financial, regulatory and customer consequences, and where human judgement must remain part of the process.
“The real test is whether the carrier can explain how a decision was reached, what information the agent used and where a person had oversight. If those answers are not clear before deployment, the workflow is not ready for that level of automation.”
Measuring the return
In production, Devalle said, the return will depend on the workflow and the problem the insurer is trying to solve. “The right starting point is to establish how that process performs today and then measure whether AI improves it.” In claims, that might mean faster handling or less administrative work for adjusters; in underwriting, faster turnaround or more time spent on complex risks. Those, he said, are the kinds of operational outcomes Guidewire is focused on.
“The important thing is to measure the outcome, not just the speed of the AI itself,” he added. “If one step becomes faster but creates more review, rework or risk elsewhere, the carrier has not really improved the process.”
The next two to three years
Looking two to three years out, Devalle expects the biggest change to be adjusters and underwriters spending less time assembling the information they need before they can make a decision. An agent can pull together the relevant policy, claims and customer context, summarise what has happened and help move the process forward, “so the person can concentrate on the judgement that actually requires their experience”.
“The risk is assuming that putting AI on top of an existing process automatically makes that process better. It doesn’t,” he said. If the data is fragmented, the workflow is inefficient, or it is unclear where a person needs to step in, an agent can simply magnify those problems.
“The carriers that get this right will start with the workflow itself: what should be automated, what should stay with the underwriter or adjuster, and what information the agent needs to do its part well. That is a much more important question than how many agents you can deploy.”
The post Guidewire on Putting AI Agents Inside Core Insurance Workflows appeared first on The Fintech Times.
