Why AI won’t replace human claims adjusters

Claims surges caused by natural catastrophes (NatCats) or other major events create bottlenecks for insurance claims adjusters. And that’s where manual work processes can collapse.

That has some firms developing artificial intelligence (AI) agents and other options to take up slack by triaging clients – freeing human adjusters to help people who are in genuine distress, say two authors of a recent Deloitte report on property and casualty (P&C) insurance claims trends.

“Automation scales. So, you’re able to have those conversations about your claim, and [it] removes a lot of that administration that bogs [processes] down when there’s a surge event,” says Colin Asselstine, a director and insurance claims leader at Deloitte.

A lot of claims-related AI development centres on creating engines to do data validation, policy and coverage checks, triage, routing, document ingestion, classification, summation, notification and status updates.

“Those are all really good candidates for automation,” Asselstine tells Canadian Underwriter. “It allows you to protect the customer experience, make sure your operations [are] resilient. You give back to the customer that really needs your support.”

Related: AI is shrinking the pool for junior hires. Is apprenticeship drying up, too?

But automation also must react to the claim type.

Chris Duvinage, a partner and national P&C insurance segment leader at Deloitte, notes that while NatCat damage is often property-specific, that’s not always the case. 

“The second you have some element of bodily injury, [the person wants to] talk to somebody. So just because you can technically automate that process [using] AI…it doesn’t mean it’s [always advisable],” he tells CU.

Adoption of AI will, though, create different work and training needs for adjusters’ workforces.

“You don’t want to move away from the existing contact centers,” says Duvinage. “In fact, you may want to train your people to be even more empathetic…around some of the people skills and take the…lower-value work like data entry, and copy and pasting between systems away from them so they can truly focus on [clients] in the moment, rather than trying to solve back-end systems and processes.”

Adds Asselstine, “You want them to look at complex coverage, complex liabilities. Look at escalations, negotiations. That’s where humans are effective, and that’s where [AI] is pushing towards using our skill sets as humans.”

Which is why adjusters, particularly older workers, shouldn’t be afraid of losing their jobs.

“From an age perspective…it’s those folks with the experience that can be more empathetic; that have a solution on the phone and are…better-trained to have some of the customer-facing conversations,” says Duvinage.

Related: Why claims pros see an influx of fire claims during deep freezes

The best adjusters help customers understand their post-NatCat-event status and can explain complex insurance processes so that clients understand their next steps, says Asselstine.

“Depending on the claim, there’s a lot of that empathy and the trust and feeling…for a customer when they’re under stress – and being able to relate to tell [them that] everything [will] be okay. That’s true today. That will be true tomorrow,” he tells CU.

Adjusting firms, insurance companies, and some brokers must also prepare for how AI augmentation will help both veterans and new hires improve work processes. Those who get comfortable working alongside AI tools will benefit from their ability to do tasks like summarizing documents and even providing suggestions for next steps.

“It’ll listen to a conversation. It’ll take a summary of that. It’ll pull up the policy. It’ll pull up your internal standard operating procedures (SOP), and say, ‘Based off what I heard, I think these are [the] relevant areas of [the] policy, and this is the relevant area of our SOP,’” Asselstine says.

“It’s [giving] suggestions for humans to then validate. When the guardrails are up [around AI] and insurers get comfortable that 99.99% of time the model is giving the right answer, you can then say, ‘For these simple claims I’m good with the model making the decision.’”