Sign in CREATE NEW ACCOUNT

← News

19

Nov 2026

Collision Industry Conference (CIC) committee tackles suggestions for AI use in claims handling

Collision Industry Conference (CIC) committee tackles suggestions for AI use in claims handling

  • Published:

  • Author: Teresa Moss via RepairerDrivenNews.com

AI is involved in the auto claims handling process, but regulations have yet to catch up, Jill Tuggle, Auto Body Association of Texas Executive Director, said during the Collision Industry Conference’s most recent meeting held in Las Vegas. The National Association of Insurance Commissioners (NAIC) approved a model bulletin in 2023 on insurer AI use that 25 states and Washington, D.C., have adopted, she said.

The National Association of Insurance Commissioners (NAIC) approved a model bulletin in 2023 on insurer AI use that 25 states and Washington, D.C., have adopted, she said. None of the states that have adopted the model require a human to make the final call on claim denials, she added. An NAIC survey also found 88% of auto insurers are using or exploring AI, she said. “We still have a long way to go,” Tuggle said. AI now touches vehicle intake, photo review, estimating, valuation, and claim closure, she added. Currently, a vehicle is often inspected by a shop and a field adjuster who agree on the required repairs, Tuggle explained. The estimate then goes to a central review team. “We’re not really sure who is behind that right now,” Tuggle said. “AI is scrubbing the files in many cases, and these algorithms are built by the companies who are reviewing them. We’re not totally sure what the parameters are that are being set and how they’re writing the algorithms to take a look at these files.” There are examples of automatic overrides for estimates that include more than eight hours of repair, Tuggle said. A shop will reach out to a field adjuster to discuss the changes to the estimate, she adds. The shop is told the adjuster is no longer on the claim and the shop must call customer service. “That’s when we start hitting brick walls,” Tuggle said. “So who actually made the final decision? So the issue here is the attribution gap. Who observed the car, the field adjuster? Who modified the estimate? The central team and their AI scrubbers? Who knows? That’s often very unclear. But the field adjuster is still named at the bottom.” The CIC Governmental Committee found an example of an AI-handled claim where the shop estimate was about $2,500, Tuggle said. It came back from the central review team at $438. The claim was closed because it was lower than the deductible, Tuggle said. She added there was no recourse for the customer on the claim decision. “What consumers should be able to count on is knowing when AI material has shaped their claim and know who made the decision and how to reach out to them,” Tuggle said. When asked, 67% of the CIC audience also believed consumers should know when AI is used in the claim handling process. NAIC’s AI model is built on the existing Unfair Claims Settlement Practices Act, she explains. It covers claims and fraud detection; states controls should scale with potential harm to consumers; and states that decisions cannot be inaccurate, arbitrary, or unfairly discriminatory. It also requires insurance companies to set clear decision authority and escalation paths, test data and models, monitor for drifts, keep records, notify consumers of AI use, and oversee vendors. Tuggle said the bulletin doesn’t require claim-level notice that AI shaped the estimate. It also doesn’t require a change log or final author on the estimate, a named accountable reviewer, or duties for shops or vendors. There should be more oversight depending on the consequence, Tuggle said. For example, disclosure isn’t necessary for administrative work such as scheduling, transcription, and document searching. Perhaps there should be AI disclosure for AI drafts with review by a qualified person and the ability to change the draft, she said. In a third scenario, insurance companies should provide a disclosure with a named approver and a reason when AI removes repairs, totals a vehicle, or flags fraud, Tuggle said. A meaningful human review needs to meet five criteria, she said. These include: “Competence: Can spot the technical error; “Evidence: Sees the photos, not just AI summary; “Time: Not racing a productivity quota; “Authority: Can override without layers of approval; and “Permission: Can disagree without penalty.” Tuggle asked, “At the end of the day, when you’ve got an estimate where so many different entities and people have influenced it, who owns it?” ”That’s kind of one of the major issues,” she continued. “You have your data provider who supplies the inputs and your AI vendor; they’re building the models, the algorithms for the scrubbing software that they’re using at the central review areas. And then we’ve got our estimating platform, which embeds it into the workflow. Insurers approve or deny the claim, and the repair facilities perform the repair.” Tuggle asked whether there could be shared responsibility that is also symmetrical in the process. The committee also suggests duties for consequential AI. This includes AI disclosing material and preserving every estimate version. It should also attribute credit to who inspected the estimate, which system was used, changes made, and who made the changes, Tuggle said. She added that insurance companies should explain the basis for each change and provide an escalation option to a human who can reverse it. The name also must mean something, she said. “The field adjuster’s name at the bottom of the file is not enough, and it’s not really getting us anywhere right now,” she said. “So, proposed again, a human decides the safety and critical items.” The committee also proposes that a qualified human must approve any cuts to procedures such as structural repairs, airbags, seatbelts, restraints, scans, ADAS calibration, fraud flags, and other safety-critical items. “AI can advise it, but it can’t be the only reason for denial,” Tuggle said. James Spears, Tractable head of automotive AI, said during a panel discussion that it would be difficult to imagine all the ways that AI touches insurance claims. “AI begins at policy acquisition,” Spears said. “AI begins before they even have a customer, right? AI is used in the way that you market. AI is used once you close a policy. AI is used in underwriting. AI is used all throughout in a number of ways.” Many assume AI has evil intent, but it can increase claim speed and accuracy, Spears said. He explained that the AI models the general public uses aren’t as advanced as the models insurance companies use. However, he said AI will advance significantly in the coming years. Jerry McNee of Ultimate Collision agreed that AI benefits the user. Before AI, a shop manually went through everything and took time pulling the repair processes. It took additional time to verify, he added. Humans also have errors and miss procedures, he said. AI is here to stay, he added. “We have to get ahead of it, and we have to understand it, and we have to put it to work,” McNee said. “There still has to be some human interaction and touch on it, in order to proof it and make sure.” Spears also agreed with McNee. “The best way to employ AI is to only use it as a tool that takes some mundane task off of your team,” Spears said. “Then push that team towards the customer. That’s where you really make a difference.” He added that he disagrees with “straight-through” processing where a human never touches a claim. The insurance and repair industries should work together to create rules AI can work within, he said. “Ninety-five percent of AI will get you to yes with shared rules very quickly,” Spears said. “The other 5%, that’s going to be your escalation, and that’s when you get humans talking about it. That’s the way I would play the game. If I was sitting in a carrier’s chair right now, I would have shared rules with every one of my DRPs.” AI can help speed up parts and rental car deliveries, Spears added. When Tuggle asked whether consumers should be notified of AI use, Spears said they should. “But here’s the deal: It’s not going to shock a consumer,” Spears said. “Because everything in their life has AI.” McNee agreed. “Absolutely, consumers should be informed, especially on the decision-making when it comes to their vehicle or their policy or returning the car back to its previous condition,” McNee said.