Insurance Claims Data Now Describes a Different Claim

An adjuster who learned the job in 2020 and came back to it this year would find the same screens, the same coverage codes and the same reserving discipline. And a book of claims that behaves differently.

Not more expensive, though it is that too. Differently composed.

CCC Intelligent Solutions publishes an annual read on this from what it calls hundreds of millions of claims-related transactions, which makes it one of the few genuinely large datasets in the field. The 2026 edition, titled Complexity Compounds, landed on 31 March. Three of its figures are worth more than the rest of the report put together.

Auto Claims Outcomes in 2026

Total loss frequency reached 23.1% of claims, an industry high. Separately, 28.3% of repairable estimates now include a calibration, the recalibration of cameras and sensors that a modern car needs after bodywork.

Put those two together and the shape of the book falls out.

Roughly 55% of claims are what the industry was built around: a car goes to a shop, the shop fixes it, the insurer pays. The other 45% are either written off entirely or need a procedure that barely existed a decade ago.

The two halves have almost nothing operationally in common. A total loss is a valuation, a title transfer and a salvage recovery. A calibration repair is a technical sign-off with a liability tail if it is done wrong. Neither is the workflow a claims system was designed around.

Why the write-offs are climbing

Part of it is repair cost, and part of it is the customer. CCC points at higher deductibles and household financial pressure changing which incidents get claimed at all.

“Insurance coverage and claim filing behaviours have dramatically shifted as consumers react to an uncertain economic landscape.”
Kyle Krumlauf, director of industry analytics, CCC Intelligent Solutions

The mechanism is worth being precise about. Raise a deductible and the small claims stop arriving, because the policyholder absorbs them. What remains is a book with the cheap end filtered out.

J.D. Power has the other half of that in numbers. In its 2025 auto claims study, fielded across 9,455 claimants between September 2024 and August 2025, 26% of customers now carry a deductible of $1,000 or more, and 7% said they had avoided filing a claim at all because they were worried about a rate increase.

Small claims of $2,000 or less were 20% of the total, down a point on the year.

So the cheap end is not shrinking by accident. It is being priced out and, in 7% of cases, deliberately withheld.

Average severity then rises without any individual repair getting more expensive. Some of the industry’s severity trend is real inflation and some of it is this selection effect, and the two are hard to separate from the outside. Anyone forecasting from a severity average alone is forecasting a mix change they have not modelled.

Bodily Injury Claim Severity

Average paid bodily injury severity rose 10.3% in a year, and 32% across four.

Four years of that arithmetic is the part to sit with. A liability book priced on 2022 assumptions is now settling claims a third more expensive, and bodily injury is where reserve development goes wrong slowly and then all at once.

It is the same pressure documented in the two-speed soft market, where casualty rates climbed while property rates fell. Claims data is where that shows up first, months before it reaches a rate filing.

The technology in the car costs three and a half days

Cycle time is the other thing claims data measures well, and it is what customers actually experience.

J.D. Power put the average repairable cycle time at 19.3 days, down from 22.3 the year before. Underneath that average is the number worth keeping.

A vehicle from 2015 or earlier with no driver-assistance systems took 17.9 days. A vehicle from 2019 or later with three or more of them took 21.5 days.

Three and a half days, or about a fifth longer, attributable to the sensors.

Which is CCC's calibration finding arriving from a completely different direction. One dataset counts how many estimates include a calibration line. The other measures how long the car sits. They agree, and neither was designed to confirm the other.

Property runs slower still. J.D. Power's 2026 property study has repairs averaging 29.6 days and final payment 40.7 days, both improved on the prior year by around three days.

Where the two datasets disagree

On total loss frequency they do not line up at all. CCC reports 23.1%. J.D. Power reports total losses rising to 27% from 24%.

Four points apart is a lot when the figure is the headline. The gap is methodological: CCC counts transactions flowing through its network, J.D. Power surveys claimants who settled in the previous nine months, and those are not the same population. Neither is wrong.

It is a useful warning about the genre. Two of the largest claims datasets in the American market disagree by a sixth on the most quoted number in motor claims, which is a good reason to treat any single published figure as a direction rather than a benchmark, and to run the measurement on your own book before you price anything against it.

The fleet aged, and the arithmetic moved with it

One number in the CCC report explains a good deal of the rest. As of the third quarter of 2025 there were 12 million fewer vehicles six years old or newer on American roads than in 2020.

An older fleet totals more readily, because the repair estimate meets a lower actual cash value sooner. It also carries more of the sensor hardware that first appeared in the late 2010s, now old enough to be damaged and out of warranty.

Both of the mix shifts above are partly a consequence of that single supply-side fact. Which is a reminder that the drivers of a claims trend often sit outside the claims department entirely.

Why the data does not turn into a decision

Every figure above is available to any carrier. Most have richer data on their own book than either source has on the market.

The gap is not measurement.

Deloitte surveyed 200 US insurance executives in June 2024, half from life and annuity and half from property and casualty, on their readiness to scale generative AI. Fragmented data and ageing systems came back as the constraint, ahead of talent and ahead of the models themselves.

That finding is usually read as an AI story. It is really a claims data story, because the same fragmentation blocks the ordinary analysis too.

The things that were never fields

Ask a claims system how many of last quarter’s repairs involved a calibration and you will often find the answer exists only inside estimate attachments and adjuster notes.

The same goes for why a total loss was called, which supplement drove a reopen, and whether a delay was the shop, the parts or the carrier. All of it is knowable from the file. None of it is queryable.

So the mix shift that CCC can see across the market is frequently invisible inside the carrier experiencing it, not because the data was never captured but because it was captured as prose.

That is the difference between having claims data and being able to use it. Structured at the point of capture, a calibration flag is a rating factor, a referral trigger and a vendor scorecard. Typed into a note field, it is a memory.

What to change first

Three things follow, in the order they pay back.

None of them is a technology project.

Make the mix a reported number. Total loss rate, calibration rate and reopen rate by month, on your own book, next to the market figures above. If those three are not on a claims management dashboard someone reads weekly, nobody will notice the next shift until it reaches the loss ratio.

Second, promote the recurring note to a field. Whatever the handlers keep writing in free text is the thing the business does not yet measure, and it is usually a five-minute configuration change rather than a data science project.

Third, keep the decision, not just the payment. Why a valuation was accepted, which estimate line was disputed, who approved the supplement. An insurance audit trail that captures reasoning is what makes a reserve review, a reinsurer question or an insurance fraud referral answerable later.

The platform part of this is narrow and worth stating plainly. Openkoda is not going to out-feature an established claims suite; those systems carry decades of accumulated functionality and would win a line-by-line comparison.

What it offers is that the data model is yours. Adding a calibration flag, a new total-loss reason code or a supplement approval step is product configuration you make and own, not a change request against someone else’s release calendar. When the claim mix moves every eighteen months, the interval between noticing and measuring is the thing that matters.

The mix will move again

Nothing in the 2026 figures looks like an endpoint. The fleet keeps ageing, sensors keep spreading down the price range, and deductibles are unlikely to fall.

A reasonable expectation is that total loss frequency and calibration rate both keep rising, and that the plain repair keeps shrinking as a share of the book.

Which makes the useful question about claims data not what it says today. It is how long it takes you to find out when it changes.