ai claims automation in 2026: what one insurer's numbers really support
2026-08-04
The most concrete ai claims automation figure on the public record, PZU taking detailed review from about 20% of body-shop claims to nearly all, comes from a single 2020 vendor release. This post argues from that one case and is honest that TIN holds it as unconfirmed.
Built on verified case files. The argument below leans on evidence The Internet Ninja validated against the public record and published in full, method included.
Insurance is where AI claims automation should be easiest to prove. Every claim is a record, every payout a number. So you would expect hard, audited figures.
You mostly do not get them. The most concrete public figure in the category, an insurer going from detailed review of a fraction of its claims to nearly all of them, rests on a single vendor press release from 2020. This post argues from that one case, PZU and Tractable, and is honest about how thin the ground under it is.
What ai claims automation means
ai claims automation is software that assesses and processes insurance claims with limited human input, commonly using computer vision to read damage photos and flag anomalies for a human to review. It changes coverage, how many claims an insurer can inspect in detail, more than it changes the final decision.
What is the actual figure on record?
PZU, the largest insurer in Poland and Central and Eastern Europe, deployed Tractable’s computer-vision AI on body-shop motor claims. Per the announcement, before AI it performed “a detailed review of around 20 per cent” of those claims, and the AI now lets it “check in detail nearly all of the body shop claims it processes, in real time,” out of “nearly 500,000 motor damage claims per year” (source).
By the November 2020 announcement the system had “handled over 150,000 claims, or 1/3 of PZU’s yearly auto claims volume, worth PLN 1.3bn (GBP 260m)” (source). The gain is coverage: from reviewing one in five body-shop claims closely to reviewing nearly all of them.
Why one case, and why the caution
Two or more sources let TIN write from firm ground. This term has one. TIN’s case file on PZU records why it stays single-sourced: the figures come from a Tractable-issued release with PZU named, and the Life Insurance International piece often treated as a second source “merely repeats that same vendor release,” so it is not independent corroboration (source).
TIN has not cleared that case to a green badge. The figures are vendor-originated, not third-party-audited, are as of 2020, and the load-bearing phrase “nearly all” is never defined. Each of those is a reason to read the number as a starting point, not a proof.
| What the record supports | What it does not |
|---|---|
| PZU reviewed about 20% of body-shop claims before AI | An audited, independent measurement |
| It now reviews “nearly all” in real time | A definition of “nearly all” |
| Over 150,000 claims handled by Nov 2020 | Any figure updated past 2020 |
| Roughly 500,000 motor claims a year | A second, independent source |
What to demand before you buy ai claims automation
Ask for the denominator and the date. A “nearly all” with no definition and a five-year-old vintage is a claim, not a result. Ask whether any figure was measured by someone other than the vendor or the insurer. In this case, none was. That does not make the deployment fake; it makes the number unproven, and those are different things a buyer must keep separate.
The bottom line
ai claims automation has one concrete public figure worth citing, PZU moving from detailed review of about 20% of body-shop claims to nearly all, and it comes from a single 2020 vendor release with no independent measurement behind it. Treat that as the shape of a real capability, not as evidence of a delivered outcome. Until an insurer or an outside party publishes audited numbers, the honest position on the category is interested and unconvinced.
Sources
- PR Newswire (Tractable release), “PZU is first Polish insurer to use Tractable’s AI to analyse auto damage,” 2020-11-27. https://www.prnewswire.com/news-releases/pzu-is-first-polish-insurer-to-use-tractables-ai-to-analyse-auto-damage-301180812.html
- Life Insurance International, “PZU to use Tractable’s AI solution to analyse car damage,” 2020-12-01. https://www.lifeinsuranceinternational.com/news/pzu-tractables-ai-solution-analyse-car-damage/
Questions
What is ai automation in insurance claims?
It is software that assesses and processes claims, often with computer vision reading damage photos, so an insurer can review more claims in less time. PZU used Tractable's AI to check body-shop motor claims in real time and flag anomalies for human review.
How much of the claims process can ai claims automation cover?
In the one figure on record, PZU went from a detailed review of about 20% of body-shop motor claims to nearly all of them, out of roughly 500,000 motor claims a year. The word 'nearly all' is undefined in the source, so treat the ceiling as unstated.
Is the PZU claims-automation result independently confirmed?
No. The figures come from a 2020 Tractable-issued release, and the trade-press piece often cited as a second source repeats that same release. TIN holds the case as single-sourced and has not cleared it to a green badge.
Sources
- PR Newswire (Tractable release), PZU is first Polish insurer to use Tractable's AI to analyse auto damage , 2020-11-27
- Life Insurance International, PZU to use Tractable's AI solution to analyse car damage , 2020-12-01
This is analysis, not a verified outcome. It carries no verification badge and never will. The proof lives in the case files, where every figure is checked against the public record and the method is printed on the page.