LO
Get told when a verified record lands on this occupation → · Mark which of these tasks are yours (VOLO Pro, free) → · For a team or a graduating class →
Insurance claims handler
Decides whether a policy pays, how much, and who has to be told no — on the worst day someone has had in a while.
Find the five largest claims your team paid last year and check what made each one large. If the pattern is 'nobody caught it early', that is the job.
This is not a probability of losing your job. It combines how much of the role's task load is exposed to automation with how far adoption has actually gone — useful for comparing occupations on one consistent basis, and for nothing else.
Written for claims handling in personal lines — motor, home, health, travel — where volume is high and most claims look like each other. Complex commercial, liability and large-loss adjusting is a different job with different exposure, and so is underwriting, which decides who gets a policy rather than what a policy pays. Regulation matters here: what an insurer must disclose about an automated decision differs sharply by market.
What is actually changing#
The unit of analysis is the task, not the job title. A role is not replaced — its task mix shifts.
Each tile is one task. Its size is how much of the job it is; its colour is where the task is heading. Click a tile to see what the judgement does not establish.
That figure is one direct-to-consumer insurer writing simple personal lines, and the same filing's next sentence says claims the bot is not authorised to settle, or where it identifies concerns, are routed to human claims experts. So it establishes that the intake end is automated at that company, not that claims work no longer needs people. It also says nothing about traditional insurers running older systems, which is most of the market by premium.
A judgement about the structure of the work, not a measurement of how often facts are contested — and that share differs enormously by line of business. It also does not say consistent is the same as correct: a rule applied consistently to a badly written policy produces consistent unfairness.
Says nothing about accuracy on unusual items or in markets where parts supply is volatile, and nothing about whether faster estimates changed what claimants receive. Neither has been measured publicly.
A judgement about where the risk sits, not a measurement of how many claims are exceptions. It also does not claim humans handle exceptions well — a tired handler at volume is exactly how an expensive claim gets missed.
It protects the answerability, not the headcount. One person can answer for many automated declines, and in several markets the explanation given is itself templated. What would settle how much the duty weighs is an enforcement action against an insurer over an automated decline, which has not surfaced.
Nothing here measures false positives, which are the whole cost of this task — a wrongly flagged claimant is a person accused of a crime by a statistical model. Regulators' survey counts how many insurers use such models and one insurer reports what it saved, but none reports how many flagged claimants turned out to be honest.
Is this your job? Say so and this page narrows to your share of it.
A job title is a bundle of tasks bought together, and no two people hold the same bundle. Nothing is sent anywhere — it stays in this browser.
Read all 6 tasks in full — direction, reasoning and limits →
Recent changes#
Two of Japan's largest non-life insurers, both in the MS&AD group, announcing with their image-analysis partner a function in their claims-payment work: it analyses the electronic features and accompanying information of the photos customers submit to prove a loss, in several layers, to detect images that may have been created or processed with generative AI, or altered or reused. The result is used as reference information in the claims departments, and the release states that the final decision on paying a claim is made by the person in charge, as before. The insurers call it a first stage. The image-analysis company co-signs the release and shares a pending patent, so it has a stake. There are no figures on volumes, detection rates or false alarms.
An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
One of the world's largest insurers reporting on its own property and casualty business in 2025. It says AI that reads complex medical documents — medical records, admission and discharge records — was applied to end-to-end automatic settlement of non-auto insurance claims covering nearly a million cases, and that with it 70% of personal injury claims were settled automatically, in as little as 51 seconds. It also reports claims savings via smart fraud detection of RMB10.51 billion in 2025, above RMB10 billion for the third year running. These are the insurer's own figures: the savings have no stated method, and the report does not say what happens to claims the system does not settle, how many are reviewed by staff, or how many claims staff it employs.
An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
United States and Europe, one insurer. The company's annual report says that as of 31 December 2025 roughly 55% of its claims were automated from start to finish, and that its claims bot triages and assigns claims it is not authorised to settle, or where it identifies concerns, to human claims experts, matched by specialty, qualifications, workload and schedule. The company's own account; it does not give the size of the exception queue or the number of people working it.
An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
European Union, addressed to national competent authorities rather than directly to firms, and covering insurers and intermediaries across the value chain. Two things have to be read together or the record misleads. What it says should be in place: following Solvency II Article 46 and Article 258 of Delegated Regulation 2015/35, effective internal control across the whole AI lifecycle, with roles, escalation procedures and named responsible functions written into policy, staff trained for their role, and guardrails so the system behaves as intended. What it is not: paragraph 2.8 says the Opinion sets out no new requirements, and paragraphs 2.4 to 2.6 name, among other high-risk AI systems insurers may use, risk assessment and pricing in life and health insurance; they do not name claims handling, and they leave the classification without prejudice to Articles 6(3), 6(4) and 7 of the AI Act. So for a claims decision the Opinion points to no high-risk regime; what it says applies is ordinary internal-control law, plus a duty to tell the customer they are dealing with an AI system and to promote staff AI literacy. Nothing here requires a person to decide any individual claim.
Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.
One US direct-to-consumer insurer writing simple personal lines on systems it built itself. The filing's very next sentence limits the claim: claims the bot is not authorised to settle, or where it identifies concerns, are triaged and assigned to human claims experts. So this establishes automation of the intake step at this company, not that claims work no longer needs people, and it says nothing about traditional insurers running older policy systems — which hold most premium in most markets.
An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
A survey issued under market-conduct authority by nine US state insurance regulators to private passenger auto insurers writing more than $75 million in 2020; 193 responded, with answers due on 28 October 2021. Of the 193, 135 reported using AI or machine learning in claims. The models were mostly an informational resource for adjusters (96 companies); others were used to determine claim settlement amounts (50), to make claim assignment decisions (58) and to evaluate images of loss (55). Few used them for claim approvals (9) and none for claim denials. For fraud, 95 used such models, mostly as a referral of claims for further investigation (83). The data describe large US auto insurers in 2021, as reported by the companies to their regulators; they do not measure how much of an adjuster's time the models replaced.
An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
What this means for you#
The work you would be hired to do first — taking claims and processing the ones that look like all the others — is the half with a company's own annual report saying a bot does 98% of it. What holds is the exception and the refusal, and neither is learned by handling routine volume. Ask, in an interview, what share of claims reach a person at all.
Your position depends on whether your day is volume or exceptions. Volume is the part with a number on it in someone's business case. Exceptions are where a claims book actually loses money, and the people who can tell an expensive unusual claim from a merely unusual one are the ones a department cannot replace — but that is a smaller department than today's.
Your options#
Four directions, each with its real constraints and one thing you can test this week. Continuing as you are is a legitimate choice — it just has to be a chosen one.
Move to the exceptions desk
Routine claims are where the automation number comes from; unusual ones are where the money is lost. A department that automates the first still needs someone for the second.
Fewer seats, and getting one usually means having handled enough volume to recognise what is unusual — which is the experience the automation is removing.
Find the five largest claims your team paid last year and check what made each one large. If the pattern is 'nobody caught it early', that is the job.
Be the one who answers for the automated decision
In the EU, insurers are expected to keep effective internal control and human oversight over the AI systems they use, with named responsible functions. As more declines are produced by a system, somebody has to be able to say why this one — and that is a defined position, not a chore.
It requires understanding what the model actually used, which most insurers cannot currently tell you — expect to spend the first year finding out.
Take one automated decline and try to reconstruct, from the record alone, why it was declined. Note how far you get.
Cross to the side that prices the risk
Claims data is what pricing is built on, and someone who has seen how losses actually arrive knows things the pricing table does not.
It is a more quantitative job than claims, and in several markets automated pricing carries obligations of its own that you would be taking on.
Ask a pricing colleague which claims pattern they most wish they had known about a year earlier.
Common questions#
Ask it per task, because the halves are moving at very different speeds. One insurer's own annual report puts a number on the intake end — 98% of the time its bot takes the claim and pays or declines without a person. The same filing's next sentence says the claims it cannot settle go to human experts, and that routing is the part that has not moved. A signal you can check yourself: of the claims you touched last month, how many would have looked like every other claim to someone who had never seen them.
No, and the gap is the useful part. That figure is one direct-to-consumer insurer writing simple personal lines on systems it built itself. Most premium in most markets sits with traditional insurers running much older systems, where the same automation would require replacing infrastructure rather than configuring it. This site keeps capability, deployment and employment change separate precisely because the distance between them is where predictions about this industry have gone wrong.
That is the right question to ask an insurer, and a better one than asking whether AI is involved. In the EU an insurer must keep effective internal control and human oversight over its AI and tell customers they are dealing with an AI system, though nothing there requires a person to decide each claim. What is worth testing is whether the explanation is specific to your claim or a template — because a templated explanation means the answerable position this page describes is not currently staffed.
It depends entirely on which door you come in through, and that is worth settling before you accept. A role that is routine volume sits in the half a company's own report puts a percentage on. A role attached to exceptions, large losses or complaints sits in the half that is answerable, and it usually requires the volume experience that the first kind of role provides — which is the uncomfortable shape of this occupation right now. Ask in the interview what proportion of claims reach a person, and what happens to the ones that do not.
What these judgements rest on#
6 of 6 task judgements on this page are backed by a verified event and 0 are platform inference, each labelled where it appears. Behind them sit 2 technology dimensions, a reconstructed trajectory since language models reached the public, and 6 verified events.
See which technologies, how it got here, and the method →
Where it sits in the official classification: skills, knowledge, related jobs →
Other roles in the same function#
A company divides its work into functions before it divides it into jobs. These sit in Customer service alongside this one — a fact about org charts, not a judgement that they are similar or that they are changing in the same direction.