LO
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.
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.
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.
Taking the claim
Automating✓ Evidence-backedFirst notice of loss: what happened, when, what is damaged, what the policy is.
Structured intake against a policy document is the ideal case for automation: the questions are known in advance, the answers are checkable against records the insurer already holds, and a wrong answer surfaces immediately rather than years later. Lemonade's own annual report states that as of the end of 2023, 98% of the time its claims bot takes the first notice of loss and pays or declines without human intervention.
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.
Deciding whether it is covered
Being augmented≈ Platform inferenceReading the policy against what actually happened, including the exclusion nobody reads until it matters.
Where the facts are clean and the policy is standard, this is rule application and machines do it consistently — more consistently than a tired person at the end of a shift. What does not transfer is the case where the facts are contested, because deciding which version of events to believe is not a rule, and the person deciding has to be answerable for having believed it.
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.
Putting a number on the loss
Being augmented≈ Platform inferenceWhat it costs to repair or replace, and whether the estimate in front of you is the real one.
Photo-based damage estimation and parts pricing are now ordinary and genuinely fast — the cost side of this task is well-suited to a model trained on millions of repairs. What is left is the argument: a repairer who says the estimate is too low, a claimant whose item has no market price, and the decision about who absorbs the difference.
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 is recorded here.
The claim that does not fit
Still human-led≈ Platform inferenceThe one where the story does not add up, or the loss is real but the policy was never written for it.
Automation gets better at the cases that look like previous cases, which is the definition of what it cannot do here. The exception is also where an insurer's money and reputation actually sit: a handful of mishandled unusual claims costs more than thousands of routine ones, and knowing which unusual claim is the expensive one is judgement about this book of business.
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.
Telling someone no
Still human-led≈ Platform inferenceDeclining a claim to a person who believed they were covered, and holding that conversation.
A system can produce a decline; it cannot be the party that answers for it. In most markets a declined claim comes with a right to an explanation, a complaint route and a regulator, and each of those needs a person who can be asked why. This is the task that turns a claims department from a processing function into an answerable one.
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. This site holds no verified record of enforcement against an insurer over an automated decline.
Spotting the invented claim
Being augmented≈ Platform inferenceFraud: the staged accident, the pre-existing damage, the loss that happened three days before the policy started.
Pattern detection across a whole book is something people cannot do at all, so this is one of the clearest cases where the machine adds capability rather than replacing it. What stays human is the decision to act on a flag: accusing a customer of fraud is a legal act with consequences, and a model's confidence is not evidence.
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. Fraud-detection performance is also rarely published by insurers, so this rests on the shape of the problem rather than on a measurement.
Which technologies matter here#
Four separate signals. They are deliberately not added together — a job exposed to two technologies is not twice as exposed.
How it got here#
The index is not a static number. This is where it would have sat at each capability checkpoint since ChatGPT — reconstructed, and labelled as such.
● 1 verified event for this occupation, plotted at the date it happened — the parts of the line near a marker are anchored to something checkable.
A high start that predates generative AI: rules engines and straight-through processing have been eating simple personal-lines claims since the 2000s. The 2023-2025 climb is the intake end — reading a free-text account of what happened, matching it to a policy and settling it — which had resisted rule engines because it needed language. One insurer's own annual report puts 98% on that step. It flattens because the remaining work is not processing: the claim whose story does not add up, and the person who has to answer for a decline. Read the level as a warning about the routine half rather than about the occupation, and note what the curve cannot see — the number of people needed per thousand claims, which is the number a business case is actually built on.
A flat line is not a forecast of safety. It says which tasks automation has reached so far — the occupations that moved least here are the ones where the constraint is physical or regulatory, and both of those can change.
Recent changes#
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.
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
A declined claim comes with a right to an explanation and a complaint route in most markets. 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 most markets a declined claim carries a right to an explanation and a route to complain, and those attach to the company regardless of what produced the decision. 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.
Method and sources#
- Assessment date
- 2026-09-13
- Basis of the task judgements
- 1 evidence-backed · 5 platform inference · 0 not enough evidence
- Verified events
- 1