Claims are a workflow, not one task
A claim passes through many steps: first notice of loss, document collection, coverage review, investigation, evaluation, communication and payment or denial. Each step has different volume, different information and different risk. That is why the useful question is not whether to use AI in claims, but which steps.
Steps that often suit AI
Intake. Reading a first report, extracting key details and checking what is missing, so adjusters don't start from a blank form.
Document handling. Pulling facts out of emails, photos, estimates and forms, and filing them against the right claim.
Triage and routing. Sorting claims by type and complexity so simple ones move quickly and complex ones reach the right person sooner.
Preparation for people. Gathering policy details, claim history and relevant notes into one view for the adjuster.
Status communication. Drafting routine updates for a person to approve or send, so customers aren't left waiting for news.
Signals for review. Flagging unusual patterns for a specialist to examine. A flag is a prompt for a person to look, not a conclusion.
Decisions that should stay with people
Coverage determinations, denials, settlement decisions and anything contested or sensitive. These affect people's finances and rights, and they are where judgment, empathy and accountability matter most. Designing the workflow so the AI hands these over, with the facts already assembled, is usually more valuable than trying to automate the decision.
Regulation and oversight
Insurers face expectations about fairness, transparency and record-keeping in how they use automated tools. In the United States, the National Association of Insurance Commissioners has published a model bulletin on insurers' use of AI systems, and many states have adopted guidance based on it. Requirements vary by state and line of business, so involve your compliance and legal teams early, and keep logs that show what the system did and what a person decided.
How to measure it
Pick measures before you build: time from first notice to first contact, time spent per claim, the number of times a claim is touched, rework and error rates, and customer satisfaction on claims. Measure a baseline now so you can compare honestly after launch.
Where to start
Choose one frequent, well-understood claim type and one or two steps, such as intake and document extraction, rather than the whole process. Keep adjusters involved from the first week, and review outputs together.
You can read more in the insurance chapter of our free Forward Deployment Playbook and on our insurance industry page, or try the free AI Opportunity Assessment to see which services fit your situation.