Insurance claims data can be licensed to AI companies when it is de-identified and your privacy notices and state insurance rules allow the disclosure. Labs value claims files because they show a full decision from first notice of loss through adjuster notes, payment or denial, and that sequence is hard to find outside an insurer, MGA or claims administrator.
This guide is general information, not legal advice. Insurance privacy law varies by state, so talk to counsel before licensing claims records.
What AI labs want from claims files
Labs training models for insurance work need to see how real claims get handled. A policy document shows the rules. A claims file shows how people applied them. The most useful records include:
- First notice of loss, with the loss type, date and coverage involved
- Reserve changes over the life of the claim, and why they moved
- Adjuster actions in order: documents requested, inspections ordered, referrals to special investigations
- The decision: paid, partially paid, denied or subrogated, with amounts
- Underwriting files showing how risks were rated and what was declined
Ordered actions with an outcome form an episode, and episodes from claims systems are some of the most valuable data we see, because the outcome (payment, denial, litigation) tells a model whether the handling worked. The insurance industry page shows a sample record and what we scrub from it.
Rules that apply before anything else
Claims records are full of personal information: names, addresses, medical details in injury claims, bank details for payments. Several layers of rules can apply.
Federal financial privacy rules
Insurers count as financial institutions under the Gramm-Leach-Bliley Act. Its privacy rules, such as the FTC's version in 16 CFR Part 313, require notice and a chance to opt out before nonpublic personal information is shared with nonaffiliated third parties, with listed exceptions. For insurers, state insurance regulators apply these requirements through their own privacy rules.
State insurance privacy laws
State insurance privacy laws generally limit disclosure of personal information gathered in connection with an insurance transaction, and some require written authorization for health information. Because each state adopted the models differently, the answer depends on where your policyholders live. The NAIC's big data and AI resources also expect insurers to govern third-party data and AI use, which matters if a buyer later uses the data to build insurance models.
Health and sensitive data
Injury, disability and health claims carry medical information. We leave health details out unless they are already de-identified under an accepted method, and we never treat removing names as enough. Our guide to de-identification vs anonymization explains the gap.
Data you handle for someone else
Third-party administrators and MGAs often process claims for carriers. If your contract says you handle the data only to administer claims, licensing it needs the carrier's written permission. Read can I sell data I hold for my customers before assuming the files are yours to sell.
How claims records get de-identified
Claims files combine structured fields and free text, which makes them harder to clean than a ledger. Our process removes or transforms:
- Claimant, insured and witness names, replaced with stable pseudonyms
- Policy, claim and bank account numbers
- Street addresses, generalized to a region
- Exact dates, generalized to month or quarter
- Exact amounts, rounded into ranges when a rare amount could single someone out
- Adjuster notes and medical narratives, dropped unless a person has reviewed them
Then we test whether any record can be re-identified from a combination of fields, such as a rare loss type in a small town in a given month. Records that can be singled out are generalized further or removed. The method behind this, k-anonymity, is explained in plain English in a separate guide.
Who buys claims data
Insurance has long-established data vendors that sell scores, reports and fraud tools built on industry claims data. AI labs are a different buyer. They want de-identified files to teach models how claims work, and they license them under terms that ban re-identification, resale and combining with other data. You approve each buyer before a license is signed and can exclude any company, including competitors and other carriers.
What claims data sells for
No public price exists for licensing insurance claims files to AI labs. Vendor products in insurance are sold through enterprise contracts, and AI training licenses for claims data are private. We won't print a number we can't source.
What moves the value, in our view:
| Factor | Effect on value |
|---|---|
| Complete files with outcomes | Lets us build episodes |
| Reserve and action history | Shows decisions over time |
| Lines of business covered | Specialty lines are rarer |
| Years of history | More claim cycles, more outcomes |
| Clear disclosure rights | Unclear rights can block the sale |
The insurance data prices page lists the public signals that do exist. For a range on your own files, the calculator includes claims and underwriting files as a source and estimates from your team size and years of history.
New claims, new batches
Claims keep arriving and old ones keep developing. After the first license sells your history, the SDK can run each month, pull claims created or changed since the last run, scrub them on your machine and upload a batch. Each license with an active refresh term pays for every accepted batch, and nobody on your team has to do anything once the schedule is set. Read how this builds on the recurring data revenue page.
Start with a valuation
Use the valuation calculator for an estimate, or read the full process on the seller hub. On the first call we ask about your privacy notices, states and carrier contracts before anything else.
Frequently asked questions
Can an insurer legally sell claims data to an AI company?
Sometimes. It depends on your privacy notices, the states your policyholders live in, whether health information is involved and whether the data is properly de-identified. Federal and state rules restrict sharing nonpublic personal information with third parties, so counsel should review the disclosure first.
We are a TPA. Are the claims files ours?
Usually not outright. Third-party administrators typically process claims on a carrier's behalf under a contract that limits use. You would need written permission from each carrier whose files you want to license.
Do adjuster notes get included?
Only after review. Notes and medical narratives are free text that often contains names, diagnoses and other identifying details, so we drop them by default and include them only once a person has checked them.
Can we exclude competitors as buyers?
Yes. You approve every buyer before a license is signed and can rule out any company.