Sometimes, but rarely on your own say-so. AI developers building claims-automation and damage-assessment tools value claim files that link photos, estimates, adjuster notes, reserves and payments to a final outcome. The catch is ownership and law: state third-party administrator laws commonly say the insurer owns the records an administrator generates, the Gramm-Leach-Bliley Act and state insurance privacy rules limit disclosure of policyholder data to outside companies, and health details need written authorization or proper de-identification. As of November 2026 we found no public closed price for a claims-data licence, so treat any figure you are quoted as a starting point for negotiation, not a market rate.
- Outcome-labelled claim files, where photos, estimates, notes and payments lead to a known result, are the part AI developers find hardest to get elsewhere.
- Check who owns the file first. Several state TPA laws say the insurer owns records the administrator generates; adjusting firms work under carrier contracts.
- Expect to de-identify. Health details need authorization under NAIC Model #672 unless the data no longer identifies anyone, and HIPAA applies to health plan claims.
- There is no public closed price for claims data yet. Value depends on years, lines, linked outcomes, photo coverage and clean rights.
Can you sell insurance claims data?
You can license claims data when you hold the rights to it and the law allows the disclosure, which usually means a de-identified copy under a written licence. Selling the raw files, with names, policy numbers and medical details, is rarely possible for anyone other than the insurer, and even an insurer faces privacy rules on disclosure to outside companies.
This guide is for third-party administrators (TPAs), independent adjusting firms, managing general agents (MGAs), agencies and small carriers. It covers what a claim file holds, who owns it, the laws that apply, how to de-identify it and what a licence looks like. Before preparing anything, see what your data is worth: the free valuation uses company-level answers, so no claim files leave the business.
Agencies usually hold far less claims detail than carriers and TPAs. For quoting, renewal and servicing records, see selling insurance agency data.
What is in an insurance claim file?
A claim file is a dated record of every step from the first report to closing. State unfair claims settlement rules require it: in West Virginia, for example, claim files "shall contain all notes and work papers pertaining to the claim in such detail that pertinent events and the dates of such events can be reconstructed" (W. Va. Code R. 114-14-3). That rule is why claim files make good training data: the sequence of decisions is written down.
| Part of the file | What it records | Why an AI developer cares |
|---|---|---|
| First notice of loss (FNOL) | Date, cause, location and the claimant's first description | The starting input for triage and routing models |
| Coverage review | Policy terms checked, exclusions considered, coverage decision | Shows how a policy is read against facts |
| Adjuster notes and diary | Calls, inspections, decisions and the reasons for them, with dates | The reasoning between input and outcome, which is rare in any other data |
| Photos paired with estimates | Damage images with line-item repair or replacement estimates | Labelled examples for damage assessment and estimating |
| Reserves | Initial and revised estimates of what the claim will cost | How expectations changed as facts arrived |
| Payments | Amounts, payees and dates | The ground truth for settlement models |
| Subrogation and salvage | Recovery attempts from other parties and salvage sales | Recovery prediction |
| Litigation | Suits, counsel assignments and results | Severity and litigation-risk signals; much of it may be privileged |
| Outcome | Paid, denied, settled, withdrawn or reopened, and the final cost | The label that turns everything above into a training example |
Source: W. Va. Code R. 114-14-3: claim file and record documentation (unfair claims settlement rule)
Why are outcome-labelled claims valuable for AI?
Insurers already use AI in claims, so the tools need examples. In an NAIC survey of 193 large private passenger auto insurers (2021 data, reported in December 2022), 135 reported using AI or machine learning for claims operations. Uses included evaluating images of loss (55 companies) and determining claim settlement amounts (50) (NAIC; summarised by InsuranceNewsNet).
The NAIC summarises property and casualty use the same way: in claims, AI is used for accident image analysis, to estimate ultimate claim settlement values and for fraud detection. Each of those tasks needs a known answer to learn from. A photo with its estimate and the amount finally paid, or a set of notes with the final decision, is that answer.
That is the difference between claim files and most other insurance data. Policy and quote records show what was sold. Closed claims show what happened and what it cost, over years, with the adjuster's reasoning in between.
Source: NAIC: Private Passenger Auto AI/Machine Learning Survey report (December 2022)
Who owns the claim file: carrier, TPA or adjuster?
Usually the insurer or the self-insured client, not the firm that handled the claim. Several states put it in statute. Missouri says "the insurer shall own the records generated by the administrator pertaining to the insurer", while the administrator keeps a right of access to meet its obligations (RSMo 376.1082); New Hampshire uses the same words (RSA 402-H:4). Independent adjusters keep records under state licensing law and must follow any longer retention policy in their contract with the insurer or self-insurer, as Nevada provides (NRS 684A.180).
- A firm that handled claims for many carriers holds many owners' data. Each carrier's consent is a separate negotiation.
- The data you clearly own, such as procedures, training and operational metrics, can still be worth licensing on its own. See licensing workflow data for AI agents.
| Your business | Who usually controls the claim file | What you may control | What to read first |
|---|---|---|---|
| Third-party administrator | The insurer, or the self-insured employer or plan that hired you | Your own procedures, workflow and cycle-time data, staffing and system configuration; de-identified statistics only if the contract allows | The administration agreement: ownership, confidentiality, data use and return-or-destroy clauses |
| Independent adjusting firm | The carrier or self-insurer that assigned the claim | Your own guidelines, training material, scheduling and productivity data | Each carrier's assignment terms and vendor agreement |
| MGA | Depends on the program agreement; claims may be handled by the carrier, the MGA or a TPA | Underwriting and program operations data the agreement assigns to you | The program or agency agreement with each carrier |
| Insurance agency | The carrier; the agency usually sees first reports and service notes only | Quoting, renewal and servicing workflows | Agency-company agreements and your privacy notice |
| Small carrier | You, subject to privacy law and any reinsurance or fronting terms | The full file, once privacy and contract rules are met | Your privacy notices, reinsurance treaties and state law |
Source: Missouri RSMo 376.1082: records maintained by an administrator are owned by the insurer
Which laws apply to licensing insurance data?
Several layers can apply at once. This is a map of the main rules, not legal advice; have insurance counsel review any licence.
- HIPAA does not reach every claim with an injury in it. Workers' compensation, automobile medical payments and liability insurance are not health plans under HIPAA, HHS says, so those claims fall under state insurance privacy law instead.
- Health information that does not identify anyone, and gives no reasonable basis to identify anyone, is outside the definition of nonpublic personal health information in Model #672 and the state rules based on it (11 NYCRR 420.3). That is why de-identification is the usual route.
| Rule | What it covers | What it means for a licence |
|---|---|---|
| Gramm-Leach-Bliley Act, 15 U.S.C. 6802 | Nonpublic personal information held by financial institutions, including insurers | Disclosure to a nonaffiliated third party needs a privacy notice and a chance to opt out. For insurers, state insurance authorities enforce it (15 U.S.C. 6805) |
| NAIC Model Regulation #672 | State rules on consumer financial and health information held by licensees | Nonpublic personal health information may not be disclosed without the person's authorization, outside a list of insurance functions such as claims administration. An authorization lasts no more than 24 months. New York's version is 11 NYCRR 420.17 and 420.18 |
| Insurance Information and Privacy Protection Acts | Personal or privileged information collected in an insurance transaction, in the states that have them | California, for example, bars disclosure unless an exception applies, such as written authorization |
| NAIC Insurance Data Security Model Law #668 | Security of nonpublic information held by insurers, producers and other licensees | Adopted in many states, with local changes. Licensees need a retention schedule and must require third-party service providers to protect the data (Ohio 3965.02, South Carolina). The model exempts licensees with fewer than 10 employees from the program; Ohio's threshold is fewer than 20 (Ohio 3965.07) |
| HIPAA | Health plan claims, and TPAs that process claims for a health plan as business associates | A business associate may use protected health information only as its business associate contract allows, and no one may sell it without authorization (45 CFR 164.502) |
| CCPA and other state privacy laws | Personal information of state residents | Information collected under GLBA is exempt from the CCPA, but data outside GLBA, such as claimant, witness or employee data, may not be |
Source: NAIC Model Regulation #672: Privacy of Consumer Financial and Health Information
What has to stay out: fraud data and third-party reports
Claim files collect reports that the claims team was only allowed to use for the claim. Remove them before any copy is made, even when the rest of the file can be licensed.
- Motor vehicle records. Under the Driver's Privacy Protection Act, insurers may use them for claims investigation, antifraud activities, rating or underwriting, and a recipient may resell or redisclose them only for a permitted use (18 U.S.C. 2721). Leave them out of any copy you license.
- Industry claim and fraud databases. Their terms limit use to claims and fraud work. For example, vehicle data reported through ISO ClaimSearch for the federal title system is held "solely for non-commercial, law enforcement, insurance investigative and anti-fraud purposes" and will not be resold without the reporter's written agreement (ISO ClaimSearch).
- Special investigation unit referrals and fraud bureau reports. Keep them out unless counsel confirms otherwise; they can name people who were never charged or found at fault.
- Medical records obtained from providers, and communications with defense counsel, which may be privileged.
Source: 18 U.S.C. 2721 (Driver's Privacy Protection Act): permitted uses and resale
How to de-identify insurance claims data
Claim files are hard to de-identify because the identifying details sit in free text and photos, not just in fields. Plan for tools plus human review.
- Remove or replace names, addresses, phone numbers, emails, policy and claim numbers, and payee and bank details everywhere, including notes, letters and file names.
- Treat vehicles and property as identifiers: VINs, license plates, house numbers and street views in photos.
- Blur faces, plates, documents and screens in photos and video. Check image metadata for location and device details.
- Generalize dates to month or year and locations to region, unless the licence needs more and the law allows it.
- For health plan claims, use HIPAA Safe Harbor or Expert Determination. The de-identification guide lists the 18 identifiers and where they hide.
- Keep a consistent token per claim if the buyer needs to follow one claim over time; that makes the data pseudonymized, which is still personal data in many laws. See anonymization vs pseudonymization.
How much is insurance claims data worth?
There is no public price to point to yet. As of 6 November 2026 we found no publicly reported closed deal for a licence of insurance claims files, only listings and sales pages without prices. Anyone quoting a per-claim or per-file rate as a market price is guessing or selling.
What we can say is what moves value up or down, based on what AI developers ask for in the published deals for other company records (deals tracker, what company data is worth):
- Structures vary: a one-time fee, a fee per year of data, per-record pricing or a revenue share. See data licensing pricing models.
- If carriers own the files, any payment may need to be shared with them or agreed with them first.
| Raises value | Lowers value |
|---|---|
| Many years of closed claims with final outcomes | Open claims or files without a recorded result |
| Photos linked to line-item estimates and amounts paid | Photos without estimates, or estimates without photos |
| Detailed adjuster notes with reasons | Short codes or blank diaries |
| Several lines of business and regions | One small book in one state |
| Clean rights: carrier consent in writing, clear de-identification | Unclear ownership across many carriers |
| Consistent fields and exportable systems | Scanned paper files and mixed formats |
What does a claims data licence look like?
A claims data licence is a contract for a defined, de-identified copy. These are the terms to settle before anything is delivered; the data licensing agreement guide covers each clause.
| Term | What to agree |
|---|---|
| Scope | Lines of business, years, states, fields and attachments included; what is excluded, such as fraud referrals and medical records |
| Rights | Who owns the files and written consent from each carrier or self-insured client whose claims are included |
| De-identification | The standard used, who does the work, and a test on a sample before delivery |
| Permitted use | Training, evaluation or retrieval, and whether the buyer may build products for insurers. See training vs retrieval licences |
| Exclusivity | Exclusive or non-exclusive, by line or region and for how long. See exclusive vs non-exclusive |
| Protections | No re-identification, no onward disclosure, security controls a Model #668 licensee would expect of a service provider, deletion or return at the end |
| Payment | Fee structure, milestones and how any carrier share is handled |
Source: NAIC Model Law #668: Insurance Data Security Model Law
First steps for a TPA, adjusting firm, MGA or agency
Start with a description, not an export. Claim files are among the most useful records an insurance business holds, and among the hardest to license, because ownership and privacy rules come first. No claim file needs to leave the business to learn whether a licence is worth pursuing: see what your data is worth with a free valuation.
Origin Data Partners is an independent data partner and is not affiliated with any AI data buyer. We help owners compare buyers and find the best fit for their records; a buyer may pay Origin when a deal closes. See what your data is worth.
- List the lines of business, years, number of closed claims and systems, and whether photos, estimates and notes are linked in the same record.
- For each carrier or client, note what your contract says about ownership, confidentiality and data use.
- Separate what you own outright, such as procedures, training and operational metrics, from client claim files.
- Mark the sensitive parts: health details, minors, fraud referrals, litigation and third-party reports.
- Use the data inventory template to record it before spending on de-identification.
Common questions
Can you sell insurance claims data?
Yes, in some cases, usually as a de-identified copy under a written licence. You need the rights to the files, which often belong to the insurer or self-insured client rather than the TPA or adjuster, and the disclosure must fit GLBA, state insurance privacy rules and, for health plan claims, HIPAA.
Who owns claims data, the TPA or the insurer?
Usually the insurer. Several state TPA laws, such as Missouri's RSMo 376.1082, say the insurer owns the records generated by the administrator, while the administrator keeps a right of access to do its job. For self-funded plans, the administration agreement with the employer or plan decides.
Can an independent adjuster sell claim photos?
Not without the carrier's agreement in most cases. Photos taken on an assignment belong to the claim file the carrier or self-insurer controls, and they often show faces, plates, homes and documents. An adjusting firm can more easily license its own procedures, training and operational data.
Is insurance claims data covered by HIPAA?
Health plan claims are, and TPAs that process them are business associates. Workers' compensation, auto medical payments and liability insurance are not health plans under HIPAA, according to HHS, but state insurance privacy rules such as NAIC Model #672 still restrict health information in those claims.
Does GLBA allow insurers to sell customer data?
Only within limits. Under 15 U.S.C. 6802 an insurer may not disclose nonpublic personal information to a nonaffiliated third party without a privacy notice and a chance for the consumer to opt out, and state rules add authorization requirements for health information. De-identified data that no longer identifies anyone is the usual route.
How much is insurance claims data worth?
There is no public benchmark. As of November 2026 we found no publicly reported closed price for a claims-data licence. Value rises with years of closed claims, linked photos, estimates and payments, detailed notes, several lines of business and clean written rights.
What is claims data used for in AI?
Training and testing claims-automation tools: triage and routing, damage assessment from photos, estimating, settlement prediction and fraud detection. The NAIC reports that auto insurers use AI in claims for image analysis and to estimate ultimate settlement values.
What is insurance data monetization?
Earning revenue from insurance data beyond its original use, for example by licensing de-identified claims, underwriting or servicing records to AI developers or analytics firms. For insurance businesses it is limited by who owns the data, GLBA, state privacy rules and carrier contracts.
Can a TPA license its own data without carrier consent?
Often yes for data that is genuinely its own, such as procedures, training material, workflow and cycle-time metrics and system configurations, as long as client claim details are not included. Client claim files, and statistics derived from them, depend on what each administration agreement allows.
Sources and further reading
Origin Data Partners is independent and not affiliated with any buyer named on this page.
- W. Va. Code R. 114-14-3: claim file and record documentation (unfair claims settlement rule)
- NAIC: Private Passenger Auto AI/Machine Learning Survey report (December 2022)
- NAIC: Insurance topics, Artificial Intelligence
- Missouri RSMo 376.1082: records maintained by an administrator are owned by the insurer
- New Hampshire RSA 402-H:4: maintenance of information by administrators
- Nevada NRS 684A.180: adjuster records and contract retention policies
- 15 U.S.C. 6802 (Gramm-Leach-Bliley Act): disclosures of nonpublic personal information
- 15 U.S.C. 6805: enforcement, including by state insurance authorities
- NAIC Model Regulation #672: Privacy of Consumer Financial and Health Information
- California Insurance Code 791.13 (Insurance Information and Privacy Protection Act)
- NAIC Model Law #668: Insurance Data Security Model Law
- South Carolina Insurance Data Security Act, S.C. Code Title 38 Chapter 99
- Ohio Revised Code 3965.02: information security program
- Ohio Revised Code 3965.07: exemptions
- 11 NYCRR 420.17 (New York): when authorization is required to disclose nonpublic personal health information
- 11 NYCRR 420.18 (New York): authorizations
- 11 NYCRR 420.3 (New York): definitions
- InsuranceNewsNet: NAIC survey, 88% of insurers are using AI or machine learning
- HHS: Business associates
- HHS FAQ 364: Are disability, workers' compensation and auto liability insurance covered under HIPAA?
- 45 CFR 164.502: uses and disclosures of protected health information, including sale
- California Civil Code 1798.145 (CCPA exemptions, including GLBA-covered information)
- 18 U.S.C. 2721 (Driver's Privacy Protection Act): permitted uses and resale
- ISO ClaimSearch: NMVTIS reporting service terms and conditions