Sell insurance agency data to AI: a licensing guide for independent P&C agencies

Why AI companies want insurance agency data

AI labs are building agents that do office work, and they need examples of real work with real results. A commercial lines agency produces that all day. An account manager gets a contractor's renewal with a new subcontractor exposure and decides which markets to approach. A marketing specialist builds a submission, chases three underwriters and records two declinations and a quote. A CSR checks a certificate request against the waiver and additional insured wording in the client's contract. Each of those steps ends in an outcome the agency management system records: bound, declined, lost, renewed, endorsed, paid.

Three things make an agency's records stand out from other businesses its size:

• Nearly everyone works at a desk. Producers, account managers, CSRs, marketing staff and accounting all write into the same systems, so a 40-person agency produces a 40-person record, not a 10-person one.
• One system holds the history. Many agencies have run Applied Epic, AMS360, HawkSoft or EZLynx for 10 to 20 years, with every activity, attachment and policy change linked to an account.
• The forms are structured. ACORD applications give buyers a schema they already understand, and the judgment around them (which markets, what coverage, what to push back on) is the part a lab cannot easily create on its own.

Agencies are also adopting AI themselves. The 2026 Agency Universe Study found 46% of agencies use AI, up from 15% in 2024 (source: Big I and Future One, 2026 Agency Universe Study release (Sep 23, 2026)), with coverage form analysis (43%) and contract reviews (35%) among the uses (source: IA Magazine, key findings from the 2026 Agency Universe Study (Oct 1, 2026)). The same work patterns are what training-data buyers ask for. To check how your own agency would come out, you can see what your agency's records could qualify for on our main page.

AI buyers that name insurance agencies and insurance brokerage

These are the programs in our registry whose own pages name insurance, shown by type. Each cell is the program's own published term; our buyer directory lists the programs and links to where each says so.

The wording matters more than the label. Broker B lists insurance agency workflows almost line by line: quoting and binding, policy servicing, endorsements, renewals, claims assistance and carrier correspondence, and says policyholder personal data is stripped. Marketplace A lists claims, adjuster notes, underwriting and policy records, and names Applied Epic, EZLynx, HawkSoft, AMS360 and Vertafore among the systems it works with. Buyer program B names statements, underwriting files and claims under finance and insurance (source: program terms, Regulr buyer directory).

Other large programs take agencies on size rather than industry. Several accept most industries if you clear their headcount gate, and one broker's owner form offers a financial services and insurance option (source: program terms, Regulr buyer directory).

P&C agency records AI buyers value for training

The pattern across buyers is simple: they want what your people wrote, decided and resolved, with the people and businesses it describes removed. What carriers and content vendors supplied usually stays out. Here is how that maps to an agency.

Two rows need a closer look. Market summaries and quotes are among the most useful records an agency has, and the ones with the most carrier information inside them: premiums, rates and appetite are what carriers protect in their agency agreements. Buyers typically take the submission and the outcome (bound, declined, reason given) with carrier names and pricing removed. Claims notes split by line: property and liability advocacy can usually be de-identified, while workers comp and bodily injury files carry health information and usually stay out entirely.

Who owns an insurance agency's records: expirations and carrier contracts

Independent agencies start from a strong position. The default rule, traced to a 1904 New York decision often called the Yonkers case, is that the independent agent owns its expirations, and courts have extended that to the records, work product and data that go with them unless a contract says otherwise (source: IA Magazine, Independent Agents Own Their Data (Mar 2, 2021)). Big I's model appointment clause keeps “the Agent's expirations, including those on direct billed business, the records thereof, and the Agent's work product and data relating thereto” in the agent's “sole ownership” (source: Big I, Guide to Agency-Company Appointment Contracts, v1.0).

Three things in real contracts narrow that picture:

• Ownership can depend on paying the carrier. In one agency agreement filed with the SEC, expirations are the agent's property on termination only if it has met its monetary and reporting obligations; until premiums are paid, ownership “shall be vested with the Company” (§10.8) (source: Crum & Forster agency agreement, SEC exhibit 10.1 (2009), sections 8.1 and 10.8).
• Carrier material is the carrier's. The same agreement treats rates, forms, manuals and underwriting guidelines as trade secrets of the company, to be “used only for the purpose provided herein” (§8.1) (source: Crum & Forster agency agreement, SEC exhibit 10.1 (2009), sections 8.1 and 10.8).
• Shared client information has its own clause. Big I's model language has both sides agree not to redisclose or use clients' nonpublic personal information “except as necessary to carry out the terms of this Agreement”, and its guide notes a carrier may have a significant interest in keeping its commissions and quote data confidential (source: Big I, Guide to Agency-Company Appointment Contracts, v1.0).

In practice that means your own work product (notes, submissions you wrote, checklists, certificates, proposals, SOPs) is the core of a license, and carrier documents and pricing are the main exclusions. Your signed agency agreements govern, and each carrier's may differ. The details are in our guide to GLBA, NAIC Model 672 and carrier agency agreements.

Who qualifies to license insurance agency data to AI

Headcount gates decide most cases. Because almost every agency employee works at a desk, the count is usually simple: total full-time staff.

How many agencies fit, for context

The channel is large and mostly small. Big I estimates about 37,000 independent agencies, with an average of 9.9 staff in 2026, up from 8.2 in 2024 (source: Big I and Future One, 2026 Agency Universe Study release (Sep 23, 2026)). Census counts for 2022 show 120,434 firms in insurance agencies and brokerages (NAICS 524210), of which 2,140 had 20 to 49 employees, 549 had 50 to 99, and 547 had 100 to 499 (source: Census, Statistics of US Businesses 2022). Those counts include captive and benefits agencies. By our estimate, roughly 3,000 to 4,000 commercial-leaning independents with 15 to 250 staff are the agencies most buyers would look at.

For a sense of scale, the 2025 Best Practices Study put revenue per employee at $228,321 (source: Big I and Reagan Consulting, 2025 Best Practices Study release), so a 40-person agency is, on that average, a business of about $9M in commissions and fees (our arithmetic, not a study figure).

What usually makes a strong candidate

• An independent P&C agency or broker with a majority commercial lines book.
• 15+ years in business and 10+ years on the same AMS, or an archive of the old system.
• Still independent and owner-run, with no LOI signed.
• Carrier accounts current, with no unpaid premium balances.

Who usually does not fit

• Captive agencies, where the carrier owns or controls the book. In our reading, these are out.
• Personal-lines-only agencies, whose records are mostly consumer data.
• Benefits-heavy, life and health agencies, whose files are mostly health information.
• Title and escrow agencies, which at least one program excludes by name.
• Agencies that migrated AMS recently and kept under about three years of history.
• Agencies already sold to an aggregator or under LOI.

Commercial vs personal lines: why the book mix decides most agency data deals

Two agencies with the same headcount can hold very different records. The difference is usually the book mix, and it cuts two ways.

Commercial files carry more judgment. A middle market account generates a multi-carrier submission, an underwriting conversation, a coverage comparison, certificates for every project, mid-term endorsements, an audit and a renewal strategy. Each step is written down by someone who decided something. A personal auto policy, by contrast, is mostly a rating transaction followed by a few service calls.

Commercial files carry fewer consumers. GLBA and Model 672 protect individuals who buy insurance for personal, family or household purposes (source: 15 U.S.C. 6809 (Cornell LII), NAIC Model 672, Privacy of Consumer Financial and Health Information Regulation). A business insured is not one of them, so a commercial file is mostly business information. It still names people (owners, officers, drivers, injured workers, claimants), and those names come out, but the core of the record survives de-identification well. A personal lines file is about a consumer from the first line, so far less of it is left once the protected detail is removed.

Some practical points follow from that:

• Split the book before you scope it. Most AMS reports can separate commercial and personal lines by line of business or department. A license can cover the commercial side only.
• Watch small businesses that are really one person. When the insured is a sole proprietor or a one-owner LLC, the business name can identify an individual. Buyers usually treat those names as personal identifiers and replace them.
• Keep the benefits division separate. If you run group health, life or voluntary benefits, those files carry health information and usually stay out as a whole, even if the P&C side is in.
• Specialty programs need a second look. Transportation, contractors, habitational and professional liability programs are rich in judgment, but trucking files are full of driver records, and program business often runs on a carrier's or MGA's own forms and guidelines.

What buyers advertise paying for insurance agency data

No buyer publishes a price for insurance agency records specifically. What exists are each program's general ranges. A few that matter for an agency:

• Buyer program H, which names insurance brokerage: “$10K to $1M+” as an illustrative value per partnership, scaling with cadence (source: program terms, Regulr buyer directory).
• Marketplace A, which names insurance agencies and their AMS platforms: “$250K+ Typical company deal” (source: program terms, Regulr buyer directory).
• Broker D, which names insurance: “$50k-$4M+” (source: program terms, Regulr buyer directory).
• Broker B, which lists agency workflows: up to $4M, with a fixed sum agreed before any data is shared (source: program terms, Regulr buyer directory).
• Broker A: an estimator illustration of $385K to $895K for a 100-person, 10-year firm (source: program terms, Regulr buyer directory).
• Buyer program L: no range; every record is priced after it reads a de-identified sample (source: program terms, Regulr buyer directory).

Treat estimators with care: several share the same formula, so they are not independent valuations. Two things specific to insurance can move a price. First, the more carrier material has to come out, the less is left to license. Second, some AMS vendors already own de-identified data built from agency records, as Vertafore's terms state (§4.1, §4.3) (source: Vertafore Standard Terms, updated August 2026), and Zywave sells “millions of anonymized premium and quote records” from its broker-facing platforms (source: Zywave, Market Basket). Buyers may ask how your set differs, and true exclusivity over aggregate figures is not on offer. Our data value checker walks through the factors for your company.

What stays out of an insurance agency data license

Insurance has more regulated and third-party material inside its systems than most industries. Here is what the rules, carrier contracts and buyers say, in summary.

• Consumer nonpublic personal information. GLBA defines a consumer as an individual who obtains financial products for personal, family or household purposes (source: 15 U.S.C. 6809 (Cornell LII)). Personal lines files are full of it. Business insureds are not consumers, but the individuals named in commercial files (owners, drivers, claimants) can be. Model 672 treats information that does not identify a consumer, “such as aggregate information or blind data that does not contain personal identifiers such as account numbers, names or addresses”, as outside personally identifiable financial information (§4V(2)(b)(iii)) (source: NAIC Model 672, Privacy of Consumer Financial and Health Information Regulation).
• Health information, including workers comp and bodily injury claims. Model 672 covers health information wherever there is “a reasonable basis to believe” it could identify someone (§4U), and a licensee “shall not disclose” it without an authorization (§18A) (source: NAIC Model 672, Privacy of Consumer Financial and Health Information Regulation). Benefits, life and health books stay out for the same reason.
• MVRs and driver records. The Driver's Privacy Protection Act lets insurers use motor vehicle records for rating and underwriting, and permits research only so long as the personal information is “not published, redisclosed, or used to contact individuals” (source: 18 U.S.C. 2721 (Cornell LII)). MVRs, driver schedules and credit-based data stay out.
• Carrier trade secrets. Rating manuals, underwriting guidelines, appetite guides, commission schedules and carrier forms (source: Crum & Forster agency agreement, SEC exhibit 10.1 (2009), sections 8.1 and 10.8).
• Licensed content. Proposal and risk-management libraries from content vendors usually cannot be redistributed. Text your staff wrote is a different matter.
• Clients' own contracts. Leases and construction contracts sent for an insurance-requirements review are the client's confidential documents. Your review checklists are yours.
• Outputs of your AMS vendor's AI features. Vertafore's terms bar using its AI features or any output to train AI models (§2.8.3(b)) (source: Vertafore Standard Terms, updated August 2026), so AI-generated fields come out.

None of this empties the dataset. Strip the people, the insureds and the carrier material, and what remains (submissions, endorsements, certificates, renewals, property and liability claims advocacy, policy checking, SOPs) is the part buyers say they want. You approve the scope, and each exclusion is named in the contract. The rule-by-rule version is in what GLBA and NAIC Model 672 keep out of an agency data license.

Your agency management system data and AI training: what the terms say

Your AMS is where the value sits, so its contract matters. Every AMS whose terms we could read today leaves the agency's data with the agency. The limits are about how you get it out and how long you have.

The practical rule follows from the table: export with the AMS's own tools, run by your staff, and keep the buyer out of the system. A plan that depends on a buyer or vendor logging in, scripting against the UI or connecting through an API runs straight into these clauses, and Applied is litigating unauthorized access to Epic in federal court. Export before you give notice to cancel, not after. Our guide to Applied Epic, AMS360, HawkSoft and EZLynx data terms goes system by system, including the Epic Data Lake and Applied's AI clauses.

If you are selling your insurance agency or planning perpetuation

Agencies change hands steadily, though the pace has slowed. OPTIS Partners counted 292 agency and broker transactions in the first half of 2026, down 15% from 342 a year earlier, and 646 in the 12 months to June, with PE-backed and hybrid buyers responsible for nearly 75% of announced deals over that period (source: OPTIS Partners, First Half 2026 M&A Report). If a sale or internal perpetuation is on your horizon, a data license is one more contract a buyer of your agency will read in diligence.

Licensing before a sale can turn years of archived records into cash an acquirer does not pay you for. Licensing after a sale usually means the records belong to the new owner and fall under its group policies. Exclusivity is the term to watch: one program's published deals are perpetual exclusive licenses, and others ask for exclusivity windows of their own (source: program terms, Regulr buyer directory). Our guide to selling an insurance agency and licensing the records first covers timing, valuation and what an aggregator will ask.

Questions to ask an AI data buyer before you share insurance agency records

Agency records raise questions that a generic data deal does not. These are the ones worth settling in writing before any sample leaves the agency:

1. Do you accept insurance agencies at all? At least one program does not (source: program terms, Regulr buyer directory), so ask first.
2. Will you take carrier correspondence and quotes with carrier names, premiums and rates removed, or should they be excluded? Buyers answer this differently, and the answer changes the size of the set.
3. What de-identification standard do you apply to commercial files where the insured's name identifies an owner? Ask how names, FEINs, addresses and policy numbers are replaced, and whether tokens stay consistent so an account still links across years.
4. Do you want any workers comp or bodily injury content? If the answer is yes, ask how the buyer would meet Model 672's health information test; if no, write the exclusion into the scope.
5. Will you work only from exports our staff run? Given the AMS clauses on bots, APIs and SDKs, a buyer that needs credentials to your AMS is a buyer to decline.
6. What happens to the license if the agency is sold? Ask whether it survives a change of control, whether the buyer of your agency must be told, and whether exclusivity binds the new owner.
7. Does your price assume exclusivity? Your AMS vendor may already hold rights in de-identified data built from your records, so ask what exclusivity really covers.

How an insurance agency data deal runs, step by step

1. Inventory. List your systems (AMS, rater, client portal, CRM, email, shared drives), how many years each holds, and your commercial and personal lines split. Nothing leaves your systems at this stage.
2. Scope. Decide what is in and what is out. Personal lines detail, MVRs, health and workers comp claim content, benefits files, carrier manuals and pricing, licensed content and AMS AI outputs are the usual exclusions. You approve the scope, and it is written into the contract.
3. De-identification. Insured names, FEINs, addresses, policy numbers, drivers, claimants, carriers and staff are removed or replaced with consistent tokens, so an account still links from submission to renewal without naming anyone. One large buyer program says it anonymizes everything inside your environment before any data is extracted (source: program terms, Regulr buyer directory).
4. Offers. Buyers review a sample or a description and make an offer.
5. Terms. Price, payment schedule, exact scope, exclusivity, permitted uses, deletion at the end of the term, what happens if the agency is sold, and the warranties you give about your rights to the records.
6. Delivery. Your staff run the export with the AMS's own tools, and the records go from your systems to the buyer you chose. With Regulr, they never pass through us.
7. Payment. Timing varies widely by program, from payment on delivery to payment after the data is accepted (source: program terms, Regulr buyer directory).

The best position is more than one written offer, because buyers price the same records differently and treat carrier material very differently. That comparison is the work Regulr does for the seller. The rules that keep client tax data out of an accounting firm's license are in our accounting firm guide, a useful parallel for agencies that also offer premium finance or bookkeeping. When you are ready, see what your agency's records could qualify for.

Frequently asked questions

Sources

Program terms checked against their own pages on October 15, 2026; our buyer directory lists the programs and links to their published terms. Statutes, NAIC Model 672, AMS vendor terms, the Big I guide, the SEC exhibit, the Agency Universe Study, the Best Practices release, Zywave and OPTIS read on their own pages on October 18, 2026. Census counts computed from the public SUSB 2022 file on October 15, 2026.

1. program terms, Regulr buyer directory
2. Big I and Future One, 2026 Agency Universe Study release (Sep 23, 2026)
3. IA Magazine, key findings from the 2026 Agency Universe Study (Oct 1, 2026)
4. Big I and Reagan Consulting, 2025 Best Practices Study release
5. Census, Statistics of US Businesses 2022
6. IA Magazine, Independent Agents Own Their Data (Mar 2, 2021)
7. Big I, Guide to Agency-Company Appointment Contracts, v1.0
8. Crum & Forster agency agreement, SEC exhibit 10.1 (2009), sections 8.1 and 10.8
9. 15 U.S.C. 6809 (Cornell LII)
10. NAIC Model 672, Privacy of Consumer Financial and Health Information Regulation
11. 18 U.S.C. 2721 (Cornell LII)
12. Vertafore Standard Terms, updated August 2026
13. EZLynx Terms of Service, updated Jan 1, 2025 (Applied Systems)
14. HawkSoft agreement, revised Sep 16, 2026
15. Zywave, Market Basket
16. OPTIS Partners, First Half 2026 M&A Report