Underwriting notes as AI training data: what they capture and who can license them

What underwriting notes are

Underwriting notes are the written reasons behind a risk decision. When an underwriter works a submission, the notes record what was examined, what raised concern, what the broker was asked for, whether the account went to someone with higher authority, and why it was quoted, conditioned, priced up or down, or declined.

Rating data shows the price. The notes show the judgment that produced it, and that judgment is the part AI developers building underwriting assistants and agents find hardest to source.

Why rationale matters more than rating data for AI

Most of what an experienced underwriter knows never reaches a rating field. Notes capture it in plain language, tied to a decision and, over time, to an outcome: whether the account bound, whether it renewed, and how its losses developed. That chain of reasoning, decision and later result is what makes the records useful for training models and for testing whether an agent would have made the same call.

It is also material public sources do not offer. Researchers at Epoch AI estimate that, if current trends continue, language models could fully use the stock of public human-generated text sometime between 2026 and 2032. The forecast carries wide uncertainty, but it helps explain why permissioned, non-public records of professional judgment draw interest.

Illustrative: what one account's notes can show

This scenario is illustrative and fictional. A commercial property account is submitted to a regional MGA three years running. In year one, the underwriter declines it over an aging roof and missing sprinkler records. In year two, the broker returns with a roof replacement invoice; the account is referred upward because of its total insured value and binds with a higher deductible as a condition. In year three, a small water loss prompts a renewal note weighing the claim against the improved building, and the account renews at a modest increase.

Read in sequence, those notes form a ready-made test: given what the underwriter knew each year, would an AI assistant have reached the same three decisions, for the same reasons?

Agencies, MGAs and carriers hold different slices

One submission can leave notes in three or four places, and each holder's rights differ.

MGAs should start with the MGA data ownership page, which walks through program agreement clauses. Agencies should read about agency management system notes first.

Where underwriting notes live

Underwriting history is often spread across more systems than the team realizes:

• The notes and activity log in the policy administration system.
• An underwriting workbench or submission intake tool.
• Rating workbooks, often saved as one spreadsheet per account.
• The shared submissions inbox and each underwriter's email.
• Referral emails or forms sent to senior underwriters or the carrier.
• Loss control and inspection reports.
• Teams or Slack channels where underwriters talk through difficult risks.

Spreadsheets and email often hold the richest reasoning, and they are the first things lost when an underwriter leaves or a system is retired.

Screening policyholder and submission data

Underwriting files hold information about the insured: financial statements, payroll, loss runs, locations and sometimes details about individuals. Commercial lines files are the better fit, because most of their content describes businesses. Personal lines files are largely consumer personal data, which is a red flag for licensing.

State insurance privacy rules vary, so the company's counsel should confirm what must be removed. De-identification and redaction rules are settled with the company before any work starts, and nothing leaves the company without a signed agreement and its authorization.

This is general information, not legal, tax or financial advice. Confirm with your own counsel before acting.

Checklist: signs of a deep underwriting archive

• Five or more years of written notes on quoted, bound and declined business.
• Referral and authority decisions written down, not only discussed by phone.
• Pricing deviations justified in writing.
• Declinations recorded with reasons.
• Renewal reviews that refer back to prior-year decisions.
• Loss outcomes that can be linked to policy or account numbers.
• A book that is mostly commercial lines.
• A clear ownership or consent position under carrier and program agreements.
• A company that reached 50+ full-time employees at peak, contractors excluded, with an executive sponsor who can approve a license.

How an operating partner raises it with an insurance services CEO

Raise it during systems planning or an add-on integration, when underwriting files are about to be migrated or archived anyway.

Insurance distribution roll-ups meet this question with every acquisition; the insurance agency roll-up screen covers it for agency deals, and the claims notes brief covers the loss side of the same accounts.

Next step

Ask where declined submissions are stored and how far back the notes go. If the answers are encouraging, register as a partner and introduce the CEO, who can begin listing systems with the data inventory builder. Portfolio-wide screening ideas are on the operating partner page.