- H2O Document AI reads broker submissions, loss runs and prior policies and returns the fields you define as validated JSON, with no templates and no labeling.
- Loss runs from different carriers, in different layouts, land in one normalized loss history per account.
- Risk-appetite matching checks each submission against your written underwriting guidelines and cites the guideline behind every flag.
- An agent assembles the file and drafts the clearance summary; your underwriter makes the call.
Your submission queue decides which risks you get to quote
Commercial submissions arrive by email in every format a broker can produce. Scanned applications, spreadsheet schedules, loss runs from several prior carriers and old policy documents all have to be read and rekeyed before an underwriter can look at the risk.
That work sits in front of every quote. While a submission waits in the queue, the broker is often shopping the same risk elsewhere, and the best risks rarely wait long. Clearance and appetite checks add more steps: someone has to confirm the account is not already in the book, then compare the risk with the guidelines, before an underwriter spends any time on it.
How underwriting submission intake works
H2O can be used to read the submission, normalize the data, run clearance and appetite checks, and hand the underwriter a complete file. Your policy system stays; the platform sits on top of it.
Step 1: take in the submission
Submissions enter a collection by upload, a connector to a shared repository, or the API. Enterprise h2oGPTe reads each page, including scans, tables and handwriting, and turns an email and its attachments into one document.
Step 2: extract the risk data
Document AI extractors turn the documents into structured output that matches a JSON schema your team defines: insured name, locations, class of business, exposures, limits requested and effective dates.
Step 3: normalize loss runs and prior policies
One extractor reads loss runs from many carriers, in many layouts, into a single loss history per account: policy period, date of loss, cause, paid, reserved and status.
Step 4: clear the submission
The agent checks the extracted insured, address and broker against your existing accounts and open submissions. Possible duplicates or broker conflicts go to your clearance team with the matching records attached.
Step 5: match the risk to your appetite
Agentic retrieval-augmented generation compares the risk with your underwriting guidelines and referral rules, and cites the section behind each flag.
Step 6: hand the underwriter a complete file
H2O AI Super Agent assembles the file: extracted data, loss summary, clearance result, appetite notes and missing items to request from the broker. The underwriter reviews it and decides whether to quote, decline or ask for more.
Risk-appetite matching and prior-policy extraction
Risk-appetite matching compares a submission with the risks your company wants to write: classes, territories, limits, construction and loss history, read from your own guidelines rather than a copy someone maintains. Prior-policy extraction pulls the terms from the expiring and earlier policies in the submission, so your underwriter sees what the account had before next to what the broker is asking for now.
Proof
H2O can be used to read broker submissions, loss runs and prior policies, check clearance and appetite, and draft the file for an underwriter. No insurer runs this workflow on H2O today, so this page describes the pattern, not a customer outcome.