Underwriting data is every piece of information a carrier uses to assess, price, and monitor commercial risk: what's on the submission, what's known from loss history and prior decisions, and what's observed about the portfolio those decisions sit inside. The NAIC's Model Bulletin defines a predictive model as the mining of this kind of historic data to identify patterns and predict outcomes that support decisions. The more complete that data is, the better the decisions built on it.

The types of data underwriters work with

Commercial underwriters draw on four broad categories. Submission data: what the broker provides on the risk itself, exposures, limits, prior losses, risk controls. Experience data: the carrier's own loss history, prior quotes, and binding decisions on that account or similar ones. Third-party data: industry statistics, compliance records, and external signals such as climatic, IoT, ESG, or reputational indicators. Portfolio data: how a risk sits against the book the carrier already holds, exposure concentration, and rate adequacy.

Each category answers a different question, and none of them is complete alone. A risk score built only on submission data misses how that risk affects the portfolio; portfolio data without submission detail can't price an individual account.

Why the broker submission is the highest-signal data a carrier has

Of the four, the broker submission carries the most weight. It's the earliest data a carrier sees on a risk, the most specific to that account, and the only one available before a pricing decision has to be made. Everything else, experience, third-party, and portfolio data, adds context to a submission; it doesn't substitute for one.

That's also why submission quality is where automation has had the most immediate impact. Ingestion agents extract and validate ACORDs, bordereaux, and loss runs; triage agents score the result against appetite. One hx customer removed 2,100 hours a year of third-party data rekeying on its Inland Marine and Property book; HDI automated data capture for its UK PI book and freed up the equivalent of two underwriting FTEs. The underwriter opens a decision-ready file instead of a PDF and a clearance task.

The value of capturing more data

More submission data reviewed means more risks an underwriter can actually weigh, not just process. Manual re-keying and chasing missing broker data caps how many specialty submissions a team can review in a day; automating that assembly redirects underwriter time toward the judgment calls that move the book. Across hx's customer base, that shift has meant a 50% reduction in submission-to-quote time. Aviva cut new policy creation from an hour to five or ten minutes; another customer freed up 92 workdays a year, a 38% productivity improvement, by eliminating rekeying into its policy admin system.

Why not-taken-up risk data still has pricing value

A bound risk shows how one decision performed. A complete record of every submission a carrier sees, quoted and declined, shows what the market is actually bringing and how the carrier's appetite compares to what's available to write.

Conduit Re captures pricing data for every piece of business it underwrites, using that record to build rate change reporting by class and benchmark loss ratios and development patterns across the whole book, not just the bound portion of it. The same principle applies to declined business: a risk that falls outside appetite still carries signal, what it would have cost, why it was declined, and how that compares to what similar risks have priced since. Retained and fed back into experience pricing, that record sharpens rate adequacy and appetite calibration in a way the bound book alone can't. It's the same principle behind Decision Trace, the mechanism that captures every pricing action and its rationale: the value isn't just in the decision, it's in keeping the full reasoning behind every decision, taken or not.

What a shared context layer is

A shared context layer is a governed record connecting submission data, pricing and appetite logic, prior decisions (including declined ones), and portfolio signals, so that whoever or whatever picks up a risk next, an underwriter, an actuary, or an agent, starts with that history already attached instead of reconstructing it from emails and spreadsheets.

Few carriers have this today. Pricing logic lives in one actuary's spreadsheet, declined risks are discarded at the point of decline, and portfolio context sits in a quarterly report. Antares saw what governing that logic alone can do: a 4x reduction in time to adjust its models, equivalent to an extra full-time employee a year. Building toward a shared context layer means extending that same governance so it's callable by any system or agent that needs it, not personal knowledge that leaves when someone does.

How a shared context layer enables agentic underwriting

Agents need an environment, tools, and a goal. A shared context layer is the environment: it's what lets an agent act on a risk with the same pricing logic, appetite rules, and decision history an experienced underwriter would use, instead of starting from a blank page.

In practice: a broker submission arrives, ingestion agents extract the data, triage agents score it against appetite (including what similar past submissions, bound or not, priced at), pricing-prep agents run it through governed Calculation Engines, and referral agents flag what needs senior review, before the underwriter ever opens the file. The same context layer feeds Portfolio Intelligence's continuous monitoring, so exposure concentration, loss-ratio drift, and rate adequacy surface while they're still cheap to act on, not two quarters late; carriers taking this approach are targeting loss ratio improvements of 1 to 2 percentage points. That's the difference between agents assisting at the edges of underwriting and agents doing the coordinated work of it: clearance, triage, risk assessment, pricing, and portfolio steering, inside the carrier's own governance.

See it in action

hyperoperator is hx's agentic underwriting workbench: it's built to coordinate exactly this, specialist agents working from a shared context layer across the full underwriting journey, for carriers trusting hx with underwriting decisions across $75bn+ in annual commercial P&C premium. Learn more about hyperoperator.