Underwriter decision support briefing

A quality briefing is a short, structured digest in which every material sentence points at a submission page, a loss-run line, or an enrichment field. It is not a quote, not a bind, and not a hazard list compiled from what usually shows up for that occupancy.

The outcome you are producing is quality of support, not a decision. The underwriter still quotes or declines. The system that holds the account (whether that is Guidewire, Duck Creek, or a Salesforce workbench sitting in front of either) must not treat a generated paragraph as authority to bind. Auto-bind from a brief is a process failure, not a time saver.

Write the brief so a second underwriter can open the same pack and find the same sentences. If a statement cannot be traced to a page, a run, or a named field, it does not belong in the brief. If the source is silent, the cell stays empty. Empty is correct. A filled cell with no source is not.

Quality also means the brief does not invent a loss or a hazard. A model that "helps" by adding a flood narrative because the ZIP looks coastal, or a theft narrative because the occupancy is warehouse, has left the job.

Keep the brief short enough to read before the first broker call. Use the desk's existing section labels so cites stay findable.

Load the full pack first. Do not draft from the broker email alone.

The pack is the submission documents after submission extraction and normalization, the loss runs as delivered (not a paraphrase), enrichment extracts and risk scores, and the account record in the policy or workbench system. Guidewire, Duck Creek, Verisk, and Salesforce are the usual homes for those pieces. Treat them as a class of systems of record, not as interchangeable products. Pull from whatever your desk actually uses. Do not assume a field exists because another carrier's stack has it.

Order of load matters. Documents and loss runs first, then enrichment and scores, then prior notes. If you reverse that order, the brief will sound like the score and then hunt for a sentence to justify it. That is the opposite of cited support.

While you load, run a pass for application vs external data reconciliation. Flag mismatches as mismatches. Do not silently prefer the enrichment occupancy over the application occupancy, or the application square footage over the SOV. Put both values in the brief, each with its cite. The underwriter resolves the conflict. The brief does not.

If appetite and complexity triage has already parked the account as out of appetite or as a specialist referral, say so in one cited line (queue, rule, or prior note) and still do not skip the pack. Triage is not a substitute for a briefing on an account that remains on the desk.

Do not generate any briefing sentence until page numbers, loss-run identifiers, and enrichment field names are in front of you. If extraction failed on a page, record "page not extracted" rather than filling from memory of similar risks.

Draft one claim per sentence, then attach the cite in the same sentence. A cite is a page, a loss-run row, or a field name a colleague can open in under a minute. "See file" is not a cite. "Per the model" is not a cite.

Use a stable sentence shape: fact, then source. Example shapes that hold up at the desk:

• Application states the building is fully sprinklered with a wet-pipe system (ACORD 140, p. 3, protection).
• Five-year loss run lists one water-damage claim with a paid amount shown on the run (loss run, 14 Mar 2022 line).
• Enrichment occupancy is coded as light manufacturing / machine shop (occupancy field on the enrichment extract).

Here is the one illustrative walk-through. A mid-market property pack arrives for a single-location light manufacturing building: ACORD 140, statement of values, five-year loss run, enrichment extract, location score. The underwriter needs cites, not a story about typical machine shops.

Write the protection line only from the application page that states sprinklers. Write the loss line only from the run that lists the water-damage claim. Write occupancy from the enrichment field, and write the application occupancy next to it if the two differ. If the SOV shows a building limit and the application shows another, put both numbers with cites. Do not average them. Do not pick the one that makes the score look consistent.

Do not add a therefore clause that turns the brief into a recommendation. "Therefore this is a clean risk" is not a cited fact. "Score is X (score field)" is a cited fact. The underwriter decides what the score means against appetite.

A sentence with no cite is the first failure mode, and it is easy to miss because it often reads well. "Operations include welding" may be true in the underwriter's experience of machine shops. If welding is not on an application page, an inspection note, or an enrichment field, delete the sentence or move welding to Open questions as "not in pack." Fluency is not evidence.

If the source is silent, leave the field empty. Empty stays empty. Do not backfill from occupancy norms, from the last similar account, or from a model prior.

Roof year, wiring update, cooking hazard, flood zone, and prior-carrier name are common blanks. A blank roof year is a question for the broker or a reason to inspect. A guessed roof year is a fabricated underwriting fact. The same rule applies to losses. If the run does not list a hail claim, the brief must not mention hail as a prior loss. Inventing a loss is the second failure mode. It is worse than a missing cite because it can drive a deductible, a surcharge, or a decline that the file does not support.

Open questions belong in their own block, labeled as unanswered, not as findings. "Roof year not stated on ACORD 140, SOV, or enrichment extract" is a valid briefing line. It cites the search. It does not invent a year.

Scores and enrichment can be silent too. A missing flood score is not "low flood." A missing crime score is not "safe area." Record "field not present on extract" and stop.

If only one source speaks, cite that source and leave the other blank. Do not fill the hole with the nearest number.

The underwriter quotes or declines after reading the brief and the pack, not because the brief exists. Treating the brief as a quote is the third failure mode. A generated digest in the activity log is not an indication, not a binder, and not authority to issue. Keep the brief in a support slot. Put the indication, quote, or decline in the transaction the desk already uses in Guidewire, Duck Creek, or Salesforce.

Do not auto-bind. Do not auto-issue. Do not auto-send a quote email from a completed briefing. Completion of the brief means the cites are in place and the blanks are visible. It does not mean the risk was accepted.

Use the brief to brief a person. If adverse selection pattern detection flags a pattern (short prior term, missing loss run, occupancy mismatch with operations), put that flag in the brief as a cited alert and leave the decision on the desk. Pattern flags are questions. They are not a decline letter.

When you hand back to the policy system, carry three things only: the structured brief with cites, the list of empty fields, and the underwriter's action. Do not carry a generated hazard that never appeared in the pack.

Read the brief once as if you were the referral underwriter. Every sentence should survive the question "where is that in the file?" If it cannot, cut it. Then quote, decline, refer, or ask the broker. The model does not do those four things. You do.