Submissions Analyst · Source traceability

Explainable underwriting AI: every value shows its source.

Explainable underwriting AI means an underwriter can trace any value back to the page it came from. Brisc stamps every extracted field with its source file, location, reasoning and confidence, and keeps every conflicting value with its origin.

Provenance

Six things ride on every value.

Provenance is stamped at extraction, on every field.

The source file

The email or attachment the value came from: named, one click away.

The location inside it

The sheet, page or region where the value was read.

The extraction reasoning

How the value was read, in writing.

A confidence

So review effort goes where the doubt is.

A plain-English reason

Why this value, stated so a reviewer can disagree with it.

Exclusions, explained

Files that were set aside carry the reason they were excluded.

Arbitration

What happens when documents disagree?

The same field is consolidated across the email and every attachment, deterministically.

  • Every value kept. Conflicts stay on record with their origin; originals are never overwritten.
  • Sources counted. The Analyst counts how many sources agree.
  • Selection explained. The most credible value wins, tagged AI-selected, rationale in writing, so the underwriter can always take the other view.

Full intake pipeline, formats and field list: Submission intake.

TIV · aggregate AI-selected · 96% USD 84,250,000

SOV workbook totals USD 84,250,000; the broker email states “approx. USD 84.3M”. The workbook is the more granular source and is selected. 2 of 3 sources agree.

SOV.xlsx · sheet 2 broker email ACORD 125 · p.2
Missing fields

What about the fields that are not there?

Every expected-but-absent field is surfaced as Not Found. Nothing is invented and nothing is quietly left blank: a missing inception date is a finding with a name.

Set aside, and said so

What it reads, and what it sets aside.

Every attachment is classified process-or-exclude: loss runs are recognised and set aside, and classification fails open, so an ambiguous file is still processed. The Analyst never accepts or declines a submission: the judgment stays with your underwriters.

Common questions

Explainable underwriting AI, answered

How do you make underwriting AI explainable?

Stamp provenance on every value. Each extracted field carries its source file, its location in the document, the extraction reasoning, a confidence and a plain-English explanation, and every conflicting value is kept with its origin. Explainability is built into the record at extraction.

What happens when two documents disagree?

Both values are kept. The Analyst consolidates the field across the email and every attachment, counts how many sources agree, selects the most credible with a written rationale, and tags that selection as AI-selected. Originals are never overwritten, so the underwriter can always take the other view.

What happens when a field is missing?

It’s surfaced as Not Found. Every expected-but-absent field is explicitly flagged, so the gaps in a submission are as visible as the values that were found.

What does it do with loss runs?

It recognises them and sets them aside. Every attachment is classified process-or-exclude, and the classification fails open: an ambiguous file is still processed. Every excluded file carries the reason it was excluded.

Does the AI accept or decline submissions?

No. The Analyst structures the data, checks it, and reconciles conflicting sources. Appetite, judgment and the decision to quote stay with your underwriters.

How is TIV captured?

As an aggregate for the submission, traced to source like every other value, with currencies and percentages normalised.

Bring your own submission. Trace any value.

30 minutes, no slide deck. Send a real broker submission ahead of the call and click through the provenance yourself.

Book the walkthrough