CASE STUDY · INSURTECH AI (CLAIMS & UNDERWRITING)
Building an AI underwriting assistant that cut quote time by 35%.
- 35%Quote time down from 10 working days to about 6.5 (35% less)
01 · PROJECT OVERVIEW
The client and the brief.
An anonymised insurer. Name withheld under confidentiality.
An assistant that extracts submission data, summarises risk and suggests pricing inputs for the underwriter.
- 1Quotes took about 10 working days.
- 2Submissions came in by email with attachments in many formats.
- 3Underwriters spent much of their time extracting figures from them and writing risk summaries before they could price anything.
- 4Brokers were losing patience and some took their business elsewhere.
02 · WHAT WENT WRONG
What went wrong, and how we fixed it.
Every engagement hits problems. These were the ones that mattered here, and what we changed so they did not come back.
- PROBLEM 01Early summaries sometimes stated a detail confidently that was not in the submission, because the model filled a gap with a plausible guess
Underwriters caught it, and trust dropped for a week. We changed the design so every statement must point to a source passage and anything without one is shown as "not found". We added a gap list at the top of the summary.
- PROBLEM 02Trust recovered because the tool now said what it did not know
- PROBLEM 03A second hurdle was adoption
Senior underwriters ignored the tool at first because it was a separate screen. We moved the summary and extracted fields into the workbench they already used, and let them reorder the sections to match how they read a file. Use went up once it saved a click instead of adding one.
- PROBLEM 04Use went up once it saved a click instead of adding one
03 · THE PROCESS
How we ran it.
The project ran as an iterative build with the underwriting team involved throughout.
- 01Step 1
Discovery: we reviewed past submissions and the underwriters' notes to see what a good summary contained.
- 02Step 2
First version: extraction and a summary, tested on closed files where the answer was known.
- 03Step 3
Pilot: a small group of underwriters used it on live submissions, with a feedback button on every output.
- 04Step 4
Rollout: wider team adoption after the pilot group agreed the summaries could be trusted as a starting point.
We reviewed feedback weekly and released changes to the prompts, rules and extraction every two weeks.
04 · RESULTS
What changed.
Measured against the way the work ran before the engagement began.
- NOW 01
Underwriters start from a structured submission and a cited summary, not from a pile of attachments.
- NOW 02
Brokers get a faster first response and a clearer list of missing information.
- NOW 03
The team reviews edits and "not found" items monthly to improve extraction.
- NOW 04
Pricing authority and final decisions remain with underwriters.
- 35%
Quote time down from 10 working days to about 6.5 (35% less)
05 · WHAT WE BUILT
How it works.
- Submission intake reads emails and attachments and extracts exposure details into a structured record.
- A summarisation step writes a short risk summary and highlights gaps and unusual items, citing the source page for each point.
- A suggestion step proposes pricing inputs from the extracted data and the insurer's own rating rules.It does not set the price.
- The underwriter can edit any field and every edit is stored.
- Access follows existing permissions, and submission data stays within the insurer's environment.
- Output is logged with the source documents so a reviewer can trace any figure.
06 · QUICK ANSWERS
The short version.
- What problem did the client have?
- Quotes took about 10 working days. Submissions came in by email with attachments in many formats. Underwriters spent much of their time extracting figures from them and writing risk summaries before they could price anything. Brokers were losing patience and some took their business elsewhere.
- What did Brandsmashers build?
- An assistant that extracts submission data, summarises risk and suggests pricing inputs for the underwriter.
- How was the work run?
- The project ran as an iterative build with the underwriting team involved throughout.
- What were the results?
- Quote time down from 10 working days to about 6.5 (35% less).
YOUR TURN
Have a similar requirement?
Tell us what is stuck. We will come back with a plan and matched, pre-vetted engineers.
