CASE STUDY · INSURTECH AI (CLAIMS & UNDERWRITING)
Automating claims processing for a regional insurance provider.
- 3 daysSimple claims settled in 3 days
- 10→6 daysAverage cycle time down from 10 to 6 days overall
01 · PROJECT OVERVIEW
The client and the brief.
An anonymised regional insurer. Name withheld under confidentiality.
Intake extraction, triage and straight-through handling for simple claims.
- 1The insurer handled 2,500 claims a month and took 10 days to settle.
- 2Handlers spent most of their time reading documents, retyping details and chasing missing items.
- 3Simple claims waited in the same queue as complex ones.
- 4Policyholders called for updates, and the calls themselves took more handler time.
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 01Extraction accuracy was fine on clean digital forms and poor on scanned repair estimates from small garages, which came in many formats
Early triage wrongly sent some of those claims to standard handling because fields were missing. We stopped treating missing fields as zero, added a "needs documents" route that asks the policyholder for the missing item automatically, and added the garage estimate formats to the tuning set. Straight-through eligibility stayed conservative until the numbers improved.
- PROBLEM 02Straight-through eligibility stayed conservative until the numbers improved
- PROBLEM 03A second hurdle was the payment limit
Finance was uneasy about paying claims without a person looking at each one. We agreed a low straight-through value limit to begin with, a sampling check by a senior handler on a share of paid claims every week, and an automatic block on any claim where the policy status or cover looked unusual. As the sample showed few errors, the limit was raised in steps with finance sign-off.
- PROBLEM 04As the sample showed few errors, the limit was raised in steps with finance sign-off
03 · THE PROCESS
How we ran it.
We worked with the claims team in four phases.
- 01Step 1
Observation: we sat with handlers, followed a sample of claims from first notice to settlement, and recorded where the days went.
- 02Step 2
Build: extraction for the common claim documents, then a triage step that sorted claims into simple, standard and complex.
- 03Step 3
Shadow run: the new flow made suggestions in the background while handlers worked as normal. We compared its triage and extracted data against theirs every week.
- 04Step 4
Staged go-live: simple claims first, with a handler spot-checking a share of them, then wider coverage as the numbers held.
We held a short weekly review with the head of claims and a monthly review of exceptions.
04 · RESULTS
What changed.
Measured against the way the work ran before the engagement began.
- NOW 01
Handlers start each day with a queue that is already sorted and pre-filled.
- NOW 02
Simple claims no longer wait behind complex ones.
- NOW 03
The team reviews overrides and low-confidence fields every month and feeds the findings back into the rules.
- NOW 04
Policyholders see their claim status without phoning, so call volume on status questions has fallen.
- 3 days
Simple claims settled in 3 days
- 10→6 days
Average cycle time down from 10 to 6 days overall
05 · WHAT WE BUILT
How it works.
- Intake accepts forms, photos and attachments, and extracts fields with a confidence score.
- Low-confidence fields are shown to a handler with the source highlighted, so correction takes seconds.
- Triage rules use claim type, value, policy checks and completeness.Anything outside the rules goes to a handler.
- Straight-through claims go to a payment approval step with a defined limit and a sampling check.
- Every step writes an audit record: input, output, confidence and the rule that fired.
- Status events feed a simple update to the policyholder.
06 · QUICK ANSWERS
The short version.
- What problem did the client have?
- The insurer handled 2,500 claims a month and took 10 days to settle. Handlers spent most of their time reading documents, retyping details and chasing missing items. Simple claims waited in the same queue as complex ones. Policyholders called for updates, and the calls themselves took more handler time.
- What did Brandsmashers build?
- Intake extraction, triage and straight-through handling for simple claims.
- How was the work run?
- We worked with the claims team in four phases.
- What were the results?
- Simple claims settled in 3 days; average cycle time down from 10 to 6 days overall.
YOUR TURN
Have a similar requirement?
Tell us what is stuck. We will come back with a plan and matched, pre-vetted engineers.
