Insurance
Knowledge & RAG Systems
Cobalt Mutual Insurance
A claims assistant that answers
with its sources.
Avg. claims handling
10 weeks
13 min
Was
21 min
−38%
+27 pts
First-contact resolution
94%
Grounding accuracy
10 wks
Kickoff to production
02 · The engagement
Published
01 · Challenge
What was actually broken
Not the symptom, the cause.
Cobalt Mutual’s claims adjusters were searching across six systems — policy wording, state-by-state regulatory guidance, prior claim precedent and internal bulletins — to answer coverage questions on a single file. New adjusters took months to ramp, and senior adjusters lost hours a week to being the de facto search engine for the team.
02 · Approach
How we scoped it
And what we deliberately chose not to build first.
We mapped every source an adjuster actually consulted, including the internal bulletins that lived only in a shared drive, and built a retrieval pipeline that treats policy wording and regulatory guidance as separate, versioned sources rather than one undifferentiated pile of text. Every answer the assistant gives carries a citation an adjuster can click through to the source clause. We evaluated the system against 340 real historical claims questions before rollout.
03 · Architecture
The system, in one diagram
What it connects to, what a person still reviews, where the guardrails sit.
Solution architecture
Knowledge
In the claims system
Adjuster question
Policy, guidance, bulletins
Hybrid retrieval
Versioned sources
Grounded answer
Refuses when unsure
Grounding check
Pass · Inline in claims UI
Answer with citations
Else · Escalated with context
Senior adjuster
Grounding score per release · weekly gold-set run
A retrieval-augmented assistant indexes policy documents, state regulatory guidance and internal bulletins as separately versioned sources, with a grounding layer that refuses to answer when retrieval confidence is low rather than guessing. It’s embedded directly in the claims-management system adjusters already use, so there’s no separate tool to open.
04 · Result
What changed
Measured in production, not in the pilot.
Adjusters now get a cited answer to a coverage question in seconds instead of pulling three systems and a colleague into a side conversation. First-contact resolution rose sharply because adjusters could resolve coverage questions on the initial call instead of following up after research.
Result at a glance
Client
Cobalt Mutual Insurance
Industry
Insurance
Service
Knowledge & RAG Systems
Time to production
10 weeks
Headline
38%
faster average claims handling time
03 · What changed
In the client’s own operating numbers
38%
faster average claims handling time
27 pts
increase in first-contact resolution
10 weeks
from kickoff to production
−38%
Avg. claims handling
21 min
→
13 min
04 · In their words
Operators, not buyers
New adjusters are productive in weeks now instead of months, because the assistant carries the institutional knowledge that used to live only in our senior team's heads. And because every answer cites its source, our compliance team trusts it as much as our adjusters do.
Lindsay Oduya, Director of Claims, Cobalt Mutual Insurance
Next case study
Let’s find out
what it’s worth in production.
Have a workflow like this one? A 30-minute call is enough to tell whether it is ready to build.