Insurance · Agentic AI
Same-day P&C claims resolution for 70% of cases — down from a 14-day average
A mid-market P&C carrier processing 3,200 claims/month was watching NPS erode as competitors settled faster. We built a multi-agent AI system — three specialized agents handling medical review, liability assessment, and financial settlement in parallel — that resolved 70% of standard claims within the same business day, while keeping adjusters in the loop on every decision.
Business Context
They weren't losing on product.
They were losing on speed.
The carrier had a solid book of business — competitive premiums, low loss ratios, a tenured adjusting team. But their claims operation was built for a world where 14-day cycle times were acceptable. That world no longer exists. Digital-first competitors were settling standard auto and property claims in 3–4 days. Some in hours.
The cost of staying still
- 14 days
- avg claims cycle
- 35%
- policyholders lost
- $340
- cost per claim
Digital-first carriers: 3–4 days
Did not renew after a slow or disputed claim
vs. $40–60 with straight-through processing
The root cause wasn't the adjusters — it was the workflow. Each claim touched four separate systems: a legacy claims management platform, a third-party medical review vendor, an internal legal triage queue, and a reserve management tool. None of them talked to each other. Adjusters were the integration layer, manually routing documents and waiting on responses that took days.
At 3,200 claims per month, that meant 18 adjusters spending roughly 60% of their time on coordination — not judgment. The ceiling wasn't headcount. It was architecture.
Scope of Work
What we were asked to build
Multi-agent orchestration layer
Designed and built the core AI system — three specialized agents (medical, liability, financial) coordinating in parallel over a shared claims context.
FNOL intake & document parsing
Automated first notice of loss ingestion from web, mobile, and email channels. Structured extraction of policy data, incident details, and supporting documents.
Adjuster decision-support UI
A lightweight interface surfacing agent recommendations, confidence scores, and flagged exceptions — so adjusters review decisions, not raw documents.
Integration with existing systems
API connectors to their legacy CMS, medical review vendor, and reserve management tool. No system replacement — additive layer only.
Constraints we worked within
- Existing Guidewire ClaimCenter could not be replaced — all work had to integrate via API
- State regulatory requirements in 12 jurisdictions governed what decisions could be fully automated
- Human-in-the-loop approval required for any claim above $15,000
- 7-month delivery window tied to their annual planning cycle
Explicitly not in scope
- Policy administration or underwriting systems
- Fraud investigation for litigated claims
- Customer-facing claims portal (handled by a separate vendor)
- Staff reductions — engagement was scoped as augmentation, not replacement
System Architecture
One orchestrator. Eight specialist agents. Zero system replacements.
How We Worked
7 months. 4 phases. One decision-maker.
Discovery & Architecture
Embedded on-site. Mapped all four systems, interviewed 12 adjusters, defined agent boundaries and jurisdiction rule sets.
Build
Medical and liability agents built and tested in parallel. Pivot in month 3 — rebuilt liability agent with per-jurisdiction logic after accuracy stalled at 74%.
Pilot
200 live claims processed through the system alongside manual review. Adjusters co-designed the exception-handling UI. Accuracy reached 91% on standard claims.
Full Rollout
System live across all 3,200 monthly claims. On-site for go-live week. Handoff documentation and adjuster training completed.
Working rhythm
- CadenceTwo-week sprints, Friday working demos
- Decision ownerSingle claims director on client side
- Escalation SLA24 hours, always with a written recommendation
- On-siteMonth 1–2 and go-live week
Results
Measured at 90 days post go-live.
of standard claims resolved same business day
Was: 14-day average cycle across all claim types
Standard auto and property claims under $15,000 with no liability dispute. Complex and litigated claims still route to senior adjusters — by design.
reduction in adjuster coordination time
Was: ~60% of each adjuster's day spent routing documents
The same 18-person team now handles 35% more monthly claim volume. Not because they work faster — because the system handles all cross-system coordination.
saved per claim in processing cost
Was: $340 fully-loaded cost per claim
Reduction driven by eliminating third-party medical review vendor fees on standard claims and cutting adjuster time-per-claim from 4.2 hours to under 45 minutes.
agent recommendation accuracy on standard claims
Was: 74% at month 3 — rebuilt liability agent before go-live
Measured against final adjuster decisions over a 90-day validation period. Adjusters override on ~12% of cases, mostly edge cases with disputed causation.
What This Means for You
The architecture that solved this problem is not unique to this carrier. It applies to any P&C operation where adjusters are the integration layer between systems that don't talk to each other.
- 01Your adjusters spend more time routing documents than making decisions
- 02Your claims cycle is measured in days while competitors are moving to hours
- 03You've evaluated automation tools but can't replace your core CMS
This engagement was scoped as an additive layer — no system replacement, no staff changes, no multi-year transformation programme. Seven months from kickoff to full rollout.
See how we approach Agentic AI for insurance