GA school-bus, $5M policy-limits demand — tender recommended
Plaintiff counsel has 3 priors, all settled within window. Decline risk: ~$3M extracontractual.
Bad-faithTime-boundedKnown·
Bitligence for Insurance · P&C claims
Atlas and Praxis. Two claims agents, governed at runtime.
Atlas handles inbound FNOL calls. Praxis runs the claim file after intake. Every decision arrives in your adjuster’s inbox with the recommendation, the priors, the confidence score, and the audit trail — and every action passes the same governance gate as the rest of Bitligence.
FNOL → filetwo agents, one trail
≥95%field completion at intake
5-axispriority on every decision
signedevery action, replayable
Live product · Assurance Observatory — agents runtime
Assurance ObservatoryICW Group / claim-workspaceSearch ⌘KLLena ADMIN
Assurance Observatory runtime · live agents
Agents
Fires today847+26/min
Policies attached125 categories
Policy eval p954.8msrolling · last 100
Audited100%WORM retention
Live·26/min
praxisf3_isoAUTOCLM-2026-018474950ms
atlascaptureAUTOCLM-2026-01847384ms
praxisf6_badfaithHITLCLM-2026-0184721700ms
atlasescalateHITLCLM-2026-01846170ms
agent connected·guidewire claimcenter·2 agents · 11 tools·100% audited
The agents
One handles the call. One handles the file.
Both are agentic services that read claim state through your API, reason over it, and write decisions, reserves, and tasks back through the same path. Your platform stays the system of record.
Voice intake · FNOL
Atlas
Atlas answers inbound FNOL calls and captures a structured claim file during the call — submitting it to your core system before the caller hangs up.
- ≥95% field completion at first contact
- Verbal confirmation on medium-confidence fields
- Full-call escalation to a human when uncertain
File handling · with trace
Praxis
Praxis consumes claim events, runs the file-handling functions per event, scores every surfaced decision, and writes decisions back through your API.
- Runs the file overnight
- Recommendation + priors + audit trail attached
- High-stakes work never auto-executes
Atlas · live intake
A first notice of loss, captured during the call.
Atlas answers, captures every field while the caller talks, verifies the policy mid-conversation, and hands a complete claim file to Praxis — no callback. The call below plays a representative intake.
Atlas · voice intakelive · structured FNOL captured during the call
Sample transcript plays here — Atlas captures the loss notice as the conversation happens.
Structured file · auto-populated
Caller
—
Injury
—
Policy
—
Loss
—
Police
—
Routing
—
The oversight surface
Every agent action lands in an inbox.
Praxis surfaces every decision that needs a human — scored on five weighted axes, with the recommendation drafted, the cost of waiting quantified, and the trace attached. Your adjuster decides; the audit log records.
Oversight inbox · Sarah Chen3 items need a human
Praxis connectedTX 5-day acknowledgment due in 4 hr — letter drafted
Standard rear-end / clear liability. TX-DOI compliant template populated. One-tap send.
StatutoryBreaches-todayKnown·
FL restaurant claimant — 3 prior ISO matches flagged
Same claimant, same loss type, 3 different insureds in 18 months. Pattern score 0.84.
SIUTime-boundedKnown·
Click any item and Praxis hands you the full recommendation, dimensions, and trace.Open inbox →
How every row is scored
Five weighted axes, set by the carrier.
Deadline · 35%
breaches-today · time-bounded · async · historical
Risk · 30%
bad-faith · statutory · SIU · litigation · reserve · routine
Reversibility · 20%
irreversible · effortful · easy · done
Source · 10%
unknown · sub-agent · external · known
Type · 5%
authorization · escalation · anomaly · confirmation
Agentic governance
After enough agreement, categories become rules.
Day one, every decision flows through the inbox. The system tracks where the agents and the adjuster agree, on which categories, with what confidence. When agreement crosses ≥95% over ≥20 samples, it surfaces the rule. Your adjuster approves it once — and that category stops hitting the inbox.
The rule graduates into a signed, versioned policy (shadow first, then active on promotion). Over time the inbox holds only novel or high-stakes work, and your adjuster’s time shifts from approving claims to approving the rules that govern categories of them.
Watch
Observe decisions
Every agent recommendation and adjuster decision is recorded with confidence and reason.
Learn
Track agreement
Agreement is tracked per category — ≥95% over ≥20 samples surfaces a candidate rule.
Graduate
Approve once
One signed attestation promotes the rule to a versioned policy; the category leaves the inbox.
Enforce
Shadow → active
Policies run in shadow first (log, don’t block), then enforce on promotion — ~8ms in the path.
Integration
Drops into the claim platform you already run.
Atlas and Praxis integrate with Guidewire ClaimCenter, Duck Creek Claims, or a homegrown core. Your platform stays the system of record; the agents read claim state through your API and write decisions, reserves, and tasks back through the same path.
Guidewire ClaimCenter
Read claim state and write back through the standard integration path.
Duck Creek Claims
Same read/reason/write loop against your existing configuration.
Homegrown core
Any system with an API — the agents adapt to your model, not the reverse.
Your boundary
Run in your cloud, VPC, on-prem, or air-gapped. Data never leaves your perimeter.
Compliance
Built for the regulatory reality of claims.
Aligned toNAIC AI Model BulletinNYDFS CL-1Colorado SB21-169Signed audit trail
NAIC AI Model Bulletin (2023)
Governance + third-party oversight in the default config.
NYDFS Circular Letter No. 1 (2019)
External-data documentation + disparate-impact analysis in the audit trail.
Colorado SB21-169 (2021)
Fairness-testing patterns applied across personal and commercial lines.
The 48-hour diagnostic
Send a representative week of your FNOL records.
Your data, your numbers, one page — back to you in 48 hours, with a 30-minute call after.