from osmoda research · insurance
Claim to decision in hours. Regulator-ready.
Voice, photo, email intake. Audit trail by default. Naga runs explainable fraud scoring against ISO ClaimSearch + local image forensics; soot pushes structured claims to Guidewire/Duck Creek; lantern auto-assembles your NAIC Model Bulletin and EU AI Act Annex IV evidence.
TL;DR
- • ~10% of P&C claims are fraudulent — $122B annual loss, 40% of all insurance fraud [Deloitte]
- • AI fraud prevention could save P&C $80B–$160B by 2032 [Deloitte 2025]
- • NAIC Model Bulletin now adopted in 24+ states; AI Systems Evaluation Tool pilot Jan–Sept 2026 [NAIC]
- • EU AI Act high-risk obligations target Aug 2, 2026 (Digital Omnibus may push to Dec 2027) [Holland & Knight]
- • os.moda runs FNOL voice/photo/email intake, fraud scoring, and adjuster routing on your box
- • Every signal that fed every decision is in a SHA-256 ledger — exportable for market conduct exams
1. The pain — fraud + a tightening regulator
P&C carriers lose ~$122B a year to fraud, roughly 10% of claims volume, and Deloitte estimates AI-native pipelines could recover $80B–$160B of that by 2032. But the path there is regulator-defined: 24+ U.S. states have adopted the NAIC Model Bulletin on Insurer Use of AI with minimal modification, requiring written AI governance, third-party model documentation, bias testing, and inventoried decisions — the NAIC's AI Systems Evaluation Tool pilots Jan–Sept 2026 across 12 states.
The legacy stack is the cost. Guidewire ClaimCenter/PolicyCenter, Duck Creek, and Majesco P&C Intelligent Core dominate the market — Guidewire was named Leader in Gartner's 2024 SaaS P&C Magic Quadrant with 500+ carriers, and the top three combined for ~$918M ARR / 15.85% market share. Implementations run 18–36 months, license costs are 7-figure, and every new line of business needs a workflow rebuild. Underwriting workbenches like Cytora, Send, Convr, and hyperexponential ship faster but live alongside — not inside — your core.
2026 makes it worse. Colorado Reg 10-1-1 already requires governance docs from life carriers; EU AI Act high-risk obligations target Aug 2, 2026 for risk-assessment and pricing AI under Annex III Area 5(b). Climate-driven parametric demand is exploding — the parametric market hit ~$23.85B in 2026 at 13.1% CAGR — and embedded insurance is on track for ~$1.1T (15% of global premium) by 2033 per Swiss Re. None of that maps cleanly onto a 2009-era ClaimCenter deployment.
Guidewire ClaimCenter / PolicyCenter
De facto enterprise standard, 500+ carriers. Powerful but rigid — 18–36 month implementations, 7-figure licenses, every new product line is a project.
Duck Creek
Cloud-first modular core for policy/billing/claims. Faster than Guidewire but still seat-priced, AI lives in the marketplace not the model layer.
Majesco P&C Intelligent Core
Strong digital engagement and cloud-native, but it's a core system, not a workflow engine — agentic claims still requires a separate orchestration layer.
Cytora / Send / Convr / hyperexponential
Excellent point tools for underwriting digitisation. Each lives outside your audit boundary; governance evidence fragments across four vendors.
2. What 2026 is bringing
- NAIC AI Systems Evaluation Tool pilot. 12-state pilot Jan–Sept 2026 — examiners will start grading insurer AI governance during market conduct exams [NAIC March 2026].
- EU AI Act high-risk obligations. Target Aug 2, 2026; Digital AI Omnibus may push to Dec 2027 but documentation expectations are already in flight [DLA Piper].
- Parametric insurance hits $23.85B. 13.1% CAGR as climate losses compress traditional indemnity timelines [The Business Research Co.].
- Consumer support for AI in P&C nearly doubled. In 2026 per Insurity survey — political cover for carriers to move faster on agentic claims.
3. The os.moda stack
- 1 · tofu (multi-channel FNOL) takes loss reports via voice, photo, SMS, WhatsApp, email. Captures EXIF, geolocation, damage photos, recorded statements with consent — timestamped into the audit ledger from second one.
- 2 · lantern (document classifier + policy lookup) parses ACORDs, police reports, repair estimates, medical records; matches loss to policy form; flags coverage gaps before the adjuster opens the file.
- 3 · frog (fraud + cross-policy scoring) scores each FNOL against your book — repeat VINs, shop concentrations, provider networks, contradiction patterns — producing the explainable feature set NAIC examiners will ask for.
- 4 · naga (governance + bias controls) enforces the NAIC Model Bulletin policy: model cards versioned, decisions logged with inputs and overrides, bias tests run on demand, EU AI Act Annex IV technical documentation auto-assembled.
- 5 · soot (workflow + integrations) pushes structured claims to Guidewire ClaimCenter, Duck Creek, or Majesco via published APIs; pulls policy data back; orchestrates SIU referral, adjuster routing, and reserves updates without manual rekey.
- 6 · haku (regulatory + insured comms) drafts reservation-of-rights letters, status updates, 14-day acknowledgement notices in the insured's language — every draft reviewable, every send signed and logged.
- 7 · lantern (underwriting intake) runs the "supersonic underwriting" pattern Cytora+hyperexponential pioneered — broker submission email → digitised risk → appetite score → quote in minutes, not hours.
4. Why it works
Sovereignty + compliance
NixOS box you run; KEYD signs every action; SHA-256 ledger gives you a tamper-evident audit log that maps directly to NAIC Model Bulletin §4 (governance), §5 (testing), §6 (third-party AI). EU AI Act Annex IV technical documentation auto-assembles from the daemon state.
Workflow integrity
Watchdog 6s median recovery means an FNOL spike doesn't lose a claim. Atomic NixOS deploys mean a model rollback is one commit. Frog's fraud signals are explainable feature-by-feature — no black-box defense to a state DOI examiner asking why this VIN was scored 0.84.
Economics
$29–$299/month replaces the underwriting workbench + claims orchestration + comms vendor + governance docs vendor stack. You keep Guidewire or Duck Creek if you have them — we sit in front and behind, not instead. BYO LLM key removes the per-claim AI tax.
5. The 3–5 year future
- 2027 · pre-claim fraud detection. Cross-carrier patterns (with consent + ISO ClaimSearch) let frog flag the staged-collision ring before the third FNOL — claims fraud detection moves from post-hoc forensic to live underwriting signal. Industry loss ratio drops 2–4 points.
- 2028 · real-time underwriting from telemetry. Connected vehicle, smart home, and commercial fleet feeds price monthly, not annually. Soot subscribes to OEM telemetry; haku regenerates policy declarations; the audit ledger logs every premium adjustment — and the insured sees why.
- 2029–2030 · parametric-by-default for catastrophe. Wildfire, flood, business interruption from utility outage all switch to parametric triggers with on-chain attestation. os.moda's signed-action model is the natural payout primitive — claim-to-decision compresses from days to seconds.
FAQ
How does this satisfy the NAIC Model Bulletin?
Naga enforces the AIS governance policy you write — model inventory, third-party AI documentation, bias testing cadence, override logging — and every artifact lands in the SHA-256 ledger. The NAIC AI Systems Evaluation Tool checklist maps to a structured export, so when your examiner shows up in Q3 2026, you hand over a signed bundle.
Do we have to replace Guidewire or Duck Creek?
No. soot's adapters write to ClaimCenter and Duck Creek via their published APIs. os.moda runs the intake, orchestration, and governance layer in front of your core — the core stays the system of record for policy and claims. Most carriers see ROI before any core change.
What about EU AI Act high-risk under Annex III?
We pre-build the Annex IV technical documentation set: risk management system, dataset description, accuracy/robustness/cybersecurity testing, human-oversight design, and post-market monitoring logs. The Digital Omnibus may move the deadline to Dec 2027 — either way, you're ready.
From FNOL to decision, on infrastructure your DOI can audit in a single export.
Spawn an insurer stack →