Use cases · Insurance carrier
Auto-insurance fraudis different. We've spent years inside it.
Banking fraud is third-party — someone impersonating the customer. Insurance fraud is largely first-party— the customer (or their broker, body shop, or attorney) participating in the scheme. The anti-fraud playbook is different. So is the data signal. We've built rules and models inside a leading US auto carrier — across both premium avoidance and inflated claims.
The auto-fraud landscape
Two fraud surfaces that require different controls.
Auto-insurance fraud spans both premium avoidance and inflated claims. The split matters: controls built for one surface do not automatically address the other.
Premium avoidance
Underwriting and policy risk
Inaccurate or incomplete answers at quote or bind that produce a lower rate than the risk warrants — rate evasion, ghost brokers, garaging fraud, unlisted drivers, chameleon carriers.
Inflated claims
Claims and provider risk
Staged accidents, inflated medical billing, PIP / IME abuse, body-shop and attorney rings, early-term claims, total-loss and theft fraud.
A healthy book runs around a 75% pure loss ratio. We find the segments running at 100%+ — and decision them differently.
Why insurance is different
First-party fraud. Money flows the other direction.
Banking / fintech
Bank pays customer first.
Fraud is mostly third-party impersonation — ID theft, card theft, account takeover. The defense is KYC, device intelligence, transaction monitoring, dark-web signal.
Auto insurance
Customer pays carrier first. Claims payout comes later — or never.
Fraud is mostly first-party — the policyholder, broker, body shop, attorney, or medical provider participating. The defense is rule-and-graph detection across the digital exhaust of quotes, policies, claims, and provider networks.
By customer journey
Five lifecycle stages. From quote to claim.
Fraud doesn't arrive in one moment — it shows up across the policy lifecycle. Centurion plugs in at every stage where an underwriter, claims handler, or SIU referral would benefit from a real-time signal.
01
Attract
Quote & channel
- Agent / broker channel quality scoring
- Quote-iteration pattern analysis
- Marketing-channel risk weighting
02
Acquire
Underwriting & policy issue
- Application underwriting & rate-evasion detection
- Ghost-broker activity & agent-fraud rings
- Garaging / territory misrepresentation
- Unlisted drivers (households, employees, related parties)
- Rate-gaming on business class / vehicle type / fleet size
- ID card fraud, sublease game, chameleon motor carriers
03
Engage
Policy term
- Endorsement & mid-term-change risk
- Payment & premium-refund-abuse monitoring
- Driver-addition pattern detection
- MVR-order-pattern signal
04
Grow
Renewal & expansion
- Renewal-underwriting risk re-scoring
- Cross-policy fuzzy-match (repeat unacceptable customers)
- Loss-ratio segmentation & treatment routing
- Book-of-business quality monitoring
05
Resolve
Claims & SIU
- Early-term claims fraud (claim within weeks of policy issue)
- Staged-accident detection & participant graph
- PIP / IME outlier detection on CPT codes & billing patterns
- Body-shop + attorney conspiracy rings
- Total-loss / theft-fraud scoring
- SIU referral routing & case prioritization
Premium avoidance — at underwriting
Six surfaces. Caught before the policy gets bound.
Premium-avoidance fraud is best fought at quote and bind — before the carrier is on the hook for a mispriced risk. The signal comes from quote-iteration patterns, cross-source contradictions, and graph linkage to prior unacceptable risks.
Ghost-broker rings
Unauthorized agents writing fraudulent policies for non-existent risks, harvested premiums, and fabricated documentation. Graph signal across phone, email, payment instrument, and broker code.
Rate gaming
Misrepresentation of business class, vehicle body type, garaging address, fleet size, or company owner — to land a lower rate than the risk warrants. Detected via cross-source contradiction and historical-pattern matching.
Garaging fraud
Policy garaged in a low-rate territory but the vehicle operates somewhere else. External-data triangulation plus quote-iteration history flags the mismatch at bind.
Unlisted drivers
Drivers operating insured vehicles who aren't on the policy — household members, employees, undisclosed relations. Detected via quote-iteration signal, MVR-order patterns, and household graph.
Sublease & ID-card fraud
Policy held under one name; the actual operator is someone else. ID-card fabrication for proof-of-insurance purposes without genuine coverage intent.
Chameleon motor carriers
Commercial-auto carriers that have been cancelled or refused elsewhere reapply under a different DOT / business entity. Fuzzy-match across name, address, contact, VIN, and prior-claims history.
Inflated claims — at FNOL and beyond
Six surfaces across the claims-fraud problem.
Claims fraud needs different signal — provider-network density, CPT-code outliers, attorney clustering, body-shop steering patterns. Deep-learning architectures on medical-code data surface the outliers that adjusters can't see by hand.
Staged accidents
Coordinated rings — driver, passenger, witness, body shop, attorney, medical provider — that engineer collisions for payout. Graph detection at FNOL surfaces ring connectivity in real time.
PIP / IME outliers
Personal-injury-protection claims with anomalous CPT code patterns, billing-frequency outliers, and provider-network density. Deep-learning models trained on medical-code + claims-data + policy-attribute features.
Inflated medical bills
Phantom procedures, overbilled units, upcoding, and unbundling — pattern detection across provider, body shop, attorney, and patient cohorts.
Body-shop & attorney rings
Coordinated steering between body shops, medical providers, and attorneys driving inflated payouts. Network signals surface the ring before each individual claim looks anomalous.
Early-term claims fraud
Policy bound days or weeks before a claim is filed — often with loss circumstances that predate the policy. Time-from-bind-to-FNOL signal plus underwriting-history match.
Total-loss & theft fraud
Reported total-loss or theft events that pattern-match known fraud rings — vehicle history, VIN matching, location signal, and prior-claim graph.
What plugs in
Three Centurion components, tuned to the insurance data shape.
ML Graph
Graph features across customer, vehicle, address, phone, email, payment, agent, and prior claim. Designed for real-time use at quote and FNOL without requiring a graph-database migration.
Learn more →Model Foundry
Custom deep-learning architectures for PIP / IME outlier detection across CPT codes, claims data, and policy attributes. Rate-evasion and ghost-broker models tuned to your book.
Learn more →FlowX
Decisioning fabric for quote-decline, verification-routing, mid-term-cancel, and FNOL-referral strategies. Versioned, A/B-testable, audit-logged. Plays with your underwriting and claims systems.
Learn more →What changes
Production work with a leading US auto carrier informed this playbook across premium avoidance and claims fraud. Applicability and measured lift depend on each carrier's book, data, and controls.
For SIU and underwriting leaders
See it on your book.
We'll run a benchmark on a slice of your historical policies and claims — quantified loss-ratio lift, segment by segment, before you ship anything.