Chapter 33 — Key Takeaways

A one-page card. Fraud, misrepresentation, rescission, the SIU, and the analytics that catch fraud — written entirely from the detection side.

The core claims

  • Fraud is adverse selection with intent. Ordinary adverse selection is passive (the people who expect a loss buy most eagerly); fraud is deliberate — manufacturing, exaggerating, or concealing a loss and lying to make the policy pay. Every underwriting defense is also a fraud defense.
  • The honest policyholder pays for fraud — twice. Undetected fraud is absorbed into the premium the whole pool pays; the controls built to catch the detected fraud are a second tax. Fighting fraud with proportion lowers that tax; treating everyone as a suspect raises it.
  • Soft fraud vs. hard fraud. Soft (opportunistic) fraud shades a real situation (padded claims, rounded payroll, a "forgotten" loss) — common, diffuse, hard to detect. Hard (premeditated) fraud manufactures a loss (staged crash, arson, phantom employee) — rarer, costlier, often more detectable because it leaves a structure.
  • A false answer is the beginning of the analysis, not the end. A material misrepresentation needs a false statement of fact + materiality (would have changed the decision) + reliance. Fraud needs one more thing data cannot supply: intent.
  • Rescission voids the policy from inception — premium back, coverage void, claims denied. It is powerful and hedged: high, state-varying standards, incontestability, innocent-misrepresentation and waiver defenses, and the burden of proof. A failed rescission is worse than the claim — it adds bad-faith exposure.
  • A red flag is a prompt to investigate, never a finding of fraud. One flag is a question; a cluster across families is a referral. Every red-flag family has an innocent explanation that is usually true.
  • The underwriter spots and refers; the SIU investigates and substantiates. Most referrals come back cleared — that is a feature. Referring is not accusing; over-referring floods the unit and buries real fraud.
  • Analytics catch fraud at scale but never deliver a verdict. Anomaly detection finds the unusual; link analysis finds hidden connections (lethal to rings); predictive models triage by likelihood. Each produces a lead, not a fact. A fraud score cannot rescind, deny, or convict.

The rule of thumb

Verify, don't accuse. Grade by materiality and intent. Ask before you assume, in writing. One flag is a question; a cluster is a referral. Catch fraud at the door (good application + up-front verification) so you never have to rescind on a pretext after the loss.

The key distinctions

Pair The line between them
Soft vs. hard fraud Shading a real loss vs. manufacturing one
Honest error vs. fraud Materiality alone vs. materiality + intent
Misrepresentation vs. concealment An active false statement vs. a material silence (needs known materiality)
Non-renewal vs. rescission Ends coverage forward vs. voids it from inception
Red flag vs. finding A prompt to look vs. a substantiated fact
Clarification vs. referral One flag, innocent explanation likely vs. a cluster warranting the SIU
Lead vs. proof What analytics produce vs. what a person must establish

Key terms

insurance fraud (soft/hard) · material misrepresentation · rescission · special investigation unit (SIU) · red-flag indicators — plus, used from their owners: utmost good faith, concealment, representation vs. warranty (Ch.4); loss run, CLUE, MVR, credit-based insurance score, FCRA (Ch.8); GLM, GBM, lift, feature engineering (Ch.32); proxy discrimination, algorithmic bias (Ch.35, previewed).

What you could defend to your manager

"The application understated the 2023 fire's cause — material, so I flagged it, but a single discrepancy on a loss whose existence and size were honestly disclosed, with no cluster of other flags and corrective controls already attached. That's a documented clarification, not a fraud referral and not a rescission issue. I sent a neutral written question to the broker, confirmed the cause is hot-work — which is exactly why the hot-work-permit program is a binding condition — documented it, and the file is now clean enough to defend on the integrity of its information, not just its price and terms."