Credit: Mdisk/Adobe Stock

Unsupervised machine learning models can detect suspicious property and casualty claims just two weeks after they’re filed — far ahead of traditional methods, according to a recent study by CLARA Analytics.

Cohort modeling across claim development periods can effectively identify cost and treatment outliers, the data showed, while mapping connections between providers and attorneys that may indicate fraudulent activity.

“By leveraging advanced analytics, we’ve shown that insurers can identify potential fraud much earlier in the claims process, potentially saving billions in fraudulent payouts, said Pragatee Dhakal, director of claims solutions at CLARA Analytics.

“What’s particularly promising about this approach is that it doesn’t rely on pre-established fraud indicators,” he added. “By using unsupervised learning techniques, the system can potentially identify novel patterns of fraudulent activity that might not match historical cases.”

Key findings from the study include:

  • Nine percent of open claims were identified as high potential for Special Investigation Unit (SIU) referral.
  • Michigan and Arizona showed the highest percentages of potential fraud indicators.
  • The model’s predictions closely matched actual SIU referrals made by adjusters but detected potential cases significantly earlier — as soon as two weeks after the first notice of loss.
  • Network analysis revealed important connections between attorneys and medical providers that traditional methods might miss.

Meanwhile, the FBI estimates that insurance fraud costs the industry approximately $40 billion annually, excluding medical insurance. These costs ultimately affect policyholders through increased premiums.

At the same time, the global AI in insurance market is experiencing substantial growth and is projected to reach $10.27 billion in 2025, attributed to data explosion, risk assessment and underwriting, fraud detection and prevention, customer experience enhancement, operational efficiency and cost reduction.

“AI enables real-time resolution for up to 70% of simple claims through end-to-end automation,” said Joe Khoury, managing director and partner on the insurance practice at Boston Consulting Group.

“Technologies such as computer vision, natural language processing and machine learning allow insurers to assess damage, validate coverage, detect fraud and issue payments in minutes—reducing operational costs by 30% to 50%,” he added. “These systems also improve fraud detection, using behavioral analytics and anomaly detection to flag suspicious patterns early. As a result, AI-driven claims platforms are not just faster—they’re more secure.”

NOT FOR REPRINT

© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.