As CEO of SecondLook Health, I’ve seen first hand that most under writing decisions in Workers’ Comp are still made without visibility into the clinical events that greatly influence claim cost, disability duration, and litigation risk. The industry prices risk using backward‑looking aggregates, while the real signals sit buried inside thousands of pages of medical records. AI can now extract these signals, from handwritten notes to laboratory tables, and use them to sharpen risk segmentation, speed decisions, and improve loss ratios.Here’s what that means for you.
Workers’ compensation underwriting is generally built on coarse, lagging indicators at the employer (book of business) level, such as:
While this approach has historically been sufficient, it arguably has three limitations. First, it is aggregate and backward-looking: it treats all “back injuries” or all “manufacturing employers” as roughly similar. Second, it is weakly connected to clinical reality: underwriters rarely see the diagnoses, treatment sequences, provider choices, or guideline adherence that greatly influence cost. And third, it under uses rich, available data - detailed medical records, provider patterns, and psychosocial risk indicators are seldom incorporated.
Moving forward, the loss environment is getting harder. This is in part due to these factors:
This can be especially true for certain injury types and psychosocial profiles.
Finally, with the coming tsunami of AI, late-adopter carriers risk adverse-selection if they price on yesterday’s aggregates while competitors use tomorrow’s granular risk signals.
Beyond insurance costs, the CDC reports that productivity losses linked to absenteeism cost U.S. employers roughly $226 billion per year, or about $1,685per employee, which makes even modest improvements in return‑to‑work outcomes economically meaningful. [4]
How can AI use healthcare data to build better predictive models of severity, litigation, and high-cost risk?
AI will improve underwriting first and foremost by unlocking clinical data, turning historical and active claims into a clearer view of risk. This will allow underwriters to “see” how injuries evolve, how quickly people return to work, and which treatment and provider patterns drive avoidable cost, disability, and litigation.
(We’ll discuss how AI ultimately evolves from predictive to prescriptive and ultimately agentic AI in another letter.)
An AI-based, clinical analytics platform such as SecondLook turns historical claim files, including full medical charts, into structured variables and timelines. These data incorporate:
These attributes can also be used to characterize network providers, e.g., fast-recovery vs. slow-recovery providers; or high-surgery vs conservative care.
Why does that characterization matter? Take for example rehabilitation plans involving surgery. WCRI and related analyses indicate that surgery is present in only a small minority of claims but can account for more than half of medical costs, and that delayed or resource‑intensive care late in the life of a claim is associated with a many‑fold increase in the probability that a claim becomes high cost. [5] In another example, recent WCRI studies find that workers’ compensation patients have a higher prevalence of psychosocial risk factors than privately insured patients, and that these“yellow flags” are strongly associated with poorer functional recovery and longer disability for common musculoskeletal injuries. [6]
Applied to underwriting, this enables:
When underwriting is powered by deep clinical insight, not just aggregate loss histories, workers’ compensation carriers can distinguish “good” from“bad” risks within the same class code; price and structure coverage more precisely and fairly; and clinically intervene earlier on claims that matter most.
The combination of clinical AI systems of record (such as SecondLookHealth), predictive and prescriptive modeling, and emerging agentic AI architectures offers a concrete path to 10x improvements in decision speed and accuracy.
[1] https://www.ncci.com/Articles/Pages/II_Insights_Medical-Costs-Then-and-Now.aspx
[2] https://www.ncci.com/Articles/Pages/Insights-2025-in-Sight-2024-in-Review.aspx
[3] https://www.wcrinet.org/images/uploads/files/wcri8364.pdf
[4] https://www.cdcfoundation.org/pr/2015/worker-illness-and-injury-costs-us-employers-225-billion-annually?utm_source=chatgpt.com
[5] https://riskandinsurance.com/the-current-state-of-complex-claims-in-workers-compensation-understanding-the-drivers-of-rising-costs-and-duration/
[6] https://www.wcrinet.org/news/detail/wcri-poorer-recovery-seen-among-injured-workers-with-psychosocial-risk-factors
