Dollars & Sense
In workers’ compensation, the organization purchasing the technology, the employer funding the losses and the injured worker experiencing the outcome may have very different definitions of success.
Consider what happens when an employer reviews a quarterly performance report from its workers’ compensation administrator. The technology vendor reports faster processing times, improved automation and millions in calculated savings. The TPA demonstrates productivity improvements and favorable claim closure statistics. The employer sees continued increases in total costs and injured workers are still experiencing delays in treatment and returning to work. Every organization may be reporting its results accurately. The troubling possibility is that each is measuring success against a different objective. The technology could be performing exactly as designed while the system fails to achieve the outcomes that matter most.
The Customer Who Defines Success
Workers’ compensation operates through a complicated chain of financial and operational relationships. An employer or captive may fund the ultimate losses while a carrier or TPA manages the claims. Technology vendors provide predictive models, automated workflows and decision-support systems through contracts negotiated with organizational buyers. Injured workers experience the resulting medical decisions, communication and return-to-work processes without participating in those purchasing decisions. Each party has legitimate interests and obligations. Problems develop when the system optimizes one organization’s preferred measure while overlooking the consequences for everyone else.
Economist George Baker examined this challenge in his 1992 research on incentive contracts and performance measurement. He demonstrated that compensation arrangements based on imperfect measures of organizational objectives can produce inefficient incentives even when participants respond rationally. His analysis helps explain why a vendor can satisfy its contract without necessarily delivering the broader outcome the customer intended. A contract rewarding processing speed will encourage attention to speed. A contract emphasizing financial savings will make savings a central measure of performance. Neither objective automatically establishes whether an injured worker received appropriate treatment or achieved sustainable recovery.
These distinctions matter because technology is not inherently aligned with the interests of every participant. A model designed to predict litigation may help an administrator allocate legal resources. A system designed to identify billing irregularities may reduce inappropriate payments. An automated workflow may free claims professionals to devote more attention to complex injuries. All three can create meaningful operational value. Leadership must still establish how that value contributes to the larger purpose of workers’ compensation rather than assuming that operational improvements constitute a complete measure of success.
When the Contract Shapes the Outcome
Financial arrangements provide a practical example. In its March 2026 Securities and Exchange Commission filing, Pacific Health Care Organization described workers’ compensation managed-care services supported by fixed fees, hourly charges and certain fees calculated as percentages of savings. These are established commercial arrangements and their existence does not indicate improper behavior. They do demonstrate how a vendor’s revenue may be connected to transaction volume or a particular calculation of savings. A percentage-of-savings agreement creates an especially important question about the baseline used to calculate performance. Employers should know whether reported savings represent genuine reductions in spending or simply differences between billed charges and negotiated payment amounts.
The importance of governing automated recommendations became apparent well before contemporary artificial intelligence. In 2010, Allstate entered a $10 million multistate regulatory agreement involving its Colossus bodily-injury claims evaluation software. Regulators identified inconsistencies in the oversight and regional tuning of the system but did not identify institutional underpayment. The agreement required stronger auditing, greater consistency and disclosure of the software’s use. It also prohibited policies requiring adjusters to settle solely at the software’s recommended value or rewarding them for settling near it. This was automobile bodily-injury claims technology rather than an AI workers’ compensation system. Its relevance is the documented recognition that software configuration and employee incentives require oversight independent of the software’s technical capabilities.
More recent scrutiny has emerged in health insurance. A 2024 majority staff report from the U.S. Senate Permanent Subcommittee on Investigations examined prior authorization practices in Medicare Advantage. The report found that UnitedHealthcare’s denial rate for post-acute care requests increased from 8.7% in 2019 to 22.7% in 2022 during a period when automated processes were expanding. The insurers disputed aspects of the report’s characterization and the findings do not establish that technology alone caused the increase. Medicare Advantage also operates under a different legal and financial framework from workers’ compensation. Nevertheless, the investigation illustrates why organizations should monitor actual decision patterns before and after automation. A technology implementation can change operational behavior in ways that deserve independent examination.
The Outcome Missing From the Scorecard
The Workers Compensation Research Institute has developed a more comprehensive understanding of what injured workers experience. In research covering 9,730 injured workers across 15 states, approximately 16.7% reported significant difficulties obtaining desired or physician-recommended medical care. An average of 10.5% had not returned to work because of their injuries at the time measured. These figures describe historical outcomes rather than consequences of AI. They demonstrate the importance of measuring access to care, physical functioning and return-to-work results alongside administrative performance. A technology procurement process that excludes these dimensions cannot provide a complete picture of whether the investment benefits the system.
Regulatory expectations increasingly recognize the responsibilities associated with third-party technology. The National Association of Insurance Commissioners adopted its AI Model Bulletin in December 2023 to guide insurance regulatory oversight. Iowa’s November 2024 Bulletin 24-04 provides a concrete example of its implementation. It addresses insurer responsibility for third-party AI systems and encourages appropriate contractual audit rights, vendor cooperation with regulators and ongoing validation. These expectations apply within the relevant insurer regulatory framework. Self-insured employers and captive arrangements require separate analysis of their regulatory obligations. Regardless of the precise legal requirements, the governance principle is commercially significant: purchasing technology does not eliminate the responsibility to understand its consequences.
The National Institute of Standards and Technology offers additional guidance through its voluntary AI Risk Management Framework. Its provisions address third-party technology risks, organizational oversight and ongoing monitoring. For employers and captives, those principles can be translated into specific purchasing requirements. A vendor should explain what its technology optimizes, how performance is validated and who can change the model’s settings. Contracts should establish access to meaningful data, independent audit rights and notification when material model changes occur. Organizations should retain the ability to question recommendations and assess results without relying exclusively on the vendor’s own performance calculations.
Buy the Outcome You Actually Want
Employers should define success before allowing a vendor to determine how it will be measured. Relevant outcomes include total claim costs, treatment timeliness, appropriate medical utilization, sustained return to work and injured-worker experience. These measures require appropriate comparisons and risk adjustment because injury complexity can influence every result. Contracts should establish baseline performance, reporting responsibilities and periodic independent evaluation. They should also address data ownership, confidentiality, model changes and the ability to discontinue a tool when performance is inadequate. Where performance-based fees are appropriate, organizations should consider multiple balanced measures with safeguards against rewarding decisions that disadvantage injured workers.
There is an important complication in assuming that the least expensive claim always represents the best outcome. A premature return to work may lower immediate indemnity payments while increasing the risk of reinjury or prolonged disability. Appropriate medical treatment can increase short-term spending while improving long-term function. Some injured workers cannot safely return to their previous positions and require legitimate long-term support. Financial efficiency remains essential because employers must maintain sustainable programs. The challenge is to distinguish responsible cost management from decisions that merely transfer costs, delay care or make the financial report look more favorable.
The greatest responsibility rests with organizational leaders who authorize the purchase and define the desired results. Vendors should be expected to explain their technology and demonstrate performance. Carriers and TPAs should ensure recommendations are applied responsibly and evaluated against the needs of individual claims. Employers and captives should demand evidence that improvements extend beyond the indicators featured in a sales presentation. Injured workers deserve a system that treats their recovery and legally entitled benefits as meaningful measures of performance. These responsibilities can reinforce one another when they are included in the technology’s design and governance. An algorithm will pursue the objective established for it. Leadership is responsible for determining whether that objective deserves to be pursued.
Research Referenced
Baker, G. P. (1992). Incentive Contracts and Performance Measurement. Journal of Political Economy, 100(3), 598–614. University of Chicago Press.
National Association of Insurance Commissioners. (2023). Model Bulletin on the Use of Artificial Intelligence Systems by Insurers. NAIC Adoption Announcement.
Pacific Health Care Organization, Inc. (2026). Form 10-Q, Quarterly Period Ended March 31, 2026. U.S. Securities and Exchange Commission. SEC Filing.
Savych, B., & Thumula, V. (2018). Comparing Outcomes for Injured Workers, 2017 Interviews. Workers Compensation Research Institute. WCRI Research.
