The Algorithm Doesn’t Know the Injured Worker 

Dollars & Sense

Predictive technology can recognize patterns across thousands of claims. Some of the most important influences on recovery emerge only when someone asks the injured worker what is happening. 

Imagine two injured workers with similar injuries, medical histories and expected recovery periods. Their claims enter a predictive system and receive comparable risk scores. One has a supportive supervisor, available modified duty and confidence in returning to work. The other fears losing their job, has not heard from their employer and cannot afford another interruption in income. Their medical information may look remarkably similar while their circumstances create different challenges. An algorithm relying exclusively on administrative claims information may have no reliable way to distinguish those experiences. The difference becomes visible when someone has a conversation. 

What the Numbers Know 

The predictive value of statistical models deserves recognition. In a landmark 2000 meta-analysis, psychologist William Grove and his colleagues found that mechanical prediction methods were approximately 10% more accurate than clinical judgment across studies of human health and behavior. Human judgment was substantially more accurate in only 6% to 16% of the comparisons. This research examined statistical prediction broadly rather than contemporary workers’ compensation AI, but its implications are important. Professional experience does not automatically produce more accurate forecasts than mathematical models. The opportunity lies in using that predictive advantage while recognizing the information and circumstances the model may overlook. 

One of the most revealing examples comes from healthcare research published in Science in 2019. Ziad Obermeyer and colleagues investigated a commercial algorithm that identified patients for additional care-management support. The system used healthcare spending as a proxy for medical need. Because less money had been spent on Black patients with comparable health needs, the algorithm systematically underestimated their need for additional assistance. Correcting the disparity would have increased the proportion of Black patients identified for additional help from 17.7% to 46.5%. The problem emerged from the measurement selected to represent need. For workers’ compensation, the parallel is worth examining because historical claim costs and disability durations can reflect injury severity alongside healthcare access, employer practices and administrative decisions. This research does not establish that workers’ compensation algorithms exhibit the same bias. It establishes why their underlying assumptions require examination. 

The Human Factors Behind Recovery 

Carol Cancelliere and colleagues reviewed 56 systematic reviews examining factors associated with return to work following injury or illness. Their 2016 synthesis identified recovery expectations, self-efficacy, workplace coordination and multidisciplinary support among the important factors associated with positive outcomes. Several of these factors are potentially modifiable through appropriate intervention. An injured worker who doubts their ability to return may need reassurance grounded in clinical progress, clear functional goals or a conversation about available accommodations. Another may need assistance addressing workplace conditions that contributed to the injury. These circumstances deserve attention because they can influence recovery without appearing as meaningful information in a traditional claims dataset. 

Consider what a claims professional might learn during an early conversation. An injured worker may explain that their supervisor has not contacted them since the accident. Another may describe uncertainty about whether modified duty actually exists or whether their position will remain available. Someone else may have received conflicting information from the employer, medical provider and insurance administrator. These concerns can influence expectations about recovery and create obstacles that extend beyond the physical injury. The conversation provides an opportunity to identify a specific problem and coordinate an appropriate response. The information becomes valuable when someone acts on it rather than simply documenting another risk factor. 

The claims experience itself deserves similar attention. Alex Collie and colleagues surveyed 10,946 injured workers across ten Australian workers’ compensation schemes. Approximately 23% reported a negative or neutral claims experience. Workers reporting positive experiences were significantly more likely to report returning to work after accounting for several injury, worker, claim and employer characteristics. Because the study was cross-sectional, it could not establish whether better experiences improved recovery or successful recovery improved perceptions of the process. Nevertheless, the association raises an important question for organizations responsible for claims administration. If the experience of navigating the system is related to recovery outcomes, communication quality deserves consideration alongside medical treatment and financial performance. 

When Professional Judgment Becomes Optional 

A practical concern emerges when predictive scores become substitutes for professional assessment. Claims departments operate under demanding workloads, complicated documentation requirements and competing deadlines. In that environment, a numerical classification can become an efficient shortcut for determining which claims receive additional attention. The danger is that the classification may be accepted without examining the information used to generate it. A high-risk score could prompt additional oversight when the injured worker actually needs improved communication or a clear return-to-work plan. A low-risk score could discourage attention to difficulties that developed after the information was collected. The system may still be functioning according to its design even when its recommendation is poorly matched to the person’s circumstances. 

This is where leadership and organizational expectations become especially important. Claims professionals need sufficient time and authority to evaluate automated recommendations rather than simply process them. Supervisors should encourage thoughtful disagreement when the available evidence supports a different response. Employers must recognize that their own communication and accommodation practices may contribute to circumstances influencing recovery. Vendors should explain the information included in their models and acknowledge important limitations. The objective is to create a working relationship in which technology supports professional decisions while preserving responsibility for the consequences of those decisions. 

Turning Information into Intervention 

The practical opportunity is to connect predictive information with deliberate human inquiry. A high-risk classification should prompt an assessment of what is driving the risk and which circumstances may be changed. Claims professionals can explore recovery expectations, communication with the employer, confidence in modified duty and obstacles to accessing treatment. Relevant information should be documented consistently without collecting unnecessary sensitive personal details. Any decision to depart from an algorithmic recommendation should have a clear rationale that can be reviewed and evaluated. These practices allow organizations to preserve valuable human information while also learning whether their predictive systems are producing useful recommendations. 

There is also an important distinction between collecting more information and improving outcomes. Asking an injured worker about concerns is valuable only when the organization has the capacity to respond appropriately. An employer may need to clarify modified-duty expectations. A claims professional may need to resolve inconsistent communication or coordinate with an authorized medical provider. A supervisor may need guidance on maintaining supportive contact without pressuring someone to return before it is medically appropriate. The effectiveness of these actions should be evaluated through functional progress, appropriate care and sustainable return to work. Adding more questions without creating the ability to address the answers provides little meaningful improvement. 

Predictive technology offers a significant opportunity to recognize patterns and allocate attention more effectively. Its greatest value may emerge when the information it produces leads professionals to investigate circumstances they otherwise would have missed. An injured worker’s expectations, concerns and relationship with the workplace can change during recovery. Those changes may influence the eventual outcome even when the original injury and medical diagnosis remain the same. Organizations should therefore evaluate whether their technology helps professionals recognize opportunities for meaningful intervention. An algorithm can estimate the likelihood of a difficult recovery. The people involved still have the opportunity to influence what happens next. 

Research Referenced 

Cancelliere, C., et al. (2016). Factors Affecting Return to Work After Injury or Illness: Best Evidence Synthesis of Systematic Reviews. Chiropractic & Manual Therapies, 24, 32. Full Study. 

Collie, A., Sheehan, L., Lane, T. J., Gray, S., & Grant, G. (2019). Injured Worker Experiences of Insurance Claim Processes and Return to Work: A National, Cross-Sectional Study. BMC Public Health, 19, 927. Full Study. 

Grove, W. M., Zald, D. H., Lebow, B. S., Snitz, B. E., & Nelson, C. (2000). Clinical Versus Mechanical Prediction: A Meta-Analysis. Psychological Assessment, 12(1), 19–30. Research. 

Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting Racial Bias in an Algorithm Used to Manage the Health of Populations. Science, 366(6464), 447–453. Research. 

Share This Article: