Artificial Intelligence and Unstructured Data in Workers’ Compensation

Over the past couple of years, the number of AI presentations and vendors at industry conferences has been almost overwhelming. Deciding which ones to attend feels a bit like my first visit to VooDoo Doughnut—so many choices, each with a different topping or filling. Are they all the same, just with different packaging? Or does one have something truly unique to offer? Will I pick the right one?

Yet, despite the excitement generated by the AI presentation titles, I often leave these events feeling unfulfilled by what I’ve learned—at least when it comes to AI’s impact on workers’ compensation. But why? A simple internet search revealed part of the problem. I asked Google, “What role does artificial intelligence have in workers’ compensation claims?” The responses were promising:

  • Identifying patterns, predicting risks, and streamlining claims processing
  • Supporting informed claims management decisions
  • Improving medical and financial outcomes

Good stuff! But I wanted more. Looking deeper, I confirmed that most AI applications in workers’ compensation focus primarily on medical treatment and disability management. For example, AI is used to assess the likelihood of surgery, analyze the impact of co-morbidities, predict opioid misuse risks, evaluate access to care issues, and estimate treatment duration and the probability of permanent impairment. These insights are valuable, but they miss an important opportunity—leveraging AI to better understand the human side of workers’ compensation.

Unstructured Data: The Missing Piece

Consider how we currently segment claims and assess initial complexity. Predictive analytics, an established science, typically relies on structured data—medical diagnoses, demographic information, and worker characteristics gathered at claim intake. But artificial intelligence could go a step further by analyzing unstructured data from claim manager notes, recorded phone conversations, medical reports, correspondence, and emails from claim parties (workers, employers, and medical providers).

Why does this matter? Because long-term disability is often driven by non-medical factors—psychosocial risks, workplace dynamics, cultural or family influences, and employer-employee relationships. Structured data alone doesn’t capture these nuances. By analyzing unstructured data, AI could help claims professionals gain a deeper understanding of a worker’s motivations, fears, and barriers to returning to work. And identify situations where the employer is a barrier to return to work. This information would allow for more targeted interventions—sometimes as simple as having a conversation with an at-risk worker, guiding them and their employer through the complexities of the claims and return-to-work process, and providing the support they need to move forward.

The Challenge for the Industry

I recognize that analyzing unstructured data is more complex than predicting cost and duration based on ICD codes and age. But the long-term value—improved outcomes and reduced costs—makes it a challenge worth pursuing. As an industry, we should start asking AI vendors: When will this be part of your service offerings?

The future of AI in workers’ compensation shouldn’t just be about streamlining processes; it should also be about supporting the people behind the claims.