The evolution of AI in workers' compensation

The evolution of artificial intelligence (AI) in workers' compensation truly begins with the rollout of predictive models. Predictive AI leverages data sets and identifies patterns to forecast outcomes or potential future events. These models are highly data-driven, low-risk and generally optional. They provide the foundation for advanced AI models across industries.

In claims management specifically, this technology predicts severity or outcomes, often identifying outliers that are most likely to become litigated. However, now that the industry and the workforce are more comfortable and more familiar with the application of AI in claims management, the technology has evolved and matured over time. Today, AI is being applied across claims management in a more actionable way. It is no longer "tell me what happened in this claim," but "tell me what I can do about it." The response is now prescriptive.

The role of prescriptive AI

Rather than replacing predictive AI, prescriptive AI has emerged alongside it, representing an important advancement in claims management. Together, these approaches enable organizations to both anticipate outcomes and actively influence them.

Prescriptive AI provides recommendations based on historical data, past experiences and predicted outcomes, allowing adjusters to make more informed decisions. For example, when a claim is predicted to carry higher severity, AI can analyze both historical and current data to suggest potential next steps. These recommendations support, but never replace, the adjuster's judgment, enabling proactive measures while keeping complete control in the hands of human decision-makers.

Across the industry, AI in claims management is often used to automate procedural or analytical tasks, rather than to make final claim eligibility decisions, which can help reduce the potential for bias to directly affect claim outcomes.

To unlock the full potential of prescriptive AI, organizations must consider:

• Data accuracy and data collection control: In addition to clean, reliable data, organizations need information that spans both macro (meta-level) and micro (granular level) views. Data collection control refers to the ability to filter out irrelevant inputs, ensuring only meaningful data informs the model.
• Operational understanding: True operational understanding is having a clear idea of best practices, deep knowledge of how things work today, and visibility into edge cases and outlier processes.

With these two dimensions in place, prescriptive AI can offer the benefits of competitive differentiation, enhanced decision-making, and provide a clearer return on investment than previous models and legacy technology.
It represents the next stage of advancement, moving from incremental improvements in predictive AI to the emerging promise of agentic AI. Looking ahead, the industry is beginning to explore this next phase.

The future of AI in claims is agentic

Agentic AI refers to emerging systems designed to handle more complex workflows with reduced human input, while still requiring oversight to ensure quality, compliance and accountability. Many industries are beginning to experiment with agentic workflows, and workers' compensation is exploring how these approaches may apply.

In claims, this evolution will occur at the intersection of three key elements: model quality, data quality and feedback automation. Achieving data quality requires a deep understanding of the individual claimant, their job, and their employer so that models can accurately interpret context and tailor their outputs accordingly. However, the adoption of agentic AI depends on consistently demonstrating that its performance surpasses that of fully human or hybrid human-in-the-loop (HITL) methods. Confidence in the system must be earned through consistent, measurable success.

Even as capabilities evolve, maintaining a HITL method remains essential, especially during the transition phase. Transparency in AI use is an industry imperative; communicating how recommendations are generated and what factors inform them helps build trust among stakeholders. HITL ensures that autonomous systems remain accountable and that outcomes are transparent, auditable and aligned with quality standards.

Human oversight will remain a cornerstone of responsible adoption, ensuring outputs are consistently reviewed, validated and aligned with regulatory and ethical standards.

Equally important, the current prescriptive models must be nearly flawless. There can be no major outlier errors, as these models form the foundation for future agentic capabilities. For example, if critical claim details are missing or if claims are inadvertently accepted or miscategorized, that undermines the trust needed to shift toward autonomous AI. These issues must be fully addressed before the next stage of evolution can begin.

Before organizations can fully embrace agentic AI, they must address key operational and cultural considerations to ensure successful adoption.

Necessary considerations before implementation

In workers' compensation, predictive AI remains critical, while prescriptive AI has emerged to complement it. The industry is now also beginning to explore agentic approaches. Each phase represents the natural evolution of these models as organizations continue to adopt advanced technology and capitalize on deep, data-driven insights.

Although it's natural to progress along the maturity model, the transition is not easy. Each successful shift requires fundamental cultural and operational changes, and it's essential to keep a human in the loop throughout it all.

To fully realize the benefits of AI in claims, organizations must promote and facilitate AI literacy, strategically upskill their workforce, and align AI initiatives with core business objectives. These are non-negotiables for the successful adoption and use of AI in claims. Consider them the table stakes of AI transformation. Understanding the benefits AI brings can help justify the investments and efforts required to navigate these challenges.

The benefits of AI in claims

Past tools relied on statistics and probability. We were effectively just "playing the odds," making overly simplistic assumptions about actions and outcomes. When results didn't fit the mold, we performed "exception processing" rather than adapting our business model to account for those differences.

AI changes that. It allows us to refine how we organize and predict our claims work, with the most impactful improvements occurring in the following areas:

• Process efficiency: Reduces errors and promotes consistency, translating into speed and accuracy, often at a magnitude of 10x or more.
• Contextual decision support: Delivers timely, specific, and context-aware insights to support more effective decisions.
• Outlier detection: Identifies anomalies or deviations from expected patterns that may require further investigation.

Despite these benefits, organizations still face challenges and risks as they expand their AI capabilities.

Managing AI's shortcomings

High associated costs and difficult-to-measure returns on investments were once two of AI's biggest downsides. However, now that costs have peaked (and are continuing to decline) and adoption is accelerating across industries, it's time to shift focus to two emerging considerations:

• A potential reliance on immature technology: Enthusiasm for the promise of AI can sometimes blind leaders to gaps in their business processes, leading to models that don't fully meet organizational needs.
• Upskilling everyone: While essential, upskilling is challenging to scale. Soon, many tasks currently performed by specialists will be handled more broadly, except in rare or highly complex cases. In essence, we will all become data analysts or agent managers, much like we began managing our own correspondence after the introduction of word processing and email. This shift represents a massive undertaking, and most organizations are not yet fully prepared to make the transition.

With these considerations in mind, the path forward for AI in claims is one of cautious optimism and deliberate action.

Considerations for AI implementation

AI has the potential to further transform the claims industry. For example, in a not-so-distant future, we anticipate greater levels of personalization throughout claims handling, as adjusters will be better positioned to understand the complete picture of the injured worker, from their job details to the social determinants of health that influence recovery. These critical factors can be the difference between successful outcomes and prolonged disability.

On the industry side, AI can positively influence employee recruitment and retention by enhancing the "coolness" factor of both the work we do and how we do it. Thirty years ago, the insurance industry was seen as modern and high-tech, with access to cutting-edge tools and technology. While that's no longer the case today, there's an opportunity to shift this perception. Companies that embrace AI and build cultures that foster innovation and growth will lead the way in attracting the next generation of claims professionals.

As X-Prize founder and innovator Peter Diamandis says, "You're missing the bigger picture if you still think that AI is about flashy demos and talking assistants. This is a civilization-scale leap. A once-in-a-generation transformation in how we work, learn, create and solve."

For claims management and the insurance industry, the opportunity now is to carefully test and adopt these technologies in ways that strengthen outcomes, ensure compliance, and build trust as the industry steadily progresses through this AI-driven transformation.