One Cupcake at a Time: Building Trust in AI

The workers’ compensation industry is in the midst of a technological revolution. From document intelligence at intake to agentic AI that helps junior adjusters make senior-level decisions, artificial intelligence is fundamentally changing how claims are managed. Yet for all the promise, claims leaders must navigate real challenges, from employee fears about job security to new governance demands.

To explore how organizations are successfully implementing AI, Risk & Insurance spoke with Christina Ozuna, Vice President of Corporate Claims at EMPLOYERS, and Matt High, Senior Vice President of Bill Review Operations at MedRisk. Their insights reveal an industry at an inflection point — one where technology is advancing faster than many leaders imagined possible.

Where AI Is Delivering Measurable Gains

Christina Ozuna, Vice President of Corporate Claims, EMPLOYERS

The productivity and quality improvements from AI in claims management are showing up in several key areas. Ozuna identified four where the impact has been most pronounced: document intelligence at intake, reserving and severity prediction, fraud and anomaly detection, and administrative workload reduction.

“It’s cutting the time an adjuster spends reconstructing a claim. They can act on it immediately instead of piecing together the history first,” Ozuna said of document intelligence. On the reserving side, she noted, “We can get reserves to the right level more quickly and get a clearer picture of the full exposure.”

“AI has significantly enhanced fraud detection by identifying patterns across providers and billers that were previously difficult to detect,” High said.

“It enables rapid link analysis and uncovers relationships that once required extensive programming and manual effort, accelerating investigations and improving detection accuracy.”

Perhaps most significant is what Ozuna describes as time-shifting decisions. “The real shift isn’t automating the decision at day ninety — it’s giving the adjuster at day fifteen the same information they’d have had at day ninety, so they can act sooner.” In a workers’ comp claim, the gap between day 15 and day 90 can be the difference between a routine recovery and one that spirals.

MedRisk’s clinical operations back this up with hard numbers of their own.

On the care management side, the data is equally compelling. “AI helps drive 76% accuracy in early identification of cases exceeding guidelines,” High said. “It can see patterns and spot issues that might otherwise be missed. This capability helps drive additional medical savings and decreases the episode duration.” Shorter episodes usually mean the injured worker is back to health, and often back on the job, sooner.

High shared a case study that puts the quality gains in stark relief. When four nurses reviewed the same set of bills, they produced four different outcomes and four different savings rates. Agentic AI, trained on that collective expertise, delivered results that were “over 99% accurate, and the nurses agreed with it.”

Ozuna pointed to another fundamental shift: from prediction to prescription. “We’ve had predictive models for years that flag when something’s starting to diverge — but then someone still has to act. These new models don’t just flag it, they tell you what to do about it. That’s the difference between information and action.”

Keys to Successful Implementation

Matt High, Senior Vice President of Bill Review Operations, MedRisk

For claims leaders looking to implement AI, both experts emphasized several critical success factors.

Leadership commitment tops the list. “Adoption depends heavily on leadership. How leadership rolls AI out, demonstrates it, and supports it across the organization is absolutely critical,” Ozuna said.

Building user feedback into every model is equally important. “Everything we’re rolling out has a way for users to feed back into the model,” Ozuna explained. “When people feel they have real input into how it’s working, and they see that feedback reflected in the results, adoption goes up significantly.”

That approach has yielded results. “We’ve achieved a 98% AI adoption rate in my claims organization. Trust isn’t built by asking employees to believe in AI. It’s built when they consistently see it making their jobs easier, improving quality, and keeping people accountable.” Ozuna said.

Ozuna also uses a “cupcake shop” analogy to make adoption approachable. “We tell our teams we’re not asking them to build, bake, and decorate a wedding cake on day one — just make a cupcake. Those small wins add up. Take the scary element out, and adoption holds at a much higher level.”

High emphasized measurable, defensible, and low-friction implementations. “You need to make sure it’s compliant and that what you get out of it is defensible. You always have to start with the outcomes in mind. If you’re going to put AI in for workflow automation or digital automation, you need to define what that success measure ultimately is.”

Governance is non-negotiable. “You’ve got to make sure you have good AI governance as a company,” High said. “Otherwise, you don’t want a situation where one person goes and uses Claude, another uses ChatGPT, and you get different outcomes because you’re using them differently.”

Data quality is foundational, especially for clinical decision-making. “It’s important to make sure that your data and the information you’re collecting is clear, documented, and vetted when implementing agentic AI,” High said.

Together, those principles reflect a discipline MedRisk has built into its bill review operations: treat AI as a rigorously governed capability, not a shortcut.

Overcoming Employee Fears and Resistance

Despite the productivity gains, employee anxiety about AI remains a persistent obstacle. Ozuna was candid about this reality.

“Jobs will continue to evolve, just as they have throughout my career. The people who learn to work alongside AI will be best positioned for future opportunities.”

Her approach is to be direct with employees about what AI cannot yet do. “AI can’t hear the fear or distrust in someone’s voice that signals noncompliance down the road, or catch a baby crying in the background and understand there’s a childcare issue keeping someone from returning to work. Those nuances are what claims adjusters are trained to pick up on — and AI still can’t. AI gives adjusters better information faster, but empathy, judgment, and trust remain human responsibilities.”

She also draws on her own career story. “I started my career in claims as a file clerk. I took papers and put them in claim files, pulled claim files out of drawers, and gave them to adjusters to work. That job no longer exists at all in claims — it’s been eliminated. And yet here I am with a successful claims career.”

Her message to employees is straightforward: “I can’t promise AI won’t someday change your job. But I can tell you this: if you adopt it, you have a real shot at staying ahead of it. If you don’t, you risk getting left behind.”

High sees AI as an opportunity for employees to level up. “This, to me, is another opportunity for people to not be the victim of AI but to become part of leading AI. The more you understand how prompting works, the more valuable you become.”

The Road Ahead: Three Eras and What Comes Next

Ozuna framed the evolution of claims technology in three eras. “The first era was workflow digitization — moving from paper files into digitized claim systems, imaging systems, and feeds,” she said. “Then we moved into era two, which was data aggregation. Era three, where we are now, is decision support. AI moves us from describing a claim to doing the early cognitive work on the claim.”

The pace of change is dizzying. “Two years ago, I couldn’t have imagined where we are today,” Ozuna said. “It’s hard to conceive what’s coming in the next year or two — AI is advancing faster than any technology area I’ve worked with.”

“In an industry facing a growing experience gap, Agentic AI serves as a digital mentor—transferring institutional knowledge to less experienced staff through real-time coaching and decision support,” High said.

“This accelerates onboarding, improves decision quality, and creates a more scalable workforce model.”

The ultimate promise, both experts agreed, is better outcomes for injured workers. “We’re making higher-quality decisions earlier in the process,” Ozuna said. “We know that the earlier we get involved with a claim and engage all necessary interventions, the better the outcome.”

For claims organizations at that same inflection point, the path forward requires thoughtful leadership, strong governance, and employees’ trust earned one cupcake at a time. Get that foundation right, and it shows up exactly where it should — in faster, better outcomes for the injured workers these systems exist to serve. The technology is here — the question is how thoughtfully it will be deployed.

To learn more, visit www.medrisknet.com.

This article was produced by the R&I Brand Studio, a unit of the advertising department of Risk & Insurance, in collaboration with MedRisk. The editorial staff of Risk & Insurance had no role in its preparation.