AI Is in ‘Hype Phase,’ Says Abarca CIO. He Sounds Notes of Caution | PBMI 2025

Serge Perras, MBA, chief information officer at Abarca Health, spoke Sept. 5 at the 2025 PBMI Annual National Conference in Orlando, Florida.

If Serge Perras, MBA, doesn’t exactly pour cold water on artificial intelligence (AI), he certainly advises letting cooler, calculating heads prevail when applying it to prior authorization and perhaps other processes in pharmacy.

“I’m not trying to be a naysayer here whatsoever,” Perras told the audience at the 2025 Pharmacy Benefit Management Institute (PBMI) Annual National Conference in Orlando, Florida, last week. “I am actually a big, firm believer that we are still in the growing pains in the emerging space of AI. This will get solved. It’s just not there right now, so be cautious as you go down that road.”

Perras, the chief information officer for Abarca Health, a pharmacy benefit manager based in Puerto Rico, spoke about some of the success Abarca has had in applying AI to medication therapy management and said the learning capabilities of AI would be useful in detecting fraud, waste and abuse that is new or outside of the norm. The AI behind chatbots, in addition to supporting interaction, “provides an ability to dynamically gather information and surface capabilities…you may not have even imagined.”

But Perras also described AI as being in a “hype phase” and cautioned against repeating the overexuberance about technology that fueled the dot-com craze of the late ’90s before ending with many companies having little or no value.

“Please look at it [AI] from a business perspective and a value proposition, not at what the technology is for the sake of technology, but at what the business problem is that you’re trying to solve,” he said.

Prras described himself as a “newbie” to the PBM sector with only four years at Abarca. He delivered a serious message about avoiding the pitfalls of AI adoptions with a light, occasionally joking touch. “My history is primarily as a software engineer, and you, unfortunately, need us technology folks to hopefully support some of the things that you do,” he said. When no one answered when Perras asked the audience whether AI had gotten a good foothold in their company’s processes, he laughed and said, “Either you’re very hard on yourselves or you’re being realistic.”

Perras pointed to medication therapy management (MTM) as one area that lends itself to AI. At Abarca, an AI bot deployed for comprehensive medication review, an important part of MTM, shaved 6.75 minutes off the reviews and also was 100% accurate. Perras said 6.75 minutes may not be impressive in isolation, but it translates into cutting the time it takes for the clinical staff to do comprehensive medication reviews from between two and two and a half hours down to just one. He said it was important to view the tasks that AI does in the context of processes to make sure they actually help with efficiency and speeding. If AI reduces the time it takes to perform from 10 seconds to five, it is not very valuable if that task is only performed five times a day, he said, because you are not going to be able to redeploy those saved seconds. He also noted that there is a cost to using AI that has to be factored into return-on-investment calculations. “It’s not just pointing AI at something. It’s about making intelligent decisions about where it’s going to provide value,” he said.

Perras also qualified the endorsement of AI for ferreting out fraud, waste and abuse. He said that between 80% and 85% of it can be identified using strategies that do not depend on AI based on certain common patterns. AI is helpful, though, in identifying the “corner cases and the things that are dynamic,” he said.

Data security is one of the challenges to incorporating AI into healthcare, said Perras, noting that OpenAI, the maker of ChatGPT, will not sign the business associate agreements that create obligations to protect personal health information under the Health Insurance Portability and Accountability Act (HIPAA). He also mentioned regulatory compliance and shortcomings in standardization and interoperability as possible stumbling blocks. The nondeterministic nature of generative AI is another obstacle that Perras discussed.

“Nondeterministic means I don't really understand how a decision was made,” he said. “That's another problem with AI today. There's no real clarity, at least not yet in most of the models that are available, to understand or to have transparency around what actually made the decision and how the decision was made. What were the data points used, etc.? What was the path through the maze? [E]very decision is ultimately a path through the maze. You got to know what the maze is, and then you got to know what the path is.”

Pro tips for implementation

Perras did not leave the audience with just a critique. He provided some advice and tips on how to implement AI. In a version of garbage-in, garbage-out, he said AI can’t produce results if it is just layered on top of data that is flawed and poorly structured. He referenced the “tribal knowledge” of unwritten and undocumented expertise and practices that accumulates in some organizations. “If your organization is primarily tribal knowledge-based, you will struggle to get AI to work because can AI read your minds?” said Perras.

Perras said pilot testing, while not rocket science, was important, especially with healthcare bristling with regulation and privacy issues. He advised “cross-functional collaboration” that can offset a tendency for tech people to see AI as a solution for everything. He said training was important to get people in an organization to actually use AI and incorporate it into their work. “If you are not a technology geek to begin with…people will get discouraged very quickly because if you don’t get results, you’re going to give up. ‘See, I can get my job done without this thing, because I tried two or three things and didn’t get the answer I expected, so I moved on.’ You need to get people to understand, and you also need to understand what they need and what the business needs.”