Adopting AI the Right Way in Healthcare

By Matt Phillion

In the rush to add artificial intelligence to healthcare, are organizations taking the right steps to ensure the integrity of their data?

A recent survey found that two out of three health systems plan to launch AI pilots or projects by the end of 2025. As reliance on large language models (LLM) grows, it’s pivotal to be sure the use is accurate. Much of the power of AI analysis depends on data integrity, but the industry has already seen that patient-matching errors alone threaten the accuracy of AI outputs. In fact, 57% of healthcare leaders surveyed believe patient-matching errors will reach a crisis level in the next five to 10 years.

Healthcare needs to tackle challenges with data integrity before ramping up AI initiatives, says Avi Mukherjee, chief product and technology officer with Verato.

“At the end of the day, there’s been a hype cycle, but what has really driven results are three things: putting the patient at the center, focusing on the quality of care, and coordination of care. The solutions that focus on these have survived the test of time.”

None of this works without knowing who is who, says Mukherjee.

“Without understanding the care continuum journey, you cannot achieve the right quality of care, the right level of safety, and the right outcomes,” he says. “I’m very excited that we’re using AI, and many of our customers are moving to cloud data platforms and building partnerships, but the central piece is understanding all that constitutes the continuum around patient care.”

This must be front and center in any use case, Mukherjee says.

“Everyone is jumping in on it, whether it’s Snowflake or Google Vertex or another platform, but without knowing who’s who and identifying the accuracy of the data, it doesn’t work,” says Mukherjee.

Verato specifically looks at identity verification and identity resolution – that core tenant of healthcare of verifying you are who you say you are.

“We need to help the machine verify that you can’t get identity in healthcare wrong. You can’t have the prescription go to the wrong person or the payment go to the wrong provider,” says Mukherjee. “Right now, the analytics are good when you look at them superficially, but you’re not fixing the back-end problems.”

It’s the old analogy of the tip of the iceberg, Mukherjee says. We see the part above the water, and it looks fine, but there is a lot happening below the surface we need to be aware of but can’t necessarily see.

“With healthcare-centric use cases, you need the identity resolution to use healthcare language. It’s its own language from how interfaces are coded to how machines talk to machines,” says Mukherjee. “There’s a digital journey and a clinical journey. Each person is unique. I’m an employee, a dad, we live in a certain part of Texas—the beauty is that my care and outcomes are very personalized, but we need to bring together all that data to help these platforms understand that digital journey.”

Driving the right outcomes

Vendors need to help their customers understand the reference architecture involved, Mukherjee says.

“Customers don’t know the right way to implement this—they’re trying to drive a business outcome—so education is the most important part,” he says. “There’s a lot of noise in the market. And there are so many AI companies. You can buy a URL with .ai in it. It is imperative to understand the business need first and foremost, and then the reference architecture.”

To reach this understanding, Mukherjee notes, there are two steps: first, understand the use case for the technology, and then understand what the ROI is.

“A lot of healthcare organizations deal with very tight margins, so what I’m seeing is more small, pet projects as fewer are prepared to go all in,” says Mukherjee.

Healthcare sets a higher bar for getting it right, Mukherjee believes.

“Identity resolution is a big thing when it comes to AI and data in the cloud, and I think healthcare is ripe for this option,” says Mukherjee. “With the growing shortage of physicians, the shortage of nurses, clinician fatigue, and more people developing chronic health problems, just on an economic, supply and demand model there’s a lot of need.”

Healthcare also has a vested interest in not just adopting this technology but doing it right, he explains.

“It’s not just writing your high school essay with ChatGPT. This is a clinical prescriber using some kind of AI model to provide the right care. You have to get it right,” he says. “Nothing is more valuable than a patient’s life. That’s the ROI.”

Physician adoption leads the charge

Mukherjee looks to a physician-led model to ensure implementation of these technologies have the right focus to provide the right care at the right time.

“We’re focused on trying to make physicians’ lives easier, and even the lives of extended caregivers,” he says. “The network of caregivers around the patient is critical, and so the technology has to have the right ease of use as well as the blessing of the physician community.”

Often, physicians are the data stewards, and their role in developing trust in a system is paramount.

“We need to make sure we’re aligning with the right thought leaders. A big part of this is engaging with physician leaders and health systems,” says Mukherjee. “These are the groups who are really pushing the conversation.”

The digital revolution in healthcare has grown exponentially in the past few years, Mukherjee notes, and the adoption of these digital tools is part of that.

“Five years ago, a digital front door to your healthcare organization was a buzzword. But especially with COVID, we’ve seen care pushed outside the four walls of your organization,” he says. “There’s more awareness of how data and AI can personalize the journey.”

Physicians, however, are acutely aware of the need to make sure new technologies are tried and tested to provide the right quality of care and patient outcomes before they adopt it.

“Some tech companies might feel it’s a barrier to entry, but I think it’s the right thing to do,” says Mukherjee.

 Mukherjee sees the bigger changes on the horizon involving LLM models.

“The hype of AI innovation in that space is fast and furious, but the issue is dominant use cases are not coming up. There’s a lot of talk about using it for marketing copy but less about actual use cases in healthcare that drive value. There’s still a lot of work remaining,” he says. “Over the next few years there will be a lot of hype, the froth on the top of the coffee, but when we start seeing things that really help the patient experience, improve patient safety, better documentation, or improve triage, those are the dominant designs I think will evolve a lot. We are starting to see Gen AI and LLMs being deployed for capturing patient notes by nurses and clinicians on their phones, editing these and then automatically entering the edited notes into the EHR, which is a compelling use case that is reducing the burden imposed by documentation on clinicians and nurses, while improving the patient experience.”

Mukherjee sees folks thinking not just in terms of AI but specifically AI and identity to reduce workloads.

“It has to be absolutely right: AI has a long way to go to get to that 98%, 99%, 99.9% accuracy rate, and that might be more than a two-year journey,” he says. “The important thing is that there’s collaboration across the healthcare ecosystem. Payers, health systems, the life sciences, practitioners. All of these people are innovative, but they’re even more innovative when they collaborate, and that’s where the patient experience really changes.”

Matt Phillion is a freelance writer covering healthcare, cybersecurity, and more. He can be reached at