Why AI Is Falling Short on Patient Engagement at Hospitals

New research suggests that health systems are struggling to effectively use AI to improve patient engagement.

The study, released earlier this month, found that while investments in tools like ambient scribes are booming, AI applications for patient engagement are lagging. For the study, patient engagement startup Lirio commissioned healthcare consultancy Sage Growth Partners to interview more than 75 health system executives across the U.S.

Only 5% of these executives said they were satisfied with the tools they have to support common patient engagement challenges such as medication adherence and missed appointments — problems that lead not only to poor health outcomes but also billions of dollars spent on avoidable healthcare costs each year.

To help hospitals address these gaps, companies selling patient engagement tools have to move toward a “N-of-1 personalization” model, said Amy Bucher, Lirio’s chief behavioral officer.

“In healthcare, standard approaches to personalization aren’t very personal,” she declared.

Often, personalization begins and ends with certain form fields like first name or age range. These approaches essentially just segment people by demographics instead of accounting for their individual motivations and behaviors, Bucher noted.

For example, a provider may send the same generic email reminder about mammography services to all women ages 40 and older. But not every woman in this broad age group needs the same type of message — and Bucher explained that an N-of-1 approach takes this a step further by generating tailored messaging that factors in each patient’s unique needs, behaviors and barriers.

“If a woman hasn’t had a mammogram in a few years, N-of-1 personalization considers why this may be the case. Is it hard to prioritize the appointment against work? Does she need childcare? Hates the anxiety of cancer screenings? Whatever the case may be, personalization that doesn’t address it won’t be as effective,” Bucher remarked.

Recent advancements in AI have created the ability to scale N-of-1 personalization, she added.

She noted that humans are able to do this well in a 1:1 format — we can take in complex information from what people tell us, as well as non-verbal signals and context clues, and adjust our approach in the moment. But humans aren’t scalable across a large patient population, nor is it affordable to use live support for every use case, Bucher said.

“Technology has been able to deal with more complex and larger datasets than humans for a long time, but it’s only recently, with the explosion of agentic AI and the use of techniques like reinforcement learning, that it can also produce meaningful N-of-1 output,” she stated.

She also pointed out that better personalization can unveil new levels of efficiency and connection.

Take diabetes for instance. The disease affects 1 in 10 Americans — but despite its prevalence, people with diabetes are often disengaged with their healthcare providers, Bucher said.

“Standard ways of trying to engage people to schedule appointments and take medications clearly don’t work for everyone. Personalizing those outreaches can help spark interest and get people thinking differently about the value proposition of taking action, and doing it with digital outreach can create operational efficiency,” she remarked.

The diabetes use case underscores why patient engagement may be one of healthcare’s most promising — and most underutilized — applications of AI. When personalization moves beyond demographics to address individual barriers to action, it can not only drive clinical improvements but also help health systems engage patients at scale, Bucher declared.

Photo: Paul Bradbury, Getty Images