Generative AI in healthcare lacks strategy: Bain & Company report

It’s early days for health system leaders interested in implementing generative artificial intelligence, according to a report published Monday from consultancy Bain & Company.  

For the report, Bain surveyed 94 health system leaders, the overwhelming majority of which haven’t fully strategized on how to use the technology. Despite this, there is still considerable excitement over how generative AI applications like ChatGPT can be used to reduce administrative headaches. 

Here are five takeaways from the report:

1. Only 6% of health systems have a generative AI strategy.

Around 75% health system executives surveyed say generative AI has reached a turning point in its ability to reshape the industry. But few have identified a clear pathway to implement the technology. Only 6% of health systems have a comprehensive generative AI strategy. 

In the absence of robust regulation from policymakers, larger health systems at the leading edge of deploying generative AI are creating their own safety and efficacy standards. But most health systems remain uncertain about how the technology works, whether to build their own applications or buy solutions from a vendor and how to deploy it safely, said Eric Berger, one of the report’s authors and a partner in Bain & Company’s healthcare and private equity practices. 

“At a minimum, one needs to set up the policy controls and governance over this new technology,” Berger said. “Everyone should have clear guidance to their employees, and to their vendors to for that matter, about the appropriate use of this technology.”

2. Resource constraints, lack of expertise lead to hesitation. 

Scarce resources will also likely stifle immediate adoption for some health systems.

Nearly half of executives reported resource constraints and a lack of technical expertise as the biggest barriers to implementing generative AI. Hiring and retaining the technology talent needed to oversee an expansive, self-regulated AI division has proven challenging. 

“Moving from intent to action takes really thoughtful work and resources, [including] upfront investment, of which there’s a limited supply these days across the industry,” Berger said. 

While some leading systems are employing professionals dedicated to AI, others have given up entirely in searching for talent. 

3. But the buzz is still high. 

While AI’s nascency might be slowing the pace of adoption, it is not necessarily tamping down interest, the authors wrote in the report. 

While health systems are historically slower to adopt technology, Berger said there’s reason to believe generative AI is different. He said it’s a foundational technology that can be deployed across multiple use cases. For health systems, he said the excitement comes from the fact it can potentially remove administrative burden and create less work for employees.

“There is such demand for new tools that have the opportunity to improve and address some of the really fundamental pain points and existential threats to health systems as a whole,” Berger said. “Microsoft Excel didn’t put the accounting industry out of business. It enabled them to do new things, different things [and] faster things.”

4. Do your homework first. 

Berger said organizations should focus on established and accepted initial use cases. He specifically mentioned the ability of generative AI to compose clinical notes after a patient visit. A number of companies, including Nuance, Abridge and Augmedix, offer solutions which input draft clinical notes directly into electronic health records for clinicians to review following a patient visit.  

Berger recommended organizations do their homework by developing early use cases on how AI can be implemented moving forward.

“What is really important and impactful is building up an internal intuition around this technology, where it can be used, where it can have impact and where it should not be used,” Berger said.

This means trying out generative AI in a controlled setting, Berger said. 

5. Systems split on shift vs long-term adoption areas. 

The survey showed providers’ highest priorities could change in the coming years. 

In the next 12 months, leaders identified patient billing, analysis of pateint data and workflow optimization and automation as the highest priority areas for generative AI adoption. In two to five years, leaders were a bit more bullish on generative AI for predictive analytics, clinical decision support and diagnostic and treatment recommendations.

“I think the administrative side of the house is getting some of the early looks but that’s not to say that some of the more clinically oriented decision making tools and support are not on the radar,” Berger said. 

Gabriel Perna contributed reporting.