Overcoming the Barriers to AI Adoption in Revenue Cycle Management

The challenges of AI for operations, RCM

AI may have the potential to streamline healthcare operations and revenue cycle management, but the technology is up against some significant and longstanding challenges of technology adoption in healthcare. First and foremost, healthcare providers are skeptical about technology adoption.

“I can’t tell you how many times there has been a new way of making the provider’s life better, their staff’s lives better,” says Kem Tolliver, FACMPE, CPC, CMOM, president of Medical Revenue Cycle Specialists, who has received countless pitches herself over her nearly two-decades-long career in healthcare. “What ends up happening is physicians become weary and they see vendors as organizations that are interested in their own financial gain.”

Skepticism clouds decision-making, especially with technology adoption in healthcare, Tolliver continued. However, provider skepticism isn’t unwarranted. After all, healthcare organizations continue to lose millions of dollars a year due to denied claims, and that’s with at least some adoption of technology aimed at improving denial management. Some revenue cycle management solutions have also failed to meet provider expectations at this point.

“Part of the skepticism in healthcare is whether the vendor has done their due diligence to ensure a favorable experience for the healthcare organization,” Tolliver stated. “We know that the vendor has done their due diligence for a favorable relationship on their end; the contract is going to be favorable towards them with the fees. But what due diligence has been done on behalf of the provider?”

AI vendors need to ensure their products can deliver results. Healthcare already has issues with data accuracy and completeness because of the nature of the industry. As some technology experts say: Garbage in, garbage out.

A recent study also found that Generative Pre-trained Transformer (GPT) 4, a multimodal large language model created by OpenAI, was 90 percent accurate at answering United States Medical Licensing Examination (USMLE) questions. GPT 4 significantly outperformed its predecessor ChatGPT, which is based on GPT 3.5. But that 10 percent could be a problem for healthcare in which percentages may represent patient lives.

“It’s important to validate any results because, at the end of the day, when we decide to use AI within healthcare operations, it’s the provider’s medical degree and reputation that is on the line. They have to do what they think is in their best interest, and most of the time, it’s not doing anything,” Tolliver said.

Top barriers to AI adoption

In addition to larger challenges, there are also barriers to AI adoption that impact healthcare providers on a more local level. The top of mind for Tolliver was integration with existing IT systems.

Electronic medical records (EMR) and practice management systems are the touchstones of healthcare delivery. These systems support the day-to-day management of healthcare services, including clinical documentation, coding, claim generation, and reimbursement. AI solutions will likely need to integrate with at least these two systems (or one, as many organizations utilize the practice management capabilities of their EMRs).

“If we are going to integrate AI into our EMR, we need our vendors who have historically not been keen to collaborate with other vendors to do that,” Tolliver explained. “And it doesn’t just have to integrate with our EMR or practice management software. We want to make sure that there are open APIs available, so we can integrate AI into whatever technology we are using.”

APIs, or application program interfaces, enable two systems to speak to one another using the language of data. These are used every day by consumers for ridesharing, mobile payment, and even changing the temperature from a smartphone. In healthcare, APIs have become key to sharing patient information between IT systems despite differences in formats.

Another key barrier to AI adoption in revenue cycle management and healthcare, at large, is the fear of job loss.

Some data has indicated that job losses are already happening because of AI, with experts predicting more jobs at risk later. Whether AI will significantly impact employment remains to be seen, but AI will almost certainly impact how humans work. OpenAI, the creator of ChatGPT, estimated earlier this year that the majority of the US labor force could have at least 10 percent of their work-related tasks affected by GPT’s introduction. Another 19 percent of workers could see at least half of their tasks impacted.

“We have providers who are very loyal to their staff and they do not want their staff to be replaced by technology, so they lean toward not adopting it,” Tolliver said.

Additionally, some providers may stick with manual processes completed by humans because that is the way they have always done it and they believe those processes have achieved the results they want.

Tolliver acknowledged workforce concerns. “They are real,” Tolliver stated, “but the goal of adoption is about using everyone to their highest skill level. We can remove some of these repetitive processes, so a director of revenue cycle management or a coding manager isn’t posing EOBs, which is something I see happen every day.”

Key considerations pre- and post-implementation

Barriers to AI adoption in revenue cycle management and operations are not insurmountable, according to Tolliver. Healthcare providers need to ensure they are partnering with the right vendor to implement the right solution for their organization’s operational needs.

“Consider who is creating your AI solution and understand their company goals,” Tolliver said.

Particularly in healthcare, a vendor’s goals should align with the end user’s mission statement or end goals. Healthcare providers should also be aware of private equity’s latest interest in revenue cycle management companies.

“It is a good idea to consider what ties them to your healthcare organization’s mission statement and end goals to determine whether or not it’s a good fit,” Tolliver said of AI vendors in general.

To gain a better understanding of alignment, healthcare providers should define key performance indicators (KPIs) they want to manage as they look to an AI solution to alleviate a specific problem area within operations or revenue cycle management.

“We want to make sure we are defining those KPIs that we want to measure in advance,” Tolliver explained. “Post-implementation, we want to go back and measure those predefined KPIs. How did we fare? How successful was our AI technology in helping us meet those goals?”

Prior to implementation of AI solutions, healthcare providers should also consider whether vendors perform peer reviews and what they use to measure their solution’s accuracy.

“Do they use physicians as peer reviewers for medical documentation functionality for their AI solution? Are they using revenue cycle management subject matter experts to build and test their RCM AI platforms? Are they using certified professional coders as peer reviewers for the accuracy of their coding elements? Having that peer review process is important,” Tolliver stressed.

After implementation, Tolliver also advised providers to look back on performance and reevaluate their KPIs. Have the organization’s goals changed since implementing the solution? If so, has the AI adapted or evolved to align with those goals?

“I would also consider the fact that now some of these decisions and processes have been eliminated from our practice,” Tolliver added. “We may need to modify our workflows and even job descriptions after we’ve been armed with all of this knowledge and information post-implementation.” 

Overcoming the challenges of AI adoption for revenue cycle management and healthcare operations is key to realizing the benefits of automation. These are two areas in healthcare where automation can significantly transform processes to achieve efficiency.

“There are still so many manual processes that could be eliminated. Hopefully, we are going to start adopting AI before we are left in the dust,” Tolliver said.