Robust analytics driven by artificial intelligence (AI) can deliver actionable information that improves collections.

The importance of an effective and efficient accounts receivable (A/R) strategy has never been greater.

With expenses at historic highs, U.S. hospitals and health systems have yet to recover from the damage of the COVID-19 pandemic. A Kaufman Hall analysis of about 900 hospitals in July showed that operating margins remain significantly below pre-pandemic levels.[1] And, although operating margins rose between June and July, hospitals were still recording cumulatively negative margins for the year.

That makes payer behavior more troubling than ever. Although most state and federal regulations mandate remittance within 30 to 40 days, many payers are now taking more than 90 days to pay. Claims denials are increasing — and staffing shortages make it difficult for hospitals and health systems to pursue reimbursement that is rightfully owed.

Health systems seeking to improve their A/R function are turning to machine learning and AI to streamline patient interactions in ways that benefit the organizations and their patients.

“Patient satisfaction is definitely impacted by revenue cycle management,” said Charles Rackley, chief strategy officer for Infinx. “We know that if a patient has had a bad experience with the revenue cycle management process, they only pay about 30% of the bill, but if they’ve had a positive experience, they pay over 70%.”

In this HFMA executive roundtable, health system executives discuss the benefits and challenges of using automation to increase the efficiency and effectiveness of their A/R processes.

Where do you see the biggest opportunity to uncover missed revenue for your organization?

Christopher Ballesteros: Our organization has been hit with a lot of front-end denials that we have to sort through, and because of staff shortages, we just don’t have enough manpower to get to every single thing.

Susan Searcy: Payers have increased request-for-information (ROI) denials. And they are short on staff, just like we all are, so where they used to turn an ROI denial around in 20 days, now it might be 50 or 60 days – and then they may deny it for medical necessity.

How are you using automation to improve your revenue-cycle functions?

Ballesteros: We are focused on our consumer platform, trying to see how we can use AI to help streamline scheduling and authorizations and things like that before a patient receives services. We’re a small community hospital, but we serve a pretty large area so when patients try to seek services, they may not be able to just because of scheduling issues. And so we’re trying to figure out the best way to use tech or AI to plug-and-play in different schedule templates.

Searcy: We are trying to use automation to gain overall operational efficiencies and reduce our cost to collect. And, like everybody, we want to give our folks meaningful work to do. Going into a payer site to check on claim status takes manual time and, if we can automate that, we don’t need to put those pending-to-pay claims in a work queue, and we’ll never have to do a physical touch. Then we can use our people for the things that need more thoughtful interactions or something that we couldn’t script.

Sam King: It’s very hard to quantify, but AI can do a lot to improve patient experience. We typically don’t equate patient experience with revenue cycle quantitatively, but we have so many interactions with patients. There’s a great opportunity to use AI to communicate first with patients and boost loyalty and retention.

Shawn Stack: When I speak with chief medical officers, I surprise them by saying, “We’re the first people who talk to your patient and we’re the last person they see. So if they have a bad experience with our team, they might not come back for follow-up care.”

I tell them that if a patient expresses concern about the financial impact of their care, we don’t expect them to have that financial conversation, but they need to connect the patient to a financial counselor. That message is reaching a lot of physicians now because of the No Surprises Act. These good faith estimates are scaring our patients, and they’re not getting care because they are afraid of a huge bill.

What challenges have you experienced with automation initiatives?

Searcy: Some of our payers have been resistant to automation for checking on claim status and other routine tasks. We turned on a bot, and we were chugging it along until the payer says, “Oh wait, that’s way too much activity” and deactivated the user ID to shut it down.

Navaneeth Nair: I can understand the payers’ concerns because their portals were built for human interactions, not robotic interactions. So if you overload it with similar users, you may actually impact the performance of the portal for other human users. There are technology approaches to solving that problem.

How do you get leadership buy-in for revenue cycle technology implementation projects? What hurdles have you encountered?

Ballesteros: We just kicked off our digital platform, which allows some online scheduling and prepayment estimates. It also allows patients to set up automated bill pay for themselves. We created a patient access task force comprised of executive leadership, physician leadership, departmental leadership and, of course, the revenue cycle leadership. By doing that, we got everybody’s buy-in on how they perceive the patient experience to be on the front end, starting with when a physician writes an order, and we were able to map out certain processes.

And we had focus groups with members of our patient advisory council.

MRI scans were our first focal point because we have a huge radiology department. Our traditional process has involved a lot of steps, so we wanted to automate it as much as we can. Because of that patient access task force, we were able to get leadership buy-in.

Searcy: Introducing the bots is our most recent project, and we did an ROI analysis for the investment. We projected that we would be able to reduce a certain number of touches and that would reduce a specific number of FTEs. And for the FTEs that remained, we would expect a higher-yield reduction in A/R. So that is the math we used to get the buy-in from our finance leadership and IT.

Rackley: Another thing to consider about the ROI calculation for AI is that, once the bot is working and you have tested it, the bot does a better job than a person. The bot never gets tired, it never miskeys something, and it usually has a better accuracy rate. I have found it is typically 0.025% better. That doesn’t sound like a lot, but if you’re doing 20,000 internet transactions a week and you’re submitting that many denial appeals, that adds up very quickly.

How have your staff members accepted AI? And what success factors can you share?

Stack: One organization told me that, when they were exploring AI options in revenue cycle and finance, they created a dashboard that allowed their staff to post their ideas about how AI could help out. While most of the ideas were not feasible, about 20% of the staff-generated ideas were adopted and staff buy-in was seamless because the staff had offered up the ideas. Also, that process prompted the management team to look at opportunities they probably would not have considered on their own.

Nair: I think showing how automation can improve metrics is useful. If staff members can see how many more claims they are able to work as a team and how this benefits them in terms of yield improvement, they can see automation as a partnership rather than as a competition with the machine.

Rackley: One thing that has worked very well is to start with an attended bot versus unattended bot. An attended bot is one that has to be kicked off every time by a person while an unattended bot works a queue without anyone touching it at all. For example, if you need to copy information from Epic and take it to the Availity site, you can have an attended bot grab all the fields and paste it for you. It saves only 40 seconds, which sounds like not a very big deal, but 40 seconds times 70 a day adds up very quickly. Those attended bots are a good way to give staff some experience with automation and get them thinking, “OK, what else can I do with the key stroke accumulation?”

Another thing to consider is that we don’t actually have to terminate staff members to realize value from AI. The first thing people think of when they hear the term “AI” is that it’s going to take away their job. Another way to frame it is that this is going to give those employees the opportunity to do something better for the organization. So, in our experience, if you have that kind of cultural awareness, you have so much more success getting buy-in.

Based on your experience with AI, what advice would you share with revenue-cycle leaders who may be wary of getting started?

King: Start small. You need a grand vision about where you want to go, but start with a small piece so you can have a tangible win. Then you can build on that and get more buy-in. And when you are ready to do a grand project, everybody will say, “go for it,” because they see the success, and they see how they can be involved and how they can learn new skills.

Rackley:
I’m sure everyone can think of some system that tried to automate the entire revenue cycle or some entire clinical department, and they got one or two failures and then no one else wants to trust them to do anything.

I think the biggest mistake people make with AI is trying to do 20 different processes at once. The truth is: It’s very time consuming to do; even [to do] one because there’s so many variations of every process you have. Every edge case — a scenario in which the application does not perform as required or as expected — happens only once every thousand cases, but there can be dozens of these edge cases so it takes a lot of time to figure them all out.

If you don’t do enough testing to find all these, you turn it on and get 60% efficacy instead of 95%, and no one wants to do it anymore.

Nair: My view is that the best way to start is to define measures of success and outcomes and get buy-in on those outcomes. At the end of the day, CFOs and CEOs don’t care about the underlying technology as much as the result that it produces. So they are buying into the result, not buying into the technology.

In conclusion, providers must develop a robust A/R strategy to ensure revenue integrity; and machine learning-based analytics and AI can be an essential component of that strategy.

To read the original article, please click here.