The Work No One Should Still Be Doing by Hand

We automated the wrong job

We created this platform to improve billing and we pretended that it was geared to help their workflow.

We've done nothing but layer on administrative burden and complexity to people who took the Hippocratic oath. We actually replaced the dictaphone and the transcriptionist with the doctor sitting in front of a keyboard.

In our infinite wisdom, we thought we should automate the capture of the information, not thinking about the fact that we were replacing the wrong job. We eliminated the transcription and we made the doctor do all the typing, too.

People don't go into medicine to spend their day charting. They do it to help people.

The work that is actually rules-based

Coding is 100% rules-based, there's no subjectivity whatsoever. And the companies that are using that are actually getting a much higher accuracy than organizations get without that.

That's where ai is best at automating things that are repetitive in rule space. The provider sector understands that and they're always playing catch up.

There's a lot of repetitious activities that are nearly 100% rules-based. So things like the charge capture process, there's a lot of interventions involved with that.

The autonomous coding capability that's having a lot of good success, I think it's going to have a much higher rate of adoption because that capability is being used and it's accelerating it.

There's autonomous coding, this is starting to be really effective as well, and these things can shorten the cycle time it takes to get that information out, so it can enable much more efficient billing and processing. But more importantly, we're giving back some time to the ones that are delivering the care.

We have a ton of regulatory requirements for quality reporting, and those metrics are based on a numerator and a denominator, with a limited combination of terminology you're looking for to define either of them. One of the most effective uses I've seen is pointing that technology at the discrete notes and the unstructured notes and getting a much more accurate score for those metrics, and taking the subjectivity out of the equation.

The things that I think are moving the quickest are really back office operations, looking at things we can streamline and shorten the time it takes to consume the information you need.

If we're focusing on the things that are really going to help drive efficiency, we drive some really obnoxiously boring work out of the queue, because it's just so routine and mechanical and rules-based. It's a time suck that no one wants to have, no matter what your job is. If we can take a lot of that grind out and let it be meaningful. Just think if analysts could actually do analysis and not data wrangling for 60% of their time.

We automated small balance write-offs because the cost of billing was far more than the cost of just writing it off. Today, that would be RPA.

In the revenue cycle, unless it's really, really broken, usually that's a place where you've got some opportunity, whether it's using RPA or just building some large language models that are really helping you to drive your scheduling in the ambulatory space, things like that.

I asked a head of a physicians organization how he was managing this. Well, I personally actually go in and I adjust my own schedule template for that. I said, yeah, but you're the head of the physicians organization, have you mandated that? He goes, no. I said, I know that, because no one's using it. And the worst part is, in the EHR they actually have the capability wired in there, where you can go into the different specialty master files and make adjustments to the protocol for different conditions. It's native to the platform, and people aren't leveraging it.

The denials machine was built on purpose

My experience over 30 years tells me that there are systems that have things built into them on purpose that cause there to be a cycle that requires at least, well, more than one submission of a claim. Before it processes.

Everyone knows denials are a big thing. And it seems to me like that's pretty automated, and it's almost a knee-jerk response when something gets submitted.

It's the same rules engines that are being used by the payers that are also being used by the providers. So it's doing a ton of these types of comparisons.

They're the secondary, they don't put things at the line item level. They put them in a header. When that payment comes through, someone actually has to sit down and work on an edit list and manually key that payment in.

I had a full production team when I was working years ago. There were eight people just dealing with the electronic payments and the charges going out the door in claims. They were supported by a team of about eight people who actually maintained the claim forms and the remits themselves. Claim forms didn't change much, but the remits changed regularly for a variety of reasons. Some of it is regulatory, but some of it is also preference of the payer in terms of how they want to operate their business. We have ANSI standards, there's a 4010 and a 5010. 5010 replaced the 4010 probably about 10 years or so ago. What that actually changed was language. It didn't actually change or improve workflow at all.

And then one day we get this thing about the Accountable Care Act. I got this long list of things that I have to put into a project plan and get my team to move on, like yesterday. I had to scramble, and it actually created a demand for me to hire two more people to manage the claim form complexity that they created, because they improved things so much.

You had a black hole. Your accounts receivable system says you generated a claim. It may have gone into a file, but it actually never left the premises. So what I did is I went through and identified the ones that had not processed. I re-queued them after I set the mapping up and I sent them all out, and I said, your 6 million dollars, it's going to be within 10 to 15% of that value on a weekly basis from that one, because you're actually going to get your claims out the door. They just weren't going anywhere.

Years ago, at a system I worked in, there was an issue just before I got there where they were seeing much bigger remittance files than they should have, and they didn't know why. What they found out is they were not billing anesthesia properly, and so it was overcharging. They were obviously getting the same rate, but just more reimbursement. That caused them to have to refund about 6 million dollars in overpayments and a full audit of an 18-month period of time, because the insurance companies demanded it.

What you do not hand to a machine

Revenue cycle is an easy one because most of them have a small number of those. But even if they had a dozen of them, it doesn't really matter because transactions are transactions. It's the clinical data sets that are really the complication.

The top-level motivation for me is really about making sure that what we design and we develop is human-centric, meaning that it reinforces. Ability of a human being to be the creative force, to be the one who can bring judgment.

There's a drift that starts to happen. And when the drift happens, it may be subtle enough that no one sees it for a while. And by it is seen, then you've got a backlog of things to fix.

Don't design for clinicians, design with them.

You need to work with providers to understand where AI can help.

If we want AI to truly benefit healthcare, we need to design it as a partner, not a replacement.

If you're a coder and you've been used to how I dictate and how my documentation is done, it's not going to take you all that long to go through it and you're probably going to get a pretty accurate score. Now introduce Dr. Joe next door, he documents a little bit differently, there's some nuance to it, and if you're used to reading my documentation you're likely to have a bit of an adjustment period. We're talking about a large number of physicians a single coder might be supporting, and there's human error always possible in that. With the level of attrition we've seen, especially with coders, the time to get someone new to that level of accuracy is not short.

It's not that people are using AI to replace a lot of what people are doing. They're really focused on trying to augment, or help, or provide a set of belt and suspenders for what they do. An additional set of eyes, even if they're artificial eyes.

The trap: do not just make bad work faster

If it's adding layers and adding complexity, adding cognitive load, that's not a solution, that's creating more problem.

We have a burnout problem that we've created for both clinicians and nursing staff. In both scenarios, we just keep layering on more things.

It shouldn't be a burden to clinical teams to have to look at this. One of the things we've got to figure out is making it really, really easy. There have got to be some really simple tools that we can put in the hands of our clinicians and the support staff. The chief data officers, the IT teams, the project managers, the legal and risk and compliance teams, those are the things we need to worry about, and we need to take care of them.

Close

I might come into an organization and realize that their revenue cycle is tight. We could automate some things there. I look at rules-based transactional work first. Taking it off people frees them to do more complex items where human judgement is required.

The measure of success is how much time it gives back for more complex things that are often higher impact and value.

Continue reading from Hutchins Data Strategy

The Prior Authorization Arms Race → https://hutchinsdatastrategy.com/the-ai-health-pulse/the-prior-authorization-arms-race

From Burnout to Breakthrough: 3 Ways AI Can Unburden Healthcare Teams Today → https://hutchinsdatastrategy.com/the-ai-health-pulse/from-burnout-to-breakthrough-3-ways-ai-can-unburden-healthcare-teams-today

On the Signal Room podcast

AI Healthcare Leadership and Strategy: From Hype to Real Value | Parth Gargish → https://signalroompodcast.com/episodes/ai-hype-to-real-value

Healthcare Data Readiness and AI Adoption: Why 85% of Organizations Aren't Prepared | Ratnadeep Bhattacharjee → https://signalroompodcast.com/episodes/data-readiness-ai-adoption

From AI Strategy to Execution: Ethical Leadership, Trust and the Operational Reality of Healthcare AI | Brian Sutherland → https://signalroompodcast.com/episodes/rethinking-healthcare-data-strategy

More from Chris Hutchins

Subscribe — one email gets you The AI Health Pulse, the Signal Room podcast, and a free chapter of Beneath the Signal → https://hutchinsdatastrategy.com/free-chapter

Book Chris to Speak → https://www.chrisjhutchins.com/

Learn about our work to improve mental health around the world → https://continuamindhealth.com/

Do you have a podcast? We offer a Free growth assessment from PodcastPull.com → https://podcastpull.com/

Read on BeeHiiv → https://aihealthpulse.beehiiv.com/

Read on Substack → https://aihealthpulse.substack.com/

Tags: AI Health Pulse newsletter · autonomous medical coding · revenue cycle AI · denials management software · healthcare automation · administrative burden in healthcare · time returned to care