top of page

Welcome to the Exception Economy

7 days ago
5 min read

When AI handles the routine work, who learns to handle what's left?



We spend a great deal of time talking about how much revenue cycle work AI can take off our hands. Topics such as automatic claim coding, denials prediction or prevention, and the discussion about how many transactions can be completed without human intervention at all are frequent fodder for blogs and articles.


Those are perfectly reasonable topics to explore, particularly for hospitals under relentless financial pressure. There’s another question, though, one that shines a light on the sometimes-uncomfortable intersection between AI and humans:


What happens to the work AI leaves behind?


As AI takes over more routine revenue cycle tasks, the cases that still require human intervention are likely to represent a disproportionate share of the difficult ones. Ambiguous coding situations, complex denials, unusual payer requirements, and circumstances where the correct answer isn't necessarily obvious.


The consequences of getting those cases wrong can be significant. An unresolved coding discrepancy or complicated payer dispute isn't simply an inconvenience. It can delay reimbursement, result in lost revenue, create compliance exposure, or leave a patient facing an unexpected financial obligation.


So, whilst we may be reducing the volume of work that requires human attention, we could simultaneously be increasing its complexity and financial significance. Now, that's an interesting proposition from an operational perspective, and it also raises a rather more troubling question.


Where are we going to find people with the expertise to handle it?


An experienced revenue cycle professional isn't simply someone who has accumulated years of technical knowledge. They've also developed judgement through exposure to thousands of ordinary situations and hundreds of extraordinary ones. They recognize when something doesn't look right, even when it technically meets the rules. They know when a payer's explanation doesn't quite add up, when a coding decision warrants another look, and when an apparent one-off problem might actually be evidence of something much larger.


That kind of judgement doesn't come entirely from a textbook or a training module. Much of it develops through doing the work — and therein lies the problem.


The irony of automation


Back in 1983, cognitive psychologist Lisanne Bainbridge published a paper titled Ironies of Automation in the journal Automatica.


Bainbridge was studying industrial automation, and this was long before anyone in the revenue cycle space was discussing AI. One of her central observations is remarkably relevant today: automation can leave human operators responsible for handling the difficult situations whilst reducing the routine practice through which they develop and maintain the skills to do so. In other words, the work we automate is the same work that helps people learn how to perform the work we can't automate. This isn't merely an interesting theoretical contradiction.


In October 2026, Becker's Hospital Review published an article examining how health systems are addressing the risk of clinical deskilling associated with AI.


At Children's Hospital of Philadelphia, educators are developing approaches that require trainees to demonstrate proficiency in certain clinical tasks before gaining access to AI tools that perform them. Johns Hopkins Medicine is also exploring ways to ensure clinicians develop independent assessments before consulting AI-generated recommendations.

The concern isn't limited to experienced professionals losing skills they no longer use. There's another possibility: that new professionals may never develop those skills in the first place. The term for this is never-skilling, and I can't help but wonder whether revenue cycle leaders should be having a similar conversation.


Are we automating the apprenticeship?


By way of example, let’s look at medical coding. Traditionally, a new coder develops expertise through a combination of education, supervised practice, and experience. Straightforward charts provide an opportunity to apply coding principles, encounter variations, make decisions, receive feedback, and gradually develop the judgement necessary to handle more complicated cases.


But if AI increasingly handles those straightforward charts, what happens to that learning process?


We can certainly train coders to review AI-generated coding decisions. However, evaluating an automated decision is not necessarily the same skill as arriving at the decision independently. How does someone recognize that AI has made a subtle error if they haven't developed the foundational understanding necessary to identify it? How do they acquire that understanding when the work that traditionally helped them develop it is increasingly performed by technology?


As AI takes on more routine coding tasks, the responsibilities of experienced coders are evolving. Their expertise can be applied to more complicated cases, quality assurance, auditing, and evaluating the performance of automated systems. That’s a valuable development. It gives experienced professionals opportunities to apply their knowledge in ways that might have been difficult when so much of their time was occupied by routine transactions.


There’s a question we shouldn’t overlook: how do we develop the next generation of professionals with the skills and judgement to take on those responsibilities? After all, experienced coders don’t materialize out of thin air. Someone has to develop that expertise, and much of it has traditionally come from doing the very work we’re now automating.

Of course, this doesn't mean we should preserve repetitive manual work simply for the sake of training people. Nor does it mean that traditional approaches to professional development are the only ones available.


Structured learning, simulation, supervised review, and deliberately designed opportunities for independent decision-making may all have a role to play, but those opportunities will have to be created intentionally. We can no longer necessarily rely on the ordinary progression of work to provide them. Developing that expertise takes time, resources, and experienced people willing and able to teach it. Those requirements deserve a place in the conversation about automation's financial benefits.


For revenue cycle executives, this creates an interesting dilemma.


The financial case for automating routine transactions is compelling. Hospitals need to reduce administrative costs, improve accuracy, and accelerate reimbursement. Nobody is suggesting we abandon those opportunities. It’s important to recognize, though, that reducing the number of transactions requiring human intervention doesn't necessarily produce an equivalent reduction in the expertise an organization needs.


In fact, the opposite may be true.


A smaller volume of increasingly complicated work may demand more experienced personnel, greater discretion, and stronger oversight. And because those remaining cases can carry substantial financial consequences, ensuring that the right expertise is available becomes an important part of protecting revenue.


There's another complication, of course. AI isn't standing still.


Some of today's exceptions will become tomorrow's automated transactions. As AI capabilities evolve, so will the responsibilities of the professionals working alongside those systems. We can’t confidently predict exactly how those roles will change, which makes the question of how we develop and preserve expertise more important, not less.


This isn't an argument against automation. It's an argument for considering what automation changes beyond the number of people needed to complete a transaction.

Hospital leaders are understandably interested in how much work AI can eliminate. But they should be equally interested in what kind of work remains, what expertise it requires, and how that expertise will be developed. More than four decades after Bainbridge described the ironies of automation, we're confronting a familiar contradiction with technology that's advancing into areas of work once considered uniquely human.


If we're automating the work that once helped people become experts, how are we planning to develop the experts we'll need when automation reaches its limits?

Comments


About Salud Revenue Partners

Salud Revenue Partners is a national model for the delivery of revenue cycle services. A technology-enabled company with experienced leadership and a high-performance culture, we help our clients achieve their financial goals.

Align SOC 2 badge
GPTW.png
  • LinkedIn
  • Facebook
  • Instagram
© Copyright Salud Revenue Partners
bottom of page