By Andrew Crook, senior advisor and John Hillery, head of new product development at Tally, Inc.
It’s become a cliché to talk about AI in breathless terms: revolution, transformation, labor market disruption, once-in-a-generation change.
Brought into the revenue cycle, that language can sound like a warning more than a promise: finance leaders are right not to dive in blindly. There are compliance and information security risks, concerns that audit trails will become a “black box”, and open questions about quality or how severe the impact of “hallucinations” could be on the collection processes.
Some leaders assume that watching others go first will help refine governance and processes. But wait-and-see has its own consequences—UnitedHealth Group on track to invest $1.5 billion in AI.
We’ve seen several practical ways that AI automation tools can be applied safely and productively in provider organizations. These guiding principles can help you transition AI from daunting to doable:
- Don’t jump to headcount reduction. The quick wins lie in using automation tools to power your existing people and processes. This is the tandem bicycle concept and it’s particularly appropriate for teams of “revenue cyclists.” The longer-term questions about staffing, roles and responsibilities and the shape of your finance function will require more deliberate planning and should not be rushed. At one medical equipment company, staff were creating hundreds of reports on payer policies and fee schedule changes but were always unable to keep up or cover most of their payers. Today, instead of creating, they are reviewing, approving and updating systems—acting on financial planning and making forecasts that they never had time for previously.
- Do your homework on risks and best starting points. Leaders who are making the most progress look at the steps of the revenue cycle and diagnose where the improvement opportunities lie, tempered by any risk that may be irreversible.
For each step, they ask: how recoverable would a mistake be before it causes permanent revenue loss? AI effectiveness is a moving target, so score it using current benchmarks or vendor data, as we’ve done in the scorecard. Start where readiness is high—a strong improvement opportunity and area where AI can be effective, with manageable downside if something goes wrong.
- Don’t just hope AI will “learn as it goes”—instead be deliberate about delivering your data and expertise to the tools. There is a notion in the market that AI just learns, “gets it,” and improves itself over time. It’s true the agents are powerful and the models are getting better all the time, but the reality is giving AI the right data context matters more than the AI itself. That’s good news for RCM leaders—you have the data and the experience; it just needs to be harnessed. Claims history, past results with payers, the tips and tricks that are in peoples’ heads—all that needs to go systemically into data sources that AI draws on.
- Probe deeply on the meaning of “human in the loop.” It’s important to note that AI can be used in many different degrees; it’s not a choice between 100% human/manual and 100% AI. On one end, AI agents can simply be “power tools” that RCM staff use as productivity accelerators. On the far other end, they can be built to do revenue cycle tasks autonomously and even linked in a chain to do RCM end-to-end, achieving what some call “touchless” RCM. Most organizations don’t leap from manual processes to AI overnight. At one Dental provider we work with, Destination Smiles, AI conducts some RCM steps and gives staff tips for AR follow-up. The CEO sees speed but “[otherwise] you can’t tell if it’s AI or human, because it’s both and the process is largely the same.”
“Human in the loop" appears in virtually every RCM AI pitch. It can mean a check-based approach, in which a human reviews the AI's output and flags errors. A production-based loop means human RCM staff are doing the primary revenue cycle work, with the AI supporting and learning from their actions. The difference in meanings has implications for compliance and risk of errors.
- Limit change to legacy systems. Many organizations worry about the cost, time and risk of overhauling legacy PMS and EHR systems. The good news is that you don’t have to tear down systems of record to take advantage of some AI processing power for revenue cycle. Most have APIs for access and write-back, and you can add human review to ensure changes to the system of record are accurate.