Is Agentic AI Transforming Revenue Cycle Management?
By Michael Berger, former senior level Revenue Cycle Officer and healthcare finance consultant.
Recently after reading health care article after article centered on the term “Agentic” I become intrigued as to its impact on the Revenue Cycle and how it is managed.
Let us start by defining agentic in the context of technology/AI. It refers to AI agents/systems that can perceive, reason, and act autonomously to conduct tasks with minimal or no human supervision. – effectively creating an autonomous environment. These systems can:
– Make decisions independently
– Interact with external tools
– Manage and execute complex workflows
– Adapt to new information
The discussion that follows is intended to serve as both an exploration and a caution for those who view AI/Agentic AI as a “magic bullet” for an optimized and flexible revenue cycle. While the potential is significant, there is a risk that we may be devaluing the importance of intuition, experience, technical acumen, and situation awareness in our analysis and subsequent decision making.
1. Have we devalued the “Art” component of Management?
2. Have we allowed “AI” to become the “End” – not just a tool to achieve an outcome/result?
3. Are you equipped to recognize “AI” misinterpreted data – and respond quickly?
4. If as Dr. Deming emphasized that successful management requires considering both measurable and non-measurable aspects – has agentic AI/AI devalued that?
1. Have we devalued the “Art” component of Management?
As a lecturer, I would begin my Health Care Administrative Issues class at Rutgers University by writing on the White Board the following, while stressing the point that high performing managers understand the difference and the need to apply both concepts:
Management Science: Planning & directing of effort and employing of resources (human and materials) to accomplish a defined objective.
Art of Management: Intuition and experience - the ability to accurately predict an outcome.
High performing leaders understand the need to apply both.
Consider this non-health care example: Over the last five years, Major League Baseball managers have become robotic requiring a strict adherence to analytics in their planning and decision-making – they rely far less on hunches, intuition, or just “rolling the dice.” If the analytics prove wrong – they blame the analytics, which is much more comfortable than questioning judgment.
Revenue Cycle leaders should not forget that their experience is a valuable analytic tool and when used in concert with AI and data analytics it allows leaders to differentiate meaningful trends from anomalies – such as the ones we experienced from the COVID crisis, short-term increases in Accounts Receivable levels, and other situational occurrences.
2. Has “AI” become the “End” – not just a tool?
As a child I remember watching a Popeye cartoon when Popeye was teaching Olive Oyl to drive – he instructed her to make a left turn, and Olive Oyl put her arm out the window to signal a left turn – but the car did not turn as she neglected to turn the steering wheel. As you suspect Popeye looked at her with bewilderment.
While not suggesting that there is a direct parallel to the cartoon story to what, we as, Revenue Cycle professionals experience – AI/Data Analytics is the turn signal, you are the one that turns the steering wheel - converting data into action plans, workflows, procedures, policies, and training modules.
3. Are you equipped to recognize an “AI” misinterpreted data – and respond quickly?
If we accept that AI is transforming management from a subjective art to a predictive science, we must also ask: are we vigilant in our review of the data and using our “on the ground” assessment of what we deem to be occurring to be skeptical of how the data as presented?
For instance:
· Customer Effort Score (CES) measures how easy it is for patients/guarantors to navigate scheduling and billing. Are you routinely performing post service surveys/soliciting patient comments and comparing them to AI generated data? Thus, allowing you to question the data when the results do not match your own assessments. Note; the AI generated data may be accurate – the patient’s perception of the process may be inaccurate, thus the need to create an action plan to either solely alter the patient’s perception or alternatively adjust the processes that directly impact the perception.
· Data collected does not accurately reflect the area of deficiency. From a personal experience – Press Ganey data erroneously blamed the Access Department for patient complaints about the time it took for a pregnant woman to be placed in a bed and seen by a clinician. The complaint was misplaced because of a combination of how the question was phrased and the patients did not differentiate the clerical part of the admission’s process from the clinical intervention.
· Negating the trend of adopting the concept of identifying a patient’s “propensity to pay” to establish future collections in favor of individualized payment plans that factor in patient’s situational factors and provide a personalized pathway to satisfying their financial obligation. I employed the later by not incorporating a means test within our financial policy and achieved a collection rate of 94% on payment plans and even higher patient satisfaction scores.
4. Has AI devalued the non-measurable?
As Dr. Deming emphasized, effective management requires considering both measurable and non-measurable factors.
The need for AI to measure the non-measurable which defy current standards of measurement and size has been a challenge to AI – so are we to accept the premise that low turnover equates to high morale or employee wellbeing / for patient-facing chatbots measuring success by a high fraction of queries that are resolved without human intervention may give a false ready as we do not easily know how many patients/guarantors refuse to use them.
While not often easy or require manual effort there is still a need to do off-line assessments, audits, person-person questionnaires/interviews to obtain real-time feedback.
My Take
While the potential of agentic AI to optimize complex revenue cycle processes, reduce costs, accelerate cash flow, and enhance patient experience is significant, it is important to recognize its limitations. AI can enhance decision-making, but it does not replace it.
The complexity of the revenue cycle especially at the individual account level results in the following challenges which may hinder achieving a “broadcasted” benefit of a 30 to 60 percent reduction in the cost-to-collect.
Issues and questions to consider in an AI agentic environment:
· Is the data utilized by AI systems of sufficient quality for effective decision making?
· Can AI/Agentic AI compatibility and functionality be achieved with the existing legacy systems?
· Availability of skilled resources to leverage AI/Agentic AI in the RCM environment.
· The degree to which we are comfortable with autonomous action.
· Compatibility with Mission Statement and Corporate Directives.
· Impact on patient satisfaction levels and Net Promoter Scores.
· Have we ensured the optimized balance between “artful” management and an autonomous environment exists - encouraging financial leaders to continue to apply their intuition, experience, situational awareness and technical acumen to the data?
About the Author
Michael Berger is a former senior level Revenue Cycle Officer and healthcare finance consultant with more than 35 years of experience in hospital operations and revenue cycle leadership. During his tenure at St. Peter’s Healthcare System, Berger led initiatives focused on improving patient financial engagement while strengthening collection performance and financial stability for the organization. Having worked directly with patient financing programs as both an operator and a client, he now shares insights on revenue cycle strategy, patient financial services, and the financial challenges facing rural hospitals.