Bridging the Gap Between AI and Humans in RCM

Healthcare providers are operating under sustained financial pressure. Reimbursement continues to tighten as labor and supply costs rise steadily. Expectations around access, experience, and quality keep expanding. In response, health systems are being forced to examine how work actually happens. The goal is maintaining stability while delivering timely, and high-quality care in an environment that leaves little margin for inefficiency.

Within that context, shared services have become a focal point and within it, the revenue cycle stands out. It touches nearly every patient interaction outside and within the clinical encounter and carries responsibility for converting care delivered into revenue collected. For decades, leaders have pursued a straightforward idea which is that of a revenue cycle that largely runs itself. Billions have been spent on automation towards that end. However, the majority of organizations still rely on large, distributed teams working through a patchwork of tools. Progress has been real, but uneven and limited to specific functions rather than the system as a whole. This has given rise to a revenue cycle that runs on constant human effort.

Health systems have been experimenting with agentic AI, a technology that can independently decide and carry out many processes, thus being more like a coworker than a tool. It can actively manage tasks across the revenue cycle, beyond just offering guidance like generative AI. The changes it brings about might be quite a lot. The cost to collect may be reduced by as much as 30-60%, cash flow could be accelerated, and staff can be freed up to work on patient value rather than the paperwork. Patients, on the other hand, could very well enjoy quicker access to care and more straightforward billing. 

Within the next several years, top health systems are anticipated to extend agentic AI from pilot programs to full, scale deployment, especially in back end revenue cycle operations, to advance both financial performance and patient experience. This article describes the steps healthcare systems can take to get agentic AI started on their back end of the healthcare revenue cycle and describes how they can start getting value from it.

Why incremental automation has plateaued and what’s the fuel for Agentic AI

Traditional automation has delivered value where work is stable and predictable because here, rules can be written. Exceptions are also highly limited with a high volume of work. The current scenario is starkly different. Revenue cycle work is no longer confined to those conditions. Policies and documentation are shifting. Payer interpretations are also differing. Exceptions are thus not rare anymore. Most existing systems respond to this complexity by stopping their entire workflow. When inputs fall outside expectations, work gets diverted into manual queues. Over time, those queues grow, concentrating the most demanding work with staff already stretched thin.

This is where many organizations now find themselves, being capped. Adding more staff keeps things moving, but at increasing cost. Efficiency gains are flattened and margins remain under pressure. The constraint is the design of the system itself.

Significant investments are made in the revenue cycle by health systems, with a typical investment of 3-4% of their revenues, resulting in costs of over $140 billion annually due to inefficiencies in processes. In addition, there are denied claims amounting to 20%, but many of these claims are not being appealed, resulting in lost revenues. Hence, there has been a focus on AI and automation in several use cases in the revenue cycle.

Agentic AI in the healthcare domain is mainly implemented by external vendors, with minimal in-house use and only for isolated tasks. Although the automation of the entire revenue cycle is a complex process with many interconnected tasks and dependencies, the back end provides a good entry point for the implementation of agentic AI for the healthcare domain.

Mid & back-end revenue cycle is a low-risk entry point for Agentic AI

What’s changing now is both the availability of new tools,  as much as the type of work technology can take on. Agentic AI is a change from advice-only AI to AI that acts, where decisions are made and tasks are completed, rather than simply identifying issues to be acted upon by humans. In practice, they behave less like dashboards and more like operational teammates.

The mid-cycle coding & back-end process is mostly administrative in nature, and this makes it a relatively safer area to implement agentic AI. These are generally structured in nature and have clear rules to follow. This makes it easier to ensure compliance and standardization. There is also minimal interaction with the patients in the mid and back-end processes, and this means that there are no chances of adversely impacting the patient experience.

Therefore, the mid & back-end provides a controlled environment where organizations can experiment with, test, and refine agentic AI. In this manner, it becomes possible to limit the risks and avoid unintended consequences while building trust in AI.

• Incremental progress without disruption: Starting with mid-cycle coding through to the back-end also allows organizations to move gradually. It is very difficult to redesign the entire revenue cycle at once. So, focus on one area, prove its value, and then expand from there. Coding may come first, Denials and underpayments next, but with each step operational confidence and internal buy-in will also build. So, it is a win-win situation.  This approach matters because large-scale change in healthcare rarely fails for technical reasons. It stalls when teams lose trust in the outcome or cannot see how early effort connects to long-term impact. Measured progress, in turn, sustains momentum.

• Test and refine: As the coding and back-end revenue cycle is primarily administrative, it offers a lower-risk environment for deploying agentic AI than other areas. Unlike front-end tasks like scheduling (which require direct patient interaction) or clinical documentation, coding and back-end processes are strictly rule-based.  This ensures clearer compliance and eliminates patient-facing friction. Ultimately, this is a safe testing ground to refine AI models with built-in guardrails against unintended errors.

• Build a strong foundation: Optimizing Coding and back end builds a good foundation and lays the groundwork for a broader transformation. As autonomous agentic AI systems demonstrate reliability within this controlled environment, organizations and stakeholders gain the confidence to connect workflows rather than treating them as isolated functions. Visibility improves and the entire workflow becomes more accurate. Over time, this opens the door to addressing more complex, patient-facing work with the same principles without forcing those areas to absorb risk prematurely.

What value actually looks like

For significant impact to be achieved in the use of agentic AI, there needs to be a balance in the short-term and long-term plan with a degree of scope that indicates advancement as well as room for growth in the future. A holistic approach might be beneficial in this case.

• While a proof of concept is essential for demonstrating potential and securing organizational buy-in, teams shouldn't judge initial pilots solely on immediate financial returns. Instead, structure pilots around success metrics that forecast long-term value. Once the solution proves its worth, pivot toward scaling it enterprise-wide. This transition ensures early momentum drives true organizational transformation rather than stalling at the pilot phase.

• Focus must be put in certain areas to get significant results and ROI. To maximize impact, organizations should align their initial AI deployments with core strategic priorities. For most healthcare providers, this means targeting high-volume, error-prone processes that offer clear, quantifiable outcomes. By selecting use cases with deliberate strategy, executives can secure immediate ROI and build the necessary momentum for a broader enterprise transformation.

• True transformation happens when humans and AI operate side by side, unlocking peak performance from both. To succeed, organizations must position AI as a workforce multiplier rather than a replacement. Ensuring long-term success will depend on continuously measuring this collaborative impact while actively identifying areas for mutual improvement.

The work ahead

Adopting agentic AI is not a single decision but a series of choices associated with design. Organizations, therefore, must be intentional about where to start. The goal is not a “touchless” revenue cycle in the abstract. Humans remain essential, for oversight, for handling exceptions and to maintain trust. However, when routine work is no longer powered by human endurance, the system gains room to improve.

Healthcare has waited a long time for meaningful improvement in revenue cycle performance aiming for structural relief. This moment matters because it offers a way to remove effort rather than redistribute it. This is to let systems handle persistence and scale, while people focus on judgment and accountability. When that happens, the revenue cycle stops being a constraint on care delivery. It becomes an enabler again.

This exact shift is central to our thesis at Serent. We engineered our solution around the conviction that true transformation requires both advanced AI and humans-in-the-loop, operating in tandem to own the outcome. We don't just automate tasks; we provide a collaborative framework where technology handles the heavy lifting and humans provide the critical oversight. It is this balance that turns the promise of agentic AI into a reliable, enterprise-grade reality.