By Emily Bonham, Senior Vice President – Product Management, AGS Health
LinkedIn: Emily Bonham
LinkedIn: AGS Health
With its speed, agility, and accuracy, agentic automation is poised to transform healthcare revenue cycle management (RCM). The confluence of advanced artificial intelligence (AI) and automation creates a next-generation digital workforce that not only eases the burden on overwhelmed RCM teams but also leverages its ability to mimic human decision-making to streamline nearly every aspect of the revenue cycle.
“Agentic” highlights AI agents’ ability to act purposefully and independently, exhibiting autonomy, goal-driven behavior, and adaptability. They understand natural language, adapt to changing rules and workflows, and make autonomous decisions to achieve tangible business outcomes. Unlike robotic process automation (RPA), which often reinforces rigid and inefficient workflows due to its limited scalability and contextual awareness, AI agents emulate human cognition. They understand processes, adapt to context, and manage exceptions with nuance.
Today’s AI agents have the potential to deliver a scalable, flexible, collaborative, cost-effective, and outcomes-driven workforce that can empower RCM teams and transform the revenue cycle for improved financial performance. However, its success is tied to a thorough understanding of what a digital workforce is, how AI agents work, and the application of best practices that have emerged from the experiences of early adopters.
Defining “Digital Workforce”
An AI workforce is a human-in-the-loop (HITL) system design in which autonomous digital agents work collaboratively with human counterparts to perform a full range of RCM functions. These agents are purpose-built to execute tasks, make decisions, and interact with users and other systems autonomously or semi-autonomously through digital interfaces.
They leverage technologies such as:
- Generative AI (GenAI)
- Machine learning (ML)
- Large language models (LLMs)
- Natural language processing and understanding (NLP/NLU)
- Advanced reasoning engines
- RPA
This enables AI agents to efficiently and accurately manage denial appeals, monitor for and adjust workflows to changes in payer rules, coordinate with other agents to manage billing processes, and interact with revenue cycle staff. These capabilities translate into significant value through accelerated turnaround times, faster collections, reduced denial rates, improved accuracy, and increased staff productivity.
Finding the Right Fit
Early adopters of AI and automation report a 63% reduction in time spent on initial claim reviews, from an average of 15 minutes to as little as five minutes, and a 30% increase in recovery rates, greater claim accuracy, and lower administrative costs and error rates. However, achieving optimal return on investment in this advanced automation requires a strategic approach to planning, design, and deployment.
The first step is to determine which forms of AI agents are the best fit for existing RCM workflows by identifying processes that are high-volume, time-sensitive, low-complexity, and have predictable decision trees. Digital agents should focus initially on simple tasks, such as eligibility checks, claims status, and standard denial workflows. More complex tasks and those requiring human interactions should be left to human agents to give machine learning models time to develop and mature through human feedback.
A clear vision of AI agents’ future role in the RCM workforce is also essential to ensure they can acquire the necessary knowledge and experience to enhance their capabilities. Over time, AI agents will learn to manage tasks with increased complexity and effectively complete more expanded workflows as more results and data come in and feedback loops improve.
The right technology partner to build the digital workforce is another critical piece of the puzzle. Along with deep industry experience in both RCM and technology, the ideal partner should have in-depth knowledge of RCM complexities, proven AI and automation capabilities, and a collaborative implementation model. Also, avoid partners that offer only siloed technology or industry knowledge. A successful AI workforce lives at the intersection of both.
Deployment Prep
Preparing the RCM team and workflows for deployment begins by determining what data is accessible to the development team or partner and whether secure access can be obtained for systems, documentation, and/or portals. This information is crucial, as a successful implementation depends on transparent data sharing and development resources that can be trusted with the organization’s system environment.
Next is ensuring that data is AI-ready. Often, critical data lives in a siloed, fragmented state that is neither AI-friendly nor capable of “communicating” with AI agents. To support their ability to analyze and act, data must be able to support autonomous decision-making, real-time feedback loops, and continuous monitoring. Clean, contextual, and connected data is necessary for goal-driven AI agents to execute tasks autonomously and effectively, learning from outcomes and adapting over time. To accomplish this:
- Standardize formats across EHR, billing, and payer systems.
- De-duplicate patient and provider records.
- Resolve any missing fields
- Map legacy codes to modern taxonomies
- Tag data by specific RCM stages
- Define decision points and outcomes
Governance and guardrails should also be established using role-based permissions and defined escalation paths, and all actions should be auditable. Set confidence thresholds for autonomous vs. HITL decisions and ensure compliance with HIPAA and payer-specific rules.
Finally, engage internal security teams early in the process, clearly define objectives and current processes, and take the time to identify and engage with the right technology and data partners.
When this phased approach is followed, a minimum viable product use case can go live in four to eight weeks and full deployment within three to six months, depending on the complexity of the environment, the selected use case, and system integrations. Agentic systems require less rigid configuration than traditional RPA, so deployment cycles are often faster and more adaptable.
The Human Element
Human intervention is essential in an AI workforce model, which should ideally combine the judgment and empathy of human agents with the speed, scalability, and adaptability of AI agents.
It is a synergistic model where AI agents learn from outcomes to improve future decisions, and human agents provide feedback to continuously train AI models and refine workflows. Digital agents handle the volume while humans handle the variance.
It is a tiered system where AI agents manage structured tasks quickly and precisely while human agents step in when nuance, judgment, or external coordination is needed. In other words, an effective digital workforce incorporates man with machine—not man against machine.
The Measure of Success
Monitoring the effectiveness of a digital workforce is crucial for performance tracking and fostering trust, ensuring safety, and driving continuous improvement. Agentic automation systems make autonomous decisions, adapt to feedback, and operate in complex, high-stakes environments, meaning success must be measured based on outcomes and processes.
Provider organizations must trust that AI agents can make safe, compliant decisions, acting within defined boundaries, and providing clear, explainable reasoning for audits and oversight. Because some AI agents also reason, plan, and adapt, it is essential to evaluate task adherence, tool use accuracy, and intent resolution.
Traditional RCM metrics such as appeal success rate, denial rate, time-to-collect, cost-to-collect, and dollars recovered or protected remain relevant with a digital workforce. However, another layer is required to measure autonomy, accuracy, and robustness. This includes:
- Percentage of tasks completed without human intervention
- Correct code selection
- Valid tool utilization
- Performance under exceptions
Also, measure the performance of the feedback loops that are essential for model training and prompt refinement. Metrics that measure how well these performance loops help AI agents to learn over time include what worked, what failed, and why.
Monitoring performance will help ensure continuous improvement of the digital workforce by providing outcomes data to refine agent rules and identify and prioritize new automation opportunities. It also sets the stage for using AI-generated insights to refine workflows and training data.
The Transformative Power of Agentic Automation
Agentic automation is intelligent, adaptive, and poised to integrate deeply into healthcare RCM. Digital agents will evolve with the business, learning from outcomes, aligning with payer changes, and adapting to new workflows in near-real time.
Embracing the shift to well-designed digital workforce positions provider organizations to reap numerous rewards by unlocking scalable, round-the-clock operations. A digital workforce can also reduce cost pressures without sacrificing quality, reassigning human talent to higher-value work and enabling the creation of a future-ready infrastructure.