Health systems have no shortage of successful AI pilots. Ambient documentation, revenue cycle automation, clinical decision support — individual use cases have delivered real results in individual departments.
But translating those wins into organization-wide results has proven far more difficult. Fragmented technologies and siloed deployments have left most organizations struggling to scale, with only 4% reporting success across the full enterprise.
However, there are evolving best practices and examples of agentic AI automating critical enterprise-wide workflows with remarkable results.
Three keys to operationalizing AI at scale
UiPath helps organizations improve efficiency and boost performance by automating processes. Through its work with organizations across industries, especially healthcare, UiPath has identified three effective practices for operationalizing AI at scale.
First, organizations need a clear picture of how work actually moves through the enterprise. In healthcare, a single workflow like denial management or referral intake can touch dozens of handoffs across clinical, financial and operational teams — often through disconnected systems.
“You’ll find a lot of rework steps,” said Peter Reischer, senior healthcare solution engineer at UiPath. “Cases go into certain stages, they go back, folks are calling a payer or finding new information that requires them to reclassify.”
Without mapping that complexity end to end, automation efforts tend to address fragments of a process rather than the whole thing.
Second, a comprehensive framework can guide healthcare organizations in putting in place the right technologies to automate these key processes.
This framework encompasses four key areas: 1) ingestion of data; 2) research and analysis; 3) data entry and 4) resolution and close.
The framework then applies technologies to automate each area. Technologies include intelligent documentation processing (IDP) to pull data from a document and put it into the system of record; agentic AI, which uses agents to perform specific tasks — and robotic process automation (RPA) for simple, deterministic, data-entry tasks.
The final “technology” is the human who must remain in the loop. “Despite the use of AI and agents and other technologies, we still need humans in the process,” Mr. Jordan observed. “The goal is to reduce their workload so they can review the output of the technology and decide whether it is correct or not.”
Third is adding agentic orchestration, which overlays and orchestrates the other technologies.
The orchestration layer decides when to deploy specific technologies at each stage of a workflow — routing cases through the appropriate sequence of tools, agents, and human reviewers. “One of the key elements of a successful AI strategy is how you orchestrate and get the right data into your AI as well as the right data out,” Mr. Reischer said.
UiPath’s orchestration platform is designed to work across a health system’s existing tech stack by integrating with EHR systems, third-party agents and other critical vendors rather than requiring organizations to consolidate around a single technology.
An orchestration platform provides the ability to look across the organization and understand what each technology is doing at any given time.
Agentic AI in practice: Streamlining denials
Claims denial is an enormous pain point for health systems, as approximately 15% of provider claims are initially denied by payers. While more than 50% of denials are eventually overturned, the process is slow, involves significant rework and is costly, with providers spending $20 billion per year on denials.
A key step in responding to denials is conducting a medical record review, where teams can spend 70 minutes per review pulling together fragmented clinical and financial data to extract relevant information. “It is a time-consuming, repetitive, highly manual process,” explained Betty Kim, director of product management in healthcare at UiPath.
UiPath has created a solution where multiple AI agents are orchestrated to streamline the medical record review process. This decreases the time spent on denials, enables faster, better decisions, and increases visibility into every aspect of the denials process. CFOs and the head of RCM can understand what’s happening with denials at an aggregate level and RCM teams can see data on why denials have occurred and how to prevent them in the future.
“Since implementing the UiPath MRS solution, we’ve reduced the average summary review time from 70 minutes to 6 minutes, a 90% improvement,” said Benjamin Smith, vice president of technology, medlitix. “Our clinicians are spending more time on direct patient care and less time digging through documentation.”
Agentic AI in practice: Expediting referrals
Historically, the referral process has been manually intensive for providers. A provider would receive a fax, have someone look at it to extract the necessary information, enter this information into an order, potentially triage the order based on urgency and then assign the referred patient to the correct clinic and work queue.
Agentic AI is transforming this process. First, data can automatically be extracted from faxes, transcribed, and classified. Then, agentic AI can use complex reasoning to determine the right provider to direct the patient to. If, for example, the patient is an oncology referral, agentic AI can assess who the right oncologist is for a given patient, according to the oncologist’s clinical practice guidelines and scope of practice. This quickly routes referrals to the most appropriate provider and decreases referral leakage.
“Organizations are using agentic and generative AI to process the wide variety of different referrals that they get . . . providing a streamlined approach,” Mr. Reischer said.
However, despite the important role that technology is playing in improving and streamlining the process, the work of bots and agents is limited to conducting research and presenting findings to clinicians or patient access coordinators, who remain at the helm of the referral process.
From pilots to AI at scale
Denials and referrals are just two examples of how health systems are going beyond using AI in a limited pilot to leverage coordination and orchestration — while keeping humans in the loop — to automate their most important processes. Results include faster decisions, improved efficiency, decreased costs, accelerated revenue recovery and improved performance of the enterprise’s critical clinical and financial workflows.
To learn more about how health systems are applying agentic AI to orchestrate clinical and financial workflows, including denial management, prior authorization, referral intake and medical record review, visit UiPath.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.