Health systems have spent the past two years stress-testing ambient AI scribes, but isolating return on investment has remained difficult due to many overlapping initiatives and a lack of clean attribution paths. Agentic AI is shifting that calculus. When an autonomous agent makes a call, completes a scheduled task and produces a transferable outcome, the value math gets cleaner.
During an April 13 featured session at Becker’s Hospital Review 16th Annual Meeting, hosted by Hippocratic AI, four healthcare leaders walked through early-stage agentic deployments and the financial frameworks behind them. The panelists were:
- Meenesh Bhimani, MD, chief medical officer and co-founder of Hippocratic AI
- Peter Pronovost, MD, PhD, chief quality and clinical transformation officer and president of the Veale Healthcare Transformation Institute at University Hospitals in Cleveland
- David Sylvan, chief strategy, innovation and marketing officer at University Hospitals in Cleveland
- Kali Ihde, director of NM Ventures & Innovation at Northwestern Medicine in Chicago
Below are four takeaways from their conversation.
Note: Quotes have been edited for length and clarity.
1. A single pharmacy use case generated $1.5 million for University Hospitals
Cleveland-based University Hospitals deployed an agentic agent to call patients who fill prescriptions outside the system’s pharmacy, a population that accounts for roughly half of UH patients. In its first run, the agent placed about 9,000 calls, connected with roughly 5,000 patients, verified 3,500 and transferred 111 to the pharmacy team, generating more than 675 prescription transfers and approximately $1.5 million to the bottom line against $10,000 in agent costs. At a follow-on planning session, UH set out to identify $100 million in revenue or cost-takeout opportunity through similar deployments and is now spinning up roughly 20 to 30 projects.
2. Northwestern Medicine is using agents to clear unfulfilled-order backlogs
Northwestern Medicine, an 11-hospital system serving Chicago and Northern Illinois, focused its first agentic deployments on two areas with significant unmet outreach demand: annual wellness visits and lung cancer screenings.
On wellness outreach, the agents placed more than 2,400 calls in days and scheduled 50 appointments, double what prior human outreach had produced. Agents also surfaced patients in distress during those calls, enabling real-time escalations to social workers for patients who expressed they were struggling.
On lung cancer screening, where Northwestern entered the year with thousands of unfulfilled orders, agents made more than 1,700 calls and scheduled 250 patients for low-dose CT scans.
Ms. Ihde noted that the majority of those patients followed through and completed their screenings, and that statistically, some likely had findings caught earlier than they would have been otherwise. Without the agents, those orders would have expired. Ms. Ihde said the contrast with ambient AI is what makes agentic deployments easier to justify.
“If the promise of agents is they can do tasks, if they can complete work, if they can make a phone call, it’s actually then very clean how to measure the value,” Ms. Ihde said.
3. The harder constraint is the scarcity mindset, not the technology
Dr. Bhimani framed Hippocratic AI’s pitch as clinical abundance where agents that scale clinical capacity beyond what health system staffing models can be matched and validated through more than 7,500 U.S.-licensed clinicians. The panelists agreed the bigger barrier to capturing that value is internal. UH has run roughly 50 to 60 AI tools through a structured down-selection process in the past year, Mr. Sylvan said, using criteria built around strategy alignment, risk testing and feasibility with intentional resistance to defaulting only to its EHR vendor’s toolset.
Dr. Pronovost added that his institute, funded by an initial $10 million venture-philanthropy gift, committed to delivering $156 million in value over five years and was at double that pace at two years, prompting a second $10 million tranche from the donor.
4. The technology is a mirror, not a fix
Dr. Pronovost cautioned that agentic AI will amplify whatever organizational culture surrounds it. “AI isn’t a technology, it’s a mirror that will amplify whatever systems we build around it,” he said. The deployments work, the panelists argued, because the systems like Hippocratic AI’s safety validation, UH’s adjudication framework, Northwestern’s business-case discipline are designed to catch what the agent itself cannot.
Why agentic AI changes the value math
Ambient documentation tools have been hard to attribute returns to because they sit on top of existing workflows and share credit with every other initiative aimed at clinician burden. Agentic AI flips that. A call placed, a screening scheduled, a prescription transferred is a discrete, attributable unit of value that finance teams will fund and quarterly reports can track.
The panelists suggested the next phase will favor systems that pair agentic deployment with explicit problem-impact accounting. “Everything we do has to be converted to a dollar,” Dr. Pronovost said of UH’s framework. For executives still piloting, these case studies make it harder to argue that agentic returns are unproven and easier to argue that delaying carries a real cost.
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.
