Cleveland Clinic is a nonprofit academic medical center headquartered in Cleveland, Ohio, with operations in Florida, Las Vegas, Toronto, London, and Abu Dhabi. The health system employs approximately 83,000 caregivers and operates 23 hospitals and 300 outpatient facilities.
In 2025, Cleveland Clinic reported $18.3 billion in operating revenue and recorded 15.9 million patient encounters, including 14.4 million outpatient encounters. At that scale, even small improvements in clinical workflows can significantly affect caregiver capacity, patient experience, and operational efficiency.
Artificial intelligence has become an enterprise priority for the organization. Cleveland Clinic appointed its first Chief AI Officer in 2024 to lead enterprise AI strategy, safety, and governance. In its 2026 State of the Clinic address, CEO Dr. Tom Mihaljevic identified staffing, scheduling, supply chain, clinical risk prediction, and paperwork automation as areas where AI could support caregivers.
Although Cleveland Clinic has not disclosed a single total investment figure for AI, its commitment is reflected in several major initiatives, including a Virtual Command Center developed with Palantir Technologies, an enterprise ambient documentation platform rolled out in 2025, and an AI-powered sepsis detection platform expanded during the same year.
These two initiatives provide a clear view of how Cleveland Clinic is applying AI directly to clinical work. Ambient AI documentation is designed to reduce the administrative burden that pulls clinicians away from patients and into after-hours charting, while AI-driven sepsis detection is intended to identify deteriorating patients earlier while reducing the false alerts that can undermine confidence in clinical warning systems.
Ambient AI Documentation for Clinicians
Clinical documentation remains one of the most persistent sources of administrative burden for physicians and other healthcare professionals.
The American Medical Association’s 2025 survey of nearly 19,000 physicians found that 41.9% reported at least one symptom of burnout, with ineffective electronic health record systems and excessive administrative tasks among the leading sources of stress. The AMA’s 2024 data also found that 22.5% of physicians spent more than eight hours per week working in the electronic health record outside their normal working hours.
For an organization that recorded 14.4 million outpatient encounters in 2025, documentation requirements can quickly become a significant operational challenge. A few minutes of additional documentation per visit may appear small in isolation, but multiplied across millions of encounters, the accumulated burden becomes substantial.
Cleveland Clinic turned to ambient artificial intelligence as one way to address that problem.
Ambient AI documentation works by listening to a patient-clinician conversation, with appropriate consent, and using AI to generate a structured clinical note. The technology Cleveland Clinic adopted uses a phone-based application to capture the encounter and then generates documentation within Epic, the health system’s electronic health record.
According to Cleveland Clinic, the platform also provides clinical documentation integrity support, point-of-care coding, and customized after-visit summaries. The objective is not simply to automate note writing but to change how clinicians interact with documentation during and after a patient encounter.
Instead of dividing attention between the patient and the keyboard, clinicians can focus on the conversation while the system captures relevant information in the background. The clinician then reviews and edits the resulting note before it becomes part of the medical record.
Cleveland Clinic’s broader AI strategy influenced how it selected the technology. Chief Digital Officer Rohit Chandra has described the organization’s approach as partnering where possible and building only when necessary. He has also characterized AI vendor selection as closer to investing in a company than purchasing a finished product because the technology remains sufficiently early that few vendors can be considered completely finished solutions.
Rather than relying solely on traditional procurement processes, Cleveland Clinic favors pilots with clearly defined outcomes. Chandra has emphasized that change management can be more difficult than the technology itself.
That philosophy was reflected in the organization’s evaluation of ambient AI scribes. Cleveland Clinic conducted a head-to-head assessment of five vendors throughout 2024, evaluating factors including documentation quality, product capabilities, provider satisfaction, ease of implementation, and return on investment.
Published accounts differ on the precise size and structure of the evaluation. The American Hospital Association reported that Cleveland Clinic involved between 25 and 35 clinicians with each vendor in pilots lasting three to five months. Cleveland Clinic’s own Consult QD publication described approximately 250 physicians across more than 80 specialties and subspecialties participating in the evaluation, while Fierce Healthcare reported that more than 300 clinicians participated over a six-month period.
Although the sources describe the evaluation differently, they agree on the central point: Cleveland Clinic tested five ambient AI vendors before selecting a partner.
Ambience Healthcare ultimately won the evaluation and received a five-year exclusive partnership with Cleveland Clinic. Fierce Healthcare later reported that Cleveland Clinic providers had logged 25,000 encounters using Ambience across 20 specialty areas.
The pilot also demonstrated why specialty-level performance matters when deploying generative AI in healthcare. Cleveland Clinic reported approximately 80% adoption among clinicians participating in the Ambience pilot, with executive director of digital health Beth Meese describing that figure as two to three times higher than the other vendors tested. Those comparisons were anecdotal, however, because Cleveland Clinic did not publish equivalent performance data for the other four vendors.
Adoption also varied significantly between specialties. Nephrology reached 99% adoption, while urology reached 55%. Cardiology increased from 50% to 71% following feature refinements.
These differences highlight an important consideration for healthcare organizations evaluating generative AI. Enterprise-wide averages can conceal meaningful differences in how well a tool performs in specific clinical environments. A system that works exceptionally well in one specialty may require additional tuning before it becomes suitable for another.
Meese cited Ambience’s documentation quality and responsive engineering support among the reasons Cleveland Clinic selected the platform.
The subsequent rollout was deliberately sequenced according to workflow complexity. A peer-reviewed account published in npj Health Systems describes four deployment waves. Clinicians who had previously participated in other scribe tools went first, followed by approximately 80% of target users with an emphasis on primary care. Selected specialties followed, while more complex specialties where the model was still being optimized came later.
Cleveland Clinic also incorporated mandatory training, live virtual sessions held three times per day, and physician super-users who supported colleagues during training.
For clinicians, the resulting workflow is substantially different from traditional documentation. Before a visit, patients receive advance notice and can opt out, while use remains voluntary for ambulatory providers. During the visit, the clinician can focus on speaking with the patient while the application captures the conversation. Afterward, the clinician reviews and edits the AI-generated draft before it is entered into the medical record, while the patient can receive an AI-generated after-visit summary.
The approach has since moved beyond experimentation into broad enterprise deployment.
Cleveland Clinic began the enterprise rollout on March 10, 2025, and reached more than 4,000 clinicians in approximately four months. A subsequent npj Health Systems paper reported that, roughly a year after deployment, more than 4,800 clinicians were using the system across more than 3.5 million encounters.
The same study reported 70% encounter-level utilization among established users, a Net Promoter Score of 60, and a satisfaction score of 96.6%. Approximately 60% of users agreed that using the technology increased their likelihood of remaining in practice.
Those results should be interpreted with appropriate context. Half of the paper’s authors are Ambience employees who disclosed equity interests in the company, and the retention figure measures clinicians’ stated intent rather than observed turnover.
Cleveland Clinic’s own figures provide additional insight into time savings. Consult QD reported approximately two minutes saved per appointment and 14 minutes per clinician per day, with active users relying on the tool for 76% of scheduled visits. Fierce Healthcare reported pilot-stage findings that included a 49.6% reduction in after-hours documentation time, a 25% reduction in note creation time, a 32% increase in patient face time, and 67% of clinicians reporting reduced cognitive burden.
These figures are self-reported and have not been independently audited. Cleveland Clinic has also not publicly disclosed a financial return on investment for the program.
Nevertheless, the scale of deployment demonstrates that ambient documentation has moved beyond a limited technology experiment at Cleveland Clinic. The organization has used competitive evaluation, specialty-specific testing, staged implementation, and clinician training to integrate the technology into a large clinical environment.
AI-Driven Sepsis Detection
Cleveland Clinic’s second major clinical AI initiative addresses a very different problem: identifying sepsis early enough for clinicians to intervene while minimizing the alert fatigue associated with traditional warning systems.
Sepsis is one of the most costly and dangerous conditions treated by U.S. hospitals. The Association of American Medical Colleges estimates that it accounts for approximately $62 billion annually in hospitalizations and skilled nursing care. According to the Centers for Disease Control and Prevention, approximately 1.7 million U.S. adults develop sepsis each year, while at least 350,000 die during hospitalization or are discharged to hospice.
Timing is critical. The AAMC identifies sepsis as the third leading cause of death in U.S. hospitals and notes that every hour of treatment delay can increase the risk of death by approximately 4% to 9%.
At the same time, simply generating more alerts does not necessarily improve patient care.
A 2021 JAMA Internal Medicine validation study of a widely deployed proprietary sepsis model at Michigan Medicine found that the system missed 67% of sepsis patients while generating alerts for approximately 18% of all hospitalized patients.
This illustrates the fundamental challenge of clinical alerting: a system can generate warnings without necessarily producing useful information. When clinicians receive large numbers of alerts that turn out to be irrelevant, they may become less responsive to the warnings that matter most.
Cleveland Clinic has not publicly identified the legacy system it replaced, but its own pilot comparison found that its previous sepsis tools generated approximately ten times as many false alerts as the AI platform it adopted.
The health system has also established formal governance around sepsis detection. James Morrison, MD, chairs Cleveland Clinic’s Enterprise Sepsis Steering Committee, which oversees the integration of detection tools into clinical care.
Cleveland Clinic adopted a platform from Bayesian Health that continuously analyzes electronic medical record data, including laboratory results, vital signs, and clinical notes, to assess a patient’s sepsis risk in real time.
Rather than relying exclusively on episodic, rules-based screening, the system continuously evaluates the available patient record. The objective is to identify patterns that could indicate emerging sepsis while minimizing unnecessary alerts.
The technology builds on the Targeted Real-time Early Warning System, or TREWS, a machine learning model studied across five hospitals and 590,736 monitored patients. A 2022 Nature Medicine study found that when clinicians confirmed an alert within three hours, in-hospital mortality fell by 3.3 percentage points, representing an 18.7% relative reduction. Among higher-risk patients, the absolute reduction reached 4.5 percentage points.
The study was peer-reviewed, although its lead author, Suchi Saria, is affiliated with Bayesian Health, meaning the evidence is not entirely independent of the vendor.
Bayesian’s platform received FDA 510(k) clearance in May 2026. The company also reported 82% sensitivity, 89% provider adoption, and detection approximately 5.7 hours earlier than standard care. Those figures come from the vendor’s research program and should not be interpreted as results from Cleveland Clinic’s own deployment.
At Cleveland Clinic, the technology is designed to integrate into existing clinical workflows rather than creating another standalone system for clinicians to monitor.
According to Cleveland Clinic’s published pilot results, the platform produced substantially fewer false alerts than the legacy approach, while increasing the number of identified sepsis cases. The organization reported approximately one-tenth as many false alerts, a 46% increase in identified sepsis cases, and a sevenfold increase in cases alerted before antibiotics were administered.
These changes matter because clinical AI is valuable only when clinicians can act on its output. Reducing unnecessary alerts can help preserve attention for patients who genuinely require intervention, while earlier identification can give care teams more time to evaluate and treat a deteriorating patient.
Cleveland Clinic has not published a detailed description of exactly who receives alerts, how alerts are confirmed, or how the system routes information between nurses and physicians. Those details are important when evaluating the broader effectiveness of clinical alerting systems because the TREWS research tied its observed mortality benefit to clinicians confirming alerts within three hours.
The Cleveland Clinic pilot at Fairview Hospital included more than 3,330 patients during 2024 and 2025. Compared with the organization’s legacy tools, Cleveland Clinic reported a tenfold reduction in false alerts, a 46% increase in identified sepsis cases, and a sevenfold increase in cases flagged before antibiotic administration.
These are self-reported results from a single-hospital pilot and have not been published in a peer-reviewed journal. Cleveland Clinic has also not disclosed mortality, length-of-stay, or cost outcomes associated with its own deployment.
The program has nevertheless progressed beyond a small-scale experiment. Cleveland Clinic reported the platform at 13 hospitals, with further expansion planned across Ohio and Florida. The health system and Bayesian Health have also planned to collaborate on AI modules for other critical conditions, potentially extending the partnership beyond sepsis.
What Cleveland Clinic’s AI Strategy Reveals
Cleveland Clinic’s experience illustrates several broader principles for implementing AI in healthcare.
One is the value of competitive pilots before making major enterprise commitments. Rather than selecting an ambient AI vendor based solely on demonstrations or vendor claims, Cleveland Clinic tested five systems against defined criteria and evaluated how they performed with clinicians. That process gave the organization evidence to support a multiyear partnership and subsequent deployment to thousands of users.
Another is the importance of sequencing implementation according to workflow complexity. Cleveland Clinic did not attempt to deploy its ambient AI documentation platform simultaneously across every specialty. Primary care and more standardized workflows were prioritized while the technology continued to mature in more complex clinical environments.
This approach can reduce implementation risk while giving technology teams and clinicians an opportunity to identify problems before expanding to more demanding settings.
The sepsis initiative demonstrates a third principle: clinical AI should be evaluated based on the quality of the signal it produces, not simply the number of cases it identifies.
In healthcare, an algorithm that identifies more potential cases but overwhelms clinicians with false alerts may not improve care. Reducing noise can be just as important as increasing sensitivity because clinician trust and attention are limited resources.
Cleveland Clinic’s approach therefore places workflow integration and clinical usability alongside model performance. Ambient documentation is valuable not simply because an AI model can generate a note, but because it can reduce documentation work without requiring clinicians to abandon their existing EHR workflow. Similarly, sepsis detection is valuable not simply because an algorithm can identify risk, but because its output can be delivered in a way that helps care teams prioritize patients.
Building AI Into Clinical Workflows
Cleveland Clinic’s AI initiatives show how healthcare organizations can move from isolated experimentation toward enterprise-scale deployment.
Ambient AI documentation addresses an administrative problem that affects clinicians every day. By reducing time spent typing and completing notes, the technology can potentially return more attention to patient interactions and reduce after-hours documentation.
AI-driven sepsis detection addresses a clinical problem where timing and signal quality are critical. By analyzing patient information continuously and reducing unnecessary alerts, the technology is designed to help clinicians identify patients who may need intervention without adding another layer of alert fatigue.
Neither initiative eliminates the need for clinicians. Instead, both place AI within existing clinical workflows and preserve human responsibility for reviewing information and making consequential decisions.
That distinction is important. The most meaningful applications of AI in healthcare may not be those that attempt to replace clinical expertise. They may be the systems that remove repetitive administrative work, surface important information earlier, and allow clinicians to devote more time and attention to the patients in front of them.
Cleveland Clinic’s experience also demonstrates that successful healthcare AI deployment depends on more than model performance. Vendor selection, governance, specialty-specific validation, training, change management, workflow integration, and continuous measurement all influence whether a technology becomes useful at scale.
As healthcare systems continue to face staffing pressures, administrative burdens, and increasingly complex patient needs, Cleveland Clinic’s approach provides a detailed example of how AI can be integrated into clinical operations while maintaining human oversight.
The broader lesson is that healthcare AI succeeds when it is designed around the realities of clinical work. The technology must produce useful signals, fit into established workflows, earn clinician trust, and demonstrate measurable improvements. For Cleveland Clinic, ambient documentation and AI-powered sepsis detection represent two different applications of that principle: one focused on giving clinicians time back, and the other focused on helping them recognize clinical risk earlier.