The AI Scribe Rollout That Actually Stuck

Six thousand clinicians, one rollout

Cleveland Clinic didn't run a pilot. It ran a deployment.

Starting March 10, 2025, the health system offered Ambience Healthcare's ambient AI scribe to 6,130 eligible clinicians across four waves, finishing in under 16 weeks. By the end, 5,748 had gone through training, more than 92% attendance in every wave, and 4,826 had used the tool at least once. A new paper in npj Health Systems documents how they did it, and a report from healthsystemCIO adds the operational detail: vendor payments were tied to onboarding volume for each wave, and a support desk fielded more than 900 inquiries with an average two-minute response time.

The number that should get a doctor's attention isn't adoption. It's retention. Sixty percent of clinicians who used the scribe agreed it increased their likelihood of staying in practice, according to both the study coverage from MedCity News and healthsystemCIO's independent reporting. Net Promoter Score came in at 60. CSAT at 96.6%, per healthsystemCIO; MedCity reported a 97% satisfaction figure from the same rollout. Encounter-level utilization among established users, those with 50 or more uses, reached 70%.

The adoption curve is not a straight line

The rollout data has a detail worth sitting with: first-use adoption fell from 100% in the initial pilot group to 72.6% by the third wave. That is not a failure, it is what real-world scaling looks like. A hand-picked pilot cohort of early adopters will always outperform the broad population it's supposed to predict for.

Specialty also mattered more than any headline number suggests. Family medicine and internal medicine cleared 80% utilization. Preventive medicine hit 90%. Radiation oncology sat at just 23%. Same tool, same health system, wildly different fit. That gap is a workflow problem, not a model problem: a scribe built around the conversational rhythm of a primary care visit will not map cleanly onto radiation oncology's documentation pattern.

Why this is a deployment story, not a model story

Most clinical AI coverage this year has been about what the model gets right or wrong: accuracy thresholds, FDA frameworks, hallucination rates. Cleveland Clinic's numbers point somewhere else. The tool that moved retention wasn't distinguished by capability. It was distinguished by how it was rolled out: staged waves, tracked training completion, a support function sized for real usage, and specialty-by-specialty attention to where the tool actually fit the workflow and where it didn't.

That is the part vendors rarely publish and hospitals rarely get right on the first try. A scribe that saves time in family medicine and stalls in radiation oncology is not a broken product. It is a product that was deployed as if every ward and every clinic worked the same way.

For a doctor on shift, none of the deployment mechanics matter directly. What matters is whether the tool shows up already fitted to how the day actually runs, list by list, specialty by specialty, instead of arriving as one interface stretched across every kind of encounter. Cleveland Clinic's numbers are a reminder that the deployment plan is not a footnote to the AI story. For clinicians deciding whether to trust a new tool, it may be the whole story.

Sources

• Ambient Scribes Improve Clinician Retention, Cleveland Clinic Research Shows (MedCity News)
• Cleveland Clinic Tied Vendor Payments to Onboarding Waves in 4-Month AI Scribe Rollout (healthsystemCIO)
• Accelerating ambient AI scribe enterprise-scale deployment: Cleveland Clinic's novel approach to health system-industry partnership (npj Health Systems)