Ambience Healthcare Unveils AI Model That Outperforms Physicians in Medical Coding Accuracy

The model’s accuracy was compared to 18 board-certified physicians, with the AI showing a 27% improvement.

San Francisco-based Ambience Healthcare has introduced a breakthrough in ambient AI technology, unveiling what it claims to be the first artificial intelligence model to surpass board-certified physicians in ICD-10 coding accuracy.

The AI, trained with OpenAI’s Reinforcement Fine-Tuning (RFT) technology, demonstrated a 27% relative improvement in accuracy compared to expert physicians.

The model aims to tackle the substantial administrative inefficiencies in the US healthcare system, estimated at $266 billion annually, including $19 billion attributed to errors in diagnosis coding and documentation substantiation.

Ambience’s AI identifies precise ICD-10 codes during patient interactions, which are used to inform its “coding-aware” scribe, a documentation tool that produces notes clinicians can verify in real time.

“Think of this technology as a scribe that fluently speaks both clinical language and the intricate healthcare billing rulebook from day one,” said William H. Morris, MD, MBA, Chief Medical Officer of Ambience Healthcare. “Our platform helps clinicians finish their notes faster and ensures that revenue cycle teams receive cleaner, more accurate, and audit-ready charts,” he added.

The platform is currently deployed across inpatient, outpatient, and emergency settings in health systems across the United States, supporting documentation in over 100 medical specialties.

Organizations such as UCSF Health, Houston Methodist, Memorial Hermann, and John Muir Health are among over 40 healthcare providers currently using the system.

How the AI-Powered Scribe Works

Clinicians initiate documentation by tapping ‘Record’ on a device and conducting the patient consultation as usual. The AI then generates a draft clinical note with ICD-10 and CPT codes for review.

Once approved, the documentation is automatically filed back into the EHR. The platform integrates with major systems such as Epic, Oracle Cerner, and athenahealth.

By addressing documentation and coding at the point of care, the technology aims to reduce compliance risks and eliminate post-visit clarifications between physicians and coding departments.

“We built for compliance at the core of our platform because health systems need assurance that AI technology in the workflow can stand up to audits, protect reimbursement, and support clinical documentation integrity," said Michael Ng, CEO and Co-Founder of Ambience Healthcare.

To validate performance, Ambience assembled a gold-standard test dataset of complex clinical cases labeled by multiple expert clinicians. The model’s accuracy was compared to 18 board-certified physicians, with the AI outperforming this group by 27%.

"Reinforcement Fine-Tuning lets developers align powerful AI models with the precision and robustness that high-stakes domains like healthcare demand,” said Shyamal Anadkat, Head of Startup Solutions at OpenAI.

Beyond ICD-10 coding, Ambience is looking to extend the platform’s capabilities to other administrative areas.

The company is already applying similar methods to other coding tasks, such as Current Procedural Terminology (CPT) coding, and is exploring potential uses in prior authorizations, utilization management, and clinical trial matching.

Founded in 2020 by Mike Ng and Nikhil Buduma, Ambience Healthcare offers an AI platform for clinical documentation, CDI, and coding. The platform is used across multiple care settings and is backed by investors including Andreessen Horowitz, OpenAI Startup Fund, and Kleiner Perkins.

Stay tuned for more such updates on Digital Health News.

Stay tuned for more such updates on Digital Health News