Cleveland Clinic

Cleveland Clinic's inpatient coders faced an overwhelming documentation burden — each encounter generates an average of 59 clinical documents and nearly 50,000 words, requiring coders to select from over 150,000 ICD-10 codes. The process was chronically short-staffed, cognitively exhausting, and over 40% of the coding team was retirement-eligible within a few years, creating a looming workforce crisis.

Cleveland Clinic partnered with AKASA to deploy a fine-tuned large language model via AKASA's mid-cycle Optimization Suite. The AI reviews coded charts, flags potential missed diagnoses, HCC captures, and quality metrics, and surfaces evidence-backed recommendations with direct quotes from the patient record — acting as a second reviewer that integrates into existing workflows without replacing them.

In the first 60 days of production, Cleveland Clinic achieved higher ICD-10 capture and case mix index, improved severity of illness, risk of mortality, and HCC capture rates. Approximately 50% of AI-flagged recommendations were accepted by coders, and every suggestion was supported by a full audit trail. The initiative improved both financial yield and quality metrics simultaneously.

• Start GenAI pilots where rules are consistent and volume is high (e.g., inpatient coding) to maximize early ROI and build internal trust before expanding to more variable workflows like prior auth or denials.
• Workforce trust requires transparent AI reasoning — showing coders the exact chart evidence behind each recommendation transformed skepticism into adoption.
• Buy vs. build decisions should favor partnership: purpose-built, healthcare-tuned LLMs outperform general-purpose models and are not feasible to maintain at scale internally.