How a U.S. Healthcare Provider Automated Phone Call QA with StackAI

How a U.S. Healthcare Provider Automated Phone Call QA with StackAI

Overview

A publicly-traded U.S.-based healthcare provider was excited to partner with StackAI to automate its call center QA process. By combining advanced speech-to-text technology with powerful LLMs, they cut quality assessment costs by 70%, improved customer satisfaction by 10%, and reduced AI build costs by 80%. The solution now powers high-volume QA processes with speed, accuracy, and full HIPAA compliance.

The Problem: Manual QA at Scale Was Slow and Costly

Quality Assessment (QA) is essential in industries like healthcare, where call center interactions directly impact patient satisfaction and trust. Traditionally, QA relies on manual reviews of transcribed phone calls, using detailed evaluation templates to assess quality. For a large healthcare provider, this method proved time-consuming, labor-intensive, and incredibly costly.

However, breakthroughs in AI, particularly in speech-to-text and LLMs, have unlocked a new level of automation. Modern transcription tools can now outperform human transcribers in accuracy, and LLMs can evaluate transcripts rapidly and consistently across thousands of calls.

The healthcare provider recognized that automating this process would improve operational efficiency, lower costs, and enhance the quality of care delivered through its call center. They turned to StackAI to make it happen.

The Solution: End-to-End QA Powered by LLMs and Speech Recognition

StackAI enabled the healthcare provider to build a HIPAA-compliant QA AI assistant with no-code tools and enterprise-grade security. The assistant was fully functional in under a month and built directly by the call center manager, eliminating the need for external vendors and adding a layer of security. 

The QA AI assistant followed a streamlined architecture, with the Deepgram Nova 2 Medical model handling high-accuracy transcription of phone conversations, optimizing for clinical language and outperforming industry benchmarks by 20% in word accuracy.

Anthropic’s Claude 3.5 Sonnet LLM then evaluated the transcripts based on internal knowledge bases and specific QA criteria. The logic of the evaluation assessed whether the doctor followed protocols, reached a recommendation, understood the patient's issue, and ensured patient satisfaction.

The results were returned in JSON format and automatically uploaded to the healthcare provider’s database via API, completing the loop with minimal human intervention.

Lower Costs, Higher Quality, Faster Deployment

The impact of the QA assistant was immediate and measurable. The healthcare provider achieved a 70% reduction in QA-related costs, freeing up resources for higher-value work. At the same time, patient satisfaction increased by 10%, directly improving outcomes and revenue.

What’s more, the assistant was developed at 80% less cost than building an equivalent solution with an internal AI team. The success of the project has since inspired further AI initiatives across other teams, including physician co-pilots and automated SOAP notes. StackAI remains proud to partner with an organization dedicated fully to advancing patient health and well-being.