No room for error: Building safe and scalable voice AI for healthcare

In the US, we are facing twin challenges: We have never been unhealthier as a nation, and at the same time, we are experiencing a significant shortage of healthcare workers that is only projected to get worse. Voice AI agents can help: they have the skills necessary to handle complex administrative and clinical tasks that can free up healthcare workers’ time. But Voice AI agents used in such high-stakes situations must be trustworthy, and also pose unique challenges: they need to follow standard operating procedures; hallucinations are a deal breaker; and all information exchanged needs to be accurate.

Infinitus Systems has addressed these challenges by deploying an innovative AI platform that has successfully automated over 100 million minutes and 5 million healthcare phone calls. Use cases include: patient-facing calls (e.g., conducting health risk assessments), payor-facing calls (e.g., verifying patient benefits) and provider-facing calls (e.g., performing medical record requests). In this presentation, we will cover critical components that were designed as a part of this platform to ensure its effectiveness in healthcare conversations: a discrete action space that limits AI responses to a set list; a knowledge graph that has learned from millions of past calls; and an AI review function that checks for accuracy and standards, escalating to a human for complex cases when necessary.