How Voice AI Assistants Are Reducing Call Wait Times for a Regional Health Network

The Challenge

The access center relied on a multi-level touch-tone IVR in front of a team of patient access representatives. Most calls ended up with a person, whatever the request.

The existing approach made it difficult to:

• Absorb demand peaks without long hold times and abandoned calls.
• Let patients say what they needed instead of guessing the right menu option.
• Handle simple requests, such as confirming or moving an appointment, without an agent.
• Serve Spanish-speaking patients without transfers to a smaller bilingual team.
• Protect patient information while sharing appointment details by phone.
• Understand why patients were calling, because call reasons were logged inconsistently.

From a technical perspective, appointment data lived in the network's EHR scheduling module, and provider availability rules differed by department. The telephony platform also had no natural language layer. As with most healthcare development work, any conversational AI had to authenticate patients without exposing protected health information and follow visit-type and provider rules. It also had to run with low enough latency to feel like a normal phone conversation.

The Solution

InnovationM developed a voice AI assistant that sits at the front of the access center line. It understands the caller's request in plain speech and resolves routine scheduling tasks end to end. Calls that need clinical judgment or a person are routed to the right queue, with a summary already on the agent's screen. The solution included:

• Natural language call intake: Replacing the touch-tone menu with streaming speech-to-text and an LLM-based intent layer, so callers can describe their need in their own words.
• Patient verification: Checking identity using date of birth and other agreed identifiers before any appointment detail is shared, with limits on what is disclosed by phone.
• EHR scheduling integration: Confirming, cancelling, and rescheduling appointments through EHR APIs against live provider availability, applying each department's visit-type rules.
• Bilingual conversations: Handling English and Spanish in the same assistant, with language detected from the caller's first response.
• Context-rich handoff: Transferring calls to patient access staff with the verified identity, request type, and conversation summary on the agent desktop.
• Safety & escalation rules: Detecting urgent symptoms or clinical questions and routing them straight to nurse triage or emergency guidance instead of automation.
• Call analytics dashboard: Reporting on call reasons, containment, transfers, and drop-off points so operations teams can see where demand comes from.

The key focus was not simply answering calls with a machine, but creating a voice AI capability that gets every caller to a resolution, or to the right person, faster.

The Impact

The implementation helped the health network move from a menu-driven line, where nearly every call waited for an agent, toward a model where voice AI completes routine requests. Staff now spend their time on the calls that need them. The solution supported:

• Shorter wait times during peak call periods.
• Fewer abandoned calls.
• Routine scheduling requests resolved without an agent.
• Faster handling of transferred calls with verified callers and summaries.
• Better phone access for Spanish-speaking patients.
• A foundation for broader conversational AI across patient access.

By bringing together speech recognition, LLM understanding, EHR integration, and careful escalation design, InnovationM showed how conversational AI development services and AI agent development can turn a busy phone line into a faster, more predictable route to care.