This article was written for Greystone.Net by Jessica Levco, a freelance healthcare writer and event strategist. 

In healthcare, the phone never stops ringing. Patients are calling to schedule appointments. Resetting MyChart passwords. Finding providers. The list goes on and on. And most of the time, patients need a quick answer. 

But what if half of those calls could be resolved instantly, without patients waiting on hold or playing phone tag with scheduling staff?

At the Healthcare Interactive Conference (HCIC) in Las Vegas, Scott Waters, Chief Information and Technology Officer at Overlake Hospital and MultiCare Health System and Rom Cohen, Co-Founder and COO at Hyro, shared how Overlake Hospital teamed up with Hyro to deploy a patient-facing AI voice agent that cut calls by 50%.    

Overlake Hospital’s location in the Pacific Northwest — neighboring Microsoft, Google and other tech giants — creates unique patient expectations.

“The patient base that we serve has some of the highest expectations you could imagine as far as how they interact with their health system,” Waters says. “They want everything to be frictionless. They expect you to meet them where they’re at.”

Even as someone who’s worked in healthcare IT for nearly 20 years, Waters says he still struggles to navigate the complexity of referrals, imaging orders, specialty visits and follow-ups. This realization became a driving force behind their AI strategy.

Rather than diving into high-risk clinical AI applications, Overlake Hospital and Hyro developed a three-phase approach that starts with lower-risk administrative use cases.

• The “crawl” phase focuses on simple interactions, like chat solutions and basic call routing, that don’t pose any risk.
• The “walk” phase tackles administrative automation that demonstrates financial ROI while reducing patient wait times.
• The “run” phase (still aspirational), envisions clinical patient-facing AI tools that go beyond basic symptom checkers.

The MyChart Support Success Story

When choosing their first major AI deployment, Overlake Hospital needed something that would deliver quick wins across multiple dimensions.

“We needed something that we felt good with from a financial perspective and an experience perspective that we could deliver quickly,” Waters says.

They chose to deploy a voice agent for MyChart support calls, previously handled by a third-party vendor. The business case was compelling: They saved 50% of MyChart support contact center costs just by doing this.

The 50% deflection rate means the AI successfully resolves half of all calls without transferring to a human agent. But measuring success required going beyond simple metrics.

“Deflection and resolution are not the same thing,” Cohen says. “We track the complete patient journey, including whether patients call back within 72 hours about the same issue and whether they were routed to the right place.”

They also acknowledge that between 8-12% of patients, upon hearing AI, will immediately request to speak with a human, but that’s a percentage they anticipated. Cohen adds that this immediate “a human please” response is part of natural patient adoption curves, and that Overlake Hospital saw this rate decline steadily as patients gained trust in the AI agent’s ability to resolve their issue.

In the next three to six months, Overlake Hospital plans to deploy outbound voice agents to solve a common patient frustration: appointment scheduling phone tag. The AI tool would check patient preferences in MyChart and proactively reach out via voice or text with available appointment options, eliminating the back-and-forth that frustrates patients and staff alike.

Beyond MyChart support, Overlake Hospital and Hyro are already expanding into higher-value operational use cases, including scheduling and appointment management. These workflows represent some of the most time-consuming call types for frontline staff, and early indicators show that AI agents can streamline tasks such as rescheduling, provider matching and handling visit instructions.

Looking one to two years out, Waters imagines “digital employees” working alongside human staff.

“These digital employees wouldn’t handle individual tasks, they would operate with role-based permissions, managing complex workflows autonomously,” Cohen says. “A human employee isn’t going to want to pick up the phone 40,000 times to remind someone to get their seasonal vaccinations. But that’s something an AI agent can do.”

Waters’ guiding motto when it comes to AI is “bend, don’t break.” Innovation is going to happen with you or around you, so balancing speed with safety is important, he says. Apply rigorous governance to high-risk use cases, particularly anything involving clinical decision support. But for lower-risk administrative applications, move quickly and iterate.

“Don’t let perfection get in the way of good,” Cohen says. “AI agents, like human agents, will never be flawless, but waiting for perfection prevents organizations from realizing meaningful ROI and experience improvements today.”