Acuity Health 如何透過 LiveKit 語音智能體改善患者服務可及性

Calls fully contained by AI agents

80%

Typical appointment booking time

<1 min

Booking calls that end with a “thank you”

81%

In healthcare, administrative bottlenecks aren't just an efficiency problem. A patient who hangs up on hold, or can't reschedule an appointment, can miss timely medical care. Meanwhile, frontline staff spend a significant share of their time answering calls, booking appointments, and verifying insurance eligibility.

Acuity Health solves both sides of this problem for hospitals and subspecialty healthcare practices. With LiveKit at the core, its team applies deep expertise in voice system design and medical workflows to deploy voice agents that automate core patient access tasks.

Practices as unique as their patients#

High-performing solutions for medical practices can't be one-size-fits-all. Each hospital or clinical network runs its own tools and procedures, and every specialty serves patient demographics with different channel preferences, comfort with technology, and familiarity with the insurance system. From an agent design perspective, it's a minefield of edge cases to account for.

Acuity Health takes a boutique approach to this complexity. It works closely with client organizations to understand their objectives and map complex workflows, then builds a highly customized agent for each deployment. That means agents that handle appointment booking, insurance eligibility triage, and provider notes with confidence, at scale, regardless of practice or requirements. The catch: this only works if Acuity Health can build high-quality, bespoke agents quickly — which is where LiveKit comes in.

An early false start#

Acuity Health initially built agents on a popular low-code provider, not for lack of technical skill but to move fast. The flaws showed up quickly. The low-code platform was rigid: any scenario it hadn't anticipated meant a time-consuming workaround. It was also a black box in critical ways — offering no visibility into an agent's context window made debugging nearly impossible.

The team looked to the provider for help working through these limitations, but support options were limited. Without a radical change, they wouldn't be able to deliver. After researching alternatives and talking to trusted colleagues, they rearchitected their approach around LiveKit and never looked back.

Taking a new approach#

Switching to LiveKit immediately improved both the team's workflow and the quality of the agents it could build. Using coding agents, the team built a functional prototype in one week and moved it to dev within a month, and the process has only gotten faster since. Building agents as code lets the team customize every aspect of an agent to a client's needs without compromises or clunky workarounds.

For QA, Acuity Health relies heavily on LiveKit Simulations to pressure-test agent behavior against real-world scenarios. Instead of testing only ideal cases, the team simulates interactions based on patient profiles — a younger caller who knows they're talking to AI and wants to move fast, an older caller who can't find their member ID, and everything in between — to surface edge cases that would otherwise only turn up after poring over transcripts from real patient calls.

Today, the Simulations suite runs before every change ships, confirming nothing breaks and agent quality holds as the team iterates. That makes shipping faster and opens the door to further efficiency gains. Reviewing simulation runs against the new-patient intake prompt, CEO Kyle Schechtman noticed the agent asking for the patient's name and callback number more than once. The team removed the overlapping instructions and consolidated the clarification into a single review at the end of the call. The prompt went from about 48 lines to 39, with roughly 24% fewer words, and still passed the same scenarios.

"Patient calls rarely follow a script. We serve a wide range of patients with different communication styles, needs, and expectations, and our agents need to adapt to each of them. That variety creates edge cases that are easy to overlook. LiveKit Simulations helps us catch them before they reach production, so we can ship releases with a lot more confidence."

Chase Fagen, CTO, Acuity Health

Acuity Health's agents typically run on LiveKit's optimized Gemma 3 27B model on LiveKit Inference, paired with a turn-detection model tuned to the nuances of each patient population. They support both voice and text, and integrate with call center tools, task management software, analytics providers, and EMR systems to execute complex, autonomous workflows. Acuity's LiveKit agents fully contain 80% of calls across all use cases, with most bookings completed in under a minute, and 81% of booking calls end in a "thank you" even when the caller knows they're speaking to AI.

"Building agents that are smart enough to make a real impact can be daunting, but LiveKit makes the process far easier. They give us the tools and control we need to tailor each agent to each clinic's needs, and with coding agents there's no reason to give that up by handcuffing yourself to a limited solution. In today's world, 'low code' really just means 'low customization.'"

Kyle Schechtman, Co-founder & CEO, Acuity Health

Continuous improvement is core to how Acuity Health builds. From expanding its agents' use cases to ongoing fine-tuning, LiveKit gives the team the tools to build with confidence, iterate quickly, and keep its custom solutions ahead of generic competition.