让账算得过来：把语音 AI 接入 ACCESS 项目

Vapicon 2026 is coming

Those working in chronic disease management know that behavior modification and care plan adherence are the holy grail to achieving clinical outcomes. Technology has played a meaningful role in supporting chronic disease management by engaging patients at the right time. To date, automation has focused on text reminders, dashboards, chatbots, and wearable integrations. The platforms work.

The problem isn't that text and chat don't work; it's that this isn't what patients want for chronic care conversations, and it isn't what moves the outcomes CMS reimburses against. 65% of patients still choose voice for complex health conversations. Historically, vendors built solutions over chat or text because humans don't scale, and voice automation didn't work. The growth of voice AI in terms of natural speech and scale means that both constraints are no longer a factor, and the data shows that voice is what your patient actually engages with.

July is two months away. The voice layer is still open to claim.

What patients actually want (and what actually moves them)

When you ask patients specifically about healthcare communication, the data is clear. Hiya's State of the Call 2025 report, which surveys more than 12,000 consumers annually across the US, UK, Canada, France, Germany, and Spain, found that 42% of consumers prefer voice over other channels when communicating with their healthcare provider. Voice was the most preferred ahead of email and text. And the trend is continuing to move towards voice: voice preference in healthcare jumped from 33% the year prior to 42% in 2025. The same pattern holds in banking (33% prefer voice) and credit card communications (32%). When the stakes get higher, people reach for voice.

But it's not just preference; it is also outcomes. Voice drives real engagement. The clearest experimental evidence comes from a Cornell research program led by Vanessa Bohns: in their original 2017 study published in the Journal of Experimental Social Psychology, identical requests made face-to-face had a 71.5% compliance rate. The same requests made by email got 2.1%. A 2022 follow-up in Social Psychological and Personality Science extended the work to phone and Zoom (running nearly 1,500 actual help requests) and found that phone calls were statistically as effective as video and significantly more effective than email. Text-based asks are likely to be ignored. The effectiveness compounds with real, human-like conversation: adaptive flow, open-ended capture, and real-time follow-up. Not press-1-to-confirm robocalls.

So if voice is what patients want and what works, why does the industry default to text?

With voice AI, the math finally works

The reason text won the last decade of healthcare engagement wasn't that anyone thought it drove better outcomes; it was the cost. It was that humans were never going to be viable at the scale chronic care requires, and vendors chose the channel that fit the budget rather than the one that fit the patient. But now, with the growth in voice agents' ability to perform natural, personalized calls at scale, for the first time, the channel patients prefer for healthcare is also an economically viable one.

For ACCESS specifically, the gap between text and voice widens further because the population the program is built around is less digitally connected than most. A widely-cited JAMA Internal Medicine study found that 41% of Medicare beneficiaries don't have a smartphone with a wireless data plan, and 26% don't have either a smartphone or a computer with high-speed internet. Pew Research reports that smartphone ownership among Americans 65+, which is the primary audience for ACCESS, is 61% with about a quarter not using the internet at all.

If you're building ACCESS workflows for text and chat, you're building for a channel patients didn't pick, a channel this population can't always access, and a channel the research shows doesn't move behavior the way voice does.

Three ACCESS workflows already running in production

The good news is that voice for chronic care isn't speculative. The workflow patterns ACCESS requires are running in production today across healthcare and other industries; they just haven't been re-pointed at chronic care yet.

Three patterns map directly to what ACCESS asks of you.

Patient enrollment and initial assessment

ACCESS vendors have a volume problem. A patient is referred, and now you have to enroll them, verify they qualify for the program, capture their condition history and medication list, and route them to the right care team. Multiplied by every patient on every panel, this is the highest-volume conversational workload in the program's first six months.

For their healthcare insurance business, Spring Venture Group built a similar pattern to automate inbound Medicare lead qualification. Their workflow:

An inbound call arrives, a Medicare prospect or a transfer from an upstream lead-gen agency; The Vapi voice agent answers and runs the qualification flow: verify Medicare eligibility, capture ZIP code (with five-digit validation), capture name and date of birth accurately, and assess fit for SVG's Medicare Advantage plans; The data is assessed against qualification criteria as it's collected; Qualified leads get transferred to a licensed sales agent with the full call context. Unqualified leads get a polite explanation and call termination; Structured qualification data is pushed into their backend systems in real time

You don't need to build a 50-person enrollment team or a custom voice infrastructure stack. Four engineers shipped this in one week, and their transfer-to-close rate grew by 3%, a massive figure when you’re operating at SVG’s scale. The combined savings and new revenue totaled over $5.8M.

Outbound scheduling triggered by a clinical signal

Imagine a patient's blood glucose readings have trended high for 4 straight days. Their wearable pushes that data into the care platform, which, after interpreting the readings as an emergency, triggers a voice agent to call the patient and set up an appointment. This gets the patient on the schedule before the trend turns into an ER admission.

UnityAI built this pattern for outbound appointment scheduling on top of Vapi. Their workflow:

A referring physician orders an imaging study, let’s say, an MRI of the knee; The order flows from the EHR into the radiology clinic's scheduling system; This order update triggers UnityAI's voice agent to call the patient: "Hey [Name], you saw your orthopedic surgeon. He's ordered an MRI of your knee. Can I get you on Wednesday at 2pm?"; The agent handles healthcare-specific nuances mid-conversation, asking about metal implants, explaining prep requirements, and more; The booking is written back to the radiology clinic's scheduling system in real time

Vapi’s solution can help create the right integrations for availability checks across multiple clinic scheduling systems, patient record updates in whatever format the downstream system requires, and SOC 2, HIPAA, and PCI compliance, without UnityAI having to build any of it.

It took UnityAI about 1 week from concept to production. Today, they run 250,000+ scheduling calls per month with confirmation and show-up rates improving in the high double digits.

Symptom screening and patient-reported outcomes

PHQ-9 for depression. GAD-7 for anxiety. AUDIT for substance use. Disease-specific symptom batteries for diabetes, CKD, and hypertension. Deterministic instruments like these are a breeze to get up and running in the Vapi dashboard.

To support their labor marketplace operations, Instawork built this pattern for their worker qualification interviews. Their workflow:

A worker applies to join Instawork as a "Pro" through the mobile app; The worker either calls into the Vapi AI interviewer or receives an outbound call; The AI runs a structured qualification interview: skills, work history, availability, shift-type preferences; Responses are captured in a structured format and stored for the matching algorithm; The worker is approved, waitlisted, or routed to additional verification based on threshold logic

Instawork integrated Vapi as a React Native SDK directly into their mobile app, supporting full-featured phone calls and in-app voice from within the worker's existing experience. Custom tool calling submits structured interview data and triggers worker ratings. Dynamic variables personalize the interview flow per worker type and location.

This matters specifically for ACCESS because PHQ-9 scores and other patient-reported outcome measures are exactly the kind of data CMS reimburses against. If you're scaling your ability to capture PHQ-9, with automatic clinician escalation when scores cross a threshold, you're directly producing the engagement and outcome data that turns into payment.

Today, Instawork runs over a million minutes of voice screening per month on Vapi, a 50x throughput increase compared to their previous workflow. They’ve also seen 30% better candidate-to-placement matching alongside 85% lower cost per call.

The first-mover stakes

The vendors who deploy voice in the first 90 days set the engagement benchmark every other vendor gets measured against. If you're an ACCESS-approved vendor and you don't have voice in your stack yet, the conversation you need to have with your team is “how do I implement this before July?”

That's the conversation we're having every day with vendors building for July. If you're working on the same problem, come talk to us come talk to us.

We’re also planning a webinar on this topic for June, stay tuned for more information on how to register.

Table of Contents

Your next customer call may come from a personal agent. Don't turn it away

How AI voice agents tell the difference between a live person and a voicemail

Best AI voice agents for CX: how to evaluate them for customer experience

How to choose a conversational AI platform for enterprise development

How to reduce wait times with AI

The best voice models now sound nearly human

Built for the Ear: Designing Conversations for Voice

How we solved latency at Vapi

What Is RCS Messaging? The Complete Guide for Developers

Launching the Vapi for Creators Program

Vapi x Cartesia: Ultra-Realistic Voice AI with Sonic 2.0

Voice AI is eating the world

Start building