How are health systems using voice AI for patient access?

Voice AI handles the routine patient access calls. Patients who call a health system about scheduling, refills or billing get their request handled on the call, and urgent callers reach a person at once. The agent books and moves appointments in your EHR during the call, checks insurance where it's connected to your eligibility source, and routes requests with details attached. You measure it against the call volume and staff hours you report.

What patient access teams own, and where voice AI fits

Scheduling, often centralized

Many health systems run centralized scheduling, one access center booking for every clinic, so a patient makes one call however many service lines they need. The agent books and moves appointments in your EHR during the call, under each service line's rules, and access leaders watch measures such as third next available appointment to see whether capacity keeps up.

Pre-registration and registration

Registration goes faster when the details are already in. The agent collects the patient's demographics and insurance information on the call, so pre-registration starts complete and the front desk confirms rather than re-keys.

Insurance verification and eligibility

A visit booked against lapsed or wrong coverage turns into a denied claim. Teams verify eligibility and benefits before the visit. Where the agent is connected to your eligibility source, usually a verification vendor and sometimes your EHR, it checks insurance during the call and passes anything it can't confirm to your team with what it has already captured.

Prior authorization

Many procedures and scans need the payer's approval before they happen, and a missed authorization is one of the most common causes of denials. Prior authorization stays with your team. The agent takes the patient's question about it and routes it with the details attached, so your staff don't start the call from scratch.

Financial clearance and cost estimates

Financial clearance means the patient knows what they'll owe and how they'll pay before the visit. Your team gives the estimate, including the Good Faith Estimates the No Surprises Act requires for uninsured and self-pay patients. The agent explains statements, takes payments where your billing system connects and offers the payment plans your policy allows.

The link to revenue cycle management

Each of these steps feeds revenue cycle management. Clean scheduling, registration, eligibility and authorization at the front end mean fewer claim denials at the back end, and a phone line that answers quickly means fewer patients who give up and book elsewhere.

Why patient access is hard to staff

Access is hard because the phone line carries all of it at once. Hold times climb when call volume spikes, and staffing shortages make the gap worse: a patient who hangs up may not call back. Most access leaders track call abandonment, hold time, first-call resolution, registration accuracy and the share of claim denials that trace back to front-end errors.

Virtual receptionists and AI agents now take part of that load. A voice agent takes the routine share of calls, so those numbers stop depending on how many people are on shift.

One agent for health system patient access, across every service line

A health system's access center answers for many clinics, and each one has its own rules for visit types. The agent applies those rules per service line and per location, so a cardiology follow-up and a pediatric well visit each land where your schedulers would put them.

• Books and moves appointments in your EHR during the call.
• Captures refill requests and routes them into your refill workflow.
• Explains statements and takes payments where your billing system connects.
• Answers questions from your approved content.
• Confirms upcoming visits by voice or SMS, and rebooks on the same call.

A caller who says "I need to move my MRI, and I've got a question about the bill from my last visit" gets both handled on one call. The agent moves the imaging appointment within your rules, then explains the balance or passes the billing question to your team with the details attached.

The agent does not diagnose, give medical advice or independently assess clinical urgency. It follows routing rules defined by your clinical team and sends those callers to the right care team straight away. If a caller describes an emergency, the agent tells them to call 911 and transfers them to a person, following your approved emergency protocol. If you run a digital clinical triage tool, the agent can pass the caller's symptoms to it through its API and act on the disposition it returns, so the clinical decision stays in your tool.

Proof from patient lines that run at scale

SafeRide Health arranges non-emergency medical transport for Medicare and Medicaid members in all 50 states. SafeRide handles 1M calls a month with PolyAI, and its agent, Steven, is the first point of contact for 95% of them. The agent has saved 47,000 agent hours on authentication alone, time its representatives can spend on the calls that genuinely need a person, and the share of callers asking for a person fell from 55% to 2 to 3%.

"The fact that we can handle a million calls a month, reliably, with a great experience for our members is a real key differentiator for us," says Ben Salter, SafeRide's Chief Product Officer and co-founder.

Howard Brown Health, a federally qualified health center in Chicago, connected its agent to MyChart. Its planned next phase adds Epic, so patients can create, reschedule and cancel appointments through the agent.

For the business case, a Forrester Total Economic Impact study commissioned by PolyAI found a 391% return on investment over three years. The study models a composite organization across industries, so read it as a general benchmark, not a healthcare result.

What your CIO and IT team will ask

English calls can run on Dialog-RSN-1, PolyAI's own audio-native dialog model, with a median model first token under 300ms in live production. The platform handles 10,000 concurrent calls.

PolyAI holds SOC 2+ Type 2 and ISO/IEC 27001:2022, two recognized security standards that independently check how a vendor protects customer data. Patient data is protected under the HIPAA Security Rule, the US rules for safeguarding electronic health information, with controls examined within our SOC 2+ report. A business associate agreement (BAA), the contract setting how a vendor must handle patient data, is available for eligible healthcare deployments. Data handling, retention and model use are governed by your agreement and deployment configuration.

You need to know what the agent will say. It gives medical information only from content your team approves, and escalates everything else to a person. Five platform guardrails come with enterprise deployments, and your team can add up to five plain-language rules of its own for each project.

Beyond Epic, PolyAI connects to athenahealth, ModMed, NexHealth, Raintree and Brightree, so the agent works inside the systems you already run. Agents act in your systems through functions you control, built in Dialog Studio (formerly Agent Studio) or as code with the Agent Development Kit (ADK) in YAML and Python, and if a system has an API, PolyAI connects to it.

Change control your team owns

Your rules keep changing: a clinic opens, a provider moves, a visit runs longer. In Dialog Studio, every project has branches. Your team makes the change on a branch, tests it against simulated patient calls, publishes it to Live, and can roll it back. On SafeRide's line, call flow and business rule changes that once took weeks are now resolved in seconds.

Wren, the agent inside Dialog Studio, reviews live conversations, raises issues and drafts fixes on a branch with tests, so every proposed change arrives already tested. A person approves every change before it reaches a patient, which we think is the right default for any patient line.