> ## Documentation Index
> Fetch the complete documentation index at: https://blogs.persistence.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Medical Call Center: Which Patient Calls AI Can Handle

> How a medical call center works at scale, the service levels that matter, which patient calls AI can finish on its own, and how to add AI to your queues.

**Product**BUYER GUIDE·8 min read

# Medical Call Center: Which Patient Calls AI Can Handle

November 12, 2026

[← Back to Blog](https://blogs.persistence.dev/index.md)On this page
[What is a medical call center?](#what-is-a-medical-call-center)
[Which service levels should a medical call center hit?](#which-service-levels-should-a-medical-call-center-hit)
[Which patient calls can AI handle?](#which-patient-calls-can-ai-handle)
[What results have health systems published?](#what-results-have-health-systems-published)
[How do you add AI to an existing call center?](#how-do-you-add-ai-to-an-existing-call-center)
[What does a medical call center cost?](#what-does-a-medical-call-center-cost)
[How this works in Persistence](#how-this-works-in-persistence)
[Who Persistence is best for](#who-persistence-is-best-for)
[Related resources](#related-resources)
[FAQ](#frequently-asked-questions)

[Persistence](https://persistence.dev) / [Blog](https://blogs.persistence.dev/index.md) / Product

![Illustration of a medical call center with patient calls routed between AI and agent queues](https://mintcdn.com/persistence-76f2dd8d/nAaSjoMX-8vROkmj/images/blog/medical-call-center/article.webp?fit=max\&auto=format\&n=nAaSjoMX-8vROkmj\&q=85\&s=3dfe91606d756c6d7ce8343ce93d023b)

Key takeaways

<ul>
  <li>A medical call center is the centralized team that answers patient calls for a health system or large group: scheduling, pharmacy and refill lines, nurse triage, billing and portal help.</li>
  <li>The one published, audited standard comes from the VA: no more than 5% of calls abandoned, and 80% of scheduling calls answered within 30 seconds. In a 2025 audit, only 3 of 17 VA contact centers met both scheduling targets.</li>
  <li>Most volume is routine. Across about 45 million VA calls in FY2024, 54% were scheduling and 30% pharmacy, the two queues where AI can finish the most calls on its own.</li>
  <li>Health systems that add AI usually start with low-risk queues (portal help, routing, directions, confirmations), measure containment against a modest goal, then move into scheduling.</li>
</ul>

**Try it on your own calls.** Build a voice agent and test it before it takes real calls. Start with \$10 in free credits. **[Start for free →](https://persistence.dev)**

A medical call center answers patient calls centrally, in queues, against service levels like speed to answer and abandonment rate. AI voice agents can now finish many routine calls end to end: scheduling within your rules, rescheduling, confirmations, directions, portal help, refill requests routed to the right queue, and billing balances with a payment link. Clinical triage, billing disputes and upset patients stay with agents and nurses. The usual result is fewer abandoned calls, because patients stop waiting in a queue for routine requests.

Every patient access leader knows the Monday 8 a.m. spike. The scheduling queue fills, speed to answer climbs, patients hang up and call back, and the abandoned calls show up in the next morning's report. Adding agents helps until they leave, and time-to-proficiency for a new scheduler is measured in weeks.

This guide covers how a medical call center is organized, which service levels actually have a published basis, which call types AI can finish versus which need a person, what results health systems have published, and how to add AI in front of the queues you already run.

## What is a medical call center?

A medical call center, also called a healthcare call center, patient access center or centralized scheduling center, answers patient calls for many clinics or sites from one place. It usually runs several queues:

<ul>
  <li>**Scheduling:** new and follow-up appointments, reschedules, cancellations, referrals waiting to be scheduled.</li>
  <li>**Pharmacy and refills:** refill requests and prescription status.</li>
  <li>**Nurse triage:** symptom calls handled by nurses using protocols.</li>
  <li>**Billing:** balances, payments, statements, financial assistance.</li>
  <li>**Portal and help desk:** MyChart and patient portal logins, password resets.</li>
  <li>**Outbound:** confirmations, reminders, recall, waitlist and referral outreach.</li>
</ul>

Calls arrive through an IVR, are routed by an ACD to the right skill group, and are measured by queue. It differs from an answering service, which mostly takes messages after hours; see our [medical office answering service](https://blogs.persistence.dev/blog/medical-office-answering-service.md) guide for that.

## Which service levels should a medical call center hit?

Most benchmark figures online are unsourced or come from vendor surveys. The one public, audited set of standards is the Veterans Health Administration's, from a VA Office of Inspector General review published in October 2025 ([VA OIG](https://www.vaoig.gov/sites/default/files/reports/2025-11/vaoig-25-00228-214_final.pdf)):

<table>
  <thead>
    <tr>
      <th>Measure</th>
      <th>VHA standard</th>
      <th>What the audit found (FY2024)</th>
    </tr>
  </thead>

  <tbody>
    <tr><td>Abandonment rate</td><td>No more than 5%</td><td>Scheduling about 11%; pharmacy about 6%</td></tr>
    <tr><td>Speed to answer, scheduling and pharmacy</td><td>80% answered within 30 seconds</td><td>Missed for scheduling and pharmacy</td></tr>
    <tr><td>Speed to answer, clinical triage</td><td>Within 120 seconds</td><td>About 82% met</td></tr>
    <tr><td>After-hours scheduling</td><td>Not set</td><td>About 11% of after-hours scheduling calls answered at the centers reviewed</td></tr>
  </tbody>
</table>

Only 3 of 17 centers met both scheduling standards. The VA isn't a typical health system, but it is the largest public dataset available, and the targets are a reasonable bar for any centralized scheduling team.

Other measures most centers track:

<ul>
  <li>**ASA (average speed of answer)** and **service level** (for example 80/30: 80% answered in 30 seconds).</li>
  <li>**Abandonment rate** (usually excluding calls under 5 seconds).</li>
  <li>**First-call resolution (FCR):** the share of calls resolved without a callback or transfer.</li>
  <li>**AHT (average handle time):** talk, hold and wrap-up time.</li>
  <li>**Transfer rate** and **occupancy.**</li>
  <li>**Containment rate,** once AI is in place: the share of calls completed without an agent.</li>
</ul>

Vendor surveys give rough ranges for the rest. In a 2023 survey of 200 US healthcare leaders by Hyro, which sells AI for health system call centers, centers reported about 16% abandonment, 52% first-call resolution, 6.6 minutes average handle time and about \$4.90 per call ([Hyro](https://landing.hyro.ai/hubfs/Hyro%20-%20The%20State%20of%20Healthcare%20Call%20Centers%202023%20Report.pdf)). Use them as a sense of scale, not a target.

## Which patient calls can AI handle?

<table>
  <thead>
    <tr>
      <th>Call type</th>
      <th>AI can finish on its own</th>
      <th>Goes to an agent or nurse</th>
    </tr>
  </thead>

  <tbody>
    <tr><td>New and follow-up scheduling</td><td>Books within your visit-type rules and templates</td><td>Complex procedures, multi-provider visits, requests outside the rules</td></tr>
    <tr><td>Reschedule and cancel</td><td>Yes, and offers the slot to the waitlist</td><td>No</td></tr>
    <tr><td>Confirmations and reminders (outbound)</td><td>Yes</td><td>No</td></tr>
    <tr><td>Directions, hours, prep instructions</td><td>Yes</td><td>No</td></tr>
    <tr><td>Portal and password help</td><td>Yes</td><td>Account security issues</td></tr>
    <tr><td>Patient ID and insurance</td><td>Verifies identity, captures insurance details</td><td>Coverage decisions</td></tr>
    <tr><td>Refill requests</td><td>Takes the request and routes it to the pharmacy or clinic queue</td><td>Every clinical decision</td></tr>
    <tr><td>Billing</td><td>Balance lookup, payment link, payment plan intake</td><td>Disputes, financial assistance decisions</td></tr>
    <tr><td>Referral and prior auth status</td><td>Reads the status back</td><td>Problems that need someone to act</td></tr>
    <tr><td>Routing (replacing "press 1")</td><td>Understands the request and routes to the right queue</td><td>No</td></tr>
    <tr><td>Symptoms and clinical questions</td><td>Never answers them; routes to triage</td><td>Always, to a nurse using protocols</td></tr>
    <tr><td>Upset or confused callers</td><td>Transfers right away</td><td>Always</td></tr>
  </tbody>
</table>

The rule of thumb: **AI finishes the routine, agents own the exceptions, nurses own the clinical.** When AI hands off, it should pass along why the patient called, whether identity is verified and what's already done, so the agent doesn't start over. See [when a voice agent should hand off to a human](https://blogs.persistence.dev/blog/voice-ai-human-handoff-escalation.md).

## What results have health systems published?

All of these come from projects with a vendor partner, so read them as what's possible, not a guarantee:

<ul>
  <li>**Tampa General Hospital** (with Hyro) reported ambulatory abandonment down 56%, wait times down 58% and daily scheduled appointments up 17%, starting with routing, confirmations, directions, refills and billing questions before adding scheduling ([Healthcare IT News](https://www.healthcareitnews.com/news/tampa-generals-call-center-ai-drives-roi)).</li>
  <li>**UC San Diego Health** (with Amazon Connect) used AI to verify patients before they reach an agent, saving about a minute per call and redirecting 630 staff hours a week; abandonment fell 30% ([CX Today](https://www.cxtoday.com/contact-center/amazon-connect-health-healthcare-contact-center-ai/), March 2026).</li>
  <li>**Relatient**, which sells scheduling software, reported across its orthopedic customers that its voice AI answered 62% of inbound calls and completed more than 60% of cancellation calls ([Relatient](https://www.relatient.com/resources/press-release/relatient-voice-ai-deflects-more-than-60-of-routine-appointment-scheduling-calls/), August 2025).</li>
</ul>

The pattern is consistent: start with non-clinical, high-volume queues, set a modest containment goal, then expand. Most of the abandonment drop comes from patients no longer waiting in a queue for a routine request.

## How do you add AI to an existing call center?

You don't need to replace your contact center platform. Most teams put an AI voice agent in front of, or alongside, the queues they already run on Genesys, Five9, NICE, Talkdesk or their phone system.

<ol>
  <li>**Pick one queue.** Portal help, directions or confirmations are the safest start. Scheduling comes next.</li>
  <li>**Pull your call reasons** from disposition codes or IVR data, and estimate the share AI could finish.</li>
  <li>**Connect the systems the queue needs:** the scheduling system for bookings, and your ACD for warm transfers to the right skill group.</li>
  <li>**Write the handoff rules:** what transfers live, what goes to triage, and what happens after hours.</li>
  <li>**Test before go-live** with real call scenarios: a new patient, a reschedule, a refill, a billing question, a caller with symptoms, a caller who wants a person, a caller who speaks another language. See our [voice agent testing guide](https://blogs.persistence.dev/blog/voice-agent-testing-and-qa.md).</li>
  <li>**Measure against a baseline:** abandonment, speed to answer, containment, transfers and appointments booked, by queue and by hour.</li>
  <li>**Expand** to the next queue and to after-hours, where calls are otherwise often unanswered.</li>
</ol>

## What does a medical call center cost?

Call centers are priced per agent, not per minute:

<ul>
  <li>**Outsourced agents:** published rates for patient-facing agents start around $10-13 an hour (<a href="https://helpsquad.com/healthcare/call-center-teams/">HelpSquad</a>, checked November 2026); US-based agents cost considerably more, and most outsourcers quote case by case.</li>     <li><strong>Contact center software:</strong> Genesys Cloud lists $75-240 per user per month, with AI virtual agent usage billed on top ([Genesys](https://www.genesys.com/pricing)); Five9 lists $119-159 per seat per month with a 50-seat minimum and AI quoted separately (<a href="https://www.five9.com/pricing">Five9</a>).</li>     <li><strong>AI voice agents:</strong> usually per minute. Persistence is from $0.10 per minute, all in, so a 3-minute scheduling call starts from \$0.30 ([pricing](https://persistence.dev/pricing/)).</li>
</ul>

*Prices checked November 2026 on each provider's public pricing page.*

The useful comparison is cost per completed call by queue: what it costs today with agents, and what it costs when AI finishes the routine share and agents take the rest.

## How this works in Persistence

Persistence is a platform for building AI voice agents that answer and make phone calls. For a medical call center:

**What it can handle, inbound and outbound:** scheduling, rescheduling and cancellations, new patient intake and insurance capture, identity verification, confirmations and reminders, waitlist fills, annual recall and care-gap outreach, referral follow-up, refill requests routed to the right queue, billing balances with payment links, portal and password help, directions and prep instructions, after-hours and overflow coverage, post-visit follow-up and surveys. Anything with an API can be connected, so the agent can look up and update the systems your team already uses.

<ul>
  <li>**Build** agents per queue, from a plain description or a step-by-step call flow: identity check, scheduling rules, FAQs, routing and transfers. [Build an agent](https://persistence.dev/features/build/)</li>
  <li>**Book into your systems:** Persistence connects to eRAD, Epic, athenahealth, eClinicalWorks, NextGen, ModMed, AdvancedMD, Tebra, Fujifilm Synapse, RamSoft, Konica Minolta Exa and other practice management, EHR and RIS systems.</li>
  <li>**Warm transfers** to the right queue in your existing contact center, with a summary passed along. Persistence connects to Genesys, Five9, NICE, Talkdesk and other contact center platforms.</li>
  <li>**Outbound campaigns** for confirmations, reminders, recall and waitlist fills.</li>
  <li>**Test before patients call:** simulated calls for each queue's scenarios, with a pass rate before go-live. [Testing](https://persistence.dev/features/testing/)</li>
  <li>**See every call:** recordings, transcripts, summaries and outcomes by queue. [Monitoring](https://persistence.dev/features/monitoring/)</li>
  <li>**Scale and languages:** as many calls at once as you need (no cap on Enterprise), so the Monday spike never builds a queue; 40+ languages, switching language mid-call, with no per-language surcharge; 99.99% uptime; HIPAA and BAA on Enterprise.</li>
</ul>

**In practice:** a US breast imaging network answers its patient calls with Persistence, more than 26,000 calls a day in English and Spanish. Its booking rate went up 85% as patients got a better experience on the phone.

See [Persistence for healthcare](https://persistence.dev/solutions/healthcare/) and [pricing](https://persistence.dev/pricing/).

## Who Persistence is best for

Persistence fits health systems, large groups and imaging networks with high-volume scheduling and routine queues, that want AI to finish those calls inside their own systems and hand everything else to their existing agents. It suits teams that want to test each queue before go-live and see the outcome of every call.

If your volume is mostly clinical triage, a nurse triage service is the better fit; AI can still take the routine calls in front of it.

## Related resources

Keep going with [healthcare contact center automation](https://blogs.persistence.dev/blog/healthcare-contact-center-automation.md), [radiology scheduling](https://blogs.persistence.dev/blog/radiology-scheduling.md), the [virtual medical receptionist](https://blogs.persistence.dev/blog/virtual-medical-receptionist.md) guide and our hub guide to [contact center automation](https://blogs.persistence.dev/blog/contact-center-automation.md).

## Frequently asked questions

> [accordion] What is a medical call center?
>
> A centralized team that answers patient calls for a health system or large group, usually in queues for scheduling, refills, nurse triage, billing and portal help.

> [accordion] What is a good abandonment rate for a healthcare call center?
>
> The VA's published standard is no more than 5%. In its 2025 audit, scheduling queues averaged about 11%, so many centers run well above 5%.

> [accordion] What is a good speed to answer?
>
> The VA standard is 80% of scheduling and pharmacy calls answered within 30 seconds, and triage within 120 seconds.

> [accordion] Can AI replace medical call center agents?
>
> No. AI can finish routine calls like scheduling, reschedules, directions and portal help, which lets agents focus on complex calls, and nurses on clinical ones.

> [accordion] Does AI work with our existing contact center platform?
>
> Yes. Persistence connects to Genesys, Five9, NICE, Talkdesk and other contact center platforms, so the AI agent answers first and transfers calls to the right queue with a summary.

## Start for free

> [card] [Build reliable voice AI with Persistence](https://persistence.dev)
>
> Design, test, and deploy production-ready voice agents.

## Continue reading

> [card] [Medical Office Answering Service: AI vs Live, Costs, HIPAA](https://blogs.persistence.dev/blog/medical-office-answering-service.md)

> [card] [Healthcare Contact Center Automation That Improves Outcomes](https://blogs.persistence.dev/blog/healthcare-contact-center-automation.md)

> [card] [Radiology Scheduling: How to Automate Imaging Patient Calls](https://blogs.persistence.dev/blog/radiology-scheduling.md)

> [card] [Virtual Medical Receptionist: Remote Staff vs AI, Costs](https://blogs.persistence.dev/blog/virtual-medical-receptionist.md)

**Persistence**Automate your calls. Connect with us.**Product**[Home](https://persistence.dev)[Pricing](https://persistence.dev/pricing/)**Solutions**[All solutions](https://persistence.dev/solutions/)**Feature**[All features](https://persistence.dev/feature/)**Resources**[Blog](https://blogs.persistence.dev/index.md)[Docs](https://docs.persistence.dev)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.