Healthcare call center AI changes one thing first. Routine calls, the reschedules, cancellations, new bookings and refill requests, stop waiting in the queue, because an AI agent answers them on the first ring and completes them in the EHR. What is left for human agents is longer, harder and more valuable, so staffing models, handle time targets and quality review all need adjusting. Clinical triage does not move: it stays with nurses.
A healthcare call center is a shared team, often called a patient access center or central scheduling, that answers calls for many clinics or locations of a medical group or health system. It typically runs separate queues for scheduling, pharmacy and refills, billing and nurse triage, and it is measured on how fast calls are answered and how many callers hang up first.
Why central scheduling call centers struggle
Consolidating clinic phones into one center is supposed to smooth out demand. It often does not, and one of the clearest public accounts of why comes from the VA Office of Inspector General.
The Veterans Health Administration expects call centers to answer at least 95 percent of calls before the caller hangs up, an abandonment rate of 5 percent or less, and to answer at least 80 percent of calls within 30 seconds. Reviewing one medical center's call center over three months in 2023, the OIG found it answered about 94,400 of 135,600 calls, a 30 percent abandonment rate, and picked up only 22 percent of answered calls within 30 seconds. Waits averaged about seven minutes. It had averaged 29 staff on the phones against an estimated need of 53.
The detail that matters for anyone planning a center is when it broke. Abandonment peaked in the first hour after the 7am opening and during lunch and break periods, when average waits rose from about three minutes to over ten. After the center was folded into a regional contact center in 2024, abandonment there was 33 percent. Consolidation alone did not fix it, because the peaks are built into how patients call.
What healthcare call center AI changes first
An AI agent does not have a first hour problem. It answers every call on the first ring, many at once, so the 7am wave of reschedules does not queue behind the person trying to book a new patient visit. Four practical changes follow.
- Routing by intent instead of by phone menu. The caller says what they need, the agent identifies the patient and location, and either completes the request or sends it to the right queue: scheduling, refills, billing, nurse triage or a specific clinic
- Transfers with context. The human agent receives who is calling, their date of birth and what they asked for, so the call does not restart from nothing
- Overflow and after hours coverage for routine calls, without a second vendor taking messages for the morning
- A transcript and summary for every call, so quality review covers all calls instead of a small sample per agent each month
For MedPhone specifically, completing a call means identifying the patient against the EHR, booking against real provider availability with the practice's appointment types and rules, rescheduling or cancelling with a reason, adding to a waitlist where configured, taking refill requests into the provider inbox as a task for a clinician to approve, and writing a call summary to the chart. At Synergy Medical, a family medicine practice on athenahealth, 65 to 75 percent of calls are handled without reaching staff.
Healthcare call center AI triage: where the line sits
Triage is the most searched phrase in this category and the one where I would be strictest. An AI agent should not triage symptoms. Telephone triage is clinical work, done by licensed nurses using published protocols, and the evidence on automated symptom assessment is a reason for caution: a 2015 BMJ audit of 23 online symptom checkers found appropriate triage advice in 57 percent of standardized cases.
What the agent can do is make the nurse queue work better.
- Emergency language stops the call and the caller is told to hang up and dial 911
- Any symptom question goes to the nurse triage queue with a summary, and never gets booked as routine
- A rule may raise a call's urgency and nothing may lower it. A rule that downgrades a call because of what the caller said has become clinical judgment
- With routine calls removed from shared queues, callers with symptoms reach a nurse sooner
What it changes for the workforce
The calls that reach people after an AI agent goes live are the ones that need a person: complex referrals, upset patients, unusual scheduling, anything clinical. Average handle time goes up as the short calls leave. Centers usually set a handle time goal; the one the OIG reviewed used six minutes. If you keep that target after the easy calls are gone, you will read a better operation as a worse one. Reset handle time targets by queue, then rerun your staffing model with the new volume and the new handle time together.
Most groups use the freed hours rather than cutting roles. The usual candidates are referral scheduling, waitlist backfill, complex new patient registration, and calls about prior authorization and billing that genuinely need a person. Some groups will hire less as volume grows. Say that openly to the team; they will work it out anyway.
Outbound calling is a separate decision. The FCC ruled in February 2024 that AI generated voices count as artificial voices under the Telephone Consumer Protection Act, so those calls need the called party's prior express consent unless an emergency purpose or exemption applies. Reminder and recall campaigns are outbound work, separate from answering inbound calls; ask any vendor what a plan includes, and take the consent question to counsel.
What should a healthcare call center measure?
Keep your existing service metrics, split them by AI and human queues, and add the ones that tell you whether the agent is doing the job properly.
| Metric | What it tells you | What to watch for |
|---|---|---|
| Abandonment rate | Callers who hang up before reaching anyone | Track the human queues separately once routine calls move |
| Service level | Share answered within your target, such as 30 seconds | Should improve at the 7am and lunch peaks first |
| Containment | Calls completed without a person | By call type; a single blended figure hides problems |
| Transfer reasons | Why calls went to people | Repeated reasons are configuration fixes waiting to happen |
| Wrong outcomes | Bookings in the wrong slot, missed escalations | Review every one weekly; this is the safety metric |
| Time to nurse | Wait for symptom calls to reach triage | Should fall as shared queues empty |
| Handle time by queue | Length of the calls people still take | Expect it to rise; reset targets |
How to roll it out across locations
Start with one queue, not one switch. Reschedules and cancellations for two or three clinics, or after hours across all of them, are the usual first slices. Review transcripts, transfer reasons and wrong outcomes weekly for the first month, then widen. Containment climbs for several weeks while call flows are tuned, so judge it on week eight, not week one.
Most of the work is configuration, not technology. Each location has its own providers, appointment types, hours and rules, and each one has to be set up and tested. A single EHR instance across sites makes this much easier.
What MedPhone has and has not done here: it is live in production on athenahealth at a single family medicine practice. We have not run a health system central scheduling center in production, and on eClinicalWorks, ModMed, NextGen and Dentrix core scheduling is covered or in active development, with capabilities varying, so ask what is live on yours. Pricing suits groups: MedPhone bills on the call minutes the organization uses, not per seat or per site, so adding a location does not add a license, and groups above the top published tier move onto an allowance sized to their real volume.
See how MedPhone routes a call
Sources
- VA Office of Inspector General, Atlanta Call Center Staffing and Operational Challenges Provide Lessons for the New VISN 7 Clinical Contact Center, January 2025
- FCC, Declaratory Ruling on AI generated voices in robocalls, February 2024
- Semigran et al., Evaluation of symptom checkers for self diagnosis and triage, BMJ, 2015
Questions people ask
How is AI used in healthcare call centers?
Mainly to answer routine calls on the first ring and complete them in the EHR: booking, rescheduling, cancelling and taking refill requests. Calls that need a person are routed to the right queue with a summary, and every call gets a transcript for quality review.
Can AI triage calls in a healthcare call center?
It should not triage symptoms. Clinical triage is nurse work using established protocols. An AI agent can route symptom calls to the nurse queue with context, give the 911 instruction for emergencies, and hand urgent calls to a nurse faster. It should never decide a call is less urgent than the caller made it sound.
Will AI replace healthcare call centers?
No. It takes the routine calls, which are often most of the volume, and leaves the complex, emotional and clinical calls with people. Most groups redeploy the freed hours to referrals, backfilling cancellations and harder registration work rather than closing the center.
What metrics should a healthcare call center track with AI?
Abandonment rate and service level, split by AI and human queues, plus containment by call type, transfer reasons, wrong outcomes and time to reach a nurse. Expect handle time on the human queues to rise, and reset those targets rather than reading it as decline.
What are the requirements for healthcare call center AI software?
At minimum: read and write access to your EHR scheduling, configurable rules per location and appointment type, warm transfers with context, transcripts and audit logs, a signed business associate agreement, and pricing that does not multiply by seats or sites. Ask to see it run on your EHR, not a demo environment.
Does AI work with an eClinicalWorks call center?
It depends on the vendor and what it has built on eClinicalWorks. For MedPhone, eClinicalWorks core scheduling is covered or in active development and capabilities vary, so ask on a demo exactly what is live, and on which workflows, before planning a rollout.
Written by
Agni Patel is the founder and CEO of MedPhone, which builds an AI phone agent for medical practices. MedPhone has not yet run a health system call center in production, and this post says so where it matters.