Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarElevenLabs Blog→ original

AI Agent Calls Patients Before Appointments: 10% More Completed Screenings Than With IVR

Nearly one in four doctor appointments falls through: the average no-show rate is 23%. ElevenLabs showed how to build a voice agent that calls the day before a visit, verifies the patient's date of birth, and confirms, reschedules, or cancels the appointment directly in the electronic health record (EHR). At Everlywell, the share of completed screenings rose 10% compared with IVR calling, the "press 1" menu system. The clinic can then offer the freed-up slot to another patient in time.

AI-processed from ElevenLabs Blog; edited by Hamidun News
AI Agent Calls Patients Before Appointments: 10% More Completed Screenings Than With IVR
Source: ElevenLabs Blog. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceElevenLabs Blog ↗
Editor-in-Chief
Ovianta68% patient engagement, 83% fewer manual staff calls
Key metric #2Everlywell's Spanish-speaking participants converted 3.5 times more often
Reminder delivered live3% no-shows; voicemail: 24%; no answer: 39%

Nearly one in four doctor appointments does not happen: a review of 105 studies put the average no-show rate at 23%. In an August survey, 32% of medical groups reported more no-shows in 2026 than in 2025 (a year earlier, 27% said the same). Patients are reconsidering visits booked six weeks out because of record-high deductibles. The clinic loses revenue, and the patient misses a routine screening that later turns into an urgent visit. ElevenLabs has published a walkthrough of how to use the ElevenAgents platform to build an outbound voice agent that addresses this problem: it calls the day before the appointment, verifies the patient's identity, and immediately edits the booking in the electronic health record (EHR).

What it looks like in practice

Take a patient named Maria, booked for 14:00 tomorrow with her doctor. In the evening, the agent dials her number, asks for her date of birth, and checks it against the record. It then reads the current appointment and offers to confirm the visit, move it to an open slot in the doctor's schedule, or cancel it. If Maria reschedules, the agent changes that same EHR record, and the freed-up time goes back into the schedule while it can still be filled.

The agent also handles edge cases. If an answering machine picks up, it leaves a message. If a patient asks not to be called again, the refusal is recorded and no more calls are made. If a patient wants to talk to a person, the call is transferred to the front desk.

The authors explain why the usual methods fall short. A receptionist with a phone does not scale: she calls only during working hours, and she is the same person who answers incoming calls. SMS and e-mail are cheap but one-way, and they work worse for older patients and for those whose native language differs from the language of the message.

IVR, an automated calling system with a "press 1 to confirm" menu, cannot reschedule a visit, does not truly verify identity, and hands the conversation off poorly to a human if the patient asks a question. An outsourced call center provides a live conversation but is expensive, quality depends on the shift, and patient data is handled by people outside the organization.

The case for a live conversation comes from a Veterans Health Administration study: the no-show rate was 3% when the reminder reached the patient live, 24% when it went to voicemail, and 39% when nobody answered and no message was left.

Customer results

Everlywell uses ElevenAgents to remind participants about scheduled screenings. The share of completed screenings rose 10% compared with IVR, and conversion among Spanish-speaking participants was 3.5 times higher. Ovianta uses the agent for calls after a treatment plan has been proposed: patient engagement reached 68%, conversion to treatment rose 25%, and clinic staff made 83% fewer manual calls. These figures come from the vendor itself; the article includes no independent verification.

Two ways to launch

The first is a nightly batch. In the evening, a script pulls tomorrow's appointments from the EHR, turns each into a table row, and sends them to Batch Calling. Only the phone column is required.

The other columns (name, date, doctor, visit type, clinic name) become variables the agent uses in its prompt and during the conversation. Times in the table are written out in words, for example "ten a.m."

, because the agent reads them aloud and the speech synthesizer should not have to guess the format. The batch takes only a limited share of concurrent calls: the lesser of 50% of the workspace limit and 70% of the agent limit, so the inbound line does not sit idle. The script handles the retry policy: no more than three attempts per patient per week, and a confirmation, reschedule, cancellation, or refusal removes the person from the list.

The second is an event-triggered call from the EHR through the outbound calling API. It catches what the batch misses: same-day appointments, a move from 8 a.m. to 15:00, a slot freed up by a cancellation. Zero Retention Mode (a mode with no data retention) is available only in this setup. In practice, the authors recommend running both: the batch for next-day reminders and real time for late changes.

The agent reads from and writes to a single system. Open slots come from the Slot resource, a reschedule is a PUT to the same Appointment record, and a cancellation changes its status. There is nothing to reconcile with the record after the call, and every change lands in the existing audit trail. The sandbox used the public HAPI FHIR server with synthetic data. For mass calling campaigns, teams can connect Campaign Manager, which accounts for calling windows, local time, retries, and voicemail.

Architect and the small details people forget

This time the workflow was not built by hand. The starting agent already had FHIR tools connected, system tools, a base prompt covering its role and manner of speaking, and nine simulation tests, but it had no workflow. The authors described the call in plain language in the Architect assistant chat, and Architect built the diagram.

The authors separately remind readers about caller ID. If the screen shows "Spam Likely", the patient will not pick up. Before the first campaign, register the number for STIR/SHAKEN, set a CNAM so the clinic's name appears, and check Twilio branded calling for your patients' carriers.

Takeaway for developers: the main value of such an agent is not the voice but the direct write access to the system of record. The material was published by a vendor, so customer metrics are worth verifying against your own data, and responsibility for compliance with medical requirements and for call consent remains with the clinic.

Story

Key Questions Answered

What it looks like in practice?

Take a patient named Maria, booked for 14:00 tomorrow with her doctor. In the evening, the agent dials her number, asks for her date of birth, and checks it against the record. It then reads the current appointment and offers to confirm the visit, move it to an open slot in the doctor's schedule, or cancel it. If Maria reschedules, the agent changes that same EHR record, and the freed-up time goes back into the schedule while it can still be filled. The agent also handles edge cases.

What are the key details on Customer results?

Everlywell uses ElevenAgents to remind participants about scheduled screenings. The share of completed screenings rose 10% compared with IVR, and conversion among Spanish-speaking participants was 3.5 times higher. Ovianta uses the agent for calls after a treatment plan has been proposed: patient engagement reached 68%, conversion to treatment rose 25%, and clinic staff made 83% fewer manual calls. These figures come from the vendor itself; the article includes no independent verification.

What are the key details on Two ways to launch?

The first is a nightly batch. In the evening, a script pulls tomorrow's appointments from the EHR, turns each into a table row, and sends them to Batch Calling. Only the phone column is required. The other columns (name, date, doctor, visit type, clinic name) become variables the agent uses in its prompt and during the conversation. Times in the table are written out in words, for example "ten a.m.", because the agent reads them aloud and the speech synthesizer should not have to guess the format.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…