USE CASE
AI Voice Agent for Healthcare Appointment Scheduling
An AI voice agent for appointment scheduling answers your phone line, books, reschedules and cancels appointments, verifies insurance details and routes triage calls to the right person. CognyX built one for a US healthcare provider that handles 2,000+ calls a week with a 92% resolution rate. Delivery starts with a 2–6 week pilot, then 6–12 weeks to production.
At a Glance
Typical figures for a project with an India-based team. Third-party facts are in the sources list.
| Item | What to expect |
|---|---|
| Typical cost | $10,000–$45,000 to build; rates from $25/hr |
| Timeline | 2–6 week pilot, then 6–12 weeks to production |
| Running cost | About $0.07–$0.31 per call minute, plus upkeep |
| Channels | Phone first; the same agent can extend to SMS, WhatsApp and web chat |
| Integrations | Twilio telephony; scheduling, EHR and practice-management APIs, FHIR where supported |
| Compliance posture | HIPAA, GDPR and SOC 2 ready; a BAA with each vendor that handles PHI |
| Handoff | Live transfer to your staff with the transcript and the details already captured |
What the Agent Handles and What Stays With Staff
The agent takes the calls that follow a clear process and need data from your systems. Your staff keep every call that needs clinical judgment.
Calls the agent handles
- New appointments. It finds open slots for the visit type, provider and location, offers options and books the one the caller picks.
- Reschedules and cancellations. It finds the existing appointment, moves or cancels it, and frees the slot for another patient.
- Insurance verification. It collects the plan and member details and checks them through your insurance verification system before the visit.
- Triage routing. It asks for the reason for the call and the symptoms the caller reports, then routes the call with rules your clinical team writes and approves. It does not diagnose.
- Common questions. Hours, locations and what to bring, answered only from content your team approves.
- Reminder calls (optional). Outbound reminders and confirmations for patients who have given consent.
Missed appointments are common: a 2018 systematic review of 105 papers found an average no-show rate of about 23%. When patients can move or cancel by phone at any hour, the slot goes back on the schedule instead of going empty.
What stays with your staff
- Clinical advice. The agent does not diagnose, advise or answer medication questions. It transfers the call or books a clinician callback.
- Emergencies. When a caller describes an emergency, the agent stops the booking flow and transfers the call to staff at once, with the emergency script your clinical team approves (for example, telling the caller to call 911).
- Everything else. Complaints, billing disputes, callers who ask for a person and calls the agent cannot resolve go to staff with the transcript and captured details, so the caller does not repeat themselves.
The rule: the agent handles the administration of care; your clinicians handle care.
How a Call Works, From Ring to Booked Slot
Every call follows the same steps, and every step is logged.
- The call connects. A patient calls your number and Twilio connects the call to the voice agent. The agent says it is an AI assistant and, if you record, that the call is recorded.
- The agent works out the request. It listens, asks follow-up questions and identifies the call type: booking, change, insurance, triage or a question.
- It verifies the caller. It uses knowledge-based questions, such as name and date of birth, or a one-time passcode, before it reads or changes a record.
- It reads only what the task needs. For a booking, the open slots for one visit type; for a change, one appointment; for insurance, the plan details.
- It completes the task. It books, moves or cancels the appointment in your scheduling system, or checks the insurance details, then reads the result back to the caller.
- It hands off when it should. On an emergency, a clinical question, a failed check or a request for a person, it transfers the call to staff with the transcript and the captured details.
- Sensitive actions wait, and everything is logged. Actions you mark as sensitive, such as changing the insurance details on file, go to a staff queue first. The audit log records what the agent read and changed, and when, with sensitive fields redacted. A call summary goes to your team.
Integrations: EHR, Scheduling and Telephony
The agent connects to the systems your front desk already uses, through their APIs. Each action has a defined schema, input checks and an allow-list, so the agent can run only the actions you approve.
Scheduling, EHR and practice-management systems
We connect the agent to scheduling, EHR and practice-management systems through their APIs, with FHIR and HL7 where the system supports them. FHIR models scheduling with a Schedule, the Slots on it that may be available for booking, and the Appointment that books a slot.
- Epic lists Appointment.$find and Appointment.$book operations (FHIR STU3) to find and book open slots.
- Oracle Health’s Millennium FHIR R4 API can list open slots, create an appointment and update its status, for example to cancel it. The health system decides which practitioners and locations it makes available to outside scheduling apps.
Access to these APIs depends on what your organization has enabled, so we confirm it in discovery. If you book from a calendar or a CRM instead, the agent uses that system’s API.
Telephony and voice platforms
The case-study agent answers the provider’s line through Twilio. We also build on Vapi and Retell AI for voice and ElevenLabs for speech, and connect voice agents to Salesforce, Zendesk, n8n and Zapier.
If we built the case-study agent today
We would run it on the OpenAI Realtime API over Twilio, and connect the scheduling and insurance verification systems as tools through the Model Context Protocol (MCP).
Compliance and Safety
Our controls are HIPAA, GDPR and SOC 2 ready. The design starts from two questions: what protected health information (PHI) the agent needs, and which vendors touch it.
Business associate agreements (BAAs)
Under HIPAA, a vendor that creates, receives, maintains or transmits PHI for you is a business associate, and so is its subcontractor. Each one needs a business associate agreement (BAA): a contract in which the vendor agrees to safeguard the PHI and use it only as the contract allows. In a voice agent, that covers the telephony, voice platform, speech and model vendors.
What the vendors we build on say about BAAs, checked on their own pages on 13 November 2026:
- Twilio signs a BAA for its HIPAA-eligible services on its Security or Enterprise Edition. Its voice AI connector, ConversationRelay, has been HIPAA eligible since March 2025. If you stream call audio to another vendor, you need a BAA with that vendor.
- OpenAI offers a BAA for its API with Modified Retention; the Realtime API is on its HIPAA-eligible list.
- Retell AI offers a BAA to sign online at no extra fee, and requires it before PHI is sent.
- Vapi signs a BAA with its paid HIPAA add-on. If you bring your own keys for speech or model providers, those accounts must be HIPAA eligible too.
- ElevenLabs offers HIPAA-eligible setups to enterprise customers, with a BAA and Zero Retention Mode.
Keep PHI to the minimum
HIPAA’s minimum necessary standard asks for reasonable efforts to limit PHI to what the purpose needs. So the agent asks for, and reads, only what the task needs. Caller data is protected with PII detection, encryption in transit and at rest, role-based access and rate limits on sensitive operations. The agent runs in the cloud, in a hybrid setup or in your own VPC.
Recording and consent
Tell callers at the start that they are speaking with an AI assistant and, if you record, that the call is recorded. Check the consent rules of each state you take calls from, and set how long you keep recordings and transcripts. For outbound reminders, the FCC ruled in February 2024 that AI-generated voices are “artificial” under the Telephone Consumer Protection Act, so these calls need the called party’s prior express consent unless an emergency purpose or an exemption applies.
Audit logs and human approval
HIPAA’s Security Rule requires audit controls that record and examine activity in systems that hold electronic PHI. The agent logs every action, with sensitive fields redacted. Sensitive actions wait for a person to approve them, and your clinicians approve the triage rules and scripts before go-live.
Results and How We Measure Them
The case study
We built a voice AI agent that answers a US healthcare provider’s phone line through Twilio. It handles three types of call: appointment scheduling, insurance verification and patient triage.
2,000+ calls a week, with a 92% resolution rate. Read the healthcare voice agent case study.
What we measure
On voice projects we track five metrics against the KPIs set in discovery:
- Hold time: how long callers wait before they get help.
- Transfers: how many calls go to staff, and why.
- Average handle time (AHT): how long a call takes from answer to end.
- Wrap-up time: staff time after a call. Automatic call summaries cut it.
- Resolution rate: calls completed without staff, with “resolved” defined for each call type in discovery.
Results depend on the call types and on how deeply the agent connects to your systems. We take a baseline from your phone system before the pilot, so the go or no-go decision rests on the numbers, not on a demo.
Cost and Timeline
Typical ranges with an India-based team, from our AI agent development cost guide:
| Item | Typical range |
|---|---|
| Build: voice bot that replaces an IVR menu | $10,000–$20,000 |
| Build: full inbound voice agent | $20,000–$45,000 |
| Build: published range, all regions | $10,000–$150,000+ |
| Call minutes (Retell all-in list price) | $0.07–$0.31 per minute |
| Upkeep of prompts and knowledge base | 10–30% of build cost per year |
| CognyX rates | From $25/hr |
- Example: 2,000 calls a month at 4 minutes each is 8,000 minutes, or about $560–$2,480 a month in call costs.
- Compliance work (audit trails, encryption and access controls) typically adds 15–25% to project cost.
- Final price depends on call volume, the number of integrations and languages. Ask for a line-by-line monthly running-cost estimate at your real volume.
Timeline
- Discovery, about 2 weeks: map your call types and current setup, set the KPIs, and confirm API access and BAAs.
- Pilot, 2–6 weeks: a working agent on one or two call types, tested on real data against the KPIs.
- Production, 6–12 weeks: security hardening, edge cases, stress tests and full sync with your scheduling and insurance systems.
- Enterprise rollout, 3–6 months: across many departments, with 24/7 monitoring.
Frequently Asked Questions
Can an AI voice agent book appointments directly into our EHR?
Yes, when your EHR or practice-management system has a scheduling API that your organization has enabled. Many EHRs support FHIR scheduling: Epic lists Appointment.$find and Appointment.$book operations, and Oracle Health's FHIR R4 API can list open slots and create appointments. We confirm access in discovery. The agent then reads open slots, books the one the caller picks and writes it back, and each booking goes into an audit log.
Is an AI voice agent HIPAA compliant?
HIPAA applies to how your organization and its vendors handle protected health information (PHI), not to a tool on its own. Our controls are HIPAA, GDPR and SOC 2 ready: PHI minimisation, encryption, role-based access, redacted audit logs and human approval for sensitive actions. You also need a business associate agreement (BAA), a contract in which a vendor agrees to safeguard PHI, with each telephony, voice and model vendor in the call path.
What happens when a caller has an emergency?
The agent stops the booking flow and transfers the call to your staff at once. Your clinical team writes and approves the emergency script, for example telling the caller to call 911. The agent does not give clinical advice. It also transfers any call it cannot resolve, and any caller who asks for a person, with the transcript and the details it has already captured.
How long does it take to launch an AI voice agent?
A typical project has about 2 weeks of discovery, a 2–6 week pilot on real data, then 6–12 weeks to production. The pilot proves the conversation logic on one or two call types. Production adds security hardening, edge cases and full sync with your scheduling and insurance systems. A rollout across many departments takes 3–6 months.
How much does an AI voice agent for appointment scheduling cost?
With an India-based team, a voice bot that replaces an IVR menu typically costs $10,000–$20,000 to build, and a full inbound voice agent $20,000–$45,000. Running cost is about $0.07–$0.31 per call minute, plus upkeep. Compliance work typically adds 15–25% to project cost. CognyX rates start from $25/hr. Call volume, integrations and languages set the final price.
Which languages can the voice agent speak?
The speech and model stacks we build on support major global languages and regional accents, and they handle barge-in and background noise. The exact list depends on the speech, model and voice vendors you choose. We list the languages your patients need in discovery and test each one in the pilot before go-live.
Start with a 30-minute review
Bring your call mix: the main call types, monthly volume and the systems you book into. In a free 30-minute architecture review, our engineers check which calls the agent can take, which systems it can reach, and the build and running cost at your volume. You own the code and IP, and an NDA applies from day one.
Book a 30-minute review
Our AI engineers check your call types, systems and volume, and scope a 2–6 week pilot.
- [email protected]
- Phone
- +91 99748 12447
- Chat on WhatsApp
Sources
All pages checked on 13 November 2026. Vendor terms change; confirm them with each vendor before you sign.
Regulations and standards
- eCFR, 45 CFR 160.103 Definitions: business associate (read 13 Nov 2026)
- eCFR, 45 CFR 164.502: minimum necessary and disclosures to business associates (read 13 Nov 2026)
- eCFR, 45 CFR 164.312(b) Audit controls (read 13 Nov 2026)
- FCC, Declaratory Ruling FCC 24-17 on AI-generated voices under the TCPA, released 8 Feb 2024 (read 13 Nov 2026)
- HL7, FHIR R4 Appointment resource (read 13 Nov 2026)
- HL7, FHIR R4 Slot resource (read 13 Nov 2026)
Vendor pages
- Twilio, Twilio and HIPAA (read 13 Nov 2026)
- Twilio, ConversationRelay is now HIPAA eligible, 21 Mar 2025 (read 13 Nov 2026)
- Twilio, Architecting for HIPAA on Twilio, v1.1, 2022 (read 13 Nov 2026)
- OpenAI, HIPAA eligible products and functionality (read 13 Nov 2026)
- Retell AI, security and compliance (read 13 Nov 2026)
- Vapi, HIPAA compliance (read 13 Nov 2026)
- ElevenLabs, enterprise (read 13 Nov 2026)
- Epic, open.epic FHIR interfaces (read 13 Nov 2026)
- Oracle Health, Millennium FHIR R4 Appointment endpoints (read 13 Nov 2026)
- Oracle Health, Millennium FHIR R4 Slot endpoints (read 13 Nov 2026)
Research
Published by CognyX AI, an AI development and consulting company in Ahmedabad, India. Ranges are typical figures as of November 2026; your cost depends on scope, volume and compliance needs. This page is not legal advice: confirm your HIPAA and consent duties with your compliance team.
CognyX took the time to fully understand our project, explained their development approach and delivered within the timeframe they gave us.
Anthony SnowellRecent Case Studies
See how we've shipped production AI for healthcare teams and beyond — with the architecture and measured outcome behind each build.
Healthcare
AI Voice Agent for a US Healthcare Provider
Handles 2,000+ calls per week with 92% resolution rate.