What a medical office AI receptionist actually does
Answers a patient's call, books or checks an appointment against the practice's scheduling system, and routes anything clinical to a nurse or provider instead of answering it itself.
A medical office AI receptionist takes patient calls the way a front-desk staffer would: understand why the patient is calling, check availability or pull up an existing appointment through a tool call into the practice's scheduling system, verify insurance eligibility against a payer API where one exists, and either complete the task or route the caller to a human. Because the call almost always touches protected health information, even just a name tied to an appointment time, everything in the chain, the voice platform, the model provider, the telephony carrier and the transcript storage, has to sit inside a signed business associate agreement before a single real patient call goes through it.
The system-of-record vendors are moving fastest here. Epic previewed an "Agent Factory" at HIMSS26 that lets health systems build their own agents on top of Emmie, its MyChart patient-facing assistant, while independent vendors Assort Health and Hyro deepened their own EHR integrations: Assort closed a $120M Series C at a $1.2B valuation in June 2026 and writes directly into Epic and athenahealth, and Hyro became a certified Epic Showroom app serving 45+ health systems including Tampa General.
Where it still breaks: anything resembling a diagnosis or clinical advice, which the agent must never attempt, and the fact that none of the vendors built for this job are scoped or priced for a single-location practice. Assort Health and Hyro sell to health systems, not solo practices, which leaves a real medical office choosing between an agency-built compliant stack and a DIY voice platform's BAA option, since there isn't yet a self-serve, HIPAA-ready receptionist built for a 2-3 provider office the way Goodcall or Smith.ai serve general small businesses.
AI capabilities involved
Real-time speech-to-text on the live call
Scheduling and rescheduling via tool calls into the EHR
Insurance eligibility checks and intake data capture
After-hours call triage and routing (never diagnosis)
Who uses this
- Independent medical practices and small clinics (2-10 providers) without a dedicated after-hours answering service
- Practices already running Epic, athenahealth, eClinicalWorks, NextGen or ModMed wanting phone coverage that writes into the same system
- Offices losing new-patient calls to voicemail outside business hours
- Practices running recall and no-show-recovery outreach that needs a documented consent and compliance trail
What the medical office receptionist handles (and what it shouldn't)
| Task | Agent handles it? | Systems it touches | Human role |
|---|---|---|---|
| Scheduling and rescheduling appointments | Fully | EHR (Epic Cadence, athenahealth) | None for standard slots |
| Insurance eligibility questions | With approval | Eligibility tool (Availity) | Confirms unusual plan situations |
| Prescription refill requests | With approval | EHR messaging | Clinical staff approves every refill |
| New-patient intake data capture | Fully | EHR intake forms | Reviews before the first visit |
| After-hours symptom-sounding calls | No | Call transfer | Nurse line or on-call provider takes over immediately |
| Billing and payment-balance questions | With approval | Billing system | Reviews disputed charges |
| Referral and records-request intake | Fully | EHR, fax/API | Processes the actual referral |
| Recall and no-show-recovery outbound calls | With approval | EHR, outbound dialer | Reviews opt-outs and consent records |
Off-the-shelf AI agents on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
Epic Emmie (MyChart patient agent)
Health systems already running Epic end to end
Included for health systems already licensing Epic; no standalone trial
Not sold separately; bundled into your Epic and MyChart license and implementation
Scoped as part of your existing Epic contract
Pros
- +Native to MyChart, so patients already using the portal see no new app
- +Epic's Agent Factory lets health systems build on top of it instead of waiting on Epic's own roadmap
- +No separate BAA negotiation, since it sits inside your existing Epic agreement
Cons
- −Only available if you're already an Epic customer
- −Not a fit for independent practices on athenahealth, eClinicalWorks or ModMed
Assort Health
Multi-site health systems on Epic or athenahealth
None; demo only
Contact sales; no published pricing
Scoped per health system; closed a $120M Series C at a $1.2B valuation in June 2026
Pros
- +Writes directly into Epic and athenahealth, including a live John Muir Health deployment
- +Four products (Concierge, Activate, Orchestrate, Empower) cover inbound, outreach and revenue-cycle work
- +Well-capitalized, which lowers the risk of the vendor disappearing mid-contract
Cons
- −No public pricing and nothing in its materials suggests it serves single-location practices
- −Enterprise sales cycle doesn't fit a practice that needs phone coverage this quarter
Hyro
Health systems wanting a certified, pre-vetted Epic integration
None; demo only
Contact sales; no published pricing
Serves 45+ health systems including Tampa General
Pros
- +Certified Epic Showroom app, so the integration is pre-vetted by Epic itself
- +Covers voice, chat and SMS from one platform
- +Raised a $45M growth round in late 2025, roughly $95M total
Cons
- −No public pricing
- −Built for health-system scale, not a single clinic
Sully.ai
Independent practices wanting one AI vendor across several front-office roles
None; demo only
Contact sales; no published pricing
About $30.5M raised as of its Series A
Pros
- +Built as a suite of "AI employees" (receptionist, scribe, nurse, coder), so the receptionist can share context with other automated roles
- +YC-backed and smaller, more likely to work directly with an independent practice than the big health-system vendors
- +Covers more than just the phone line if you want to expand later
Cons
- −Smaller company with less proven track record than the Epic-integrated competitors
- −No published pricing to compare against a custom build
The AI stack
A medical receptionist needs the same voice and reasoning layers any phone agent does, plus an EHR integration and a BAA signed at every hop that touches protected health information.
Voice (speech-to-text + text-to-speech)
Converts the patient's speech to text and the agent's reply back to speech, inside a BAA-covered pipeline
Bland AI (bundled)
$0.11-$0.14/min all-in, with a BAA included because the stack is self-hostedThe cheapest path to a fully BAA-covered voice stack
Vapi plus HIPAA add-on
$0.05/min hosting plus a flat $2,000/mo HIPAA add-on, with STT/TTS/telephony pass-through on topPractices with enough call volume to amortize the flat fee
Retell AI
Voice engine $0.055/min plus LLM, TTS and telephony, with a custom BAA on EnterpriseTeams wanting a predictable rate card with a custom-negotiated BAA
Our pick: Bland for the simplest fully-bundled BAA path, or Retell if you want more control and are willing to negotiate enterprise BAA terms.
Telephony
Carries the call and connects it to your voice pipeline
Vendor-native telephony (Bland, Retell)
Included in the per-minute rateFast, compliant production builds
Twilio direct
$0.008/min inbound, $0.014/min outboundFully custom builds with an existing Twilio relationship
Our pick: Use your voice platform's native telephony so the BAA chain stays as short as possible.
Reasoning model
Understands the patient's request, decides when to book, route or escalate, and never attempts a diagnosis
Claude Sonnet 5.5
$2/M input, $10/M outputThe main call-handling loop
Claude Haiku 5.5
$0.10/M input tokens (up to 100K), $0.50/M outputRoutine confirmation and reminder calls
Our pick: Sonnet 5.5 for anything that touches scheduling logic or eligibility, with a hard-coded rule layer (not the model's judgment alone) enforcing the never-diagnose boundary.
EHR integration & eligibility
Reads and writes real appointment and eligibility data instead of a generic calendar
Direct EHR API or FHIR integration
Developer time plus EHR vendor API feesCustom builds
Availity or a comparable eligibility API
Per-transaction or subscription fee, varies by payer mixAny path that handles insurance questions by phone
Our pick: Prioritize EHR scheduling write-back first; eligibility checking is valuable but can launch as a phase two.
Observability & evals
Reviews call transcripts for diagnosis-boundary violations, missed verifications and scheduling errors
Vendor call logs
Included on most platformsA starting point, not a substitute for human review
Langfuse plus a manual weekly audit
Free up to 50K units/mo, then $29+/moAny production medical deployment
Our pick: Have clinical staff, not just an ops person, review a weekly sample of calls specifically for diagnosis-boundary and triage-routing mistakes.
Reference architecture
The hardest engineering problem is the never-diagnose boundary: the agent has to recognize a symptom-sounding call and route it to a nurse line or on-call provider every time, with a hard rule rather than the model's own judgment, because a single missed escalation on a genuinely urgent call carries consequences no cost calculator captures.
Call connects
Telephony (BAA-covered)The patient's call routes to the receptionist's number through a carrier covered by the signed BAA
Speech-to-text
STT engineThe patient's audio is transcribed in real time inside the BAA-covered pipeline
Intent and symptom check
Reasoning model plus a hard-coded rule layerThe model identifies the request type; any symptom-sounding language trips a hard rule that routes to a human regardless of model confidence
Scheduling or eligibility tool call
EHR/eligibility integrationA booking, reschedule or eligibility check runs against the real EHR or payer API
Text-to-speech
TTS engineThe agent's reply is converted to speech and streamed back to the patient
Human handoff
Nurse line or on-call providerSymptom-sounding, urgent or ambiguous calls transfer immediately with a spoken summary
Logging and audit
Observability toolEvery call is logged inside the BAA-covered storage, with clinical staff auditing a weekly sample
Estimated cost per request
~$0.12-$0.30 per minute of call depending on which BAA path you choose, before any EHR integration maintenance cost
Cost calculator
Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.
Estimate your monthly cost at your own call volume and average call length.
Estimated monthly cost
$203
≈ $2,436 per year
Calculator notes
- This calculator models a Bland-style bundled voice stack at roughly $0.12/min, which already includes the signed BAA; Vapi's HIPAA add-on instead charges a flat $2,000/mo regardless of call volume, which only pays for itself at high volume.
- Model and orchestration cost layers on top of the bundled voice rate to cover scheduling and eligibility tool calls.
- A custom agency build typically lands in the same $150-$700/mo running-cost range once you include EHR integration maintenance.
- The enterprise vendors (Assort Health, Hyro, Epic Emmie) aren't modeled here since none of them publish per-call or per-minute pricing.
Build it yourself with vibe-coding tools
Retell lets a technical practice assemble a voice receptionist with a published rate card and a negotiable enterprise BAA, rather than absorbing Vapi's flat $2,000/mo HIPAA surcharge.
Time to MVP
A few days for a basic version, 4-6 weeks to get the BAA chain, scheduling write-back and never-diagnose rules production-ready
Total cost to MVP
$0-$150 to start, plus whatever Retell's negotiated enterprise BAA terms add once you're ready for real patient calls
You'll need
Starter prompt
Create a new Agent in Retell using Claude Sonnet 5.5 as the model, with your EHR's scheduling API and an eligibility-check tool connected. Set the greeting to: 'Thanks for calling {Practice Name}, this is an AI assistant, how can I help you today?' Give the agent this system prompt: 'You are a medical office receptionist for {Practice Name}. You schedule, reschedule and confirm appointments, answer insurance-eligibility questions using the eligibility tool, and collect new-patient intake information. You never discuss symptoms, never suggest a diagnosis, and never give clinical advice of any kind, even if directly asked. If the caller describes any symptom, says anything urgent, or asks a clinical question, immediately say: "I want to make sure the right person helps you with that," and transfer the call to the nurse line without asking further questions about the symptom. For refill requests, collect the medication name and pharmacy, then transfer to clinical staff rather than confirming the refill yourself. Never claim to be human if asked directly, and never use a title like "nurse" or "doctor" to describe yourself. This call is recorded for quality purposes, consistent with the greeting already given.' Add three tools: schedule_appointment and check_eligibility pointed at your EHR and payer APIs, and transfer_to_nurse_line pointed at your clinical triage number. Turn on call recording inside your BAA-covered storage, and have clinical staff review a weekly sample of transcripts specifically for any symptom-boundary violation.
Paste this into Retell AI
Follow-up prompts (run in order)
- 1
Add a recall-outreach flow that calls patients due for a check-up and books them directly
- 2
Add a Spanish-language path that detects the caller's language and switches voices
- 3
Add a no-show-recovery flow that texts a rebooking link after a missed appointment
- 4
Add a nightly report listing every call transferred to the nurse line and why
Expected output
A phone number that books and confirms appointments directly into your EHR, answers eligibility and intake questions, and immediately routes anything symptom-sounding to a nurse line without attempting to answer it.
Known gotchas
- !A single unsigned BAA anywhere in the chain, the voice platform, the model provider or the transcript storage, breaks your HIPAA compliance even if everything else is correct
- !Vapi's $2,000/mo HIPAA add-on is a flat fee regardless of call volume, so it's a poor fit for a low-volume solo practice
- !The never-diagnose boundary has to be a hard rule, not something left to the model's judgment, or it will eventually answer a symptom question it shouldn't
- !CA AB 489 bars using titles like "nurse" or implying a licensed provider is on the line; the agent's own self-description has to avoid this
- !Forgetting the AI-disclosure line at call start is both a compliance problem and the most common patient complaint about these systems
- !EHR scheduling APIs vary enormously by vendor, so the integration timeline is often the slowest part of the build, not the voice layer
Compliance & risk reality check
A medical receptionist carries the heaviest compliance weight of any voice agent on this site, because almost every call touches protected health information.
HIPAA business associate agreement
Any vendor that creates, receives, maintains or transmits PHI on your behalf, the voice platform, the model provider, the telephony carrier and the transcript storage, is a business associate and needs a signed BAA.
Mitigation: Verify a signed BAA with every hop in the chain before a single real patient call goes through it; Bland bundles this into its standard rate, Vapi charges $2,000/mo for it, and Retell negotiates it on Enterprise.
Never implying a licensed provider
California's AB 489, in force since January 1, 2026, bars AI from using titles like "nurse" or "doctor" or otherwise implying a licensed human is providing care.
Mitigation: Have the agent identify itself plainly as an AI assistant and never use a clinical title to describe itself.
Never diagnosing or giving clinical advice
The agent's job is scheduling, intake and routing, not clinical judgment; any symptom-sounding call needs an immediate, rule-based transfer rather than an AI-generated answer.
Mitigation: Hard-code a transfer trigger on symptom or clinical language, independent of the model's own confidence score.
TCPA for outbound recall and reactivation calls
Recall and no-show-recovery calls are outbound AI voice calls under the FCC's February 2024 ruling (24-17), which requires prior consent and disclosure.
Mitigation: Capture written consent for recall-style outbound calls, disclose the AI at the start of every call, and honor opt-outs immediately.
Call recording and wiretap law
Ambriz v. Google and Taylor v. ConverseNow both survived motions to dismiss in 2025 on the theory that an AI vendor capable of using call data to train its own models can be an unlawful eavesdropper.
Mitigation: Disclose recording and AI at the start of every call, and contractually bar every vendor in the chain from training on your call data.
Build vs buy: the real math
3-6 weeks for a RapidDev build; 4-6 weeks for a careful DIY version given the BAA review
Custom build time
$13K-$25K for an agency build (up to $35K for multi-system or compliance-heavy work); $150-$700/mo in running voice-stack and BAA costs
One-time investment
There's no clean breakeven against a packaged SaaS vendor, because Assort Health and Hyro don't price for a single practice at all
Breakeven vs buying
Because the health-system vendors (Assort Health, Hyro, Epic Emmie) are scoped and priced for multi-site systems, not independent practices, the real comparison for most readers isn't buy versus build, it's DIY versus agency. A DIY build on Vapi's $2,000/mo HIPAA add-on at low call volume (say, 300 calls a month at 4 minutes each) adds up to roughly $2,150/mo before you've written a single line of scheduling logic, while an agency build using Bland's bundled BAA rate or Retell's negotiated enterprise BAA typically runs $150-$700/mo once live, against a $13K-$25K upfront cost amortized at $1,083-$2,083/mo in year one. For most independent practices, the agency path's total year-one cost lands close to or below the Vapi-DIY path, with a properly verified BAA chain instead of one you assembled yourself.
Skip the DIY — RapidDev builds the production version
A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship through our AI app development service.
Discovery call (free)
30 minWe map your exact medical office AI receptionist use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.
AI-accelerated build
3-6 weeks for a RapidDev build; 4-6 weeks for a careful DIY version given the BAA reviewOur engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.
Launch + handoff
1 weekWe deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.
What you get
Timeline
3-6 weeks for a RapidDev build; 4-6 weeks for a careful DIY version given the BAA review
Investment
$13K-$25K for an agency build (up to $35K for multi-system or compliance-heavy work); $150-$700/mo in running voice-stack and BAA costs
vs SaaS
ROI in There's no clean breakeven against a packaged SaaS vendor, because Assort Health and Hyro don't price for a single practice at all
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
How much does an AI receptionist for a medical office cost?
The enterprise vendors (Assort Health, Hyro) don't publish pricing and generally only serve multi-site health systems. A custom RapidDev build runs $13K-$25K upfront (up to $35K for complex, multi-system work), plus $150-$700/mo in ongoing voice-stack and BAA costs. A DIY build runs $299-$499/mo on a bundled HIPAA platform like Bland, or adds a flat $2,000/mo on Vapi's HIPAA add-on.
How long does it take to set up?
An enterprise health-system deployment with Assort Health or Hyro takes months to scope and implement. A custom agency build takes 3-6 weeks to wire into your EHR and write compliant scripts. A DIY version can work within a few days, though getting the BAA chain and never-diagnose rules production-ready takes 4-6 weeks.
Can RapidDev build this for us?
Yes. RapidDev has shipped 600+ apps and offers a free 30-minute consultation to scope your EHR, call volume and BAA requirements before you commit to anything.
Does every vendor in the chain need to sign a BAA?
Yes. The voice platform, the model provider, the telephony carrier and the transcript storage all count as business associates if they touch protected health information, and a single unsigned BAA anywhere in that chain breaks your HIPAA compliance even if everything else is handled correctly.
Can the receptionist discuss symptoms or give medical advice?
No, and it shouldn't even try. The agent's job is scheduling, intake and routing; any symptom-sounding call should hit a hard-coded rule that transfers immediately to a nurse line or on-call provider, independent of how confident the model sounds.
Will patients know they're talking to AI?
They should hear it in the first line of the greeting, and California's AB 489 specifically bars using a clinical title like "nurse" to describe the agent. Stating plainly that it's an AI assistant satisfies both AB 489 and the broader disclosure expectations under California's BOT Act and the EU's Article 50 rule.
Can it work with Epic, athenahealth or eClinicalWorks?
Epic's own Emmie only works inside an existing Epic license, and Assort Health and Hyro both integrate with Epic and athenahealth at the health-system level. A custom build can write directly into any EHR with an API or FHIR interface, including eClinicalWorks, NextGen and ModMed.
Want the production version?
- Delivered in 3-6 weeks for a RapidDev build; 4-6 weeks for a careful DIY version given the BAA review
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.
