An AI phone agent for the clinic phone line that never stops ringing

An outpatient clinic's phone carries almost every kind of request the practice receives: new patients trying to get in, established patients rescheduling, refill requests, insurance and billing questions, referral paperwork, and the occasional caller who should not be on hold at all because they need emergency care. The front desk answers that mix while checking patients in and verifying coverage—so calls roll to voicemail, hold times stretch, and messages get transcribed by hand into the wrong queue.

ThunderPhone gives a primary-care or specialty clinic an AI phone agent that answers immediately, works from the scripts and escalation rules the clinic writes, collects structured information, uses connected scheduling and practice systems, and transfers the calls that need a person. It is a front-office worker, not a clinician: it never makes a clinical judgment, never gives medical advice, and never authorizes a prescription. Those boundaries are the design.

What a clinic agent can be configured to do

Every workflow below runs on clinic-written instructions. The agent follows the script, collects the fields the clinic requires, and escalates when a clinic-defined trigger is met.

Appointment scheduling, rescheduling, and cancellation

The agent can answer scheduling calls during busy hours or around the clock, collect the identifiers the clinic specifies, the reason for visit in the caller's own words, and preferences, then check availability and book, move, or cancel through a connected calendar or practice-system API—and confirm the result back to the caller.

Reminder and no-show-reduction outbound

Outbound campaigns can call patients ahead of scheduled visits, confirm attendance, capture a reschedule request on the spot instead of leaving it to a missed appointment, and repeat approved preparation instructions—fasting requirements, arrival time, documents to bring. Voicemail detection can leave a configured message or follow the campaign's retry policy.

Prescription-refill request intake

The agent can run a structured refill-request line: patient identifiers, medication name as stated by the caller, pharmacy, and callback details, delivered to the clinical team through an integration or webhook. It states plainly that the request goes to the care team for review. It never approves, denies, or promises a refill, never discusses dosing, and never advises on a missed dose—those calls route to licensed staff under the clinic's protocol.

Insurance, eligibility, and billing questions

For general questions the clinic has approved—which plans the practice accepts, how to submit updated insurance information, where to send records requests—the agent answers from the knowledge base. For patient-specific coverage, balances, or disputes, it collects the details and routes the call to the billing team rather than guessing.

New-patient registration

The agent can collect the registration fields the clinic defines—demographics, contact details, insurance information the clinic chooses to take by phone, referral source, preferred pharmacy—and deliver a structured record to the intake queue, so the first in-person minutes are not spent on a clipboard.

After-hours answering with clinic-defined escalation

After hours, the agent answers instead of a voicemail box. Its first job, per the clinic's script, is to screen for the symptoms the clinic has listed as emergencies. If a caller reports one—chest pain, for example—the agent does not assess it; it immediately follows the clinic's written protocol: the approved emergency guidance and, where configured, a transfer to the on-call line. For everything else it takes structured messages, books permitted next-day slots, and queues requests for the morning.

Recall and preventive outreach

Outbound recall campaigns can reach patients due for annual physicals, wellness visits, chronic-condition follow-ups, or screenings the clinic tracks, using clinic-approved wording, and book the visit in the same call. Campaign retry rules govern unanswered calls and voicemail outcomes.

Referral coordination follow-ups

When the clinic refers a patient out, or receives a referral in, the agent can place follow-up calls: confirm the referral was received, check whether the visit was scheduled, collect obstacles, and log the outcome for the coordination team.

Post-visit check-ins

The agent can place clinic-scripted follow-up calls after visits or procedures—confirming a medication pickup, asking the clinic's approved check-in questions, and flagging any response on the clinic's escalation list for human callback. It does not interpret answers; it records and routes them.

Multilingual patient communication

ThunderPhone supports 47 voice languages. A clinic can set a primary language plus additional languages, and the agent switches mid-call when a caller changes to a configured language—useful when the patient panel speaks languages the front desk cannot cover on every shift.

Sample clinic call flow

Agent: Thank you for calling Cedar Park Family Medicine. This is the clinic's automated assistant. If you are experiencing a medical emergency such as chest pain, trouble breathing, or severe bleeding, please hang up and dial 911. Otherwise, are you calling to schedule, request a refill, or something else?

Caller: I need to reschedule my appointment on Thursday, and I'm also almost out of my blood-pressure medication.

Agent: I can help with both. May I have your full name and date of birth to find your appointment?

Caller: Maria Alvarez, March 12, 1971.

Agent: Thank you. I see Thursday at 2:15 p.m. with Dr. Okafor. I have openings Monday at 9:40 a.m. or Wednesday at 3:00 p.m. Which works better?

Caller: Monday morning.

Agent: Done—Monday at 9:40 a.m., and Thursday is cancelled. For the refill, which medication and which pharmacy do you use?

Caller: Lisinopril, at the pharmacy on Lamar Boulevard.

Agent: I have recorded a refill request for lisinopril at your Lamar Boulevard pharmacy. The care team reviews all refill requests—I cannot approve medications myself. Is the number ending in 0148 the best callback number?

Caller: Yes.

Agent: You are set. Monday at 9:40 a.m. with Dr. Okafor, and your refill request is with the care team.

Had the caller reported a symptom on the clinic's emergency list at any point, the agent would have skipped everything else and followed the configured path: the approved emergency instructions and, where configured, a transfer to the on-call number.

Platform capabilities behind the clinic workflows

GDPR and HIPAA compliance posture — and the clinic's configuration responsibility

ThunderPhone's public compliance posture is GDPR and HIPAA compliant. That posture does not make any particular clinic workflow compliant by itself, and in a medical setting this responsibility deserves the most attention on this page. The clinic decides and configures: which patient information the agent may collect and which it must not; whether calls are recorded and what notice callers receive; how long transcripts and recordings are retained; which integrations receive protected health information and under what agreements; what the agent's disclosure script says; and who can access call data. Limit the agent to the minimum information each workflow needs, attach only approved documents, and have your compliance owner review the prompt, recording settings, and data destinations before launch.

Inbound and outbound calling on your numbers

Agents handle inbound calls and place outbound campaigns. Clinics can bring existing numbers through supported direct VoIP connections or manual SIP configuration for any SIP trunk; verified imported numbers support inbound and outbound calling, so patients keep dialing the number they already know.

Warm and cold transfer

Cold transfers route a caller immediately. On eligible production numbers that can place an outbound SIP leg, warm transfers let the agent put the caller on hold, privately brief the nurse, billing specialist, or on-call clinician, and connect the call only after the recipient accepts—so an escalated patient is never dumped into a cold queue.

Hold, keypad menus, and voicemail handling

The agent can navigate hold queues and keypad phone menus on outbound calls—useful when confirming referrals with other offices. Voicemail detection can let the prompt decide, hang up, or leave a configured message, and campaigns retry voicemail outcomes under their retry policy.

Knowledge grounded in clinic documents

Attach clinic policies, visit-preparation instructions, accepted-plan lists, and service descriptions—text files up to 5 MB, PDF or DOCX up to 50 MB—or import pages by URL. The agent searches these approved sources mid-call instead of improvising an answer, and can be instructed to route any question the sources do not cover.

Integrations that fit the clinic's stack

Connected apps, HTTP APIs, and remote MCP servers let the agent act during a call—check a calendar, create an appointment, file a refill request. HMAC-signed webhooks deliver call events to clinic systems with verifiable integrity. A web widget can put the same agent on the clinic's website for patients who prefer to start there.

Testing before a patient ever hears it

This vertical demands rehearsal. Run AI-caller simulations of the risky scenarios—a caller mentioning chest pain mid-scheduling, a refill caller pushing for dosing advice, an angry billing dispute—with graded scenarios and transcripts. Build the passing cases into regression suites so a prompt edit cannot silently weaken the emergency script, and use A/B experiments to compare wording on live traffic. Recordings, transcripts, and grading make every production call reviewable.

What would this cost each month?

Suppose a clinic handles 1,000 AI-assisted calls per month, averaging 4 minutes each:

1,000 calls × 4 minutes = 4,000 minutes per month

  • Spark: 4,000 × $0.02 = $80/month
  • Bolt: 4,000 × $0.05 = $200/month
  • Storm: 4,000 × $0.09 = $360/month

That is base engine usage, not a quote. Premium voice or language paths can add up to 3¢ per minute, and phone-number or carrier charges depend on the telephony setup. Hold time bills a flat 2¢ per minute; voicemail and phone-menu minutes bill at the engine rate. Optional add-ons such as verbal acknowledgements, call supervision, and long prompts carry their own published surcharges. The agent builder shows the current all-in rate before calls are placed.

Guardrails a medical clinic should write into the agent

Each rule belongs in the prompt explicitly, with a simulation that proves it holds:

  • No clinical judgment, ever. The agent never assesses symptoms, estimates urgency, or interprets a patient's condition. Urgency handling is a lookup against the clinic's written list, not a decision.
  • No medical advice. Approved, clinic-written instructions (visit preparation, directions, policies) may be read back; anything resembling advice about a health question is routed to clinical staff.
  • No refill authorization. Refill calls produce a structured request for the care team. The agent says so in plain language and never predicts the outcome.
  • Emergency script first. The emergency screening and redirection language runs at the top of every after-hours call, exactly as the clinic wrote it.
  • Minimum necessary information. Collect only the fields each workflow requires; do not let the agent free-form probe into health details a scheduling call does not need.
  • Disclosure and recording notices. The agent identifies itself as an automated assistant and delivers whatever recording or privacy notice the clinic's compliance owner requires.
  • Escalation is always available. Any caller who asks for a person gets a transfer path or a callback message—the agent never becomes a wall between a patient and the clinic.

Frequently asked questions

Is ThunderPhone HIPAA compliant for a medical clinic?

ThunderPhone's public posture is GDPR and HIPAA compliant. The clinic remains responsible for configuring its own compliant workflow: what patient information the agent collects, recording and notice settings, retention, which integrations receive protected health information, and who can access call data. Review the configuration with your compliance owner before launch.

Can the agent tell a patient whether their symptoms are urgent?

No, and it should not be configured to try. The clinic writes a list of emergency triggers and the exact response—typically approved emergency guidance and a transfer to the on-call line. The agent matches and follows; it never weighs symptoms or offers an opinion.

Can it approve prescription refills?

No. It can take a structured refill request—patient identifiers, medication as stated, pharmacy, callback number—and deliver it to the clinical team for review. It tells the caller explicitly that the care team makes the decision.

Can it book, reschedule, and cancel appointments in our system?

Yes, when connected to a calendar or practice-system API that permits those actions, through an app connection, HTTP API, or MCP server. The agent collects required fields, offers times, makes the change, and confirms it back to the caller.

What happens after hours?

The agent answers instead of voicemail: it runs the emergency screening script first, then takes messages, books permitted next-day slots, or escalates to the on-call number per the clinic's rules. Everything else is queued as structured records for the morning team.

Can it call patients who are overdue for physicals or screenings?

Yes. Outbound recall campaigns can work through a list with clinic-approved wording, book visits during the call, and handle voicemail through detection plus the campaign's retry policy. The clinic is responsible for ensuring its outbound calling and messaging comply with applicable consent rules.

Can one agent serve patients in multiple languages?

Yes. Configure a primary language and additional languages from the 47 supported, and the agent can switch mid-call when a caller changes language, provided the selected voice supports the configured set.

How do we verify the agent is safe before patients call it?

Test it like clinical software. Browser mic tests support fast iteration; AI-caller simulations rehearse graded scenarios such as an emergency mention mid-call; regression suites with pass-rate gates keep prompt changes from weakening the guardrails; and recordings, transcripts, and grading make live calls auditable after launch.