An AI answering service for a doctor's practice runs from $0 for 25 calls a month (then $3 a call) to $150 for 75 calls on Smith.ai's AI plans; Moneypenny's live per-minute service starts at $265 for 100 minutes. After hours, the AI should take routine messages and follow a clinician-approved escalation route, never judge symptoms.
The price that matters is the bill for your own mixture of short requests, long explanations, transfers and failed calls. A cheap plan without a reliable on-call hand-off may leave the practice paying staff to rework messages anyway. Ask every supplier to price the same four-week call sample and to show the exact steps when a caller says something urgent.
Price the call pattern, not the headline subscription
Two published rate cards illustrate why a single industry average is misleading. Smith.ai's AI Receptionist page lists Free at 25 real calls a month, with subsequent calls at $3 each, and Pro at $150 for 75 calls, with extra calls on that tier at $2.50 (its 150-call tier is $270). It describes round-the-clock AI answering, recordings, transcripts, summaries, routing and scheduling. Moneypenny's live receptionist plans are billed by the minute instead: $265 a month for 100 minutes with extra minutes at $2.65, or $422 for 200 minutes with extra minutes at $2.39, with no contract and 30 days' notice to cancel. These are different service models, not interchangeable promises of clinical coverage.
Check the billing unit before comparing prices. A call-based product counts answered calls according to its own rules. A minute plan counts operator time and may have separate rules for connected transfers, outbound contact, hold time or voicemail. A flat monthly quote may include a defined number of calls or minutes and charge overage beyond it. None of those units tells you whether the vendor will page the right person when the rota changes on a public holiday. The operational requirement comes first; the price comparison follows.
| Cost line | Published example or quote item | Question to ask |
|---|---|---|
| AI per call | Smith.ai Free: 25 calls; then $3 per call | Which calls count, and can overage stop service? |
| AI bundle | Smith.ai Pro: $150 for 75 calls; extra calls $2.50 on that tier | Does a larger tier cost less than paying overage? |
| Live per minute | Moneypenny: $265 for 100 minutes, then $2.65 a minute; $422 for 200, then $2.39 | Are transfer and hold minutes billed? |
| Setup and changes | Neither pricing page prints a setup fee | What changes to scripts or the rota trigger a charge? |
| Practice labour | Internal estimate, not a vendor fee | Who reviews, corrects and closes each message? |
Do not read a general-purpose AI receptionist's feature list as evidence that it is suitable for your practice's patient data or emergency pathway. Obtain the agreement, security information, data-retention schedule and escalation specification in writing. A business associate agreement or data-processing agreement may be needed depending on your practice's obligations. Have the appropriate adviser check the contract. If a vendor cannot show how recordings, transcripts and messages are stored and deleted, exclude it from a patient-data pilot.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA four-week call log makes the quote comparable
Export or manually record four representative weeks of calls, separating opening hours, evenings, overnight and weekends. For each call, note the arrival time, duration, purpose, outcome, whether anyone was transferred, and whether a staff member called back. Keep identifiable patient details out of the comparison sheet. If the phone system reports only totals, sample a manageable set of recordings under your existing policy and count categories manually. Seasonal peaks need a second scenario: an ordinary month and a busier month.
Use categories that change the bill or the risk. One useful set is appointment administration, prescription or results request, new-patient enquiry, clinician message, urgent-sounding request, misdialled or sales call, and abandoned call. Split ordinary booking from calls in which a person describes symptoms. The latter may begin as an administrative request but must follow the approved clinical instruction. Track which categories require a transfer, a page, a secure message or no response outside opening hours.
Here is an illustrative practice calculation, not a claim about typical usage. A five-clinician practice receives 120 after-hours calls in four weeks: 65 are routine requests, 25 are booking enquiries, 12 are repeat or misdialled calls, 10 request the on-call clinician and eight are unclear. Suppose the 120 calls average three connected minutes, or 360 minutes altogether. Smith.ai's stated Free arithmetic would be 25 included calls and 95 extra at $3, giving $285, provided the plan's billing rules allow that usage. Its Pro plan at the 75-call tier would be $150 plus 45 extra calls at the published $2.50, or $262.50, just under the $270 of the 150-call tier. Moneypenny's live plans would be $265 plus 260 extra minutes at $2.65 ($954) on the 100-minute plan, or $422 plus 160 extra minutes at $2.39 ($804.40) on the 200-minute plan. The products do not deliver identical work, so the lower figure is only a starting point.
Now change one assumption. If half those calls last six minutes because the service keeps callers connected while seeking a clinician, the minute count rises sharply while the call count does not. Ask whether the supplier bills the entire connected duration, a lower post-connection rate or a capped transfer. Conversely, if the practice already has staff returning long calls and wants only a short message, the duration may fall. Put these assumptions on the quote request rather than accepting a marketing calculator's default.
A second illustrative calculation shows why internal time matters. If 30 routine messages each need four minutes of receptionist review and re-entry, that is two staff hours in the month. If unclear transcripts create 15 further six-minute callbacks, add another 90 minutes. The owner should assign an internal hourly cost to those three and a half hours, including supervision, before declaring the AI plan cheaper. A service that sends clean structured messages to the existing work queue might reduce that work; one that emails a long transcript with no category might increase it.
Specify the after-hours service in writing
The AI can tell callers the practice's opening times, collect non-urgent callback requests, identify which clinician or team should receive a message, and give only approved instructions. It should not diagnose, rank clinical severity, reassure a caller that waiting is safe, change medication advice or imply a clinician has reviewed the call. The protocol should be signed off by a clinical lead and rehearsed with reception staff. This is the cost-critical part of procurement because escalation mistakes create both risk and rework.
Write a short matrix: call type, information to collect, destination, response window, and fallback if delivery fails. For a routine appointment request, the message might contain callback number, preferred time and whether the caller is an existing patient; it goes to the morning office queue. For an urgent-sounding request, the system should stop routine questioning, read the practice's approved emergency instruction and attempt the nominated on-call route. The AI must not ask a long sequence of symptom questions to invent a triage result.
An illustrative script for a caller saying they have severe symptoms could be: “I cannot assess symptoms or give medical advice. If you believe this is an emergency, contact emergency services now. I can connect you to the practice's on-call service if one is available.” The exact words, red-flag phrases and local emergency instruction need clinician approval. Test that “I don't know if this is serious” does not produce false reassurance. The service should also say what happened when a transfer failed rather than falsely promising that somebody has been alerted.
On-call destinations change. Keep the rota in one owned system with a named updater, a backup contact and a scheduled verification before each handover. The AI vendor should explain whether it reads the current rota automatically or depends on someone changing a dashboard. A missed update can make a flawless voice agent page last week's clinician. Record who made each change and when, then make the weekly rota test a standing administrative task.
For a fuller design of the symptom and routing path, use the separate doctor's office after-hours call rules. The price decision here is whether the chosen product can perform that path, and what each step costs. If the vendor's AI cannot reach the on-call clinician reliably, paying for a human answering option or retaining the existing cover may be the sounder choice.
Check what the plan includes before signing
Ask for a line-item demonstration of call forwarding, message delivery, transfers, transcripts, recordings, caller verification, booking, language support and support availability. A vendor may list “routing” while leaving the practice to configure every number and fallback itself. Another may charge for account changes or human intervention. Get the current plan schedule, overage terms and implementation statement in one document. A summary of features on a public page is not a substitute for the actual service agreement.
A useful quote request contains the measured calls and minutes for ordinary and busy months, service hours, routine message categories, on-call rota size, number of fallback contacts, existing phone number, and destination for each message. Ask for one price covering the pilot and one covering full use. Require the vendor to identify every one-off charge, cancellation notice, minimum term, recording charge, training allowance, and fee for changing a protocol. If a feature requires an integration, name the system and ask for a live demonstration with a test record. “We integrate with calendars” is not proof that the practice's appointment rules will be enforced.
One quote may look cheaper because it excludes staff review. Put a line called “practice handling” beside the supplier's monthly fee: number of messages multiplied by minutes to read, verify, enter and close them. Add the time the clinical lead spends approving scripts and investigating exceptions. The practice may choose to spend that time for safety; the point is to budget it honestly. A supposedly automatic booking that requires reception to correct visit types can cost more staff time than a simple, accurate callback message.
When the practice receives proposals, request a sample invoice with the same 120-call, 360-minute scenario. Ask the vendor to mark which 10 urgent transfers, 12 repeats and eight unclear requests count as billable events. Ask what would happen if the on-call clinician answered on the third attempt, and whether those attempts generate separate units. This is a better comparison than a one-line “from $150” claim because it reveals where the vendor expects your team to take over.
Sample one: an existing patient asks for a repeat prescription at 22:00. A safe administrative flow collects a callback number and the requested item only if the practice has approved that data collection, then routes the message to the right morning queue. It should not promise a refill or claim a prescription is ready. If the practice's policy forbids leaving medication details with the service, capture a generic request to call back instead. That decision affects transcription and storage requirements as well as staff time.
Sample two: a caller asks to book a first consultation and mentions worrying symptoms. A booking-focused sales script may try to offer the next available slot. The practice's rule should interrupt that flow and provide the approved emergency or on-call instruction. In a trial, listen to the recording and inspect the delivered message. If the symptom was only in the transcript while the queue label says “new booking”, the routing is unsafe even though speech recognition worked.
Sample three: the on-call phone rings but nobody answers. Require a second contact, an attempt count or time threshold agreed by the practice, and a message to the caller that does not imply success. Ask for a delivery log showing each attempt. A plan that includes a “transfer” feature but lacks an escalation ladder can be poor value for this use. The emergency answering setup tutorial explains the general call-failure pattern; adapt its checks to your clinicians' policy.
Sample four: a caller hangs up after the greeting and calls back three times. Will the vendor bill four calls, one conversation or only answered calls? Will those calls appear as four patient messages? Ask the vendor and check the trial invoice against the call log. A small number of repeated calls can make a per-call plan less attractive than its headline price. Spam and wrong numbers need their own billing rule. Smith.ai's page says every answered call counts by default, calls from known spam numbers are filtered out before the AI answers, and customers may remove up to 10% of calls from each bill as spam; a misdialled patient is not spam, so ask how wrong numbers are treated and keep the contract wording for disputes.
Sample five: a patient asks what a test result means. The AI can acknowledge that a clinician must interpret results and collect a callback request according to policy. A wrong answer may sound confident and plausible, so include this exact request in testing. A transcript showing a “helpful” explanation is a failed call, not a successful deflection. Keep approved administrative facts separate from clinical material in the knowledge base.
Use the trial to test cost and safety together
Give the supplier an approved instruction sheet and a set of 12 to 20 test calls. Mix ordinary requests with ambiguous wording, silence, background noise, repeat callers and urgent phrases selected by clinicians. At least two staff members should score recordings independently: Did the AI identify the request, collect only permitted information, give the approved words, choose the correct destination and record whether the hand-off succeeded? Count a call as passed only when the whole path works. A pleasant voice is not the measure.
For example, a test caller says, “Can I come in tomorrow? I feel dizzy and my chest hurts.” The desired outcome is the approved emergency instruction and immediate routing attempt, not a calendar link. Another caller says, “Please cancel my appointment; my child is at school.” The service should not infer a patient's identity from that sentence. A third asks to talk to the clinician by name while the rota lists a different clinician; the current rota must win. These examples expose policy and configuration errors that generic vendor demos rarely reveal.
Run a small live pilot during a defined set of hours. Compare every AI record with the telephone system's record and the practice's work queue. Look for unlogged calls, duplicated tickets, wrong callback numbers, stale on-call destinations and messages delivered outside the promised window. The free-trial call checklist gives a broader test pattern. For a doctor's practice, add clinical lead sign-off and an incident route before real patient calls begin.
At the end of the pilot, recalculate the bill using actual counted calls or minutes. Add the staff time needed to review messages and fix errors. Compare the result with your current cover on equal hours and the same call mix. If the provider will not disclose how a transfer or extra call is billed, write that as an unresolved cost rather than inserting a guessed rate. Keep a copy of the quote, the scored calls and the rota handover procedure with the contract.
Choose the cheapest model that passes your hand-off test
A low-volume practice seeking routine message capture may start with a call-based AI plan, but only after the patient-data agreement and test calls pass. A practice with long on-call transfers should model the connected minutes explicitly. A practice needing a person to interpret complicated administrative context may compare human or hybrid cover. The AI receptionist and answering-service comparison helps distinguish these models without treating them as the same product.
Do not buy capacity for a hypothetical future call volume. Start with the smallest plan that covers the measured ordinary month, establish what happens at the cap, and set an alert before usage reaches it. A service that silently stops answering at a cap is unacceptable unless the practice's own phone route takes over. Recheck the arrangement after a rota change, a new clinic session or a sustained shift in call mix. The final approval should come from the clinical lead and the person accountable for the contract, not from a price table alone.
Questions to settle before buying call cover
Does a low-cost AI receptionist replace an on-call clinician?
No. A receptionist can collect a message and follow an approved routing instruction. Your practice must still nominate the clinician, keep the rota current, define the response window and specify what happens when a transfer fails. Ask the supplier to demonstrate that sequence in a test call.
Should we compare per-call and per-minute quotes using the same call count?
Use the same call log, but retain both call count and connected minutes. A practice with many short routine calls can see a very different result from one with lengthy transfers. Include failed transfers, spam, holiday calls, overage and any platform or setup fee in both scenarios.
Can the AI book appointments without staff review?
Only if the practice has explicitly approved the appointment types, eligibility rules and live calendar connection. Start by collecting a callback request, then test booking against double-booked slots, wrong visit types and cancellations. Urgent symptoms should follow the clinical escalation script instead of entering a routine booking flow.
Further reads
- Virtual Receptionist for Doctors: A Worked Call Flow and Script Set — Turn the chosen service model into call scripts.
- Paediatric Answering Service: Staff, AI or a Nurse Line? — See a different practice's staffing and escalation choice.
- Healthcare AI Call Centre: When a Group Practice Should Centralise Calls — Assess shared call cover across several practices.
- AI Chatbot for a Doctor's Website: Scope, Disclosure, Handover — A practice owner's set-up guide: scope sheet, disclosure line, hand-over rules, logging and named chatbot tools with costs.
- AI Platform for Doctors: One Suite or Separate Scribe and Phone Tools? — A buying rule for small practices: when one suite wins, when separate scribe and phone tools win, and the integration test to run first.
- Cliniko for Doctors: Fit, Cost and the AI Tools That Sit Around It — Cliniko is built for allied health. What that means for a doctor's practice, what it costs, and the AI scribes and phone tools that connect to it.
- What's the Best AI Tool for Therapists? A Rule by Practice Size — A rule for choosing a therapy AI tool by practice size, with list prices, BAA checks and worked costs for a solo, three-clinician and nine-clinician practice.
- AI Receptionist Pricing for Medical Offices: Per Minute or Booking — Per-minute, per-call and per-caller AI receptionist pricing compared for a medical office, with worked monthly costs, overage rules and BAA gating.
- AI Receptionists for GP Practices: A Four-Week Trial and Stop Rules — A GP practice's four-week AI receptionist trial, week by week: what to measure, what counts as a failure and the stop rules agreed before it starts.
- AI Scheduling for Doctors: Appointment Types and Waitlist Rules — How to configure AI scheduling for a doctor's practice: appointment types, buffers, overbooking limits, waitlist fill rules and a safe link to the EHR calendar.
- AI Scribes for Doctors: Six Note Errors to Catch Before You Sign — The six mistakes an ambient scribe makes in a consultation note, with a worked example and a two-minute check to run before you sign.
- How Can AI Help GP Practices? Eight Admin Jobs Beyond the Scribe — Eight admin jobs for a GP practice beyond the scribe, from letters and coding suggestions to rotas and complaint drafts, each with its check step.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Smith.ai AI Receptionist pricing and FAQ (smith.ai/pricing/ai-receptionist); Moneypenny answering-service pricing and FAQ (moneypenny.com pricing page, people-answering plans). Both checked 4 November 2026.