For after-hours paediatric calls, use staff or a nurse line for anything a parent says about symptoms, injury, medicines or a baby they are worried about, and use AI only for non-clinical calls such as appointment requests, directions and messages. A hybrid usually costs least. AI must never triage.
The cost question is secondary. The first decision is which calls the practice is willing to let a machine answer at all, because that sets the route for every call after midnight. Sort last month's calls into groups, convert every quote into a cost per call on that mix, put the routing and callback wording into the contract, and run test calls before a worried parent finds the gap for you.
Sort your own calls before you ask for a single quote
Pull your last month of after-hours messages from your current voicemail, answering service or on-call log. Do not guess. Sort each into one of five groups. The share in each group changes which service suits you.
| Group | Examples | Who should answer |
|---|---|---|
| A. Clinical concern | Fever, breathing, rash, vomiting, injury, medicine question, "something is wrong with my baby" | Trained nurse or on-call clinician, live |
| B. Urgent but not clinical | Parent cannot reach the clinic at all, lost prescription pickup, safeguarding concern raised by a caller | Human, with a clear route to the on-call clinician |
| C. Appointments | Book, move, cancel, run late | AI or human; AI handles it well when the rules are fixed |
| D. Admin messages | Forms, letters, billing questions, results enquiries | AI or human takes a message for the morning |
| E. Information | Opening hours, location, parking, vaccine clinic times | AI from an approved fact sheet |
Illustration only: say a three-clinician practice finds 220 after-hours calls in a month, and sorting them gives 70 in group A, 10 in B, 60 in C, 50 in D and 30 in E. Group A is 32 percent of calls and 100 percent of the risk. Groups C, D and E, 140 calls or 64 percent, are the ones AI can handle without anyone getting hurt if it fails. That split, not a vendor's brochure, is the reason a hybrid is worth pricing.
What each model can and cannot do
Three models exist, and vendors mix them. A human answering service uses live operators who follow your script. A nurse line connects the caller to a registered nurse trained in paediatric triage. An AI service answers the phone with a voice agent that follows rules you set.
- Human answering service. NotifyMD describes a human-staffed medical answering service whose platform guides live agents through each call and lists integrations with Epic, athenahealth, eClinicalWorks and NextGen. Its specialty list on the page does not include paediatrics, so ask for evidence of paediatric experience. MAP Communications describes general answering services with health-privacy-compliant support where needed, month-to-month agreements and billing only for the time its staff spend speaking with your callers; its home page does not mention paediatrics either.
- Nurse line. AccessNurse's page describes a paediatric programme called TeamPeds, with an RN trained in paediatric triage. The page gives no price, contract term or integration details, so those are questions for sales.
- AI voice agent. Smith.ai lists, as of November 2026, a free AI receptionist plan with 25 calls a month and then $3 a call, a Pro plan from $150 a month for 75 calls with $2.50 overage, and human receptionist plans at $300 for 30 calls, $810 for 90 and $2,100 for 300. It is a general-purpose service, not a paediatric nurse line, so treat it as an example of how AI and human pricing are published rather than a recommendation. Ask any vendor whether it signs a data agreement for health information.
None of the pages above publishes a paediatric nurse-line rate, and neither NotifyMD nor AccessNurse shows any price at all. That is typical: nurse-line pricing is usually quoted per practice. Do not accept "we will send you a quote" without asking for the unit it will be stated in (per call, per minute, per month, per patient on your list).

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 nowConverting every quote to a cost per call
With the 220-call example and these placeholders, here is the sum. Illustrative figures only, except the Smith.ai prices above.
| Route | Monthly cost on 220 calls | Cost per call | Note |
|---|---|---|---|
| Human service at $9 per call (the per-call figure from $810 for 90 calls) | $1,980 | $9.00 | Plans have call limits; check overage |
| AI only at $3 per call | $660 | $3.00 | Cheap, but it cannot do group A |
| Nurse line, placeholder $25 per call on group A | 70 x $25 = $1,750 | $25.00 on those calls | Replace $25 with a written quote |
| Hybrid: AI for C, D, E (140 calls at $3) plus nurse line for A (70 at $25) plus human for B (10 at $9) | $420 + $1,750 + $90 = $2,260 | $10.27 blended | Most coverage, clear routing |
Compare the hybrid with a nurse line and a human service handling everything. If group A goes to the nurse line at $25 and the remaining 150 calls go to the human service at $9, the cost is $1,750 plus $1,350, which is $3,100. The hybrid saves $840 a month in this made-up case, about $10,000 a year, by letting AI take the 140 routine calls. Change the nurse-line quote and the whole result shifts, which is why the placeholder must be replaced before you decide. If your own call mix has 60 percent in group A, the saving shrinks and a nurse line with a human backup may be simpler.
Per-minute pricing needs a second conversion. MAP Communications is one service that publishes its rates: its Business plan lists $179 a month for 125 minutes, then $1.30 a minute. If your calls average 3.5 minutes, those 125 minutes cover about 35 calls and each call after that costs about $4.55; if parents of newborns talk for eight minutes, the allowance covers about 15 calls and each extra one costs $10.40. On the 150 non-nurse calls in the example, at 3.5 minutes each, that is 525 minutes: $179 plus 400 minutes at $1.30, or $699, about $4.66 a call. Add the $1,750 nurse line and that human-plus-nurse route costs about $2,449, only $189 more than the hybrid, which shows how far the answer moves with the quotes you actually receive. Ask for your own average from a pilot month and compare against the call-based quotes, not a brochure.
What must reach the nurse line, written as a rule
Your clinicians should own this list, and the service should apply it without interpretation. A starting draft (to be edited by your clinical lead) might read:
ALWAYS ROUTE TO NURSE LINE (no AI advice, no reassurance):
- Any mention of: fever, temperature, breathing, wheeze, rash, vomiting,
diarrhoea, dehydration, injury, fall, burn, bite, seizure, medicine,
dose, swallowed, allergic, "not feeding", "floppy", "very sleepy".
- Any caller aged under 1 year, or a parent who says "newborn" or "baby".
- Any caller who says they are worried, scared or "something is wrong".
- Any caller who asks "should I go to the hospital?".
- Any repeat call within 30 minutes about the same child.
IF THE NURSE LINE IS BUSY: keep the parent on the line or offer a
guaranteed callback within [X] minutes; never offer a symptom message
for the morning.
EMERGENCY LINE (fixed wording, approved by the clinical lead):
"If you think your child's life is in danger, call [emergency number] now."
This list has two safety features. It is broad, so over-routing is the failure mode, not under-routing. And it contains no advice: the emergency line is a fixed sentence, approved in advance, that the AI reads but never edits. An AI that is allowed to improvise reassurance is the failure you are designing against.
For the general rules on emergencies in an AI answering setup, see setting up an AI answering service that never misses an emergency. For paediatric practices that also run a nursery or childcare partnership, is it safe to use AI with children's data? sets out the data questions.
Parent callbacks: the wording that keeps trust
Parents who are told "someone will call you" want a time and a person. Write the callback promise as a fixed sentence and the practice as the party responsible: "A nurse will call you back within [X] minutes on the number you gave. If you have not heard from us by then, call this number again and say 'callback'." Then measure it. The number that matters is the share of group A callbacks made within the promised time, taken from the service's logs. If the vendor cannot give it, you cannot manage it.
For group C and D calls that AI takes, the morning follow-up should arrive before the practice opens, in a shared list, not in an inbox someone has to find. A sample AI-drafted morning list (illustrative) reads: "Call 1, 20:14, rescheduling Tuesday 09:30 for [child first name], booked Thursday 11:00, confirmation sent. Call 2, 21:40, form request for school, left for admin." Short, with the time, the action and the status.
Contract terms to ask for in writing
- Term and exit: month to month or a fixed term? MAP's public page says month-to-month with no long-term commitment; confirm any such statement in the contract itself.
- Rate card: unit, minimum, overage, and charges for holidays, after midnight, extra scripts and extra integrations.
- Response times: pick-up time for live calls and callback time for nurse calls, with a remedy if missed.
- Clinical scope: who is on the nurse line, their qualifications, the protocols used, and who is accountable for advice given.
- Data: a signed data agreement, where recordings and notes live, how long they are kept, and how you delete them.
- Handoff to your on-call clinician: how, in what format, and what happens when the on-call phone is not answered.
- Reports: monthly call counts by group, response times, abandoned calls and repeat calls.
Have a qualified adviser read the contract and the data agreement; the clinical scope clause in particular is the one to have your insurer see. The group-practice call-centre guide covers how a practice with several sites should run this centrally.
The routine-call script AI can safely run
Once group A is protected, the AI side can be tightly scripted. Here is a filled-in example for appointment requests (illustrative):
OPENING: "You have reached [practice name]. I am the practice's automated
assistant. If your child is unwell or you are worried, say 'nurse' now and I
will connect you."
IF the caller says "nurse", mentions any symptom or worry: go to the nurse
route. Do not continue the appointment script.
IF the caller wants to book, move or cancel:
1. Child's first name and date of birth.
2. Parent's name and best number.
3. Offer up to three slots that the booking system shows. Never invent one.
4. Confirm the slot, say a text will follow, and ask "Is there anything
else I can pass to the team?" Record the answer word for word.
NEVER: ask about symptoms, comment on a child's health, give a dose, or
tell a parent whether to come in.
Notice the first sentence: the caller can say "nurse" at any time and escape the script. That small door is worth testing in every call. A script with no exit pushes anxious parents into a loop, and a parent in a loop at 02:00 is a complaint waiting to happen.
Peak nights, holidays and the day the nurse line is busy
After-hours demand is lumpy. Winter evenings, school holidays and the first night of a public holiday weekend produce more group A calls than a normal Tuesday. A service that is fine at 220 calls a month may queue at 40 calls in a single night. Ask each vendor for three things: how many nurses are on shift on your busiest nights, what happens when all are busy (hold, callback, or overflow to another team), and whether an overflow team has the same paediatric training. If the answer to the last one is no, the overflow should go to your on-call clinician.
Plan your own holiday cover too. A hybrid where the AI answers group C, D and E calls on a public holiday can still look quiet because parents book fewer routine appointments. The real test is group A, so run a scripted test call on the first night of every long weekend and read the log the next morning.
A monthly review you can do in 30 minutes
Whichever route you choose, put the same review in your calendar. Illustrative numbers for one month are shown below so you can see what a useful report looks like.
| Measure | Target you set | This month (illustrative) | Action |
|---|---|---|---|
| Group A calls reaching a nurse | 100% | 70 of 70 | None |
| Nurse callbacks within promised time | 95% | 64 of 70 (91%) | Ask vendor for the late six and the reason |
| Abandoned calls (caller hung up) | Under 3% | 9 of 220 (4%) | Check hold message and peak nights |
| AI calls that needed a person after all | Under 10% | 11 of 140 (8%) | Read the transcripts, fix the script |
| Repeat calls about the same child in 30 minutes | Watch | 6 | Read these first; they show unmet worry |
| Complaints from parents | Zero tolerated | 1 | Call the parent and review the call |
The targets are examples, not benchmarks; set them with your clinical lead. The most useful column is the last. A review that produces no action is a review that is not reading transcripts. For the first ninety days, measuring an AI receptionist's return gives a way to track the cost side alongside this.
An email to send to each service before you sign
Subject: After-hours service for a paediatric practice: written answers
Hello [first name],
We are a paediatric practice with [number] clinicians and about [number]
after-hours calls a month. Please answer in writing:
1. Which of your services are live humans, which are AI, and which are
nurses? Who employs the nurses and what is their paediatric training?
2. Your rate card: unit (call or minute), minimums, overage, holiday and
overnight surcharges.
3. How do you route a call that mentions a symptom? Can we supply our own
trigger list, and will you apply it exactly?
4. Your callback time for nurse calls and what happens when it is missed.
5. Your data agreement, recording storage and retention, and deletion on request.
6. Contract term, notice, and what we receive when we leave.
7. A sample monthly report with names removed.
Thank you,
[your name]
If a vendor will not put the answer to question 3 in writing, do not proceed. Everything else can be negotiated; the routing rule cannot.
Test calls to run before you go live
- A parent says their toddler "has a bit of a temperature and is sleepy". The call must go to the nurse line with no AI reassurance.
- A parent asks to move tomorrow's 09:00 check to the afternoon. The AI should offer valid slots and send a confirmation.
- A parent calls twice in 20 minutes about the same child. The second call should be flagged as a repeat and go to a nurse.
- A caller asks for the clinic's opening hours and where to park. The AI answers from the fact sheet only.
- A caller says only "I do not know what to do". It must be routed to a nurse, not treated as an unknown topic.
- The nurse line is busy. The caller gets the fixed holding message and a callback promise, not a message for the morning.
Run these scripted calls in the evening and at 03:00, because services are often staffed differently overnight. Keep a log with the time, call, result and the fix. If a test fails, do not go live until it passes twice in a row.
A realistic mistake: the AI that tried to help
Illustrative. A practice launches an AI receptionist and instructs it to "be warm and reassuring". A parent calls about a child who has vomited twice, and the agent says "that is very common and usually nothing to worry about, but we will pass your message on". There is no nurse transfer because the trigger list did not include "vomited". Nothing bad happens, but the practice reads the transcript the next day, rewrites the rules so that "warm" never includes any statement about symptoms, adds the trigger words and tests again. The cost of the lesson was a transcript review the practice should have done in week one. Read ten transcripts a week for the first month.
Telling parents what you have changed
Parents notice a new voice on the line. Tell them before it happens: a short note on your website, in your new-patient pack and in the last line of your voicemail greeting. Illustrative wording: "After hours, your call is answered by a service that can book appointments and take messages. If your child is unwell or you are worried, say 'nurse' and you will be put through to a nurse. In an emergency, call [emergency number]." Keep it to three sentences, and make sure it says the same thing as the script.
Also tell your own staff what the morning list will look like and who clears it by 08:30. A good service with no owner in the morning produces a queue of unanswered messages by Friday. Name one person, give them a checklist, and review the list in the weekly meeting for the first month. If the same message type appears every night, such as school form requests, turn it into an automated reply on your website rather than a call.
A starting choice by practice type
Small practice with one or two clinicians and an on-call rota. Use a human answering service for everything, add AI for group C, D and E calls after a month of listening, and keep the nurse line or on-call clinician as the destination for group A.
Practice that already contracts a nurse line. Add AI in front of it for the non-clinical calls only, and keep the nurse line's own rules for symptom calls.
Multi-site group. Centralise the after-hours route, because the nurse line and the AI both work better with a single script and a single log. The comparison for other sectors, such as AI versus human answering for HVAC firms, uses the same cost-per-call sum with different stakes.
If you would rather map your call groups and routes with someone, that is what my AI implementation consultation covers.
Paediatric after-hours calls: questions to settle first
Can an AI answering service triage a sick child?
It should not. Symptom triage is a clinical judgement, and a wrong answer to a parent about a feverish baby is not an acceptable error rate. Use AI for non-clinical calls such as appointment requests, directions and message-taking, and route anything about symptoms to a trained nurse or on-call clinician.
What should always go to the nurse line?
Every call where a parent describes symptoms, an injury, a medicine problem or worry about a baby. Let the answering service sort by a fixed list of trigger words and questions, and when in doubt send it through. Your clinicians decide the list; the service applies it without interpretation.
How do per-call and per-minute prices compare?
Per call is simple to budget, and per minute rewards short calls but punishes anxious parents who talk for ten minutes. Convert each quote to a cost per call using your own average call length. Nurse-line prices are rarely published, so ask for a written rate card with every extra fee.
Do I need a business associate agreement or processing agreement?
Almost certainly, because calls contain children's health details. Ask each vendor whether it signs one, where recordings are kept, for how long and who can listen. Have a qualified adviser read the agreement, and keep the signed copy where your compliance lead can find it.
Is a hybrid worth the extra complexity?
It can be. A hybrid sends routine calls to AI and symptom calls to a nurse line, which cuts the number of calls that need paid human time. The extra complexity is the handoff, so test it with scripted calls before you go live and again after each change.
Further reads
- How to Set Up an AI Phone Agent for After-Hours Calls — How to set up an AI phone agent for after-hours calls.
- What Is the Cheapest Way to Handle Enquiries Out of Hours? — Cheaper ways to cover out-of-hours enquiries.
- AI Receptionist or Answering Service for a Booking-Based Business? — Which suits a booking-based business.
- AI Answering Service for a School Office: Absence Calls and Cost — Absence calls and cost for a school office.
- After-Hours Dental Answering Service: Can AI Take Emergency Calls? — The same question for dental emergencies.
- AI or Human Answering Service: Which Suits a Property Manager? — Human versus AI answering for a property manager.
- Answering Service for Towing Companies: Night-Call Cost, Human or AI — Night-call cost for towing firms, human or AI.
- AI Answering Service for Doctors: What It Costs and What You Get — Compare per-call and per-minute bills, then specify the on-call hand-off your practice actually needs.
- AI Answering Service for a Doctor's Office: After-Hours Call Rules — Build an after-hours call matrix, approved emergency script and test set before forwarding patient calls to AI.
- Best AI for Paediatricians: Vaccine Recalls and School Forms — Build a vaccine recall list, letter templates and a school and camp form queue with AI, keeping parent data safe and symptom advice out.
- AI Tools for Paediatricians' Offices: Seven Admin Jobs — Seven front-office jobs for AI in a paediatric practice, from phones to rotas, each with an example and a rule for using it safely.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: AccessNurse and NotifyMD home pages (no prices); MAP Communications home and pricing pages; Smith.ai human and AI receptionist pricing pages (all checked 4 November 2026). Placeholder rates for a nurse line are illustrative and must be replaced with written quotes.