An AI receptionist is worth it for a small business when missed calls are costing you bookings: it answers every call, takes messages, answers common questions and, with some services, books straight into your calendar. It is weaker on unusual requests, upset callers and anything that needs your judgment, so the safe way to decide is to test it for a week with your ten most common calls and a clear way to hand a caller to a human. Pricing is usually per call, per minute or per customer, so work out what your own call volume would cost before you sign up.
Who it is worth it for (the missed-call math)
You need your own numbers, not industry averages, and you can get them in a week.
- Count missed calls. Check your phone’s call log or your phone provider’s dashboard for one normal week. Write down how many calls you did not answer and did not return within an hour.
- Estimate how many were real customers. Some will be spam or existing clients. Be honest. If you listen to voicemails, you will get a rough idea.
- Estimate how many would have booked. Use your own experience. If about one in three new callers usually books when you do answer, use that.
- Multiply by what a booking is worth to you.
Missed real customers per month, times your booking rate, times your average job value, gives you a rough monthly figure for lost work. If that figure is much bigger than the price of the plan you would need, a test is worth your time. If you miss two calls a month and call both back within the hour, it probably is not.
This math matters most for businesses where the owner is the one working: a hairdresser mid-cut, a plumber under a sink, a therapist in a session. Our guides for salons and spas and tradespeople look at those situations in more detail.
What it does well
AI receptionists are good at calls that follow a pattern. The three services in the pricing section below list features like these on their plans, and you should check any service you try for the same:
- Answer 24 hours a day, and on some services handle several calls at once.
- Take a message and send you a summary, transcript or recording by text or email.
- Answer common questions you have taught it, such as where you are, what you offer and how to book.
- Send a booking link by text, or book directly into a connected calendar on some plans.
- Filter out known spam numbers.
- Transfer a caller to you or a colleague when a rule says so.
The pattern is the point. If most of your calls are “Are you open Saturday?”, “Can I book a trim?” and “Do you do boiler servicing?”, an AI receptionist can deal with them. If most of your calls are long conversations about a complex job, it will mostly take messages.
Where it fails
This is the part vendor pages skip, so plan your test around it. Watch for these problems:
- Questions you did not teach it. When a caller asks something outside its instructions, a badly set up AI may guess. A guessed price or a guessed opening time is worse than “Let me take a message.”
- Upset or confused callers. Someone who is angry, distressed or hard to understand needs a person quickly. Check how fast the AI hands over and whether it gets stuck repeating itself.
- Judgment calls. “Is this leak an emergency?” or “Can you fit me in before my wedding?” need you. The AI should take details and pass them on, not decide.
- Names, numbers and addresses. Spelling a name, catching a phone number or an unusual street name over a bad line is hard. Read the transcripts to see what it gets wrong.
- Calendar mix-ups. If it books into your calendar, check for double bookings, wrong appointment lengths and bookings outside your hours.
- Callers who will not talk to a machine. Some people hang up. Your call log will tell you whether that happens often.
Cost: how pricing is structured
Pricing changes often, and services count usage in different ways. Here are three real examples, each from the vendor’s own pricing page, checked on 25 October 2026. They are examples of structure, not recommendations, and prices may have changed since.
| Structure | Example | What the pricing page said on 25 Oct 2026 |
|---|---|---|
| Per call | Smith.ai AI Receptionist | Free plan with 25 calls a month, then $3.00 per call. Pro from $150 a month for 75 calls ($2.00 per call). |
| Minutes included in a monthly plan | Rosie | Professional $49 a month with 250 minutes; Scale $149 with 1,000 minutes; Growth $299 with 2,000 minutes (monthly billing). |
| Per unique customer | Goodcall | Starter $79 a month for 100 unique customers, then 79 cents each. Growth $129 for 500. |
What to check before you compare prices:
- What counts as usage. A call, a minute and a unique caller are very different units. A 30-second wrong number costs differently under each.
- What happens when you go over. Look for the overage rate on the pricing page.
- Which features sit on which plan. On Rosie’s page, for example, booking directly into your calendar and live call transfers are listed on Scale, not Professional.
- Contract and notice period. Smith.ai’s page says it is month to month with 30 days’ notice to cancel. Check this for any service you try.
Take your call count from the missed-call math above and work out what one month would cost on each structure.
A one-week test plan
All three services in the table offered a free plan, free trial or free demo agent when we checked on 25 October 2026, so you can often run this before paying for a full month. Check the current terms on the vendor’s site.
Day 1: Write the script. List your ten most common calls. For each, write the answer you want given, or the rule (“take a message”, “transfer to me”, “send booking link”). Add your hours, services, location and what to say when it does not know.
Day 2: Make test calls yourself. Call it ten times, once for each common call. Then call five more times with awkward ones: a question you did not teach it, a vague caller, an upset caller, a request for a price you have not given it, and a caller asking for a human.
Day 3: Fix and retest. Read every transcript. Tighten the instructions where it guessed or got stuck, then repeat the awkward calls.
Days 4 to 7: Overflow only. Send it only the calls you would have missed: after hours, or when you do not pick up after a few rings. Keep answering the rest yourself.
Each evening, track four things:
- Calls it answered.
- Calls it handled fully (answered, booked or took a complete message).
- Calls it handed to a human, and whether the handover worked.
- Mistakes: wrong information, wrong booking, wrong name or number.
At the end of the week, compare what it caught with your missed-call math, and decide based on your own numbers.
Rules to set before you switch it on
Write these into its instructions, whatever service you use:
- Never promise refunds, discounts or exceptions.
- Never quote a price you have not written into its instructions.
- Never give medical, legal or safety advice. Take details and pass them on.
- Never confirm a booking it cannot see in the calendar.
- Always offer to take a message or transfer when unsure.
- Always say how and when you will get back to the caller.
The same thinking applies to AI help with written messages. Our guide to answering customer emails faster with AI covers where you still need to check the draft.
Telling callers it is AI
Some services let you choose the greeting and the agent’s name. A plain greeting that says what it is can help, for example “You’re through to the virtual assistant for [business name]. I can take a message or help you book.” It sets expectations and gives callers who want a person a reason to ask.
Rules about telling callers they are speaking to an automated system, and about recording calls, differ between countries and regions. This guide does not give legal advice. Check the rules where you and your callers are, and check what the service records and how long it keeps it, before you go live.
For other tools that save time without being a chatbot, see useful AI tools that aren’t chatbots, or browse our customer guides.
FAQ
Is an AI receptionist worth it for a small business?
It can be if you regularly miss calls from new customers. Count your missed calls for a week, estimate how many would have booked, and compare that with the price of a plan that covers your call volume.
What can an AI receptionist actually do?
Answer calls around the clock, take messages, answer questions you have taught it, send booking links or book into a connected calendar on some plans, filter spam and transfer calls to a person.
How much does an AI receptionist cost?
It depends on the service and how it counts usage: per call, per minute or per unique customer. On 25 October 2026, published plans we checked started at $0 (with a call limit), $49 and $79 a month. Check current vendor pricing pages.
Will customers know they are talking to AI?
They may. Some services let you set a greeting that says so. Telling callers up front sets expectations. Rules on disclosure and recording vary by place, so check the ones that apply to you.
How do I test an AI receptionist before I commit?
Use a free plan, trial or test calls. Make test calls with your ten most common questions plus a few awkward ones, fix the instructions, then send it only overflow calls for a few days and track what it gets right and wrong.