Usually not for single coffee orders, and sometimes yes for catering and large orders. It pays when missed calls that were real orders, such as office catering or pre-orders, are worth more than the monthly fee: about three extra large orders a month covers a $150 plan. If you miss fewer than six order-type calls a month, skip it.
Most vendor pages sell the idea without asking whether a coffee shop gets enough phone orders to matter. Many do not, since orders arrive at the counter, in an app or by a link. So the question is not "can an AI answer my phone" but "what is a missed call costing me, and is it more than the fee". The calculation is short, and one section below covers when to leave it alone.
Step one: count what is actually happening on your phone
Do not buy anything yet. For 30 days, keep a tally of every call you miss or cut short, and write one word for why: order, catering, question, supplier, sales, other. A phone provider's call log shows missed numbers; ring a few back to find out which were customers. A paper sheet by the till does the rest.
An illustration for a café that gets around 18 calls a day on busy weekdays. Over 30 days the tally might read: 540 calls, 86 missed. Of those 86: 22 were opening-hours questions, 17 supplier or delivery calls, 15 sales calls, 12 pre-orders, 9 catering or large-order enquiries, 11 other. That is 21 missed order-type calls out of 540 (4%), which is the number that feeds the calculation, not the 540.
Two checks make the tally honest. First, how many of the opening-hours questions would vanish if your Google Business Profile hours were right and your voicemail said them? Second, how many callers rang back or came in anyway? You cannot know the second number precisely, so assume half of them were lost.
The payback calculation, with numbers
The formula has four inputs.
- M, missed order-type calls a month (from your tally).
- C, the share you would convert if someone answered well. Be cautious: 40 to 60% for catering enquiries is plausible, but it is your number to test, not a fact.
- V, the average value of an order.
- G, the share of that value left after ingredients, cups and labour: your contribution margin.
Monthly gain = M x C x V x G. Compare it with the monthly cost.
The worked example uses the list price from Smith.ai's page as recorded in my facts sheet (September 2026): its AI receptionist is free for 25 calls a month, then $3 a call, or $150 a month on the Pro plan at $2 a call (about 75 calls). Say the café's average large order is $85 with a 60% contribution margin, which is $51 per converted order.
| Café | Missed order-type calls (M) | Converted at 50% (M x C) | Gain at $51 each | Plan cost | Result |
|---|---|---|---|---|---|
| Quiet neighbourhood café | 3 | 1.5 | $76.50 | $150 | Loses $73.50: do not buy |
| Café with a regular office trade | 10 | 5 | $255 | $150 | Gains $105 a month |
| Café with catering and pre-order trade | 21 | 10.5 | $535.50 | $150 | Gains $385.50 a month |
The break-even is where M x C x $51 = $150, so M x C is about 2.9 converted orders; at 50% conversion that is six missed order-type calls a month. Below six, the fee is larger than what it recovers. These are illustrative figures, so replace V, G and C with yours. If your average large order is $40 rather than $85, break-even rises to about 12 missed calls.
The table ignores two things in opposite directions. It ignores the owner's time saved, which is real: a missed call costs a ring back at 4pm. It also ignores bad calls: a voice agent that mishears a pickup time costs a refund and a regular, so subtract an allowance for errors. For the broader ROI method over three months, see how to measure an AI receptionist's return in the first 90 days.
Which calls should the receptionist take, and which should it not?
Catering and large orders
This is where the money is. A good script collects: date and time, number of people or items, drinks versus food, dietary needs, pickup or delivery, delivery address, payment method, contact name and number. The AI should read the details back, say a team member will confirm by message within a set time, and never quote a price outside your list. Anything over a threshold, such as 30 items or under 24 hours' notice, goes to a person.
A filled-in intake, illustrative: "Office order for Thursday 9 January, delivery at 8:30am, 20 coffees (8 flat white, 6 latte, 4 oat cappuccino, 2 decaf americano), 12 pastries assorted, one nut-free request, pay by invoice, contact [first name], [number]." The AI's output should look like that, in the same order, every time. If it drifts, tighten the script.
Pre-orders
Phone pre-orders for the morning run (a bag of a dozen sandwiches, a birthday cake for pickup) are fine to take, with confirmation by text. For a plain coffee order, a link to online ordering is cheaper. See what AI adds over a basic POS in a coffee shop, since your till's ordering tool may already do the job.
Supplier and delivery calls
Route them to the owner's mobile, or to a message with the caller's name and company. A milk supplier changing a delivery slot should reach a person. If the receptionist cannot recognise a supplier, set a rule: unknown business callers get "please leave a message for the owner" and a text goes to your phone.
Questions, complaints and allergens
Opening hours, location and menu questions are safe. Complaints go to a person, with the receptionist taking a name and number. Allergen questions need a human or a fixed scripted line pointing to your written allergen information; a generated answer is not acceptable. The reasoning is covered in whether an AI chatbot should answer allergen questions.
The no-buy case, and the cheaper fixes to try first
You should skip an AI receptionist if your tally shows fewer than about six missed order-type calls a month at current values, if most calls are hours and location questions, or if you rarely handle phone orders at all. Before spending anything, fix the leaks that cost nothing.
- Voicemail greeting: "You've reached [café]. We're serving customers, so we can't pick up. For opening hours and our menu, see [website]. For catering and large orders, text [mobile] or use the form at [link]. We reply within the hour." Thirty seconds, no fee.
- Missed-call text-back: many phone systems can send an automatic text with the ordering link when a call is missed. Check what your provider already includes.
- A catering enquiry form on your website with the same fields as the intake above. The 21-calls-a-month café above may find half of them move to the form once the link is on the voicemail and the door sticker.
- Google Business Profile: correct hours, holiday hours, an ordering link and a catering link, so fewer people call to ask.
- A fixed callback time: ring back unanswered numbers at 2pm after the lunch rush, when the shop is quiet.
Run these for a month and re-tally. If missed order-type calls fall below six, you have saved the fee. If they stay high, you have a clean baseline to evaluate a vendor against.
If the sum says buy: a 30-day test before you commit
- Pick a vendor with a free allowance or trial. The Smith.ai page I used listed a 25-call free allowance for its AI receptionist in September 2026; confirm the current terms. Human-answered plans on its main pricing page began at $300 for 30 calls, which changes the sum: at $51 per converted order that plan needs six converted orders a month.
- Write the call rules down first: what it may answer, what it must pass on, the hand-off number, and a clear "do not" list (no prices outside the price list, no allergen answers, no refunds).
- Run twelve test calls yourself, including a catering order, a pre-order with a changed time, a supplier call, an allergen question and an angry caller. The twelve test calls for a restaurant voice agent list the set to adapt.
- Go live on missed calls only first, with your phone ringing as normal and the AI taking overflow and out-of-hours calls.
- Read every transcript for the first two weeks, and count: orders taken, orders confirmed, errors, hand-offs.
- Re-run the sum with real numbers at day 30: gain from confirmed orders against fee and error costs.
The same vendor trial also teaches you how it feels for customers. If you would not want to be on the other end of the call, change the script or stop. For the comparison with a human answering service, AI receptionist or answering service for a booking-based business sets out the trade-offs.
The costs the plan price leaves out
The fee is only part of what an AI receptionist costs, and a fair sum counts the rest.
- Your setup time: writing the rules, loading the price list and menu, and testing calls. Allow six to ten hours the first time. At a notional $20 an hour that is $120 to $200, once.
- Review time: reading transcripts for the first fortnight and a sample each month. Thirty minutes a week is realistic. Over a year that is about 26 hours, so roughly $500 at the same notional rate, which is why a plan that only just clears break-even is not worth the bother.
- Phone changes: call forwarding or a new number, and a short period where staff are not sure which line rings where.
- Error costs: a wrongly taken order. If one in twenty confirmed catering orders needs a correction and each correction costs $20 of goodwill and time, ten orders a month costs about $10.
Put the setup and review time into the sum as a monthly figure. A $150 plan with $40 of your time attached is a $190 plan, and the break-even moves from six missed order-type calls to about eight at the $85 order value. That is still low for a café with an office trade, and high for one with a quiet phone.
A busy week, call by call
An illustration of what a week with an AI receptionist on overflow looks like, for the middle café in the table. The phone rings 112 times. The team answers 91. Of the 21 that would have been missed, the assistant handles them like this.
- Monday, 8:10am: a regular asks whether the 8am pastry delivery has arrived (a supplier call). It takes a message and texts it to the owner. Good outcome.
- Tuesday, 12:40pm: an office coordinator wants 24 sandwiches for Thursday. The assistant takes the details, flags the quantity as over the 20-item limit and texts the owner for confirmation. The owner replies at 2pm, and the order is confirmed.
- Wednesday, 9:05am: a caller asks whether the oat flat white contains nuts. The assistant says a team member will call back, and the owner does within the hour.
- Thursday, 4:50pm: a call about a lost bag. The assistant takes a number and a description, and nothing else.
- Friday, 7:30am: a pre-order for a birthday cake for Saturday. The assistant takes the details, the owner confirms by text.
Count the value: Tuesday's catering order and Friday's cake are real money. The other three are the owner's time saved, or neutral. At the end of the week the owner reads five summaries over a coffee, which takes ten minutes. That is the scale of benefit to expect: a handful of rescued orders and a quieter afternoon, not a transformed business.
Confirmation texts that keep the order honest
A phone order is only as good as its confirmation. Whether a person or the system sends it, use a fixed format and let AI draft the wording once. Before: "Got your order, see you Thursday." After: "Hi [first name], confirming your order for Thursday 9 January: 20 coffees (8 flat white, 6 latte, 4 oat cappuccino, 2 decaf americano) and 12 assorted pastries, delivered to [address] by 8:30am. Total $[amount], payable by invoice. Reply YES to confirm, or tell us what to change. [café name]". The reply is a second check, and catches misheard drink counts before the kitchen does.
A short prompt to produce this format from the call summary:
Turn this call summary into a confirmation text under 70 words. List items with quantities, state the date, time and pickup or delivery, show the total only if it is given, and end with "Reply YES to confirm or tell us what to change." Do not add items, prices or promises that are not in the summary.
Summary: [paste]Check that the total and item counts match the summary before sending; a model sometimes rounds a quantity or adds a friendly extra the café does not offer.
Three café types, three answers
A small kiosk with a queue and no seating. Calls are rare and mostly "are you open?". The answer is no. Fix the hours online and put a QR code for ordering on the window.
A café on a business street with a lunch trade. Calls spike at 12:15 when staff are busiest. A missed-call text-back with an ordering link handles most of it, and the overflow assistant is worth trialling only if your tally shows at least six lost order-type calls.
A bakery café that does celebration cakes and office catering. Calls are fewer but valuable, and they come in during service. This is the closest fit for an AI receptionist, as long as the intake form and the human confirmation are in place. A single $120 cake order recovered each week more than covers the fee.
If your sums say "buy", also read what AI phone answering costs and pays back for a restaurant for a view of how per-call pricing behaves when volumes rise. If catering is the main driver, corporate catering proposals and AI outreach covers the proposal side once the call has landed.
A call script that works for a café
Greeting: "Thanks for calling [café]. I'm the café's phone assistant. I can take catering and large orders, pre-orders and general questions. How can I help?"
Rules:
- Collect: name, phone number, date, time, quantity, items, dietary needs, pickup or delivery.
- Read the order back and ask "is that right?".
- Say: "A team member will confirm by text within [two hours]."
- Never quote a price that is not on the price list below.
- Allergen or ingredient questions: "I'll ask a team member to call you back, as that needs a person."
- Complaints, refunds and anything unclear: take name and number and say the owner will call back today.
- Supplier or delivery callers: take company, name, message; do not discuss orders.A realistic sample output from the system after a call (illustrative): "Catering enquiry. [first name] from an office team, Friday 24th, pickup 8:15am, 15 coffees (mixed), 10 croissants, one nut allergy mentioned. Asked about price per head; told a team member will confirm. Phone [number]." What to fix: the nut allergy line must trigger a human call, not a text; add that to the rules. The summary should reach you as a text or email within a minute of the call, not wait until close.
Mistakes that make an AI receptionist cost more than it earns
Counting all calls as value. Opening-hours questions answered by an assistant do not earn anything. Only order-type calls enter the sum.
No hand-off. A customer who asks for "a person" and cannot reach one tells friends. Always provide a callback route.
Quoting from memory. A model that has never seen your price list will invent one. Give it the list, and ban prices outside it.
Letting it confirm what the kitchen cannot make. A 60-sandwich order for tomorrow at 7am needs a human check against capacity. Set a size limit that triggers confirmation by a person.
Forgetting to retire it. If a month's data shows fewer than six confirmed order-type calls, cancel. Note the cancellation terms before you sign; plans differ on notice and refunds.
Start the call tally before you price a receptionist
Start the tally today, set the voicemail greeting, build the catering form and check your Google profile. In four weeks, you will know your own M, V and G, and the answer will be a number, not a sales pitch. If you would like help working through the figures, my AI implementation consultation can go through the call log with you. For the wider picture of where AI saves café owners time, the best AI tools for cafés and coffee shops, grouped by task puts phones in context beside rotas, stock and social posts, which for many shops are bigger wins than the phone.
Questions café owners ask before buying
How many calls does a coffee shop need before an AI receptionist pays?
Count missed calls that were orders, not total calls. With an $85 catering order and a 60% contribution, about three extra converted orders a month cover a $150 plan. If you miss fewer than six order-type calls a month, a text-back and a good voicemail greeting are enough.
Can an AI receptionist take a coffee order over the phone?
It can take structured pre-orders such as drinks, quantity, pickup time and name, but a person or your till system should confirm them. For single walk-in style orders, an online ordering link is cheaper and more accurate than a voice agent.
Should it answer supplier calls?
No. Route suppliers and delivery drivers to your mobile or take a short message. A supplier who hits a bot may hang up and move your delivery slot.
What about allergen questions on the phone?
Send them to a trained member of staff, or give a fixed scripted answer pointing to your written allergen information. Do not let a generated answer decide what is safe for someone with an allergy.
Is there a free way to test the idea?
Yes. Log every missed call for 30 days with its reason, and use a free trial or free-call allowance if a vendor offers one. For example, Smith.ai's page listed a free allowance of 25 calls for its AI receptionist in September 2026; check the current page.
Further reads
- What to Do When an AI Receptionist Gets a Booking Wrong — What to do when it gets something wrong.
- Coffee Shop Inventory With AI: Ordering Milk, Beans and Pastries — A job that saves more money than phones.
- How Can a Coffee Shop Use AI? Start With Supplier Prices — Another place where café AI pays.
- AI Receptionist for Restaurants: Yelp Host, Slang, Popmenu, Maple — Compare Yelp Host, Slang.ai, Popmenu AI Answering and Maple on real call handling, with an eight-call test you can run yourself.
- Virtual Receptionist for a Small Bookstore: AI or Human Answering? — AI or human phone cover for a bookshop: monthly costs, what each does with stock checks, special orders and events calls, and a worked sum.
- ChatGPT Prompts for Coffee Shop Owners: 10 for Menus, Posts, Rotas — Ten ChatGPT prompts for a coffee shop, from cost per cup to rotas and posts, each with example inputs and a check before you use it.
- AI Receptionist Kiosk or Phone AI: What a Walk-In Business Needs — Phone AI or a counter screen? A four-week tally, a music school worked through, list prices and the failure modes to test before you pay for either.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Smith.ai pricing page and facts verified in September and October 2026. All call counts, order values and margins in the examples are illustrative.