Scenario · Coffee shop
AI receptionist for coffee shops.
An AI receptionist for a coffee shop reads the catering, room-hire and wholesale enquiries that arrive while the bar is three deep, drafts each reply from your own menu and house rules, and leaves the send button with you. One AI employee costs $149 a month on the Starter plan.
It runs over text: a form or quiz posts to the lead endpoint and the enquiry is stored before anything else happens. A Saturday-night catering request is read and drafted on a Saturday night, and the reply waits for whoever opens the shop. Phone answering is custom scope on an Enterprise plan, which the section on calls sets out in full.
Unistaff · Coffee shop
Counter
What arrives at this desk
The mix below is the deployment shape this scenario is written around. The timing column is the reason the work slips: most of it lands while the room is busy or after everyone has gone home.
Office catering for 20 people and up
Weekday mornings
Private hire of the back room
Evenings, weekends
Wholesale beans for an office nearby
Anytime
Hours before a long weekend
The week before
Oat milk, allergens, dogs, wifi, power outlets
All day
Seating for a group of twelve
Friday afternoons
“Are you hiring?”
Anytime
Something left on a table this morning
Same day
What it takes on, what stays with a person
Read the person column as the job that remains, and staff for that job rather than for the whole desk. Every row matches what the platform does today.
| Task | AI employee | Person |
|---|---|---|
| First reply to a catering enquiry | AI employeeDrafts it on the next run with your lead time, your minimum order and the headcount and dietary notes the form captured | PersonConfirms the date and sends it |
| Repeat questions — hours, milk, allergens, dogs, wifi | AI employeeAnswers from the menu, the service list and the house rules in one pass | PersonKeeps that material current |
| A price for a tray or a hire | AI employeeMarks every number as pending your approval — the validator rejects a draft that states a price without the marker | PersonConfirms the number before it goes out |
| Room hire | AI employeeCollects the date, the headcount, the window and what they need on the table | PersonChecks the diary and answers on availability |
| A wholesale enquiry | AI employeeQualifies volume, grind and delivery cadence into structured fields | PersonDecides whether to take the account |
| A complaint about this morning's order | AI employeeStops, says why it stopped, and hands over what the customer wrote | PersonAnswers it personally |
| The month's enquiry volume | AI employeeJournals every run with its inputs, its outcome and its error text | PersonReads the journal and corrects the material |
Where it stops
A specific date's availability, a complaint about an order, a custom roast or cake, anything that needs the roaster or the owner's judgement.
What the owner keeps
The send button, the diary and the final number.
A Saturday catering enquiry, hour by hour
The loop runs on a timer — every fifteen minutes in the reference deployment — and each pass takes at most five enquiries that have no draft yet. Here is what that looks like on the weekend the shop is busiest.
- Sat 22:07
The form arrives
Thirty people, Thursday morning, oat and regular, one nut allergy. The enquiry lands in the database before anything else runs; the body is capped at 10 KB and a hidden field absorbs bots.
- Sat 22:15
The draft is written
The next pass picks the enquiry up and writes two to four sentences from your material: your catering lead time, the headcount and dietary notes read back, and a price band marked pending your approval.
- Sun 07:40
A person sends it
Whoever opens the shop reads the draft, confirms Thursday against the diary, adds the customer's name and hits send. The full name, phone number and email never entered the prompt.
- Mon
The run is in the journal
Inputs, outcome and error text are stored, so the share of drafts that went out unedited is a query rather than an impression.
Phone calls, honestly
Most pages selling an AI receptionist put a phone number in the hero. Text intake is what runs here today: a form or quiz posts to the lead endpoint, the enquiry is stored, and one email notification goes out with a five-second timeout — the enquiry reaches the database first, so a mail failure costs a log line while the record stays.
Voice is custom scope. The published plans are Starter at $149 a month, Scale at $599, and Enterprise at custom pricing; custom roles and custom workflows appear on the Enterprise column of the price table only. A voice desk starts as a scoping conversation about telephony, recording consent and who answers when the role stops.
The practical read: if most of your enquiries already arrive in writing, a text-first desk covers the volume today. If the desk lives on the phone, the first step is moving the repeat questions to a text channel and measuring what is left.
Web form and quiz intake
LiveThe enquiry is stored on submit, and the first reply is drafted on the next run.
Lead notification by email
LiveOne attempt with a five-second timeout; the outcome is logged either way.
Phone answering (voice)
Custom scope · EnterpriseCustom roles and custom workflows appear on the Enterprise column of the price table. A voice desk starts with telephony, recording consent and the handover rule.
Calendar booking
Scoped per deploymentNo calendar connector ships today, so live availability is scoped before launch.
SMS and messenger intake
Scoped per deploymentNo SMS or messenger connector ships today; the provider is chosen with you.
Full connector list with its evidence: /integrations
What it costs against the hours
The table is the site’s cost model run on this scenario. Ten hours a week is the assumption to change first: it is the time a café spends on enquiries rather than a full front-desk shift.
Read it as a comparison of hours. The share left over is the part of the desk that carries judgement — the exception, the upset customer, the decision — and it still needs a person. What changes is that the person spends the shift on those hours instead of retyping the same answers.
Model, except the price. Hours and share are assumptions; $149 is published
The brief this desk writes
Setup is a writing job. The role reads this material on every run, so the quality of the brief sets the quality of the drafts. An afternoon covers the first version, and the shadow week corrects it.
The catering menu that answers an enquiry: tray sizes, minimum order, lead time and what travels well.
Room-hire rules — hours, deposit, what is included, what is off the table.
The allergen answer you already give, written down once.
Holiday and long-weekend hours for the next quarter.
The stop rules: dates, complaints and anything about one specific order go to a person.
Week one runs in shadow: every draft is read before it goes, and the material is corrected wherever the draft was wrong. At the end of it you have the share of drafts that went out unedited, which decides how much of the desk moves across next.
Questions people ask before they buy
It prepares the order and a person confirms it. The draft carries the headcount, the date, the dietary notes and your lead time, with any price marked as pending your approval. Unistaff ships no calendar or point-of-sale connector today, so the confirmed order goes into whatever the shop already uses.
On Unistaff, no. Text intake runs today — a web form or quiz posts to the lead endpoint and the first reply is drafted on the next run. A voice front desk is custom scope on an Enterprise plan and starts with telephony, recording consent and the rule for handing a call to a person.
One AI employee is $149 a month on Starter, with a 14-day pilot and cancellation at any time. Up to five AI employees with custom briefs and priority support is $599 a month on Scale. Enterprise is custom-priced.
You write them once and the role reads that material on every run: the service list, the brief and the knowledge items. Anything outside the material is a stop — the draft says so and hands the enquiry over rather than inventing an answer.
The language of the enquiry. Answering in the language the request arrived in is a standing instruction in the role's brief, so a message written in French comes back drafted in French. For a bilingual city that makes the material behind it worth keeping in both languages.
The draft is written ahead of the reply, so it is waiting rather than being composed while the enquiry cools. The timer in the reference deployment runs every fifteen minutes and takes five enquiries a pass. Sending stays with a person, so your response time is the draft plus whoever is on shift.
The same desk, another room
02
Salon or barbershop
Bookings and quote questions arrive while every chair is full. The reply that wins the booking is the one that goes out first.
03
Dental practice
The desk is with a patient, the phone is ringing and the inbox holds new-patient enquiries that decide next quarter's schedule.
04
Auto repair shop
The advisor is under a hood, the phone is on the counter, and every quote request needs six facts before anyone can price it.
05
Medical clinic
One person covers phones, check-in and scheduling. Requests that arrive during a consultation wait for a gap that never comes.
All five scenarios on one page: /resources/ai-front-desk
Continue reading
Hub
AI front desk for small business
Blog
AI receptionist for small business: what it handles, what stays human
Deep dive
AI customer support assistant
Tool
ROI calculator: AI employee vs. human hire
Integrations
What connects today, and what is planned
Pricing