An AI receptionist that answers every call, and knows what it must never say
A call you miss at half past seven is a job that goes to whoever picks up next. An AI receptionist answers every call, and never invents a price or a date to do it.
Five missed calls a week, one in ten a job — a job a fortnight. Your numbers, not mine.
Every call picked up, the job booked before they hang up
Tell me what should happen with your calls
Two lines is plenty. What comes in, and what should happen to it. I read it myself and answer within one working day.

Sergey Avakyan. You are writing to me, not to an inbox — I am the whole company, and I am the one who builds it. More about me →
A person handed anything that needs one — the same on a Sunday at eight in the evening.
Projects go in the order they arrive — the earlier we start, the earlier yours is built.
- Answered on the number you already give out — nothing changes on the van or the website.
- Booked straight into the calendar you already keep.
- A written list of things it must never say — a price, a date it cannot know, a promise about money — enforced in the code, not asked for in a prompt.
- Everything said, written down and searchable afterwards.
The part that checks an answer before it is given is not new work. It has been reading every reply to a real buyer in my own business before it goes out, every day, for months — one of the thirteen systems of mine running daily, out of sixty-five built.
A quarter of whatever figure we agree is due at the start, and that quarter is the only money at risk under the two-week promise — nothing else is due until it is running. What puts you at the bottom of that band: one calendar, one number, and rules you can already say out loud. What puts you at the top: several calendars or locations, a customer system that has to be written into, and rules nobody has written down yet. It lands in the middle band because answering is the easy half — the work is reaching into the calendar you already keep and the place you already track customers, and agreeing what it must never say.
- Two weeks, or the first stage is free. The first stage is one named piece, written down before we start. If it isn't running two weeks after the start, you pay nothing for it. The clock pauses on any day I'm waiting on you for an access or an answer.
- You write the pass mark before I write the code. Before anything is built we agree one sentence that says what working means — for example, "it takes the call, puts the booking in the calendar and never promises a price, on nine of every ten of fifty real calls". If it misses that, you choose: I finish it at my own cost, or you take that stage's money back and the work stays with me.
- Three months of fixes included. For three months after handover I fix whatever breaks in what I built, already paid for. Something you ask for later, or a service someone else changes underneath it, is new work — and I'll tell you which it is before I touch it.
“I” here is the whole company: AVAROPA SLU, one founder and one employee, both of them me. The guarantee is the company's, and I am the person who honours it — you deal with the same one throughout.
Every stage you have paid for is yours the moment it is paid — the code and the accounts, handed over as we go. Nothing here depends on you staying.
Prices exclude VAT — 21% in Spain, reverse charge for an EU business with a valid VAT number, and normally none outside the EU — confirmed on your invoice. The band is a starting point; the exact figure is agreed and fixed before any work begins, and it does not move after that. The two-week first piece is the part I guarantee; the finish windows are what a job this size usually takes.
What changes on Monday
It rings out. Maybe they leave a message, usually they call the next name on the list.
Answered, the job described back to them, and a slot in your calendar before they hang up.
Whoever picked up guesses, or promises to call back and forgets.
It says what it can say and refuses the number it cannot know — then puts you in front of a caller who is already qualified.
Six voicemails, three of them the same person, and an evening spent calling back.
One list at the end of the day: who called, what they wanted, what was booked, what needs you.
None of this needs your team to change how they work. The phone number stays the same.
How it works, what it refuses, and when you don't need it
What this is built out of
Every part of this is already running somewhere. The part that checks an answer before it is spoken — a separate check that reads every reply and blocks anything the business has not authorised — runs today in my own customer assistant. Booking into a real calendar, with rules about who does what and how long it takes, runs on this site: the free hour you can book below is written into a calendar by a machine. Turning speech into words and words back into speech is something you buy off the shelf; it is not where a job like this succeeds or fails.
So the build is not research. It is fitting proven parts to your phone number, your calendar and your rules — and then being strict about what the thing is never allowed to say. The systems on this site run from a single chat bot up to one that answers questions about a whole business; answering a phone sits comfortably inside that.
The calls that are hard, and how they are handled
- Speech, not text. Accents, background noise, a bad line and people talking over each other. Accuracy on a quiet demo call tells you nothing about a busy Friday. If a vendor will not let you test on your own recordings, that is the answer.
- It answers something it should not. A price, a delivery date, a refund, a clinical question. One invented answer costs more than a month of missed calls, and it will sound completely confident — which is exactly why the reply is checked by something before it is spoken, not after a complaint.
- The booking is the hard part. Anyone can take a message. Actually writing into your calendar, respecting who does what and how long it takes, without double-booking — that is the integration, and it is where these projects stall.
- Handover to a person is an afterthought. The caller who needs a human should get one without repeating themselves. If the handover loses the context, you have added a step rather than removed one.
- Nobody measures the calls it lost. The number that matters is not calls answered, it is bookings made and callers who hung up. Both are measurable, and both are usually missing from the dashboard.
What I would build, and in what order
- The safety check before the voice. A second, separate check that reads every answer before it is spoken and blocks anything promising a price, a date or an outcome the business has not authorised. This is the piece already running on the text side.
- One real integration, not five. The calendar or the booking system you actually use, written to properly, with the rules of who and how long. Everything else waits.
- A named handover. A short list of situations where it stops and passes the call on, with the context attached, and a person who owns that queue.
- The two numbers, from day one. Bookings made and callers lost, measured before and after, so the thing can be judged rather than admired.
Where the call should stop, in your own words
This list is the actual deliverable, and it is written with you before anything is built. A version of it, for a clinic:
- Anything about price beyond the published one — stop, take a number, promise a call back within the hour.
- Anything clinical — stop immediately, no paraphrasing of what a doctor might say.
- A caller who says they are unhappy — stop on the first sign, hand to a person with everything said so far.
- A booking outside the normal rules — take the details, do not confirm, flag it for a human to confirm or refuse.
You don't need this if
- You miss a handful of calls a week. A shared inbox and a call-back rota is cheaper and works.
- Your booking rules live in somebody's head. Write them down first — that alone will win you some of the bookings.
- A wrong answer on the phone has real consequences and you have nobody to own the exceptions queue.
What actually gets built
Open the parts list — 10 steps, and which of them already run
Not a drawing of a product. This is the shape of the system for this job, and what each part is made of. Pick a situation and the path through it lights up.
The call arrives
- The call arrivesyour number, forwardedYour existing number, forwarded. Nobody has to change what is on the van or the website.
- Turned into textspeech to text, keptSpeech to text, cleaned up, kept.Working today →
It works out what they want
- What they actually wantbooking, a price, or youBooking, a price, an existing job, or something that needs you.
- The answer is writtenin your wordsIn your words, from what you have said before.Working today →
The line it cannot cross
- Checked before it is saidthe list it cannot crossA second, independent check reads every intended answer against the list of things it must never say — a price, a date it cannot know, a promise about money.Working today →
What happens next
- Booked into your calendarthe slot you actually haveThe slot you actually have free, in the calendar you already use.
- Written down where you can find itsearchable laterWho called, what they wanted, what was said. Searchable later.Working today →
- Handed to a personwith everything already saidWith everything already said, so nobody starts the conversation again.
Someone is watching it
- Someone notices if it stopssilence is the costly failureSilence at three in the morning is the failure that costs most, because nobody sees it.Working today →
- One summary each morningcalls, bookings, refusalsHow many calls, how many booked, what it refused and why.Working today →
Follow a situation
Tap one to light up the path. Tap it again to see all of it.
6 of the 10 parts here are already running in my own businesses every day — they come from 4 of the thirteen systems of mine in daily use, and each box says which. The rest is the work: your calendar, your customers, and the list of things it must never say.
The parts of this that already run
- Customer-chat assistant — the closest thing to this already running: routine questions answered in the seller's voice, with a separate check signing off every reply. It has been doing that against real buyers every day for months.
- Personal AI assistant — the other half — booking a table, chasing an answer, transcribing a voice note and filing it where it can be found.
- Team briefing bot — the boring part nobody demos: the same summary landing in the team chat every morning, whether or not anyone is free.
- Systems watchdog — the monitor that notices the moment any of it stops running.
Honest answers
How much does an AI receptionist cost?
A one-off build of €4,800 to €8,300, higher with several calendars or locations and rules nobody has written down. A quarter is due at the start and is the only money at risk under the two-week promise; nothing else until it is answering calls. The first hour, to work out whether it is worth it at all, is free.
Is an AI receptionist worth it?
Count the calls you miss. At five a week and one in ten a job, that is a booking lost every fortnight — if that pays back the build within a season, it is worth it. If you only miss a handful of calls a week, it is not: a shared inbox and a call-back rota is cheaper and works.
What does an AI receptionist do?
It answers the call, handles the routine questions, books into your calendar and takes a message when it cannot. The part worth judging is the opposite: what it refuses to answer, and how fast it gets a real person on the line when it should.
Will callers know it is not a person?
They should be told, plainly, at the start. Beyond the law in some places, it is simply the version that does not blow up: a caller who was told is annoyed by a bad answer, and a caller who was not is angry about being tricked.
Can it book appointments by itself?
Yes, when it is properly wired into the calendar or booking system you already use, with your rules about who does what and how long it takes. That integration is the real work — taking a message is the easy part that every demo shows you.
What should an AI receptionist never do?
Quote a price, promise a delivery date, agree to a refund, or answer anything clinical or legal on its own. Those need a check between the model and the caller, not an apology afterwards.
What already exists and what gets built?
The two hard parts already run: the independent reply check that stops a machine promising a price or a date, live in a real customer assistant, and machine booking into a real calendar, live on this site. What gets built for you is the voice front end, the wiring to your phone number and your rules, and the list of things it must refuse. The voice is the fast part; your rules are what take the time.
Tell me what your phone is costing you
Two lines: how many calls come in, what most of them ask, and what happens now when nobody picks up. I will tell you straight whether a voice agent is worth it for you, or whether something far simpler wins back most of those bookings. The first conversation is an hour and it is free.
Book that hour →