The problem it solves, in one scene
A plumber is under a sink at 2pm with both hands occupied. The phone rings. It rings out. The caller does not leave a voicemail, because almost nobody does any more. They call the next plumber on the list, and that job is gone.
The plumber never knows it happened. There is no record of the job that was lost, only a vague sense that the phone is busy and business is not growing.
The same scene runs at a clinic at 8pm, a salon on a Sunday, and a roofing company during a storm when demand spikes and nobody can pick up.
What it actually does
Two versions, and you should usually start with the second.
Missed-call text-back, the simple version
A call comes in and is not answered. Within seconds, an automatic text goes to that number: an apology for missing them, a question about what they need, and an offer to book. The conversation continues by message, which most people prefer anyway.
This is simple to build, hard to argue with, and often produces a visible result in the first week. It is the ideal first system for a first client.
The conversational version
An AI actually handles the exchange, by voice or by message. It answers common questions, works out whether the person is a fit, checks availability, books the slot, sends a confirmation, and sets a reminder. It hands over to a human when it hits something it should not decide.
More valuable, more to get wrong, and worth more monthly. Build it once you have delivered the simple version and understand what people actually ask that business.
How to build one that does not get switched off
Most failed deployments fail the same handful of ways. Design against them from the start.
- 01Give it a narrow job. Booking appointments and answering five specific questions. Not being helpful in general, which is how you get confident wrong answers.
- 02Write the escalation path first, before the happy path. What happens when it does not know? It should say so plainly and hand to a human, not guess.
- 03Make reaching a person easy and obvious. What customers hate is not AI, it is being trapped with something that cannot help them.
- 04Feed it the business's real answers. Prices, hours, service area, what they do not do. Most bad AI receptionists are bad because nobody gave them the information.
- 05Test it as the most annoying possible customer. Vague, impatient, asking things it was never briefed on, changing their mind mid-conversation.
The failure modes worth knowing
- It answers confidently and wrongly about price. This is the one that genuinely damages a client relationship. Either give it exact pricing or forbid the topic and escalate.
- It cannot recognise an existing customer, so a regular gets treated like a stranger. Check the CRM before responding.
- It books outside working hours or into slots that are blocked. Calendar integration has to be real, not approximate.
- It keeps talking when the person has clearly asked for a human.
- It sends the same follow-up to someone who already replied. Nothing looks more automated, or more careless.
How to sell it
Lead with the scene, not the technology. Ask an owner how many calls they miss in a week and watch their face. Ask what a typical job is worth. Let them do the arithmetic themselves, out loud.
Then say what you would build, in plain language: when someone calls and you cannot pick up, they get a text within seconds, and by the time you are out from under the sink, the job is booked.
No mention of AI, workflows, or platforms unless they ask. They are buying the outcome, and the technology is only interesting to you.
Which businesses to offer it to
The profile matters more than the industry:
- Phone or message is how customers first make contact
- The owner or staff are physically busy doing the work, so the phone is genuinely unattended
- One job is worth a meaningful amount, so a missed call has real cost
- The business runs on appointments, so booking is the natural outcome
Home services, clinics, salons and med spas, dental, veterinary, legal, real estate. Not because those industries are special, but because that is the profile.
What to charge
Price against the value of a recovered job. If one job is worth a meaningful amount and the system recovers even a few a month, the monthly fee is easy to justify and easy to defend when questioned.
Structure it as a setup fee for the build plus a monthly fee for running it, and remember that usage costs for calls and messages bill to the client's own account when you resell the platform.