The same questions all day, and the phone still rings out.
Your team answers the price question, the parking question and the "are you open on Saturday" question again and again. Evening enquiries sit in an inbox until morning. Somebody spends an afternoon a week retyping forms into the CRM. AI integration takes one of those jobs off them: a chatbot that answers from material you have approved, a workflow that ends the retyping, or a voice agent for the calls that ring out. Each is built into the systems you already use.
Sound familiar?
Three leaks you can name, and could measure.
Your website at nine at night. Someone has read the treatment page and wants to know the consultation fee and the next free Saturday. There is nobody to ask. The enquiry form waits until morning, and by then the enquiry may have gone cold.
The front desk when every line is busy. Three lines ring at once with questions that have a settled answer, and the next call rings out.
An afternoon a week of copying. Enquiries from the website form are typed into the CRM by hand, and bookings are copied from one system into another. Nobody would call it a job, but it takes that afternoon every week.
How it works
- Name the leak, then one job. We start with where your business loses something you can measure, not with the technology. Together we agree the one job it does, the point where a person takes over and the subjects it must never attempt.
- It answers from copy you have approved. It can also take actions in your systems, such as booking into your calendar. A narrow job is deliberate: an assistant allowed to hold any conversation is one nobody can test.
- It is tested on your own questions first. The awkward questions, the price questions and the ones asked before anyone will book become a fixed test set, run before launch and after every change. Go-live is your call, not a milestone in our plan.
- It lands where your team already works. The enquiry goes into your CRM or your calendar, not a second inbox. The chat widget sits on the site you have now, and switching it off puts everything back.
- Thirty days of tuning, then your choice. For the first thirty days after go-live, what it says and when it hands over are adjusted. After that, an optional Maintenance care plan picks up monitoring.
What the starting price buys
Enough for one job: the chatbot or the workflow, not both, and one system for it to connect to. That is a working capability, not three things half-finished.
Add when you need it: an inbound voice agent built to your own systems and answer sets, on a divert from the number you already publish · more systems for it to read from and write to · a second job, such as a chatbot and a workflow together · outbound calling, only against a consent record you can evidence. Each is scoped in writing when you need it; none has to be there on day one.
AI chatbot, AI receptionist or workflow — or no AI at all
| Where the business leaks | What fits | The test |
|---|---|---|
| An enquiry at 9pm waits until morning | A chatbot that answers from your approved copy | Can every answer come from material you have already signed off? |
| Calls ring out, out of hours or when every line is busy | A voice agent built to your own systems, behind a divert | Is there a person to hand over to, or a callback to promise? |
| An afternoon a week moving data between two systems | An n8n or Make workflow | Would two competent people produce the same output from the same inputs? |
| A form, a booking page or a rule is the real problem | No model at all | Does the plain fix cost less and break in fewer places? |
A workflow calling a language model to do arithmetic is a slower, dearer, less reliable spreadsheet.
What keeps you safe
Callers know what they are talking to. The voice agent identifies itself at the start of the call and offers a real route to a person: a transfer during opening hours, a callback request outside them. How UK law classifies an AI call is not settled, so it is built for the safer reading rather than the convenient one. Our guide to whether AI receptionists are legal in the UK sets out what the law asks and what is still open. If calls are recorded, callers are told in the opening line and the recordings land somewhere you own.
Treat an answer your assistant gives a member of the public as your own marketing copy. So it answers from reviewed copy and declines treatment questions, which matters for aesthetic clinics and dental practices.
Nothing that needs consent loads before it. The chat widget waits behind the consent gate, with reject as prominent as accept.
None of this is legal advice.
What it costs
From £4,500 excl. VAT, a fixed fee against a written scope. A voice agent is an add-on and is priced in that scope; whatever the platforms charge for, such as model and call usage or workflow runs, is not included: it is passed through at cost, based on what you use. How to work out whether a voice agent pays for itself is in the AI voice agent ROI guide; starting figures for this and everything else are on the pricing page.
When something else fits better
If enquiries already arrive but nobody follows them up, the gap is the pipeline, not the conversation: that is CRM, and it is also where a chatbot's qualified leads should land. If the workflow needs people signing in, roles and records of its own, it is an application, and that is Web Applications. If the website itself is the weaker link, the free audit is the cheaper place to begin. Once a build is live and the thirty days of tuning are over, Maintenance picks up the monitoring.
Get in touch with what your phones and your admin actually look like. Within three working days you have a written scope; where the honest scope is "do not build this", that is what arrives instead. Or read how a project runs first.