Answers the usual questions
Opening hours, indicative prices, location, timeframes, payment methods, what the customer needs to bring. With your answers, not ones it makes up.
AI implementation
An assistant that answers on WhatsApp and on your website with the real information of your business: it replies to questions, books appointments, filters what deserves your time and alerts a person when needed. It works at three in the morning and in August.
The real cost is not the chatbot: it is the enquiries you lose today. In a small business, the person who answers the phone is also serving the customer in front of them, invoicing and closing the till. When three messages arrive at once, two go unanswered.
That customer does not try again: they write to the next business. And that loss shows up nowhere, because nobody counts what never happened.
If a new customer is worth €400 on average and you lose three enquiries a week, that is more than €60,000 a year leaving without anyone writing it down.
A well-built assistant does not replace anyone: it takes care of the usual questions so that people can handle what needs judgement. It is the same idea I apply in the rest of the process automation work, applied to the first contact with the customer.
Try it
Bramwell Family Dental does not exist: it is a made-up practice so you can try the assistant here. Ask what a patient would ask, or request an appointment and watch it being logged. Notice above all what it does NOT do: it does not diagnose, it does not quote a price for your case and it does not promise that it will not hurt.
Hello, I am the virtual assistant of Bramwell Family Dental. I can tell you about treatments, prices and opening hours, or request an appointment for you. How can I help?
Demo of a fictitious dental practice: appointment requests made here are not sent to anyone, so feel free to use made-up details.
What it can do
Opening hours, indicative prices, location, timeframes, payment methods, what the customer needs to bring. With your answers, not ones it makes up.
Checks the real free slots in your calendar, suggests options, confirms and sends the reminder. No more back-and-forth messages.
Tells the person who asks out of curiosity from the one who wants to buy, and passes you only the second, with the context already gathered.
Asks the necessary questions, checks what is missing and leaves you a sheet ready to quote from instead of a loose conversation.
At night, at weekends and in August. Most of the enquiries that get lost arrive when nobody is there.
If it detects a complaint, an urgent case or something off script, it stops and alerts you straight away. It does not improvise.
Where it answers
They do not have to learn a new tool. The assistant sits in the channels they already use.
How it is built
The questions you get every day and the answers you give. That material is 80% of the work and you already have it.
In two weeks it answers the most frequent enquiries. You try it with real customers before anything is extended.
Calendar, CRM, catalogue or invoicing, so that besides answering it can do things: appointments, records, alerts.
I review the conversations of the first month and fix what fails. That is where an assistant goes from correct to good.
We have all argued with a bot that understood nothing. Those are trees of fixed answers: if you do not type exactly what was expected, they get stuck. The difference lies in three build decisions:
And something that is not always said: there are businesses where it is not worth it. If you receive four enquiries a week, the assistant will not pay back what it costs. At the first meeting I will tell you.
Common questions
A classic chatbot runs on a tree of fixed answers: if the customer does not type exactly what was expected, it gets stuck. An AI assistant understands what it is asked even when it is badly written or phrased another way, and answers with the real information of your business. The customer notices the difference within two messages.
Only if it is built badly. The assistant answers from your documents, prices and rules, not from whatever the model knows on its own. When a question falls outside them, it says so and alerts a person instead of improvising. There is also a point of human approval before any commitment: where that point sits is up to you.
Between three and six weeks depending on complexity. The first version, answering the most frequent questions, is usually ready in two weeks so that you can try it with real customers before extending it.
Very little: a WhatsApp Business line and the answers you already give a hundred times a day. If you have a digital calendar or a CRM, it gets connected; if not, one can be set up. The first meeting is precisely for seeing what you have to work with.
The history and contact details are stored on European Union servers that I manage, or on yours if you prefer. The language model is supplied by an outside provider under a contract that forbids training on your content: I tell you which one it is and what is sent to it. We sign the data processing agreement required by the GDPR (the EU data protection regulation, RGPD in Spanish) and I document for you which data passes through where and for how long it is kept.
At any time and with no lock-in contract. The assistant is yours, the code too, and you can turn it off or take it to another provider. I do not work with contracts that are hard to leave.
Start here
Half an hour at no cost. We leave with a clear idea of how many enquiries are being lost and what can be recovered.