AI Automation

The problem is the gap, not the answer

Someone messages your business at nine in the evening asking whether you have a thing in stock and what it costs. You see it at eight the next morning. By then they have asked two other people the same question, and one of them replied.

Almost nobody measures that gap, which is why almost nobody fixes it. It is not in a report anywhere. It just quietly decides which enquiries turn into work.

What the assistant actually does

It answers on WhatsApp the moment a message arrives, asks three questions, and then stops.

• What do you need?
• Which city are you in?
• Do you already have a website, and what is the address?

Three, because three is enough to know whether this is worth a call, and because a bot that asks a fourth question starts to feel like a form. The answers are written into a proper database as a row you can sort and filter, rather than left buried in a chat thread nobody scrolls back through.

Then a person takes over. That is the point of it. It is not there to close anybody.

It is measured, and the number is on record

The system stamps the moment a message first arrives and the moment the lead is qualified, and stores the difference. On the first live run that difference was 64 seconds, with each individual reply going out two to three seconds after the message that triggered it.

This runs on my own enquiries. I am not describing a product I have installed for a dozen businesses, I am describing the thing that answers my WhatsApp, which is why I can tell you what it measures rather than what it promises.

What it deliberately will not do

These are the three decisions that separate an assistant from the chatbots everyone has learned to hate, and all three are refusals.

• It does not guess. If someone names a city it does not recognise, it stores exactly what they typed and flags it for a human. It never quietly maps an unfamiliar answer onto a familiar one, because a wrong guess buried in a database is worse than a blank.
• It does not interrupt. Once a lead is qualified, or once it has decided a person is needed, it stops sending automatic replies. Permanently. A bot talking over a real conversation is worse than silence, and this is the branch most automations get wrong.
• It does not start conversations. Every message it sends is a reply inside the window a customer opened themselves, so nothing needs pre-approval as a marketing template and nobody gets messaged out of the blue.

What it costs to run

Because every message is a reply inside the window the customer opened, Meta does not charge for that traffic. The logic runs without a server sitting idle waiting for messages, so there is nothing to pay for between enquiries.

The build is the cost. Running it is close to nothing, and I would rather say that plainly than sell you a monthly fee for something that mostly sits still.

Whether this suits you

It suits a business where enquiries arrive faster than someone can pick up a phone, where the same three questions get asked every time, and where the answers currently live in a chat history rather than anywhere you can count them.

It does not suit a business whose enquiries all need a real conversation from the first message. If that is you, an assistant in front of it will annoy people, and I will tell you so rather than build it.