What the AI does in a system we build
We design every system so nothing the AI writes goes out without your OK, and a message you approve once goes out as written after that.
13 October 2026
Contents
The fair worry about AI in your business is a bot texting one of your customers the wrong thing, in your name. In what we build, the AI reads every enquiry and reply and fills in the customer’s card, and anything it writes waits for you.
- Messages that are the same every time, like the text back after a missed call, get your OK once and then go out as written.
- The AI never writes a price, and a booking is only confirmed when you OK the time.
- When a detail is missing, your system sends the question you’ve already approved for it, as written, without asking you again. Turn off auto-send in Settings and the AI drafts each question instead, and each one waits for your OK.
Where this stands on 11 October 2026: this is how we design every system, and none of our clients has the AI or the automatic texts switched on yet.
What the AI does
When a customer texts, emails or fills in your form, the AI reads the message and fills in their card: the job, the suburb, how soon they need it, any budget they mention, when suits them and a one-line summary. It only fills empty fields, so it never writes over something you typed.
When a customer is ready for a price, your phone gets a text like this, with a link to their card:
“Mia wants a price for the hedge trim in Mona Vale. Reply with the price, just the number (e.g. 380).”
The hedge trim and Mona Vale came from the AI’s read of Mia’s message. You reply with your number from the ute, and the quote goes to Mia in the wording you’ve OK’d, with the follow-ups lined up behind it.
When a customer replies, the AI labels the reply in their thread: keen to go ahead, a question, pushback, a no thanks, a request to stop, or unclear.
When it isn’t sure, it says so. An error, a timeout, a garbled answer or low confidence flags that customer, and your phone gets “Worth a look: the details it read from Mia’s message may be off.”
It can also be sure and wrong, say by filing Mia’s hedge as a lawn mow. Your wording fills in the job from the card, so the wrong job would show up in your questions that name it, the quote and the follow-ups. You see the AI’s read before any quote goes, in the price text and on the card you price from, and the link opens the card to fix it.
Everything that acts, like sending a message or moving a customer to the next stage, is plain code following fixed rules, or it’s you.
Messages you approve once
Some messages are the same every time, so there’s nothing for the AI to write. You OK the wording before your system goes live and can change it in Settings. After that it goes out as written, with details like the first name filled in from the card.
| Message | When it goes | What stops it |
|---|---|---|
| The text back after a missed call | Straight after a call you couldn’t take | You’ve taken that customer over. You get a text to call them back instead. |
| A question about a missing detail | One at a time, while a detail you need is missing | They answer it. After two asks that don’t get it, you get a text about them instead. |
| Follow-ups after a quote | On the days you pick, inside the hours you set | The customer replies, or you mark the quote accepted or declined. |
| A reminder for a booked job | The day before, inside the hours you set | You’ve taken that customer over. It waits until you hand them back. |
A second question only goes after the customer replies. If they go quiet, nothing more is asked, and they sit in your list waiting on their reply.
What the AI writes
When a detail is missing, your system sends the question you’ve already approved for it, as written, without asking you again. Turn off auto-send in Settings and the AI drafts each question instead, and each one waits for your OK. It writes each one to fit what the customer said, as it does for a customer you’ve taken over.
Your wording, OK’d once
The missed-call text, your questions, the quote around your price, the follow-ups and the reminder, sent as written. The AI doesn’t touch them.
A question the AI wrote
It waits in the customer’s thread under Approve & send, Edit and Dismiss, and goes only when you send it or text back Y.
Code throws out any draft that isn’t exactly one question, or that has a dollar figure in it, and flags the customer for you instead.
You don’t have to open the app for it. Your system texts your own mobile, one thing at a time, with the question it wrote and “Reply Y to send it.”
| You text back | What happens |
|---|---|
| Y | The question goes to the customer as written. |
| N | Nothing sends, and you get “Left it. It’s here when you want it,” with a link. |
| Anything else | Nothing goes to the customer. You’re asked for Y or N again, with a link to change the draft. |
A reply sent to the wrong person can’t be taken back, so your texts back are only ever read as Y, N or a price. Nothing else you type reaches a customer.
What stays with you
The price is a number you type, and the AI never writes one. If the number you text looks like a typo, your system asks you again before it sends.
A booking is only confirmed when you OK the time, and the customer’s confirmation goes out when you do.
A customer who replies STOP is taken off your list by code, and the AI never sees that message. Every send checks that list first, and every email carries an unsubscribe line. Any reply from a customer stops their follow-ups. Take a customer over and nothing sends to them on its own until you hand them back.
What gets recorded
Every approval is recorded, and the customer’s thread marks each one “draft edited” or “draft as suggested”. Once that record shows you’ve stopped changing a kind of draft, that kind could earn the right to send on its own, and only with your yes. None of our systems does that yet.
Every call your system makes to the AI is logged with what came back, or why it failed. The rules in this article live in the code, with automated tests that check them.
We test the AI on a business’s real enquiries before we switch it on, and the automatic texts go on once you’ve OK’d every one.
Why we build it this way
AI reads a messy enquiry the minute it lands, at any hour, and we build with it every day.
It also gets things wrong. The same model can answer one message well and the next one badly, and you can’t always tell which in advance. Your name is on every message, and a customer who gets a wrong price or a time you can’t make holds it against your business.
We’ll tell you which parts of a system call an AI model, and if a spreadsheet would fix it, we’ll say so. Where AI does run, identifiable patient data never goes to an overseas AI model, and the AI providers inside our systems don’t train on your data. Where your data lives and who owns it has the rest.
Bring the message your business sends most often to the free 20-minute call. We’ll show you where the AI would read it and where your OK would go.