Where AI should never act alone.

What the AI does

IntakeDesk reads every incoming enquiry, sorts it, and gives a reason and a suggested next step. EstimateFlow turns a technician’s voice note into a draft estimate. ReviewPilot writes a reply to a new review in your voice.

All three produce drafts. That is where the time goes in a small office: reading, sorting and writing the first version. The AI takes that part.

The four things it never does alone

Without approval from a named person, the AI in a Sirrius system never:

1. 01Sends a reply, a booking or a quote to a customer.
2. 02Sets a price.
3. 03Publishes anything, including a review reply.
4. 04Declines an enquiry.

A person makes each of those calls. These are the steps where a mistake costs you a customer, money or your name. They are also the steps where a person decides in seconds what an AI would get wrong some of the time.

A named person, not “a human in the loop”

“A human in the loop” is too vague. If the approval belongs to everyone, nobody checks. Before we build, we write down which outputs need review and who approves them. The owner approves prices. The office manager might approve replies. Whoever it is, it has a name.

A person can also override the AI. If IntakeDesk sorts an enquiry the wrong way, someone moves it, and the scoring rules are there to read, not hidden.

When the AI doesn’t know

It says so. If an estimate is missing information, EstimateFlow turns the gap into a question instead of guessing a number. If a review is hostile or unsafe, ReviewPilot hands it straight to a person. Nothing is made up to fill a blank.

Every change we make to a system is tested against a fixed set of examples before it touches live work.

One question to ask any vendor

If you are looking at AI for your office, from us or anyone else, ask this: what will it send, price or publish without asking me first? The answer should be nothing. If it isn’t, you should know exactly what, and decide whether you are comfortable with it.