What an AI operator actually does for a small business

You already have someone doing your admin. It's you, after dinner, working through the enquiries that came in while you were on a job, the two quotes you promised, and the invoice that's three weeks late. Nobody pays you for that shift.

The phrase "AI operator" gets thrown at that problem a lot now, usually next to a screenshot of a chatbot. That's not what we mean by it, and it's not what we run in our own company.

So here's a plain walk-through. What an operator actually does across a normal working week, where you stay in charge, what it will not touch, and what it costs against the wages you'd otherwise pay.

"Operator" is not another word for chatbot

Four things get lumped together under "AI for business" and they're not the same.

A tool is software you drive. Xero, your CRM, your calendar. Useful, and it does nothing until you sit down and use it.

A chatbot answers questions on your website. It can't send a quote from your email, can't see that the same person enquired twice last month, and can't tell you on Friday which jobs went quiet.

A virtual assistant is a person, usually offshore, on hourly rates, working set hours. Good VAs are worth having. They also need training, sleep, and a manager, and the manager is you.

An operator sits inside your actual accounts. Your inbox, your CRM, your calendar, your accounting software, WhatsApp if that's where customers message you. It reads the enquiry, drafts the reply, sends the quote, logs it, chases it, raises the invoice, chases that too. When something needs a decision, it asks you on Telegram and waits for a yes or no. You don't log into a new system. The work gets done inside the systems you already have.

One week, five seats

The wages table on azlar.ai lists five seats a service business ends up paying for one way or another: reception, admin and quotes, bookkeeping follow-up, sales follow-up, and after-hours. Here's what the operator does in each across a normal week. No made-up customers, no dramatised timestamps. Just the routine.

Reception

Enquiries arrive by email, web form, ad lead form or message. The operator reads each one, works out what they're asking for, and replies with the questions you'd ask yourself: where's the job, what do you need, when. It logs the person in the CRM, notes the source, and flags it if they've enquired before.

Where you approve: for the first two weeks, every reply. After that, standard enquiries get answered without you. Anything odd, a complaint, a request outside what you do, gets a Telegram nudge before anything goes out.

Admin and quotes

Once an enquiry has enough detail, the operator drafts the quote from your price list and your templates, in your wording. It sends it, logs it against the lead, and sets a follow-up date. When the customer accepts, it books the slot in your calendar and sends the confirmation.

Where you approve: the quote itself. Prices are your call, always. In practice most owners approve a quote from their phone between jobs. The typing, the formatting and the sending are done.

Bookkeeping follow-up

When a job's marked done, the operator raises the invoice in Xero (or whatever you run) and watches the due date. When it ticks over, it sends the reminder in your tone. Then another one. When it's clear a reminder won't cut it, it tells you the account needs a phone call from you, with the history in front of you. It also keeps the CRM honest against what's actually been paid, so your pipeline isn't fiction.

Where you approve: anything to do with money outside the routine. A payment plan, a write-off, a credit. It asks, you answer.

Sales follow-up

Every quote that hasn't been answered gets chased on a schedule until the customer books or goes cold. When they book, the calendar and CRM update. When they go cold, they're marked and archived rather than sitting in your pipeline making it look healthier than it is. End of the week, you get a plain list: what's open, what's gone quiet, what needs you personally.

Where you approve: nothing routine. The follow-up messages are ones you signed off on in the first two weeks. If a customer replies with something the operator hasn't seen before, it comes to you.

After-hours

Nights and weekends. An enquiry that lands on a Saturday gets read, replied to and logged the same way a Tuesday one does. You see it Monday morning already sorted, or Saturday afternoon if you'd rather. In our own shop we get a morning action list on Telegram and an end-of-day reconciliation of the CRM, so the day starts with a list and ends with the books square.

Where you approve: same rules as daytime. The operator doesn't get braver after 6pm.

Here's the week in one table.

Seat	What the operator does	Where you approve
Reception	Reads, replies, logs, flags repeats	Every reply for two weeks, then only the odd ones
Admin and quotes	Drafts, sends, logs, books	Every quote, always
Bookkeeping follow-up	Raises invoices, sends reminders, reconciles	Anything outside the routine (plans, write-offs)
Sales follow-up	Chases until booked or cold, weekly list	New situations it hasn't seen
After-hours	Same as daytime, plus morning list	Same rules as daytime

What it does not do

Worth being blunt here, because most of the disappointment with "AI" comes from people expecting it to do the wrong job.

It doesn't make judgement calls. Whether to take on a job, whether to discount, how to handle the customer who's angry about something that wasn't your fault. It flags those, gives you the history, and waits.

It doesn't do the work. It's not on site, it's not in the clinic room, it's not designing anything. It handles the admin around the work.

It doesn't set prices. It uses yours. If your price list changes, you tell it and it uses the new one.

It doesn't do anything you haven't approved. New type of action, new template, new customer situation: it asks first. Sensitive actions, meaning money, cancellations, or anything client-facing outside the routine, go through Telegram approval every time. That's the front door and it stays shut until you open it.

And it doesn't replace your accountant or your bookkeeper's judgement. It does the chasing and the data entry they'd rather not.

What it costs against the wages

Now the math. These are the wage anchors we use on azlar.ai for the five seats above.

Seat	Roughly what a hire costs per year
Receptionist	about $60,000
Admin and quotes	about $75,000
Bookkeeping follow-up	about $65,000
Sales follow-up	about $70,000
After-hours	overtime you can't hire for

Nobody hires all five. Most owners hire nobody and do the lot themselves, which is the free shift after dinner. But even one admin hire is about $75,000 a year once they're trained and staying.

The operator side:

Plan	Per month	Per year	What you get
Setup Only	none	one-off build	The build, no retainer
Partner	$800	$9,600	Light support, check-ins, small tweaks
Managed (recommended)	$2,000	$24,000	Monthly operation and tuning, new tasks added
Full Ops	$3,800	$45,600	Deep hands-on ops for teams past $100K/yr in admin wages

Every plan starts with a one-off setup build, quoted after the 20-minute call.

So the comparison most owners actually make: an admin hire at $75,000 versus Managed at $24,000. That's $51,000 back per year, and the operator also covers the reception, follow-up and after-hours slices the admin hire wouldn't. You can do the math yourself. It's not close.

To be fair about it: the operator does the admin part of each seat, not the whole job. A receptionist also answers the phone and greets people at the counter. If you need a human at the front desk, this doesn't replace them. If what you need is the enquiries answered and the quotes out and the invoices chased, it does that.

How it gets set up

Here's the deal, start to finish.

You book the 20-minute call. We ask what comes in, how you quote, what you use for accounting, and where the admin actually piles up. From that we quote the setup build.

We connect the operator to your accounts. Your CRM, Gmail or Outlook, calendar, Xero, WhatsApp, Telegram. It runs on your systems, not on ours, which is why you can export the lot and leave any time.

Then two weeks supervised. Every message it wants to send comes to you on Telegram first. You approve, edit, or reject. This is where it learns your tone and your rules, and where you learn what it's actually going to do. Most owners stop approving the routine stuff by the end of week two because it's right, and keep approvals on quotes and money.

Worried it'll say something dumb to a customer? That's what the two weeks are for. Nothing goes out unseen until you decide it can.

After that, on Managed, we operate and tune it monthly and add tasks as your business changes.

We ran it on ourselves first

We built this for Intelligent Living Solutions, our own Canberra smart-home company. Two founders, no admin staff, and a Meta ads pipeline that doesn't care what time it is. The operator has run there since early 2026: hourly lead sync into the CRM, a morning action list and end-of-day reconciliation on Telegram, quotes generated from a priced bill of materials, follow-ups scheduled, stale leads archived after 30 days, review requests sent, backups run nightly. As of 17 August 2026 that's 337 leads through the system since December 2025, 267 of them from Meta ads. We could put an "hours saved a week" figure here like every other agency does. We haven't measured it properly yet, so it stays out until we have.

The full write-up is in the case study below.

Related reading

• Case study: the company we built it for first
• What a done-for-you AI agent costs in Australia
• AI receptionist vs human vs voicemail
• Plans and pricing
• Book the call

If you want to know what it would do in your business and what the build would cost, book the 20-minute call at https://azlar.ai/#contact.