A pattern I keep running into: an owner has done the homework. They've written a proper brief, often a long one, covering everything they want their systems to do over the next couple of years. Fewer manual reminders. Follow-ups that sound like a person wrote them. Then they send it to a few developers and expect the quotes to land in roughly the same range.
They don't land in the same range. The same brief, sent to several competent people, produces answers that look like they were written about different projects. Not because most of them are cowboys, but because the brief describes what the business wants to happen and says nothing about what the existing software will let anyone do.
That gap is where the money goes.
The AI is the cheapest part of your AI project
Getting a model to write a decent follow-up message is close to free and close to solved. Classifying an enquiry, drafting a reply in your tone of voice, pulling structured data out of a messy email: a few lines of code and a few cents per run. It's also the only part most owners have personally touched, which is why it's the part they budget for.
The cost is everywhere else. Getting data out of the system you already pay for. Getting permission to send anything on your behalf. Working out what happens when it gets something wrong at 9pm on a Saturday and nobody notices until Monday.
There's a second thing going on with briefs like these, and it's newer. AI writes very good specifications now. An owner with an afternoon can produce a document describing a complete operating platform, internally consistent and genuinely thoughtful. It reads like a product team spent a month on it.
What it isn't is a build order. That document is a three-year target. The work you're actually paying a good developer for is deciding which small piece to build first so the second piece doesn't require throwing the first one away. Nothing in the document tells you that, because the model writing it had no idea what your booking system can do.
Four things that set the price, none of which are AI
None of the four below are hypothetical. They're the things that have actually moved the number on recent jobs, and in each case they cost more than the AI did.
1. Some software can't tell you when something happens
Say you want something to happen the moment a customer cancels. Reasonable request, obvious value.
Well-behaved software can tell you the moment something changes. It pings your system to say "this just changed" and your automation reacts within seconds. Developers call these webhooks. When they exist, real time is easy and cheap.
Plenty of popular business software doesn't have them. Then the only way to know something changed is to ask, over and over, forever. Every minute, across every location: anything changed?
That works, but it's different software with a different cost shape. It runs constantly rather than only when something happens, and it needs monitoring so you find out when it silently stops. It also has a ceiling, because most systems cap how many times an hour or a minute you're allowed to ask. Once you're checking several locations alongside everything else the integration does, you start rationing.
So "real time" is a question about your vendor, not about your budget. Ask it before you get quoted.
2. Which email account you have changes the project
Whether a business is on Google Workspace or ordinary Gmail sounds like an accounting detail. It isn't.
If an outside system needs deep access to a mailbox, Google requires an annual security assessment by a Google-approved third-party assessor under the CASA framework. That's not a form. It's an audit, it recurs every year, and it costs real money.
The important exception is internal use: an app used only by people inside your own Google Workspace organisation, on a project that organisation owns, skips it. So a Workspace business building something for itself is often fine. A business on standard Gmail wanting a third party in its inbox is looking at weeks versus months before a single email auto-sends.
The other lever is how much you ask for. Google tiers its permissions, and sending mail sits in a lighter tier than reading a whole mailbox. Asking for less is usually cheaper than asking for everything and paying to be audited on it.
Sources: Google's restricted scope verification guidance and the Gmail API scopes list.
3. Text messages got stricter in July
If your SMS shows your business name at the top instead of a phone number, that name is an alphanumeric sender ID, and since 1 July 2026 it has to be registered with ACMA. Unregistered ones get over-stamped as "Unverified" on the recipient's phone, grouped with other unverified traffic including scams. ACMA has warned those messages are more likely to be ignored or deleted even when the sender is legitimate.
Registration goes through a participating telco, and your ABR details need to be current first.
The part worth knowing: this only covers branded sender IDs. Sending from a normal phone number is outside the rules, and it has a side benefit that suits plenty of small businesses better anyway, which is that customers can reply.
So "should our reminders show our business name?" turns out to be a compliance question with a lead time attached. Answer it early. Source: ACMA, SMS Sender ID Register.
4. Being small doesn't always keep you out of privacy law
Most Australian businesses under $3 million turnover are exempt from the Privacy Act. Most, not all, and the carve-outs are wider than people assume.
The one catching people this year: from 1 July 2026, real estate professionals, lawyers, conveyancers, accountants and trust and company service providers came into the Privacy Act through the AML/CTF reforms. The Act applies to personal information they handle in connection with those obligations regardless of turnover. If you're an accounting firm that assumed the $3 million threshold covered you, check again.
Longer-standing carve-outs catch businesses that trade in personal information, contracted service providers under a Commonwealth contract, and private-sector health service providers, all bound by the Australian Privacy Principles no matter how small they are. Broader removal of the exemption is still government policy with no Bill and no date, so plan for it as direction rather than a deadline.
Whichever bucket you're in, the question is the same: where does your customer data go once an automation touches it, and does it pass through a model hosted somewhere you haven't thought about? Decide before you build. Retrofitting it is miserable.
Sources: OAIC privacy guidance for AML/CTF reporting entities and the OAIC announcement.
The biggest cost lever is one you control
Nobody asks about this one, and it moves the price more than anything above. How much are you willing to let the thing do without you?
Three settings. It drafts and you send. It sends to the cases it's confident about and everything else waits for you. Or it just sends.
The first is cheap. A human catches mistakes before they leave the building, so the software doesn't need to be defensive about much. The third is expensive, and not because the AI is harder. Every failure mode has to be designed for in advance: duplicate messages, the customer who cancels and rebooks in the same minute, the loop that texts someone four times. You're paying for guardrails, logging, and someone's judgment about what happens at 2am.
That's a business decision, not a technical one. Make it before you ask for a quote rather than discovering it halfway through a build.
Sometimes the honest answer is that a task should stay manual. If it happens six times a week and takes two minutes, automating it properly might cost more than it saves for years. I'd rather tell you that early than build it and watch you regret it.
Before you ask anyone for a quote
Work through these yourself. They cost an afternoon and they make every quote you receive comparable.
- List the software you won't replace. That list constrains everything else.
- Ask each vendor whether it can notify other systems when something changes. Use the word webhooks and their support team will know what you mean.
- Ask whether there's a cap on how often outside software can talk to it, and whether the cap is per business or per location.
- Check whether your email is Google Workspace, Microsoft 365 or a personal account, and who administers it. If nobody knows, fix that first.
- Decide whether your texts need to show your business name. If yes, start the ACMA registration now, separately from any build.
- For each automation, write down whether it may send on its own or must wait for you. Be honest rather than aspirational.
- Check whether privacy obligations apply to you, including the AML/CTF ones that started in July.
- Pick the task costing you the most hours per week. Not the most interesting one.
Take those answers to whoever quotes you. If they don't ask about half of them unprompted, keep looking.
Diagnose before you prescribe
We run a paid diagnostic for exactly this. Fixed fee. It maps what your systems will and won't allow, what each automation would really require, and which piece to build first so the rest still fits around it. You own the document whether or not you build anything with us. Hand it to another developer if you'd rather, and their quote will be better for it.
Free quotes are guesses dressed up as prices. Get the diagnosis before the prescription.
If you're in Adelaide, we'd rather do this over coffee than over Zoom. Happy to come to you.
OrionX Technologies builds software and AI automation for Australian small businesses and accounting firms. The regulatory positions above were verified in August 2026. Rules change, so check the linked sources.
