Ask what an automation project costs and you’ll usually get “it depends”. True, and useless when you’re trying to work out whether to bother.
So here’s the straight version, starting with my own numbers, because a post about pricing that doesn’t contain any prices is part of the problem.
What I charge
Australian dollars, excluding GST.
| Engagement | Range |
|---|---|
| Discovery, fixed price and standalone | $3,000 to $6,000 |
| One AI agent, built and in production | $10,000 to $20,000 |
| Each additional connected system | from $6,000 |
| Multi-process programmes | $30,000 and up |
| Advisory and troubleshooting, hourly | $500 per hour |
Discovery is separate and it comes first. You get a process map, the decision rules written down, and an agreed measure of success, whether or not you build anything with me afterwards. If you do proceed, it comes off the build. If the map shows the process shouldn’t be automated, you’ve spent the smaller number to find that out, which is the entire point.
An agent in production means in production. Connected to the systems it needs, tested against your real cases rather than tidy demo ones, with the boundaries configured, your team shown how to use it, and someone named as responsible when it needs changing. A proof of concept is a fraction of that and worth roughly what you’d expect.
Additional systems are where the number moves. An AI receptionist that answers questions is one thing. The same receptionist grounded in a knowledge base, writing into your CRM and taking a deposit, is three more integrations, each with its own edge cases and its own way of failing.
Ongoing support is a monthly retainer, from $900 for monitoring and break-fix up to $2,800 for ongoing optimisation, sized to what’s running and how much it would cost you if it stopped overnight. The tiers are set out on the pricing page alongside what each one includes.
Who this isn’t for
Worth saying plainly, because it saves us both a call.
If you want two apps connected for a few hundred dollars, that’s a legitimate need and I’m the wrong person. Zapier or Make will do it, their own templates are good, and a generalist freelancer will charge appropriately. Paying consultancy rates for that is a waste of your money.
If you want to try something and see how it goes, without a defined process or an agreed measure of what better looks like, we’ll both be disappointed. That’s not caution about the technology, it’s the single most reliable predictor of an automation project failing, and I’ve written about why that happens.
The work I do well is the kind where the process is genuinely yours, several systems have to agree with each other, and getting it wrong has a cost. That work needs discovery, it needs decisions made and stuck to, and it’s priced accordingly.
The four things you’re actually paying for
Most quotes bundle these together. Separating them tells you a lot about whether a quote is realistic.
1. Discovery
Working out what the process actually is. Mapping it including the exceptions, extracting the decision rules, agreeing what success looks like.
This is the part people want to skip and the part that determines whether the rest works. It’s also the most variable. If you can already describe your process end to end, discovery is short. If the rules live in someone’s head and have never been written down, this is the bulk of the engagement.
A provider who proposes no discovery at all is quoting on the happy path, and you’ll meet the exceptions later at your own expense.
2. Build
Configuring the workflow, connecting the systems, writing whatever logic is needed, and testing against real cases rather than tidy examples.
Build cost scales with how many systems are involved and how many ways the process can branch. Two tools and one decision point is straightforward. Five tools, conditional routing and an approval step is a different job.
3. Running costs
Two components, and both are usually smaller than people fear.
The first is a platform subscription for something like n8n or Make. Entry tiers are modest, and self-hosting n8n brings it down further if you’ve got somewhere to run it.
The second is AI model usage, charged per use. For text-handling workflows the amounts are typically small. A workflow drafting a few dozen replies a day generally costs less per month than one professional software licence.
The failure mode here isn’t expense, it’s surprise. Ask for an estimate of monthly running cost before you commit, and ask what happens to it if your volume doubles.
4. Maintenance and change
The one most often left out of quotes.
Automations sit between systems you don’t control. Tools change their interfaces, your process changes, a supplier changes their file format. Something will need adjusting, not constantly, but predictably.
Work out who does that and on what basis before you need it. An automation nobody maintains eventually becomes an automation nobody trusts.
What makes a project cheap or expensive
Almost none of it is the technology.
Cheaper when the process is already documented or simple enough to describe in one sitting, when there are only one or two systems and both have proper integration support, when the rules are clear with few exceptions, and when one person owns the decision and can answer questions quickly.
More expensive when the rules are undocumented and vary depending on who’s doing the work, when legacy systems have no clean way in, when there are many exceptions each needing their own handling, and when several stakeholders disagree about how the process should work.
That last one is worth pausing on. Disagreement about the process isn’t a technical problem, but it burns project time faster than anything technical, because nothing can be built until it’s settled. In large organisations it’s why projects stall for months. Small businesses have the advantage: the decision-maker is usually in the room.
Working out whether it’s worth it
You can do this on the back of an envelope, and you should, before talking to anyone.
Count the hours. How many times a month does the task run, and how long does each take? A twenty-minute job done daily is about seven hours a month. A two-hour job done weekly is about eight.
Value them. Use what the person doing the work actually costs, not a notional hourly rate. If it’s you, and your time would otherwise go to billable work or winning clients, the number is higher than you think.
Add the error cost. How often does this go wrong, and what does it cost when it does? A missed follow-up on a job you’d have won isn’t a small number, and it belongs in the calculation.
Compare against the build cost. That gives you a payback period. Under six months is usually an easy yes. Beyond twelve, be sceptical unless there’s a strategic reason.
Frequent tasks dominate this maths because the setup cost is paid once while the saving recurs. A workflow running twenty times a week repays its setup far faster than one running twice a month, even if the monthly one feels more painful each time you do it.
Where the money usually gets wasted
Automating something that should be deleted. Mapping a process turns up work with no remaining purpose: reports nobody reads, approvals never once declined, data entered twice because two systems couldn’t talk a decade ago. Removing that costs nothing. Automating it makes it permanent.
Buying a platform before understanding the problem. Annual licences signed for tools that turn out to fit badly. Choose the tool last.
Paying for a system nobody adopts. A working automation the team abandons under pressure has returned nothing at all. Training and a one-page reference cost very little against the build, and they’re what protects the whole investment.
Questions worth asking any provider
- What does discovery involve, and what do I get out of it?
- What are the monthly running costs, and what changes them?
- What happens when one of the connected tools changes?
- What does this look like if my volume doubles?
- What would you recommend I don’t automate?
The last one tells you the most. Anyone who says everything you mentioned is a great candidate is selling rather than advising. Some of what you described almost certainly isn’t worth automating, and a provider who says so before invoicing you is the one to trust.
The short version
The technology is rarely the expensive part. Understanding your process is, and it’s also what determines whether the money achieves anything.
If you want a straight answer for a specific process rather than general ranges, describe what’s eating your week and I’ll tell you what’s involved, including if the answer is that it isn’t worth doing.
Frequently asked questions
How much does AI automation cost for a small business?
My own rates, in Australian dollars excluding GST: discovery is $3,000 to $6,000 as a fixed standalone piece, one AI agent built and running in production is $10,000 to $20,000, and each additional connected system starts around $6,000. Advisory work is $500 an hour. Cost is driven far more by how well understood your process is than by the technology.
What are the ongoing costs of an automation?
Usually a workflow platform subscription plus per-use charges for any AI model calls. For a typical small-business workflow both are modest, often comparable to a single software subscription. The bigger ongoing cost is change: when your process or a connected tool changes, the automation needs updating too.
How do I calculate the return on an automation?
Multiply how often the task runs by how long it takes to get hours per month. Value those hours at what the person doing them costs, then add the cost of errors the automation prevents. Compare that monthly figure against the build cost for a payback period. Under about six months is usually an easy decision.
Is it cheaper to use an off-the-shelf tool?
Often, and it's worth checking first. If a product already does the job, buy it. Custom automation earns its cost when your process doesn't match any product, when several systems need to work together, or when the decision logic is specific to how you operate.
Why won't automation providers publish their prices?
Because the same request can mean two very different pieces of work depending on how well defined the process is. That's a genuine difficulty, but it's not a reason to accept a quote with no breakdown. Any provider should be able to separate discovery, build and ongoing costs for you. Mine are published on this page.
Should I pay separately for discovery?
Yes, and be wary of anyone who throws it in. Discovery is where the process gets mapped and the decision rules get written down, and it's the step that determines whether the build works. Paid separately it stays honest: you get a usable process map either way, and if the map shows the process shouldn't be automated, you've found that out for the price of an afternoon rather than a project.
Wondering what this would look like in your business? A short chat is usually enough to tell.
Let’s chat