Every week someone asks us whether they should "use AI" for something. Our answer is usually maybe, but only once you've done the sum.
The five-minute test
You need four numbers:
- H: hours per week the task takes, across everyone who does it.
- R: the loaded hourly cost of the people doing it (wage plus super, leave and overheads; roughly 1.3 times their hourly rate).
- B: the cost to build the automation.
- M: the monthly cost to run and maintain it.
Annual saving is roughly H × R × 48 (working weeks), minus M × 12.
Payback in months is B ÷ (annual saving ÷ 12).
A worked example
A trades business spends 6 hours a week chasing quote follow-ups. The admin person costs about A$45 an hour loaded.
- Annual cost of the task: 6 × 45 × 48 = A$12,960
- Automation build: A$6,000, running costs A$80 a month
- Annual saving: 12,960 − 960 = A$12,000
- Payback: 6,000 ÷ 1,000 = 6 months
That's a good candidate. A task that takes 40 minutes a week would take years to pay back the same build, and probably isn't worth it.
When the sum says yes but the answer is still no
Some processes look worth automating on paper but will cause trouble. Watch for these:
The process is different every time. If three people do the task three different ways, automating it just bakes in one of those ways. Agree on the process first.
Mistakes are expensive and hard to spot. Sending the wrong invoice to the wrong client, or quoting the wrong price, costs more than the hours saved. These need a human approval step, which reduces the saving. Include that in the sum.
The inputs are messy. If the data arrives as photos of handwritten notes, PDFs in five formats and voicemails, the first job is tidying the inputs, not automating what happens after.
Nobody owns it. Automations break when a tool changes its settings or a form gets a new field. Someone needs to notice. If nobody will, budget for monitoring.
Where AI helps
AI is good at reading messy text and deciding what it is. Is this email a new enquiry, a complaint or a supplier invoice? What job is this customer describing? Which of our documents answers this question?
It's not good at being reliably right about numbers, dates and money without checking. For those, ordinary automation (rules, integrations, scheduled jobs) is cheaper, faster and more predictable.
Most useful systems are a mix: AI to read and sort, plain automation to act, and a person to approve anything that matters.
Start with an audit, not a tool
The most expensive automation is the one that solves the wrong problem. Before building anything, time the work properly for a week. You'll usually find the biggest waste somewhere you didn't expect.
We can go through this with you on a discovery call, and we'll tell you if something isn't worth building yet.