What to automate first in a small business, and what to leave alone
Start with the work that is already costing you money you can name. The invoice nobody chased. The quote that sat for three days while the customer rang someone else. The call that rang out at 4:40 on a Friday. Then work down that list: money already owed first, then quote speed, then calls answered, then reporting.
That order is the whole argument. It is not a preference and it is not a fashion. It is the order the returns actually arrive in, and it means the first build pays for the second one instead of competing with it.
Nearly a third of Australian business owners still do their own financial admin, and one in five spends between 21 and 40 or more hours a month on it. That is close to a full working week a month, spent on work that repeats and requires no judgement. Somewhere in that week is the first thing worth automating.
What should a small business automate first?
The work that holds money you have already earned, before the work that might earn more.
Unpaid invoices are the clearest case, because the money is yours and it is sitting outside your bank account. An overdue invoice queue can be chased on a schedule without anyone deciding to do it: a reminder on day seven, a firmer one on day fourteen, a human escalation at day thirty. No build is needed to know it works. Either the payment lands sooner or it does not.
Quoting is next. A quote that goes out at the visit wins more often than the same quote sent that night, and a quote sent that night wins more often than one sent on Sunday. Anything that assembles the quote from the details you already captured on site removes the gap between the visit and the send, and the gap is where the job goes.
Calls answered comes third. A missed call is a lead with a short life. A booking link, a text back inside a minute, or a reception layer that takes the details all beat a voicemail box nobody checks. Answering first matters more than answering perfectly.
Reporting comes last, and it is the one that gets built first most often because it is the easiest to demonstrate. A weekly dashboard is pleasant. It rarely changes a bank balance by itself, so it can wait.
The field is still early and still arguing about this. ABS figures on business use of AI put reported use at 12 per cent of businesses in 2024-25, against 1 per cent in 2022-23, which means most of your competitors have not decided either. That is an advantage while it lasts, not a reason to buy something.
How do you rank two candidates against each other?
You score both, in writing, on your own week. Not on the pitch, not on the demo, on the arithmetic.
The scorecard has two sides. The return side is hours a week multiplied by your loaded labour rate. The effort side is the hours it takes to build, plus what it costs to run each month.
Weekly value is hours saved in a week times the loaded rate. Loaded, not award. Add superannuation, leave, insurance and the time it takes to supervise. If the work is your own, use what an hour of your labour earns you, because that is what the task is costing while you do it instead of selling.
Payback weeks is the build hours divided by the hours saved each week. A build that takes twelve hours and saves five hours a week pays back in under three weeks. A build that takes forty hours and saves one hour a week pays back in forty weeks, and by then you will have forgotten why you started.
Work it through on two real candidates. Chasing overdue invoices takes three hours a week and the build is six hours. Quoting from site notes takes forty minutes a week and the build is twenty hours.
| Overdue invoice chasing | Quote assembly from site notes | |
|---|---|---|
| Hours a week | 3 | 0.7 |
| Loaded rate (your own figure) | your rate | your rate |
| Weekly value | 3 x your rate | 0.7 x your rate |
| Build hours | 6 | 20 |
| Payback weeks | 2 | 29 |
The invoice chase wins on both sides, and it is not close. Ranking candidates like this takes ten minutes and it settles arguments that otherwise run for months, including arguments with a salesperson.
Score every candidate that survives the next section. Anything with a payback past about eight weeks (your own figures, your own week) is not a first automation. It is a second or third one, and it should be written down and left alone until then.
What should stay manual for now?
More than the pitch decks admit.
Keep pricing manual. Judgement on a job, a margin, a customer who haggles, is the part of the business that is actually yours. A price built by rule loses the jobs it should lose and wins the ones it should have walked away from.
Keep hiring manual, and keep the first conversation on a complaint manual. Both need a person to hear something that is not in the form.
Leave anything rare alone. A task you do twice a year has almost no hours to save, so the build never pays back, whatever the demonstration looked like. Same for one-off jobs and anything with a long tail of exceptions: the exception handling costs more than the task ever did.
Leave alone the work that is already working. If the current way is fast, cheap and not resented, automating it buys you a project and a subscription.
And never automate a process you have not first fixed. Automating a broken process means making the same mistake faster, more reliably, and in more places at once. The government's productivity guidance for small business puts reviewing your processes and standardising the routine steps ahead of exploring digital tools, in that order. Follow it. Standardising first also makes the build smaller, because a task that always runs the same way is cheaper to automate than one that runs five ways.
What does the first month of an automation look like?
Boring, if it is done properly.
Week one is the fix, not the build. You write the task down step by step, cut the steps that exist only because nobody ever questioned them, and settle the wording. That document is the build specification and it is worth more than any software choice.
Week two is the smallest version that works. One trigger, one path, no branches, run beside the old way rather than instead of it.
Week three is comparison. You count the hours actually saved against the hours you predicted, and you count the mistakes. Most first automations need one fix here, usually at the handover point where a person still has to check something.
Week four you decide. If the hours are real, keep it and move to the next candidate on the list. If they are not, switch it off and keep the written process, which cost you almost nothing and is still an improvement.
There is a second half to week four that matters more than the first. Write down what you learned about how long a build really takes you. That number is what the next scorecard needs.
When is the second automation the right one?
When the first one is quiet.
Quiet means it runs without you watching it, the exceptions have settled, and the hours it was meant to save have turned up for a month. That is the signal to start the next candidate, and the first one should by now be paying for it.
Starting the second while the first is still noisy is the most common way to end up with two half finished systems and a subscription for each. The order in this article exists to prevent that: each build funds the next, and nothing is built on top of something that has not yet proved itself.
Questions people ask before they start
What is the cheapest useful first automation?
Whatever already exists in the tools you pay for. Most job, booking and accounting systems have reminders, templates and follow up sequences switched off by default. Turning them on costs nothing, and if it works you have learned the same lesson for free.
Should quoting or follow up come first?
Quoting, if quotes leak. Follow up on invoices first, if the leak is in getting paid. Whichever one is holding money right now is the one to build, which is why the scorecard ranks your candidates rather than a list someone else wrote.
What should I never automate?
The judgement work: pricing, hiring, the unhappy customer, and the apology. Also never automate anything that sends to a customer without a person able to see what it said, at least for the first month.
How long before it pays for itself?
Whatever the payback weeks number on your scorecard says, and if you have not worked that number out you cannot answer the question. A payback under about eight weeks (your own figures) is worth building. Past about twenty, it is usually a project rather than a fix.
Run the free automation plan at /start/. Bring your own week, the candidates you are weighing, and the hours each one eats. You will leave with a ranked list instead of a shortlist, and the first build will already be paying for the second. The audit is where that ranking gets priced.