How do I find automation opportunities in my business?
'I've seen people build a seven-step automation to save themselves four minutes.'
I say that in workshops and someone always laughs a bit too knowingly... so I finish it: 'Congratulations. You've automated yourself into a second job.' You probably know the kind of build I mean. A form feeds a spreadsheet, the spreadsheet pings Slack, Slack fires off an email, and the chain falls over every time someone types an emoji into the form, so now you spend Thursday afternoons keeping it alive.
Finding automation opportunities is mostly about avoiding that. Score each task you do from 1 to 3 on how often it happens, how annoying it is and whether AI can reliably handle it, then multiply the three numbers. Anything scoring 18 or more is worth building, and anything on 6 or under can stay exactly as it is. Before any of that, check whether the task should exist at all.
How do I find automation opportunities in my business?
Run every task through Frequency × Friction × Feasibility. It's the filter I teach for this, and it comes down to three questions:
- Frequency: How often?
- Friction: How annoying/time-consuming?
- Feasibility: Can AI reliably handle it?
High/high/high = build it. That's the short version, and for most tasks it's enough.
Each question kills a different bad idea. Frequency catches the task you do twice a year, where building the automation costs more than doing it by hand. Friction catches the task that happens daily and takes nine seconds, which isn't hurting anyone. Feasibility catches the one that's frequent and painful and still needs a human brain, like pricing a custom job or deciding which customer gets the last delivery slot before Christmas.
How do I score a task?
Give each task a 1, 2 or 3 on each question and multiply, so the top possible score is 27. I multiply instead of adding because one weak score should drag a task down. A job that's constant and maddening but that AI can't do reliably shouldn't sneak through on the strength of the other two.
My thresholds:
- 18 to 27: build it.
- 8 to 12: maybe. Test it by hand with an AI assistant first.
- 6 or under: leave it.
- A 1 for feasibility: leave it, whatever the total says.
Here's how a small service business with a busy inbox, a pile of supplier paperwork and a monthly board meeting might score its week.
| Task | Frequency | Friction | Feasibility | Score | Call |
|---|---|---|---|---|---|
| Turning meeting notes into actions and a follow-up email | 3 | 3 | 3 | 27 | Build, with a person reading before it sends |
| Copying website enquiries into the CRM | 3 | 2 | 3 | 18 | Build |
| Keying supplier invoices into the accounts | 3 | 2 | 2 | 12 | Maybe, test on one supplier first |
| Scheduling the week's social posts | 2 | 2 | 2 | 8 | Maybe |
| Writing the monthly board report | 1 | 3 | 2 | 6 | Leave, use AI as a drafting partner |
| Pricing a custom job | 2 | 3 | 1 | 6 | Leave, it needs your judgement |
Two patterns tend to show up when I run this with a team. The winners are rarely glamorous (meeting notes and CRM data entry often beat the clever ideas). And the board report scores low even though everyone hates it, because it only comes round once a month, so the better use of AI there is as a drafting partner, with nothing to build or maintain.

Should I fix a process before I automate it?
Yes, and it's the step people skip. Before we automate it, I'd like to know whether the process deserves to exist in its current form.
Plenty of business admin is a fossil. Think of the spreadsheet that gets copied into another spreadsheet because of a software change years ago, the form that needs three signatures for a $40 purchase, or the weekly report nobody has opened since the person who asked for it left. Automate any of those and you've made the waste faster and harder to see.
Don't automate the bureaucracy. Cut it or simplify it first (or ask whoever owns it whether it still matters), then score whatever survives.
Where should a human stay in the loop?
Put a person between the automation and anything that touches a customer, money or a decision about someone. In practice that's a review step: the AI drafts the follow-up email and a person reads it before it goes, or the AI reads the invoice and suggests the coding and someone approves it.
The National AI Centre's SME AI Pulse found that checking AI outputs before they affect customers is the most common safeguard among Australian SMEs using AI, and roughly half have it in place, which leaves the other half finding out from their customers. The Australian Government's essential AI practices make the same point at a bigger scale, with oversight and accountability built in from the start.
The framework I teach is Human → AI → Human. A person sets the direction, AI expands and challenges it, and a person chooses and shapes what goes out. The human doesn't disappear from the loop, even when the middle runs on its own at 2am.
How much time should I spend building an automation?
Less than it will save you, which sounds obvious until you do the maths. Randall Munroe's xkcd chart 'Is It Worth the Time?' works it out across five years, and shaving five minutes off a daily task buys you roughly six days of building before you're behind.
Now run my four-minute example through it. Four minutes saved once a week adds up to about 17 hours over five years. A seven-step build that takes a weekend to set up and an hour a month to babysit eats all of that before the first year is out.
What can I do this week to find my first automation?
Six steps, and most of the effort is noticing what you already do.
- Keep a task log for one week. Every task and roughly how long it took, on your phone or a notepad by the kettle.
- Cross out anything that shouldn't exist, starting with reports nobody reads.
- Score what's left from 1 to 3 on frequency, friction and feasibility, then multiply.
- Take the single highest score and write its steps out as if you were training a new starter on their first day.
- Mark where a person checks the output before it reaches a customer, an invoice or anyone's record.
- Run it by hand with an AI assistant for two weeks, and automate it once the steps stop changing.
Step six is the one people want to skip... and it's the one that saves the second job. Doing it by hand for a fortnight shows you the exceptions (the supplier who sends invoices as photos of a printout, the enquiry form someone fills in with 'see attached') before you've built anything around them.
Weave, our build service on the services page, is for the task that scores 18 or more when you'd rather not build it yourself. We build around how you work, test it on your data and hand it over running. For anyone still working out where AI fits at all, the place to start is what your business should use AI for first.
The four-minute jobs can stay yours (some of them are almost relaxing!)
Questions people ask
What's the difference between using AI and automating with AI?
Using AI means you open a tool, ask for something and check the result. Automating means the work starts without you, triggered by an email, a form or a time of day. An agent is the model as the brain, connectors as the hands, plus instructions and the ability to take several steps, so it can act inside your other systems.
Do I need a developer to automate a task?
Usually not for a first automation. No-code tools such as Zapier and Make connect common business apps without writing code, and an AI assistant can draft the steps with you. Bring in help once the automation touches money, customer records or several systems at once, because that's where a silent failure costs the most.
How do I know if an automation is working?
Time the task by hand before you build anything, then time it again a month after. Check a sample of the outputs every week for the first month, and note every time a person had to step in and fix something. Regular fixes mean the process is still changing and needs a person closer to it for now.
What should a business never automate?
Anything where the point is that a person did it, like an apology to a customer or telling someone they didn't get the job. Final decisions about people, pricing and money stay human too. AI can draft and suggest in all of these, and a person decides and puts their name on the result.