Quick answer
Automate the work that is high volume, rule-based and needs no judgment: moving data between systems, chasing quotes and invoices, sending follow-up emails, assembling the same report every week. Score each candidate on four things: how often it runs, how many minutes it eats each time, how rule-based it is, and what an error costs. Tasks that run at least weekly, take more than five minutes, follow the same steps every time and fail harmlessly are your first automations. Anything involving judgment, exceptions or an unchecked client relationship waits.
Choose the next workflow to test
I would shortlist work that repeats and is straightforward to check.
- 01
Find repetition
Look for tasks that happen often and take meaningful time.
- 02
Check the difficulty
Assess input quality, judgement, exceptions and system access.
- 03
Choose a small trial
Pick a bounded task with an owner and a visible review step.
A frequent, simple task is often a better first trial than an ambitious process with unclear inputs.
Score every candidate on four things
Pick ten tasks that annoy people in your business and run each one through the four factors below. This takes about half an hour with a spreadsheet and it beats any tool comparison you could read.
| Factor | What you are measuring | Scores high when | How to get the number |
|---|---|---|---|
| Frequency | How many times the task runs in a month | It runs daily, or once per job or per customer | Count a normal week and multiply by four |
| Time per run | Minutes of human attention each time | It takes more than five minutes of clicking and copying | Time yourself three times and take the middle number |
| Rule-based | Whether a new starter could follow written steps with no judgment | The steps are identical every time | Try writing the steps down. If you cannot, it is not ready |
| Error cost | What happens when it goes wrong | Score this one low. Low means a mistake is obvious, internal and cheap to fix | Ask who sees a bad output: you, a supplier, or a client |
Frequency multiplied by time per run gives you monthly hours. That number is the whole business case. A task that eats twenty minutes once a quarter is not an automation project no matter how much it irritates you. A task that eats four minutes forty times a week is nearly three hours a week, and that is worth building something for.
Then use rule-based and error cost as gates rather than scores. If the task is not genuinely rule-based, automation will produce confident nonsense at speed. If the error cost is high and nothing catches mistakes before a client sees them, you either add a human check step or you leave it alone for now.
The automations that usually win first
Across the businesses I have sat down with, the same handful of workflows come out on top. None of them are exciting. That is the point.
- Data entry between two systems. Anything a person retypes from one screen into another: enquiries into the CRM, approved quotes into the accounting system, timesheets into payroll. High frequency, purely mechanical, and errors show up quickly.
- Follow-ups and quote chasing. Quotes that sit unanswered for a week are the cheapest revenue in most small businesses. A scheduled sequence that nudges at day three and day ten, and stops the moment someone replies, is one of the highest-return things you can build.
- Report assembly. Pulling the same numbers from the same three places into the same weekly document. Automate the assembly and the distribution, keep the human reading and interpreting.
- First-draft documents from structured inputs. Job notes into an invoice draft, a site visit form into a quote draft, a call recording into a summary with actions. The draft is automated. A person still approves it before it leaves the building.
- Enquiry triage and routing. Classify an inbound email or form, tag it, route it to the right person, acknowledge receipt. This is where AI earns its keep over plain rules, because it can read messy human writing that a keyword filter cannot.
What not to automate first
The list of things to leave alone is as useful as the list of things to build. These are the ones I most often talk people out of.
- A process that is already broken. Automating a bad process gives you a faster bad process and makes it harder to change. Fix or simplify it on paper first, then automate what is left.
- Low-volume work. If it runs a few times a month, the build, the maintenance and the day it silently breaks will cost more than the task ever did.
- Work that changes every month. Anything tied to a process still in flux, or to a system you are about to replace, is a build you will pay for twice.
- Unchecked client-facing output. Automated drafting is fine. Automated sending of anything a client will treat as a commitment, a price or advice is not, until a human is reliably in the loop.
- The task one person does differently every time. That usually means the rules live in their head. Get the rules out of their head first. Sometimes that alone is the whole win.
Do it yourself, hire an agency, or get advice first
If you have one or two candidates and they connect mainstream apps, build them yourself. Zapier and Make are both designed for non-developers and you will learn more about your own process in an afternoon of building than in any workshop. Start with the smallest version that works, run it alongside the manual process for a fortnight, then switch.
Hire an agency when the workflow spans several systems, when failures cost money, or when nobody in the business can maintain what gets built. Expect a written scope naming each workflow, a fixed price for the build, and a clear statement of what any ongoing fee covers. If a proposal is a monthly retainer with no defined deliverable, that is a subscription, not a project.
Get independent advice first when the shortlist is long, when quotes vary wildly, or when you are not sure the process is worth automating at all. AI Consulting is A$1,195 per month + GST, with a call each fortnight, clarity on what to build in Claude or ChatGPT and practical guidance on how to do it. No lock-in. Custom development is A$150/hour + GST, with a separate project scope, estimated hours and terms before work starts.
A 30-minute version you can run today
- List every task your team repeats at least weekly. Twenty lines is plenty.
- Against each one write how many times a month it runs and how many minutes it takes. Multiply for monthly hours.
- Cross out anything under an hour a month.
- For what is left, try to write the steps as instructions for a new starter. Cross out anything you cannot write down.
- For what survives, ask who sees the output if it is wrong. Put anything that reaches a client straight into a second column marked needs a human check.
- Sort the remainder by monthly hours. Your first automation is at the top. Build that one, prove it for a fortnight, then come back to the list.
If the top of your list is obvious after that exercise, go and build it. If three things are tied, or the list is full of work you suspect should not exist at all, that is the conversation to have with someone who is not quoting on the build.
Official resources
Explore the tools and check current details with the people who make them.
