Find repeated work
The best automation candidates are repeated, rule-heavy, and already documented by habit, even if not formally written down. Ask the team which task they would happily never do again, then watch how it is actually done. If the steps are stable from week to week, such as copy the data, rename it, summarise it, send it on, it is a candidate. If every instance needs a judgement call, it is better suited to AI-assisted drafting with a person finishing the job. Frequency matters more than size: a ten-minute task done daily is worth more than an hour-long task done quarterly.
Automate the preparation step
The first win is often drafting, summarising, classifying, routing, or pre-filling work rather than making the final decision. Preparation automations carry less risk because a person still signs off, and they are easier to build because tools like n8n, Make, and Zapier already connect to most common business apps. Start where the data already lives, such as the inbox, the CRM, or the project tracker, because integration effort is the real cost in most builds. The n8n documentation is a useful way to see what a workflow tool can reach before committing.
Source notes: n8n documentation
Keep a human checkpoint
Automations should prepare work. Sensitive customer, finance, compliance, publishing, and employment actions need review. This is also where Australian obligations apply: workflows that handle personal information should line up with the OAIC's privacy guidance, and anything granted system credentials should follow Australian Cyber Security Centre basics such as scoped access and multi-factor authentication. The checkpoint is not a bottleneck when it is designed as a queue with a clear approve-or-edit decision.
Source notes: OAIC privacy guidance, Australian Cyber Security Centre
Measure the saved loop
Track time saved, error reduction, lead response speed, publishing velocity, reporting quality, or fewer handoffs. Baseline first: time three or four real runs of the manual process before automating, so the saving is a measured fact rather than a guess. Pick one number before the build and measure it the same way afterwards. A claim like it feels faster does not survive a budget conversation. Lead replies went from four hours to ten minutes does.
Make ownership explicit
Every automation needs a named owner who knows how to run it, pause it, update it, and explain it to the team. Unowned automations fail silently: an API changes or a form field gets renamed, and nobody notices until a customer does. The owner's job is a monthly check that the automation is still running, still accurate, and still worth keeping. Ownership also covers the prompt and the template: when output quality drifts, the owner is the person who notices and adjusts.
A worked example: the Friday report
An agency spent about five hours every Friday building client status reports: pulling numbers from three systems, pasting them into a template, and writing a summary. The automation now collects the numbers on a schedule, fills the template, and uses an AI step to draft the summary paragraph. The account manager reviews and edits each report in about ten minutes. Five hours became under one, reports go out on time every week, and the review step means a wrong number has never reached a client. The build took two days and paid for itself inside a month.
Common automation mistakes
Automating a broken process, which only produces mistakes faster. Skipping the human checkpoint on customer-facing output. Leaving automations undocumented, so the business depends on one person's memory. Connecting tools with shared admin logins instead of scoped credentials. Chasing full autonomy on day one instead of starting with preparation steps. And stopping measurement after launch: the value case should be re-checked monthly, because volumes, prices, and processes change.