Every business has a list of things that “someone should automate one day”. Most of those projects never happen — or they happen, break after two weeks, and nobody fixes them.
After building automations for my own work and for clients, I’ve ended up with a very unglamorous rule: automate the boring task first.
The three questions
Before automating anything, I ask three questions about the task:
- How often does it happen? Daily beats monthly. A 5-minute task done every day is more than 20 hours a year.
- Is it the same every time? If you could write the steps on one page, it can be automated. If every case is “it depends”, it probably can’t — yet.
- What happens if it goes wrong? Sending a weekly report with a typo: harmless. Charging a customer twice: not harmless.
The perfect first automation is frequent, repetitive and low-risk. It’s rarely the exciting idea. It’s usually something like:
- copying form submissions into a spreadsheet or CRM,
- sending a confirmation or follow-up message,
- collecting numbers from three places into one weekly summary,
- renaming, resizing and filing photos or invoices,
- posting the same update to several channels.
Why automations die
Most automations don’t fail on day one. They fail quietly, months later. The usual causes:
- Nobody owns it. It was set up once, the person who built it moved on, and when a password changed it simply stopped.
- It fails silently. No alert, no log — you find out when a customer asks why they never got a reply.
- It was too clever. One giant workflow that does ten things is impossible to debug. Five small ones that each do one thing are easy.
- It automated a bad process. If the manual process was messy, the automated version is the same mess, just faster.
So every automation I build has three things from the start: one clear owner, an alert when it fails, and a way to see what it did.
Where AI changes the picture
Classic automation is great at moving data from A to B. It’s bad at anything that needs judgement — reading a message and understanding what the person wants.
That’s the gap AI now fills. A modern automation can:
- read an incoming email and decide which of five workflows it belongs to,
- pull the important details out of a messy message or PDF,
- write a first draft of a reply, a description or a summary.
The combination is powerful: plain automation for the plumbing, AI for the judgement, and a human for the final yes on anything that matters. I wrote more about that last part in AI agents that do real work.
A small example
A common one: a business gets enquiries from a website form, Instagram and email, and someone copies each one into a spreadsheet, then replies by hand.
The automated version:
- Every enquiry, from every channel, lands in one inbox.
- AI reads it and fills in the details — name, dates, what they want.
- A draft reply is prepared with the right information.
- A person checks it and sends it. The spreadsheet updates itself.
Nothing in that list is magic. Each step is small and boring. Together they give someone back a part of every single day.
Your homework
Write down everything you did on the computer yesterday that you also did the day before. Circle the most boring one. That’s your first automation — and if you’d like help building it, get in touch.