What Should I Automate First?

Everyone says "automate it". Nobody says what.

You've been told AI will save you hours. Probably by someone selling something. What you haven't been told is which hours — so you either do nothing, or you automate the thing that seemed most exciting and it turns out to save you eleven minutes a month.

The problem is the question. "What should I automate?" is impossible to answer cold, because it's asking you to imagine a solution before you've properly described the problem.

Don't ask what you could automate. Ask what you're doing over and over that follows the same steps every time.

That's a question you can answer from memory, in about ten minutes, without knowing anything about technology. And once you can answer it, the automation part is the easy bit.

How to use it

List the week

Everything you'd
rather not have done.

Run the four tests

On each one.
Takes seconds.

Fix the top one

Just one.
Not all of them.

Do this before you read on

Open your calendar and your sent emails from last week. Not a normal week — last week, whatever it looked like.

Write down every task you did that you'd hand to someone else tomorrow if you could. Don't filter. Aim for fifteen. Memory alone will miss half of it — that's the whole reason we're looking at the record instead.

The four tests

Run every task through these. Yes or no. Don't overthink it.

01 · Do I do this at least once a week?

If it's monthly or rarer, cross it off. It stays on your plate for now.

Why it matters — frequency is the whole economics of this. A task taking ten minutes a day is over 40 hours a year. The same ten minutes once a month is two hours. People almost always try to automate the big annoying quarterly job and ignore the small daily one, which is backwards.

02 · Does it follow the same steps every time?

Roughly is fine. If you could describe it as "first I…, then I…, then I…" — that's a yes.

Why it matters — anything with steps can be written down, and anything written down can be handed over, to a person or a machine. If you can't describe the steps, that's not a failure of the task. It usually means it involves judgement, which test four deals with.

03 · Is the information already written down somewhere?

In an email, a spreadsheet, your calendar, a form, a message. Anywhere that isn't only your head.

Why it matters — this is the test most people skip and it's the one that decides whether something is a two-hour job or a two-week one. A machine can only work with what it can read. If the crucial detail lives in your head or in a phone call nobody wrote up, the first fix isn't automation — it's capturing the information. Do that and the rest gets easy.

04 · Would it matter if it was 90% right?

Be honest. Some things genuinely have to be right. Most don't.

Why it matters — this is the safety test, and it's the one that stops you doing something stupid. A first draft that's 90% right saves you an hour. A price quote that's 90% right costs you money. A legal document that's 90% right costs you a lot more than that.

If the answer is no, it doesn't mean leave it alone — it means a human checks it before it goes out. That's still most of the time saved.

Reading your answers

4 yeses · Fix it this week, yourself

Weekly, same steps, information already written down, and 90% is fine. Usually solved by a good saved prompt and a template — no software, no spend.

Typically: first-draft emails, turning notes into summaries, rewriting the same message for different customers, drafting social posts, formatting quotes from a set price list.

3 yeses · Worth a few hours of setup

Usually failing test three — the information exists but it's scattered. Fix the capture first and this drops into the tier above.

Typically: anything where you're re-typing something that already exists elsewhere, or hunting through emails to find a detail you know you've been sent.

2 or fewer · Leave it alone for now

Not because it can't be done — because the effort won't pay back yet. Come back once the easy wins are running.

Being told not to automate something is more useful than being told to automate everything. Most of the money wasted on this gets wasted here.

A worked example

Fictional four-person branding agency. Their real list from one week.

Writing the weekly client update emails — 5 clients, ~15 min each, every Friday

Weekly ✓ Same steps ✓ Info written down ✓ 90% fine ✓ — 4 yeses. 65 hours a year. This is the one to fix.

Chasing overdue invoices — ~30 min, weekly

Weekly ✓ Same steps ✓ Info written down ✓ 90% fine ✗ (tone matters, these are clients) — 3 yeses. Draft it automatically, send it by hand.

Writing proposals — 2–3 hours each, ~2 a week

Weekly ✓ Same steps ✓ Info written down ✗ (the discovery call is never written up) 90% fine ✗ — 2 yeses. The real fix isn't automation, it's writing up the call properly. Do that and it becomes a 4.

The quarterly board pack — a full day, four times a year

Weekly ✗ — fails at test one. Four days a year and it feels enormous, which is exactly why they'd been planning to start here. Wrong call.

The boring job — five update emails — was worth 65 hours a year. And the job they were about to spend money automating was the worst candidate on the list.

Start with the boring one

Your list will have one task that's obviously the most interesting to automate, and one that's the most frequent. They're almost never the same task.

Do the frequent one. It's duller, it works, and it'll be running by Friday — which matters more than you'd think, because the first one working is what makes you do a second.

Don't automate a bad process. You'll just get bad results faster.

If a task is annoying because it's badly designed, fix the design first. Half the things on your list will disappear entirely once you look at them properly — and that's a better outcome than automating them.

Run this audit again in three months. Your list will have changed, and so will the answers — that's the point. It's a habit, not a one-off.

Where this goes next

Most people find two or three four-yes tasks on their first go. Those are genuinely fixable yourself, this week, with a saved prompt and twenty minutes — and you should do that before spending anything.

The three-yes ones are where it gets interesting, because they're usually a capture problem rather than an automation problem, and solving that once tends to unlock several tasks at the same time.

Do the easy ones first. If you get stuck on a three, that's a fair conversation to have.