Automate first the job that happens often, follows clear rules, costs little if it goes wrong and produces something you can check in thirty seconds. That is not usually the job that annoys you most. It's the job that is safest to get wrong while you're learning.
You can find yours in about ten minutes with a five-question score, and I've put the whole method below. No list of ideas. You already have more ideas than hours.
I run a paid AI membership, so I have an obvious interest in you building things. I'd still rather you picked a dull first job and finished it than picked a clever one and abandoned it.
Why does the first automation matter so much?
Because the first one decides whether you do a second. A first build that works teaches you how triggers work and why you test before switching anything on. A first build that misfires teaches you that automation is risky and not for you.
The job most people want to automate first is the one they hate most. It's usually the messy one: lots of exceptions, a client-facing output, a judgement call in the middle. Hating a job and being able to automate it are two separate things.
What are the five questions to score a job on?
Score each job from 1 to 3 on each question. Higher is better for automating.
| Question | 1 point | 3 points |
|---|---|---|
| How often does it happen? | A few times a year | Weekly or more |
| How long does one run take you? | Under 5 minutes | 20 minutes or more |
| How rule-based is it? | I decide case by case | Same steps every time |
| What does a mistake cost? | A client or money is affected | Nobody sees it, easy to fix |
| Can I check the output quickly? | I'd have to redo the work | I can see it's right in 30 seconds |
Frequency and time per run together tell you the size of the prize. Multiply them in your head: something that takes 20 minutes and happens ten times a month is over three hours. Something that takes 45 minutes once a quarter isn't worth building first, however much you dislike it.
The last three tell you how likely the first build is to work and how safely it can fail.
How do you score your jobs in 10 minutes?
Set a timer. This is the whole exercise.
- List 6 to 8 repeating jobs from the last two weeks. Pull from your sent emails and calendar, not from memory. Memory only offers up the dramatic ones.
- Score each on the five questions above. Go fast. Gut scores are fine.
- Add them up. Out of 15.
- Apply one veto. Any job that scored 1 on "what does a mistake cost" can't be your first, whatever its total.
- Pick the highest remaining score. On a tie, take the one with the cheaper mistake.
A made-up example, to show the mechanics. A bookkeeper lists four jobs:
| Job | Often | Time | Rules | Mistake | Check | Total |
|---|---|---|---|---|---|---|
| Send intake form after a booking | 3 | 2 | 3 | 3 | 3 | 14 |
| Weekly summary of overdue invoices | 3 | 2 | 3 | 2 | 3 | 13 |
| Reply to client questions | 3 | 3 | 1 | 1 | 2 | 10 |
| Write quarterly newsletter | 1 | 3 | 1 | 2 | 2 | 9 |
Replying to client questions is the one she dreads, and it scores 10 with a mistake score of 1. It's vetoed. The intake form wins at 14. It's a dull first build, and it's the right one.
What if my top job needs AI to make a judgement?
Then split it. Automate the rule-based part first and keep the judgement call as a person's step, or as an AI draft you approve.
In my own setup, a booking triggers an intake email, and the form answers land in a sheet. Only after that do I let an AI agent research the client, because that step needs judgement. I built the plain version first. It took me about twenty minutes, and the research step came later. The whole thing is in 4 business automations for non-technical founders.
If you aren't sure whether your job needs an automation or an AI agent, the quick test is this: if you can write the steps as "if this, then that", it's an automation. If you'd have to say "it depends", it's a judgement step.
What should you check before you switch it on?
Run it on three real examples while you watch, with the sending step switched off. Read what it produced. Then turn it on for real.
Two checks I'd never skip:
- Test the odd one. Use a booking with a missing field or a weird name. Real life sends those.
- Decide how you'll notice it breaking. Have it message you on failure, or look at the output every Friday for a month.
Before you build, write down the time it takes now, in minutes per run. You can't claim you saved time later without a starting number. Count the checking too.
When the first one is live, the question of what's next is answered by the same score, because your list is still sitting there. If you want to see how a handful of automations join up into a system, the STACK framework covers mapping your processes before you build.
Quick answers
Should I automate my content first?
Only if it scores well. Content is often high in time and frequency but low on "rule-based" and "mistake cost", because it's public. Automate the admin around it first, like scheduling and file naming, and keep the writing as a draft you approve.
What if every job scores about the same?
Pick the one with the cheapest mistake. When scores tie, safety is the tiebreaker, because your second build will be better than your first anyway.
Should I automate a job I only do monthly?
Only if each run is long, say an hour or more, and the steps are identical. Otherwise you'll forget how the automation works by the time it next runs.
Which tool should I use?
Whichever you can get running this week. Check whether software you already pay for has a built-in automation before you add another subscription.
Where live help speeds this up
Scoring is the easy half. The slow half is the build, the testing and the "why did it send twice" moment.
One member wrote: "OMG I have built my first automation using N8N (had never heard of it before here!) pulling in data from Karbon and Xero so I can see exactly what jobs and clients are taking the most time and who is the least profitable." That's one person's result, not a promise of yours. In Wright Mode you bring your top-scoring job to the fortnightly Co-Work and Q&A, watch me build in the monthly live Build, and have someone to ask when it breaks.
