How to Pick Your First Automation (A Scoring Method)

People almost always automate the wrong thing first.

Not the most frequent task, or the cheapest to build. The most annoying one — the thing they've been resenting for months. And annoyance is a terrible selection criterion, because the tasks that irritate you most are usually the ones requiring judgement, which are exactly the ones automation handles worst.

Here's a scoring method that takes about ten minutes and consistently picks better.

The short version

Score every candidate on three things:

Frequency — how often does this happen?
Time cost — how long does each instance take, including context-switching?
Error cost — what happens if it goes wrong and nobody notices for two weeks?

You want high frequency, meaningful time cost, and low error cost.

The trap is high frequency, high time cost, and high error cost — which looks like the biggest win and is the most dangerous place to start.

Step 1 — List candidates properly

Don't work from memory. Memory over-weights the annoying and forgets the frequent.

For one week, note every repeated task. Anything you did more than twice. A note on your phone is enough.

Then add the ones you'd forget: the weekly report you assemble, the monthly invoice chase, the thing you do every time a client signs.

Typical list for a small marketing operation:

• Copying form submissions into a spreadsheet or CRM
• Formatting a blog post into social variants
• Sending a follow-up after a call
• Assembling a weekly performance report
• Adding new customers to an email list
• Chasing unpaid invoices
• Filing receipts
• Checking whether a scheduled post published
• Answering the same five questions in your inbox

Aim for ten to fifteen. You're looking for a field to choose from, not a perfect list.

Step 2 — Score frequency

How many times per month?

Frequency	Score
Daily or more	5
A few times a week	4
Weekly	3
A few times a month	2
Monthly or less	1

Anything scoring 1 is almost certainly not worth automating. Something you do once a month will take longer to build and maintain than it will ever save. This eliminates a surprising share of candidates immediately, and it's the most useful thing this exercise does.

Step 3 — Score time cost

How long does each instance take? Include the context-switching — the two minutes of reorienting before and after, which people systematically forget and which often exceeds the task itself.

Time per instance	Score
Over 30 minutes	5
15–30 minutes	4
5–15 minutes	3
2–5 minutes	2
Under 2 minutes	1

A caution about low scores: a two-minute task done fifteen times a day is thirty minutes daily, which is genuinely worth automating. Frequency and time cost multiply — read them together rather than separately.

Step 4 — Score error cost (inverted)

The one people skip, and the one that prevents disasters.

Ask: if this ran wrong fifty times before I noticed, what would it cost?

Fifty, because silent failure is the norm rather than the exception. Automations break quietly and stay broken until something forces you to look.

If it goes wrong	Score
Nobody notices, trivial to fix	5
Some rework, no external impact	4
Internal confusion, a day to unpick	3
Customers or prospects affected	2
Money, legal exposure, or reputation	1

Note this is inverted — low error cost scores high, because low error cost is what you want in a first automation.

Where things land:

Filing form submissions into a spreadsheet: 5. Worst case is some missing rows you can backfill.

Sending automated outreach: 1. Fifty wrong emails went to fifty real people and can't be recalled.

Publishing social posts: 2. Public, hard to unsee, and represents you.

Assembling a report you review before sharing: 4. You'd catch it.

Step 5 — Multiply and read the result

Frequency × Time × Error tolerance.

Worked examples:

Copying form submissions to a CRM. Frequency 4, time 2, error 5 → 40.

Formatting posts for social. Frequency 3, time 3, error 4 → 36 (assuming you review before posting; drops to 18 if it publishes directly).

Sending post-call follow-ups. Frequency 4, time 3, error 1 → 12. High frequency and real time cost, but it goes to a person unsupervised.

Monthly invoice chasing. Frequency 1, time 4, error 2 → 8.

Start with the highest score, and check it isn't scoring high only because of one inflated term. A candidate scoring 40 on genuine balance beats one scoring 45 on frequency alone.

Step 6 — Sanity-check before building

Four questions before you open Zapier.

Is the process actually good? If the manual version is a mess, automating it produces a faster mess. Fix it by hand first.

Can I specify it exactly? Write the rule as an if-then sentence. If you can't, you can't build it — and that's usually a signal the task needs judgement.

Do the tools connect? Check both systems have an integration or an API before designing anything around them.

What's the manual fallback? When it breaks — and it will — what happens? If the answer is "everything stops and nobody notices," build the alerting first. How.

The easiest thing to automate

If you want a concrete starting point: move data from one system to another.

Form submission to CRM. New customer to email list. Purchase to spreadsheet. Enquiry to Slack notification.

It scores well on every axis — frequent, tedious, and low-consequence when wrong. It's the standard first workflow for good reason, it takes under an hour, and you'll learn how your tools behave on something where mistakes are cheap.

Deliberately not first: anything that sends, publishes, or spends. Those are the ones where the error cost score is 1 or 2, and they're where beginners get hurt.

Calculating whether it's worth it

Rough arithmetic, before and after.

The estimate: time per run × runs per month × 12, minus build time, minus estimated maintenance.

Twenty submissions a month at four minutes each is 16 hours a year. Building it takes two hours. Even at an hour a month of maintenance — which would be high — you're net positive by a wide margin.

Where it goes wrong: a workflow saving five minutes a month against three hours of build time and ongoing maintenance. That's net-negative and will stay running for years because nobody checks.

Measure the real result after a month, not the projection. The audit.

Frequently asked questions

What's the easiest thing to automate?
Moving data between systems — form submissions into a CRM, new customers onto an email list, enquiries into a Slack channel. It's frequent, tedious, requires no judgement, and costs almost nothing if it goes wrong, which makes it the ideal place to learn how your tools behave. Most people build one in under an hour. What to avoid first is anything that sends, publishes, or spends money.

How do I calculate automation ROI?
Time saved per run, multiplied by runs per year, minus build time, minus maintenance time, minus the cost of errors. Maintenance is the term people omit and it's rarely zero — APIs change, fields get renamed, services deprecate endpoints. A workflow built in three hours can quietly cost an hour a month to keep alive, which changes the arithmetic considerably on anything low-frequency.

Should small businesses automate at all?
Yes, but selectively and starting small. The advantage a small business has is that its processes are simple enough to automate cleanly, and one person doing five roles benefits enormously from removing the mechanical parts. What doesn't work is trying to automate the judgement — a small business usually competes on being personal and responsive, and automating that away removes the thing customers were paying for.

What to do next

For the next week, note every repeated task. Don't judge them, just write them down.

At the end of the week, score the list. The winner is usually something dull that you'd never have picked from memory — which is exactly the point, because memory over-weights annoyance and automation rewards frequency.

Free: The automation planning checklist.

---

Related guides

• AI marketing automation — the wider framework
• Is your automation actually saving time? — measuring the result
• Building automations that fail safely — before you build
• Zapier vs Make vs n8n — choosing the tool