What to automate first: three tests before hiring

The short answer

Automate first the task that passes three tests:

1. It repeats with the same steps, at least weekly.
2. It can run without you: every input is reachable without a person logging in or clicking something at run time.
3. Someone uses the result: the output lands where a person already looks, and they act on it.

Most disappointing automation projects fail one of these, not the build. I know because I have built a couple of dozen scheduled jobs and tools for myself this year, and the ones that disappointed me failed the second or third test, even though the code worked.

Three tests, in this order

A task has to pass all three to be your first automation.

1. It repeatsSame steps, at least weekly.
2. It runs without youNo one logs in or clicks anything at run time.
3. Someone uses itThe result lands where a named person looks, and they act on it.
4. Automate it firstFail one, and the build can be perfect and still disappoint.

Why this is harder than it looks

When you list things you would like automated, you list what is annoying. Annoying and automatable are different. The most annoying tasks are often rare (the yearly tax pack), or need judgement at every step, or need you to be logged in to something. A freelancer can build any of them. You will still be disappointed, because the automation will not save what you hoped.

Two of the three tests exist because I paid for skipping them, in time rather than money. Freelancers make you pay in money.

Test 1: it repeats with the same steps

Frequency is the multiplier on everything. A task that takes 20 minutes and happens every weekday is about 80 hours a year. The same task once a quarter is about an hour. No quote makes sense for the second.

Frequency is the multiplier

The same 20-minute task, hours spent per year.

Every weekday≈80 h Once a quarter≈1 h

“Same steps” matters as much. Write the task down as it runs today, step by step. If two people would write different steps, or the steps change every time, you do not have an automation yet; you have a process to agree on first.

Test 2: it can run without you

This is the test people skip, and the one that bit me hardest.

I built a tool that collects company figures from a paid data site I use. It works: give it a list of names and it searches, opens each page and pulls the numbers. But the site’s login dies the moment the browser closes; nothing that survives between runs keeps you signed in. So the tool can never run on a schedule. Every run needs me to sit down, log in, and start it. It saves typing, not time.

Another one: I wanted product details pulled from a large shopping site. Every automated route was blocked, by the site or by the tools. What worked was me copying the whole page and pasting it in, which took a few seconds. No automated route would have beaten that manual step.

Before you hire, list every source the task reads from and ask of each one:

• Does it have an official way in (an API, an export, a shared sheet), or only a web page behind a login?
• Does the login last, or does it expire when the browser closes, or ask for a code on your phone?
• Is automated access allowed? Some sites forbid it in their terms and block it in practice.

If any source fails, the freelancer can still help, but the job changes. It becomes “a tool I run” rather than “a job that runs”, and the price should be lower to match.

Ask three questions of every source

One source failing changes what you are buying.

Official way in? API, export, shared sheet→Login lasts between runs?→Automated access allowed?

All yes

A job that runs

It starts on a schedule and saves you the time, not just the typing.

Any no

A tool you run

You sit down, log in and start it each time. Still useful; the price should be lower to match.

Test 3: someone uses the result

I built a daily digest that read a handful of public channels, picked out what mattered and sent me a summary each morning. It ran well. I switched the schedule off after one day and made it on-demand, and later retired it. The summaries were fine; I found I wanted them when I asked, not on a timer, and after a while I stopped asking. A result nobody acts on is a cost with no benefit.

I also built two side tools, one that combined several weather forecasts and one that summarised videos on a topic. Both did what they were built for, and I stopped both once they had shown me what I wanted to see. That is fine for a side project. For a paid automation, it is the sign you bought a one-off answer at the price of a machine.

The test is simple: name the person who will act on the output, and where they will see it. “The weekly sales figures appear in the team sheet on Monday by 9am, and the manager reads them in the Monday meeting” passes. “It would be nice to have a dashboard” does not.

Name the person, and where they see it

Passes

"The weekly sales figures appear in the team sheet on Monday by 9am, and the manager reads them in the Monday meeting."

• A named person
• A place they already look
• An action they take

Does not pass

"It would be nice to have a dashboard."

• No person
• No place, no time
• No action

How to score your list

Put your candidates in a table and score each test yes or no.

Start with the row that has three yeses

Example list, scored yes or no on each test.

RepeatsRuns without youSomeone uses it Copy new form entries into the CRM ✓ yes✓ yes✓ yes Weekly overdue-invoice reminders ✓ yes✓ yes✓ yes Pull figures from a portal that logs you out ✓ yes✗ no✓ yes Daily news digest ✓ yes✓ yes✗ not really

See the scores as a table

Task	Weekly or more, same steps?	Runs without a login at run time?	A named person uses the result?
Copy new form entries into the CRM	Yes	Yes (both have APIs)	Yes, sales follows up
Weekly overdue-invoice reminders	Weekly, same steps	Yes	Yes, customers pay
Pull figures from a portal that logs you out	Weekly	No	Yes
Daily news digest	Daily	Yes	Not really

Start with a row that has three yeses. Copying data between two tools, scheduled reports and notifications usually win. The rows with a “no” are not forbidden; they are just not first.

Then do the rough maths: runs per year × minutes saved per run, plus the errors you stop making by hand. Compare it with the quote plus a year of fixes, because connected tools change and something will break. The maintenance guide covers that last part.

Where it goes wrong

• Automating the exception. People pick the task that went wrong last month. Pick the one that goes right every week and just takes time.
• Automating a broken process. If the steps are unclear by hand, a machine makes the confusion faster.
• Forgetting who fixes it. Every automation needs an owner on your side who notices when it stops. Without one, it will stop quietly. The acceptance checklist is how you check it tells someone.

Wrap-up

Pick the task that repeats, runs without you and feeds someone who acts on it. Write its goal as one measurable sentence, such as “every new form entry is in the CRM within five minutes”, and you have the first line of a brief. The rest of the brief, and sellers who build this kind of job, are on the workflow automation page.