Skip to content
Ioannis Ouzounis
All articles

Automation

Is a 20-Minute Task Worth Automating? Do the Math Before You Build

Small tasks hide big numbers. Here's a simple way to put a price on a repetitive task, work out your break-even budget and spot the tasks that aren't worth automating, with a worked example.

6 min read

On this page

"It only takes twenty minutes."

That sentence keeps small manual tasks alive for years. Twenty minutes doesn't feel like a problem, and nobody puts it in a budget. But twenty minutes every working day adds up to more than 75 hours a year, for one person. If three people do it, that's about 230 hours, close to six working weeks.

Whether that's worth automating isn't a matter of opinion. It's arithmetic, and it takes about five minutes. Here's how I do it.

Step 1: Put a price on the task as it is today

You need five numbers:

  • Minutes per run. Time it once or twice; don't guess.
  • Runs per year. Per day, week or month, times the working weeks in a year. I use 46 working weeks, which leaves room for holidays, sick days and quiet periods.
  • People who do it.
  • Hourly cost per person, fully loaded: salary plus taxes, benefits and overhead, not just the figure on the pay slip.
  • Mistakes, and the time to fix them. How often does something go wrong, and how long does it take to find and fix it?

Then:

  • Task hours = minutes per run × runs per year ÷ 60
  • Rework hours = mistakes per year × minutes to fix one ÷ 60
  • Cost today = (task hours + rework hours) × hourly cost

A worked example (illustrative figures)

Two people each spend 20 minutes every working day reconciling the same export against a spreadsheet. About twice a month something slips through, and it takes an hour to track down and fix. Their fully loaded cost is €40 an hour.

  • Runs per year: 1 × 5 days × 46 weeks × 2 people = 460
  • Task hours: 460 × 20 ÷ 60 ≈ 153 hours
  • Rework hours: 24 mistakes × 1 hour = 24 hours
  • Cost today: about 177 hours, or about €7,100 a year

Nobody would approve a €7,100 budget line for "reconciling an export". But that's what it already costs.

Step 2: Estimate what automation would actually remove

Automation rarely removes all of a task. Someone still starts it, glances at the result and handles the odd exception. So be conservative. In this example, assume the automation removes three quarters of the time and prevents three in four of the mistakes:

  • Hours saved = task hours × share of time removed + rework hours × share of mistakes prevented
  • Here: 153 × 0.75 + 24 × 0.75 ≈ 133 hours a year
  • Money saved: 133 × €40 ≈ €5,300 a year

That's almost three hours a week handed back to two people who were hired to do something else.

Step 3: Find your break-even budget

The money saved in the first year is your break-even budget: the most you can spend and still get the money back within twelve months. Here, that's about €5,300.

If you have an actual quote, the payback period in months is the quote divided by one month of savings (the yearly savings ÷ 12). A €3,000 build would pay for itself in about seven months, and come out about €13,000 ahead over three years.

Don't trust the assumptions? Make them pessimistic: say the automation removes only half the time and prevents no mistakes at all. The savings drop to about €3,100 a year, which still pays back a €3,000 build within a year. When even the pessimistic version works, the decision is easy.

What the stopwatch misses

The calculation above counts only time. Three costs of manual work never show up on a stopwatch.

Errors that travel. A step that works 95% of the time sounds safe. Repeat it twelve times, say once per language in a release, and the chance that all twelve go right is 0.95¹² ≈ 0.54. The chance that at least one goes wrong is close to one in two. One broken placeholder in a client delivery costs far more than the minutes it took to make. (More on this in The Hidden Costs of Manual Localization.)

Knowledge in one head. When only one person knows how the task works, every vacation and every resignation puts a deadline at risk.

Experts on clerical duty. The people doing the copy-paste are often your most experienced staff, because they're the ones who know where the traps are. Every hour they spend reformatting is an hour they don't spend on the work you hired them for.

None of these are in the formula, which makes the formula conservative. That's the right way round.

When the answer is no

Sometimes the numbers say don't automate, and that's a useful answer too. Be wary of automating a task when:

  • It's rare. A task done four times a year seldom earns back a build.
  • The rules aren't settled. If the process changes every month, fix the process first. Automating chaos gives you faster chaos.
  • Every run needs real judgment. Some steps are the job itself, not the overhead around it.
  • It's about to disappear. If next year's system migration removes the task anyway, the payback has to beat that date.
  • The payback is longer than the tool's useful life. As a rule of thumb, if it takes more than three years to pay back, look for a cheaper option or leave it alone.

Why smaller tasks are now worth it

For a long time the break-even point sat high, because building was the expensive part. Writing a small tool, testing it on messy real files and documenting it for the team took days of developer time. Only big, painful tasks cleared the bar, and twenty-minute tasks stayed manual forever.

That's changing. With AI agents drafting code, generating test cases and hunting for edge cases in parallel, building and testing a small, well-scoped tool costs less than it used to. Lower the cost side of the equation, and tasks that never used to clear the bar start paying for themselves.

What doesn't get cheaper is the checking. The code still has to be tested and reviewed like any other code, with validation checks and a manual fallback, before anyone relies on it. That's where the effort goes now, and where it should go.

Measure it, don't promise it

Whatever you decide, write the baseline down before anything gets built: minutes per run, runs per month, people involved, errors per month. After go-live, measure the same things again. That turns "it feels faster" into a number you can show whoever approved the budget, and it tells you honestly whether the next automation is worth doing.

Want to run your own numbers? The savings estimator takes about a minute, runs in your browser and stores nothing. Not sure your task is a candidate? Start with the checklist.

About the author

Written by Ioannis Ouzounis, an independent AI and automation consultant. He coordinates teams of AI agents to explore, test and build solutions, and personally signs off on every result.

More about Ioannis

Keep reading

Is this costing your team?

If this post sounds like your week, put a number on it, then let's talk about removing that work.

I read and answer every message myself.