Three hours a week, back: the boring math

People ask me whether automating something is worth it. The answer is arithmetic, and it takes about two minutes to do yourself before you talk to anyone.

Here is the whole calculation.

The four numbers

How long does it take, per go. Be honest and include the setup. Opening the file, finding the reference, checking it, switching windows. If you time it, it is usually longer than you guessed.

How often. Per day, per week, per month.

What an hour of that person costs you. Not their hourly wage. Their loaded cost, which is wage plus tax plus everything else, and for most small businesses that lands somewhere between $25 and $45 an hour for admin work.

What it costs to fix. For a single workflow, usually $2,500 upward.

Multiply it out

Say the job takes 20 minutes and happens nine times a week. That is three hours a week. At $30 an hour loaded, that is $90 a week, or about $4,700 a year.

Against a $2,500 build, you are square in six or seven months, and from then on the work simply stops costing anything.

That is the whole model. No spreadsheet, no discounted cash flow, no consultant.

The number people get wrong

The one everybody underestimates is frequency, and the one everybody overestimates is duration.

A five minute job done twelve times a day is an hour a day. That is five hours a week, more than 250 hours a year, and it never feels like a big job because each individual go is trivial. Meanwhile the two hour monthly report feels enormous and is worth 24 hours a year.

Frequency beats duration nearly every time. If you are choosing what to fix first, count how often something happens before you count how long it takes.

What the arithmetic does not capture

Three things, all of which push in the same direction:

Errors. A person retyping numbers gets one wrong occasionally. The cost of that is not the 20 minutes, it is the invoice sent for the wrong amount and the conversation that follows.

The work that does not happen. Quotes that never get chased. Reports that stop when someone is on leave. This is usually worth more than the hours, and it is the reason follow-up automation tends to pay back faster than anything else: its return is closed work rather than saved time.

Where the hours go. Saved time only counts if it becomes something. If three recovered hours a week turn into three hours of the owner doing sales instead of data entry, the return is much larger than $90. If they turn into three more hours of scrolling, you have bought nothing.

When the answer is no

Run this honestly and it will sometimes tell you not to bother.

If the job takes ten minutes a month, leave it alone. If it is going to change completely next quarter, wait. If it only happens because a process is wrong somewhere upstream, fix that instead, because automating a broken process just makes it produce wrong answers faster.

I would rather tell you that in a free half hour than build you something that saves nine minutes a month.

The version to hold in your head

Hours per week, times fifty, times what the hour costs. Compare it to the build. If payback is under a year, it is almost always worth doing. Under six months, do it now.

Anything more precise than that is false precision anyway.