Before you automate a process, make sure the process works

Somebody on your team is going to bring you a tool this quarter. It will demo beautifully. It will promise to take a task that eats four hours a week and reduce it to zero. And the honest answer, most of the time, is: not yet.

Not because the tool is bad. Because automation is a multiplier, and multipliers work in both directions. Point one at a clean process and you get leverage. Point one at a broken process and you get broken faster, at scale, with a subscription fee attached and one more system nobody fully understands.

The tell

Here is the fastest diagnostic I know. Ask three people who do the task to describe it, separately, start to finish.

If you get three different answers, you do not have a process. You have a habit that three people have each independently invented, plus a shared belief that it is the same habit. Automating that means choosing one person’s version and quietly breaking the other two — usually without telling them, and usually discovering it six weeks later when something downstream goes wrong.

A process you cannot describe is a process you cannot automate. You can only automate your guess about it.

What to do instead, in about a week

• Write the process down as it actually happens. Not as it should happen. Not as the policy says. As it happens, including the workaround where someone re-keys data into a spreadsheet because the two systems don’t talk. That workaround is information.
• Count the steps that exist only to fix an earlier step. These are pure waste and they are usually the majority of the pain. Removing them often solves the problem without buying anything.
• Find the decision points. Anywhere a human judges something — is this urgent, does this need review, is this the right code — is a place automation will either need a real rule or need to hand it back to a person. Both are fine. Pretending the judgment isn’t there is not.
• Run the cleaned-up version manually for two weeks. If the manual version doesn’t hold, the automated version won’t either. This step gets skipped constantly and it is the one that saves you.

Only then go shopping. And when you do, you’ll find something useful: you now know exactly what you need, which makes the demo call about twenty minutes long instead of an hour, and makes it very hard for anyone to sell you something adjacent to your problem.

The part nobody says out loud

Sometimes you go through this and discover the process shouldn’t exist at all. The report nobody reads. The approval step added after one incident in 2019. The double entry that exists because two departments each wanted their own record.

That is the best possible outcome and it costs nothing. It is also the outcome you can only reach by looking first — which is precisely what buying the tool lets you skip.

The order matters more than the tool. Explore what’s actually happening, examine the evidence, choose the most useful direction, then build. Reverse those and you’ll spend the next year maintaining something you never needed.