2026-08-31 · Wallion

When to automate a process (and when not to)

One stable step in a line, versus a flow knotted with exceptions.

Automation is not "putting intelligence" on top of a mess. It is removing a step you already know how to do, that repeats, and that shows when it fails.

If the step changes every week, automation freezes the error. If nobody looks at the result, it freezes the error in silence.

Automate when

The step fits in one line. "When the customer accepts, send the order to the warehouse." Not: "when it feels right, depending on the customer, if there is stock, if María is in."

The data already lives somewhere. Not in someone's head. Not in a chat. If the trigger is "I remember", there is nothing to connect.

The failure is visible the same day. An order that does not leave. An email that does not go. If the failure shows up at month-end, make it visible first. Then remove the click.

One person does it more than ten times a week. Less than that, the click is cheaper than the bridge. You do not need an industry statistic. Count yours.

Do not automate when

The process is not decided. You are using automation to avoid arguing the rule. That does not save time. It hides the argument.

The exceptions are the real work. If half the cases leave the line, it is not a flow. It is judgement. Judgement does not chain.

Nobody will watch the queue. Automation with no owner is an error that copies itself.

You want it to "think". That is not this article. The job here is a repeated step, not a judgement.

What a first step looks like

One trigger. One action. One place you can see whether it happened. Nothing else.

Examples that fit: an accepted quote creates a row. A closed job pings whoever invoices. A record created in one system appears in the other.

Examples that do not: "prioritise the customers". "Write the quote the way Juan would." That needs a person, or a tool built for your rule — and then you are not automating a step, you are building.

How this relates to integrate

If A and B already do the work and only the trip is missing, integrate. If the step exists nowhere, there is nothing to automate: write it or build it.

Do not link a fashionable product. You do not need a new URL. The test is the same with the cable you have or one you still have to make.

After

If the step is one, the line is written, and someone owns it, it can be commissioned. The commission path is custom software. This page only says whether you need to remove a click or you need to think.

Thirty minutes. No pitch, just a diagnosis.

Tell us what's breaking. We'll tell you if custom software fixes it — and exactly what it would cost.

Book a 30-minute call