What is automation anyway?

The word has been stretched far enough to mean almost nothing. A recurring calendar invite is automation. So is a mail merge, a scheduled report, a hundred-node flow with retry logic, and a phone tree. If a single word covers all of that, it is not describing a technology — it is describing a decision.

Here is the decision, stated plainly: automation is choosing which parts of a job do not need a person’s attention, and arranging for those parts to happen anyway.

Everything useful follows from taking that definition seriously.

The test: would a person add anything?

For any step in a process, ask what a human contributes. Sometimes the answer is obvious — copying numbers from one screen to another adds nothing, and a person doing it will be slower and more error-prone than a script.

Sometimes it is subtle. Deciding whether a reply sounds defensive looks mechanical from the outside and is not. Judging whether a post is on-brand is the same. These steps look automatable precisely because the person doing them is good at them.

The steps worth automating share a property: the correct output is fully determined by the input. If two competent people would produce different results from the same information, you are looking at judgement, and judgement automated is judgement discarded.

Three honest costs

Automation is usually presented as pure saving. It has three costs that show up later.

It has to be maintained. Every flow is code, whether or not it looks like code. It breaks when an upstream format changes, it needs updating when the process changes, and it needs someone who understands it.

It hides the process. A manual step is visible; anyone watching can see it happen. An automated step is invisible until it fails, and a process nobody can see is a process nobody can improve.

It calcifies decisions. The rules you encode become the rules, including the arbitrary ones you meant to revisit. Six months later somebody asks why posts never go out on Thursdays and nobody remembers.

None of these are reasons not to automate. They are reasons to automate deliberately, and to keep a list of what you have automated somewhere a new colleague can read.

Start where the work is repetitive and reversible

The best first automations share two traits: they happen often, and a mistake is cheap to undo.

Collecting numbers into a report is a perfect candidate — it happens weekly and a wrong number can be corrected. Sending an apology to a customer is a terrible candidate: it happens rarely and cannot be unsent.

Ordering your backlog by frequency times reversibility will get you most of the way to a sensible plan, and it will keep you away from the impressive-looking flows that cause incidents.

The part that stays human

In marketing specifically, the durable split has been the same for years. Machines are good at collection, formatting, timing, routing and summarising. People are good at deciding what to say, judging whether it is true, and answering the person who replied.

Teams that get this right feel less automated, not more. The tooling disappears into the background, and what is left is a small group of people doing the parts of the job that are actually interesting.

That is the whole goal. Not fewer humans — fewer humans doing things a loop could have done.