Map the Workflow Before You Automate It

Almost every AI rollout I have been called in to fix failed at a step nobody had written down. Not a technical step. A judgment call somebody made every time without noticing they were making it.

Mapping sounds like the boring part. It is the part that decides whether the build works.

Why unmapped work breaks

A workflow in someone's head is smoother than the real one. They remember the path they take when nothing goes wrong.

The real version has branches: the exception they handle without thinking, the check they only do on Fridays, the one client whose file is formatted differently. Automate the remembered version and it works beautifully until the first Tuesday it does not.

The map, in one page

Sit with the person who actually does the work, not their manager, and write down four things.

The trigger. What starts this, honestly. If the answer is "when I remember", say so, because that is a finding.

The steps. Every one, including the ones that feel too obvious to write. Especially those.

The decisions. Anywhere a person chooses between two paths. This is the part that never makes it into a brief and the part that breaks builds.

The definition of done. What good looks like, specifically enough that two people would agree on whether it was met.

The four questions that decide readiness

• Does it happen often enough to be worth it? Weekly at minimum. Anything rarer, do it by hand.
• Is the input consistent? If every instance arrives in a different shape, you are automating the wrong end.
• Can you tell quickly if the output is wrong? If checking takes as long as doing it, you have not saved anything.
• What does a mistake cost? This decides whether a person stays in the loop, watches it, or steps out entirely, and it is a per-workflow decision rather than a company-wide one.

• The trigger

  What actually starts it, including "when someone remembers".

• The steps and the decisions

  Especially the judgements people make without noticing.

• Done, defined

  Specific enough that two people would agree it was met.

• The cost of being wrong

  Which sets how much oversight the workflow needs.

Why this is the cheapest hour you will spend

A map takes about an hour per workflow. A build based on a guess takes weeks and then gets quietly abandoned, which is the expensive outcome nobody puts in a case study.

You also find things worth knowing regardless. Half the workflows I map turn out not to need AI. They needed someone to notice that two teams were doing the same task twice.

That is the exercise working, not a failure of it.

Why do AI automation projects fail?

Usually because the workflow was never mapped, so the build was based on the version in someone's head rather than the real one, which has exceptions and judgment calls in it.

What should a workflow map include?

The real trigger, every step including the obvious ones, every point where a person makes a decision, and a specific definition of what finished looks like.

How do I know if a workflow is ready to automate?

It happens at least weekly, the input arrives in a consistent shape, you can check the output quickly, and you know what a mistake costs.

Who should you map a workflow with?

The person who does the work, not their manager. The manager describes the intended process; the person doing it knows the exceptions.