Volume, variance, consequence

Three properties decide whether a process is a good candidate, and you can assess all three over a coffee.

• Volume. How often does it run? Twice a week is a hobby project. Forty times a day is a business case.
• Variance. How similar is each instance to the last? Low variance is automatable. High variance is a research project wearing an automation costume.
• Consequence. What happens when it is wrong? This is not a reason to avoid a process — it is the thing that sets how much checking you have to build, and therefore how long the project takes.

High volume, low variance, low consequence is the beginner's sweet spot. High volume, low variance, high consequence is where the real money is and where you need the shadow period this course insists on.

The three that punish

The process nobody owns. If you cannot name the person who decides how it should work, you cannot get a success condition agreed, and at the end there will be nobody to hand it to. You will build it and it will quietly stop being used.

The process that is about to change anyway. Automating something three weeks before the department reorganises means building a very good solution to a problem that no longer exists.

The process that is really a symptom. People spend four hours a week reconciling two systems because the systems disagree. Automating the reconciliation makes the disagreement permanent and cheap to live with, which is the opposite of fixing it.

The third one is the hardest to see, because from inside it looks exactly like a great candidate: high volume, low variance, universally hated.

Score three candidates, not one

Write down three real processes and score each on volume, variance and consequence. Three, because scoring one in isolation always produces "yes" — you have no comparison and you already want to build it.

For each, also write the name of the person who owns it. If you cannot, write "nobody", and note what that means.

Which of these would you take?

Four candidates from a real business. All four are genuinely annoying to the people doing them.

Monthly volume of your chosen process

Instances per month, taken from a system of record rather than from anyone's estimate. Note where the number came from.

If you cannot find a system that knows, that is itself a finding: a process nobody counts is a process nobody can prove you improved.

Publish the selection, including the rejects

All three candidates, their scores, which you chose and — the part that matters — why you rejected the other two. A selection document with only the winner in it is a decision nobody can audit.

Before you get attached

You are about to spend six weeks on this. Ten minutes of doubt now is cheap.