In 30 seconds

A decision is any point where what happens next depends on something. Write each one as a question with observable inputs and clear paths. This is where you learn what needs a rule and what needs judgment.

Why this comes up

Decisions hide inside innocent-looking steps. "Check the invoice" is really several decisions: does the supplier match, does the amount match, is it a duplicate. If you map it as one box, you hide the work that matters most for automation.

Write a decision as a question

Replace the action label with the question the person is actually answering.

Instead of "validate the request," write:

A good decision question has observable inputs (you can point at what it looks at) and named outcomes (yes, no, and often unclear).

Draw every path, including "unclear"

Most decisions have more than two outcomes. There is yes, no, and the case where the answer is not clear from the available information.

The "unclear" path is often the most important one. It is where judgment or escalation lives, and it is exactly the part automation gets wrong if you ignore it.

Mark each decision: rule or judgment

For each decision, ask whether it can be stated as a stable rule.

  • If yes, it is likely a business rule suited to deterministic automation.
  • If no, it involves interpretation and may be a place where AI helps, with review.

Marking this now saves confusion later. It tells you which decisions need a model and which do not. See deterministic vs probabilistic systems.

A small example

"Approve the discount" turns out to be:

  • Within limit: approve automatically.
  • Above limit: route to the sales manager.
  • Customer type unclear: check the account record first.

Three paths, one clear rule, and one small sub-decision. Now it can be designed properly.

What this does not solve

Mapping a decision does not tell you how often each path is taken. That frequency matters when you assess AI fit, so gather it from real cases when you can.

Good enough to move on when

  • Each decision is written as a question with observable inputs.
  • Every decision shows its real outcomes, including unclear.
  • Each decision is marked as rule or judgment.