It needs a target and a floor

Most people write the target and forget the floor.

• The target is what would make this worth doing. "Median handling time under two minutes."
• The floor is what must not get worse while you achieve it. "Error rate no higher than the 4% baseline."

Without a floor, every automation succeeds — you can always make something faster by making it worse. The floor is what turns the target into a claim somebody should believe.

Add a third clause people forget: what share of volume it has to cover to count. Two minutes on 20% of instances is not the same result as two minutes on 80%.

From a wish to a condition

Three versions of the same intention, ending with one you could be wrong about.

Get it acknowledged in writing

Send it to the process owner and ask them to confirm. Not approval — confirmation that this is what you both think success means.

If they want to change it, that conversation now is worth ten times the same conversation in six weeks, and it frequently surfaces a goal you did not know they had.

Which of these can you fail?

Four proposed success conditions for the same project.

Publish the baseline package

Distribution, error rate with its definition, cost per instance, and the agreed success condition, all dated before any code exists. This is the artifact almost nobody has, and it is what makes everything you claim in six weeks believable.

The last moment before building

Everything from here is construction. This is the last point at which the project can be redirected cheaply.