Transaction Versus Judgment

Work where the output is the point, and any qualified party would produce the same result, separates from the person and does not come back. Work where whose call it is forms part of what is being bought does not separate.

The same sort has run twice before, on time and then on place. Activity left the schedule, then it left the room, and now it is leaving the person. The mechanism has held each time. The expensive errors are misclassifications, usually caused by sorting a whole category by its most visible activity rather than its most common one.

Where it came from

A regional property manager produces a monthly variance report and, later that quarter, decides which of three capital projects gets funded. Same expertise, same data, same person, both arriving as documents. Sorted by difficulty or by job title they land in the same column. They do not belong there. Any two competent people working the same ledger produce the same variance report. A year from now somebody will ask why that project and not the other one, and the answer has her name on it.

When to run this

Before an AI roadmap is built, and before anyone commits to what a system will take over. After the roadmap exists, the classification is buried inside decisions that have already been made.

The steps

List the things you are planning to hand to an AI system this year. Not the categories. The actual tasks.
Split anything that is a category rather than a task. Customer support is not a task. Financial reporting is not a task. If an item is the name of a function it contains both columns.
Put each item in a column. Transaction, where the output is the point and any qualified party produces the same result. Or judgment, where whose call it is forms part of what is being bought.
Apply the accountability test to the transaction column. For each item, ask what happens when it is wrong.
Apply the visibility test to the judgment column. For each item, ask whether you sorted it that way because of its most visible activity or its most common one.

What a bad answer looks like

A judgment column that is too long. Everyone believes their own work is judgment, and a framework that lets people declare themselves unautomatable will be used that way in every room where it is run. The check is not introspection. Ask whether a customer would notice or care if somebody else did it.

Items that will not sort at all. Usually that means the classification depends on somebody else's rules rather than on the nature of the work. A licensing body, a regulator, or a liability regime may require a named human regardless of what the columns say. That is a different conversation with different people in the room.

What to do with the output

Run it with the people who do the work rather than the people who approve it. The variance report and the capital decision look identical from two levels up.

Then set a review date. Substitute quality moves. An item correctly filed as judgment today can become transactional in eighteen months without your classification having been wrong when you made it.