The Four-Criteria Use-Case Screen for the CCAO-F Exam

Why a screen comes before the prompt

Before you decide how to prompt Claude for a task, there is an earlier question the Claude Certified Associate - Foundations (CCAO-F) exam expects you to ask: should this task go to Claude at all? Deciding that is a structured evaluation, not a gut feeling. The same four Delegation criteria that help you map individual steps of a workflow also screen whole use cases, and running them every time is what turns a vague unease into a call you can defend to a colleague or a compliance reviewer.

The four questions are deliberately simple to remember because they are meant to be run often, in seconds, on routine decisions rather than reserved for obvious high-stakes moments. Reversibility, consequence of error, need for human creativity or empathy, and accountability. Each is a lens on a different kind of risk, and together they cover the ground that matters before any data or instruction reaches the model.

Reversibility and consequence of error

The first two criteria are about what happens when the output is wrong, because the interesting question is never whether Claude can be wrong but what it costs when it is.

Reversibility asks whether a wrong output can be caught and undone before it causes harm. A draft that a person reads before anything happens is highly reversible: the mistake is caught in review and corrected at no cost. An action that fires automatically and cannot be recalled, such as a payment sent or a final decision recorded, is not. Irreversible consequences raise the bar sharply, because there is no second chance to catch the error.

Consequence of error asks a related but distinct question: how costly is a wrong output? A typo in an internal brainstorm costs almost nothing; a wrong figure in a filing that a decision depends on costs a great deal. Higher consequence demands more human control, or rules the use case out entirely. The two criteria pull in the same direction but are not the same: a task can be reversible yet high-consequence, or low-consequence yet irreversible, and you want to notice both.

The human element and accountability

The last two criteria are about people rather than error costs, and they are the ones candidates most often skip.

The need for human creativity or empathy asks whether the task requires judgment, relationship, or care that AI cannot supply. Some work should stay with a person regardless of how capable the model is: a message of condolence, a sensitive personnel conversation, a piece of work whose value is precisely that a human did it. Capability is not the test here. The test is whether the task is the kind of thing a person should own.

Accountability asks who remains answerable for the outcome, and whether that accountability can actually be exercised over an AI-produced result. This is the criterion that cannot be waved away, because responsibility for an outcome cannot be transferred to a model. Someone in the organization stays on the hook for the result whether or not Claude produced it. If no human can meaningfully own and stand behind the output, that is a signal on its own, independent of how cheap or reversible the task looks.

What the exam trips candidates on

Two traps recur on this knowledge point, and both come from misusing the four criteria.

The first is treating the criteria as a simple pass/fail checklist rather than factors that interact. The exam is explicit that the four do not work like a checklist where any single failure ends the discussion. They are weighed together. A task can trip one criterion and still be fine with the right arrangement, or clear several and still be wrong because of the one that matters most. Reading the four mechanically, tallying passes and fails, misses the point of the screen.

The second is assuming a task is automatically fine for AI because it is low-cost or fast, while ignoring who owns the outcome. Speed and cost are real benefits, but they do not answer the accountability question. A cheap, quick task with no clear owner for its result has not passed the screen just because it is efficient. The credited reasoning always keeps accountability in view even when the task looks trivially safe on the other three criteria.

Common misreadings to avoid

How this shows up on the exam

Domain 6 questions on this knowledge point give you a proposed task and ask which considerations decide whether it belongs with AI. The reliable move is to name the four criteria - reversibility, consequence of error, human creativity or empathy, and accountability - and treat them as interacting factors rather than a checklist. Watch especially for the low-cost distractor that quietly ignores who owns the result.

This is the foundation the rest of Task 6.1 builds on. Once you can run the screen, you sort the result into three use-case classifications, learn to identify the load-bearing criterion when the four point in different directions, and, for the middle classification, specify a real human review gate. The four questions are where every one of those later judgments starts.