EU AI Act Orientation

Name the outcome and its boundary

Organise the use description, participants and evidence before selecting a control checklist. The same product can raise different questions in different deployments.

Use this guide to prepare the question. Legal applicability and obligations require evaluation of the actual facts by appropriately qualified advisers.

Compare different causal paths

Compare process redesign, changes to an existing system, bounded rules, AI assistance and automation. Each option needs a reason it could change this piece of work, and assumptions that can be tested.

Keep the workload, acceptable quality, human review, integration and ongoing costs visible. A modelled capacity gain becomes financial value only under explicit conversion conditions.

Design a test that can change the decision

• Describe the system, organisational role, intended purpose and relevant deployment context.
• Identify missing documents and responsible reviewers without declaring a compliance outcome.

Agree the metric, baseline, threshold, observation source and person responsible for the next decision. A result may support a bounded pilot, more evidence, redesign or stopping.

Keep a testable decision record

Metric and baseline

Describe the system, organisational role, intended purpose and relevant deployment context.

Threshold and failure signal

Agree what would change the decision before observing results. Include unacceptable quality, rights or operating cost, and the evidence that would disprove the hypothesis.

Measurement and responsibility

Name the observation source, period, reviewer and person who accepts the outcome. Preserve assumptions separately from observed results.

Next action and review

Decide whether to test, redesign, expand or stop. A further operating step requires its own permissions and evidence; revisit it when the context changes.

Continue with this question

Explore the calculator · Research method · All research guides