GLOSSARY
Validation
Validation uses appropriate evidence to determine whether a system meets the requirements of its intended use. The standard depends on the task, user group and potential consequences of error. Passing a technical functionality test therefore answers only part of the question.
From a convincing example to a dependable workflow
An AI system may produce a strong negotiation briefing in a demonstration. Everyday work includes incomplete customer information, conflicting interests and different contract versions. Assessment should reflect those conditions. Relevant criteria include preserved qualifications, identified knowledge gaps, suitable options and correction effort.
Criteria agreed in advance reduce the risk of selecting only favourable results afterwards. Previously unused cases and assessment by knowledgeable people outside the immediate development team strengthen the evidence. A credible baseline shows whether the additional effort is justified compared with the existing process.
What a pilot can establish
A small pilot provides evidence within the scope it examined. It does not establish unrestricted suitability for other industries, languages or tasks. Changes to the model, data or workflow may require renewed assessment. Continuing evaluation should reveal deviations and define when to improve, restrict or roll back the system. Approval and remaining limitations belong together in a traceable decision.