Fitness for purpose
Data is 'fit for purpose' when it is good enough for the specific job you need it to do, quality is judged against the task, not in the abstract.
The Simple Version
Data is 'fit for purpose' when it is good enough for the specific job you need it to do, quality is judged against the task, not in the abstract.
Detailed Explanation
Fitness for purpose is the overarching data quality concept in ISO/IEC 5259 and DAMA frameworks: quality is inherently contextual, and the same dataset may be fit for one AI application but not another. For example, a transaction dataset aggregated to monthly totals may be fit for annual forecasting but unfit for real-time fraud detection. Assessing fitness requires explicit documentation of intended use, operating conditions, user population, and performance requirements, then evaluating whether the dataset meets those requirements across relevant quality dimensions. This concept underpins the EU AI Act's expectation that training data must be 'relevant, representative, free of errors and complete' for its intended purpose.
Key Characteristics
- Context-dependent, fitness is always assessed relative to a specified use
- Requires explicit statement of intended use before quality can be assessed
- Applies across all data quality dimensions, weighted by their importance to the use case
- Central concept in ISO/IEC 5259 data quality for AI
Why It Matters
Data acquisition teams should document the intended AI use case before procuring or preparing datasets, enabling fitness-for-purpose assessment that prevents costly late-stage data rework.
Real-World Analogy
A hammer is fit for driving nails but unfit for tightening screws, the same tool may be excellent quality yet inappropriate for a particular task. Data quality works the same way.
Common Misconceptions
- High-quality data is automatically fit for any purpose, fitness depends on the match between data characteristics and use-case requirements.
- Fitness for purpose is a vague concept, it can and should be operationalised through specific quality dimension requirements and measurable thresholds.