Data Quality

Data quality refers to how well a dataset is fit for its intended use. It spans five dimensions: accuracy (are the values correct?), completeness (are records and fields present?), consistency (do the same things mean the same thing across systems?), timeliness (is the data current enough?), and relevance (does it represent the population and conditions the model will operate in?). Data can be high-quality in one dimension and completely unfit in another — a complete dataset full of inaccurate values is still bad training data.

Model quality is bounded by data quality. A sophisticated model trained on poor data will produce unreliable outputs; a simpler model trained on clean, representative data will consistently outperform it. This matters for resource allocation: investing in model complexity before fixing the underlying data is an expensive way to get the same bad answers faster. The most common AI project failure mode isn't a bad model — it's a bad dataset that nobody audited before the project started.