"Good data" means data that is fit for a particular purpose. Six dimensions help assess it.
Accuracy
Does the data correctly describe reality? Wrong addresses, mistyped amounts and incorrect labels reduce accuracy. Check against trusted sources or samples verified by hand.
Completeness
Are required values present? Measure missing rates per column, and check whether whole categories of records are missing — for example, a region that wasn't captured.
Consistency
Do values agree across systems and over time? The same customer with different birth dates in two systems, or a metric whose definition changed mid-year, are consistency problems.
Timeliness
Is the data up to date enough for the decision? Yesterday's inventory may be fine for reporting but not for real-time ordering.
Validity
Do values follow the rules and formats expected? Dates that exist, codes from the allowed list, ages within a plausible range.
Uniqueness
Are records duplicated? Duplicates inflate counts and, in machine learning, can leak between training and test sets.
Putting It Into Practice
- Define quality rules for important fields.
- Measure them automatically in data pipelines.
- Track quality over time with dashboards and alerts.
- Assign owners who fix problems at the source rather than patching downstream.
Fitness for Purpose
The same dataset can be good enough for trend analysis and not good enough for billing. Judge quality against the use.