Consistency problems may not look like errors. One system stores revenue in thousands and another in units. The sales team codes a closed deal as WON, the finance team as 4. Dates arrive as day-month in one export and month-day in another. Each source is valid on its own; the problem appears when they are combined.
Places to compare:
- schemas and field names for the same data
- code values for the same status or category
- units, currencies, and date and time formats, including time zones
- definitions that teams use differently, such as what counts as an active customer
Fixing consistency means mapping each representation to one rule. It does not mean merging differences that are real: two business units that define a churned account differently need both definitions kept and labelled, not one overwritten by the other.
How is data consistency different from data integrity?
- The broader term data integrity covers the accuracy, completeness, and consistency of data over its life cycle. As a readiness axis, Integrity asks whether values are valid and complete: missing values, invalid entries, duplicates, broken references.
- Data consistency asks whether the same meaning is expressed the same way across datasets, teams, and time.
- A table can pass every Integrity check and still be inconsistent with the table it is joined to.
Where does data consistency break in AI work?
A forecasting model is trained on two regions’ order histories. One region records quantities in cases, the other in units, and both columns are named qty. The merged table validates and the model trains, but its forecasts mix cases and units, so the two regions’ forecasts cannot be added or compared. A type or null check does not flag it, because every value is a valid number.
The check is a comparison across sources before they are combined: schemas, code lists, units, formats, and each owner’s definition of the field. In Syntitan, consistency is one of the six axes Ready for Raw Data diagnoses, alongside usability, integrity, context, reproducibility, and traceability. Data quality for AI explains how the six axes fit with a before-and-after test on the task.