Data quality: definition and framing
Definition
Data quality describes how well data fits a given purpose. This includes completeness, correctness, currency, consistency and traceability of origin. For AI, data quality is decisive because a model can only work as well as the data it uses. Faulty or incomplete data leads to faulty results, regardless of how capable the model is.
Why the Term Matters
AI projects rarely fail on the model and often on the data. A tool shown off spectacularly collapses in daily use when the underlying data is contradictory. For neutral consulting, data quality is therefore a touchstone before budget flows. Clarifying it is uncomfortable but honest, because it decides whether a project is viable at all. Skipping it means paying twice later.
When It Is Relevant
- An AI project should build on existing company data.
- Results must be reliable and repeatable.
- Data from several sources should be merged.
- An investment decision should rest on a realistic data foundation.
When Not
When a project does not rest on your own data at all, such as a pure text summary, data quality is secondary. And when it is already clear that a data clean-up is needed first, it need not be proven again but planned in. Then it is about sequence, not about whether quality matters.
Industries That Use It
- Retail and e-commerce: unify product data so recommendations and information are reliable.
- Manufacturing: check machine and sensor data for completeness before analysis builds on it.
- Healthcare: secure data origin and correctness because errors have immediate consequences.
- Finance: ensure consistency and traceability for auditable analysis.
Typical Use Cases
- Before an AI project, the data foundation is checked for gaps and contradictions.
- Data from several systems is unified before it is merged.
- Duplicates and outdated entries are spotted and cleaned.
- The origin of important data fields is documented to make results evidenced.
- A small sample is checked to estimate the clean-up effort.
Decision Guide
| Signal | Clarify data quality first | Alternative |
|---|---|---|
| Project builds on your own data | check the data foundation | building on an unchecked base is risky |
| Data from many sources | unify first | merging amplifies errors |
| Result must be evidenced | document origin | a result without origin is delicate |
| Project uses no own data | rather secondary | proceed directly |
Example
A retailer wants AI-supported product recommendations. We check on a sample how consistent and current the product data is and estimate the clean-up effort from it. Only once the foundation holds does the build begin. No model is set on bad data. This describes the method, not an asserted client metric.
Service Connection
We check the data foundation in the AI readiness check and frame it in the AI potential analysis; later use is guided by AI implementation.
Sources
This framing describes common review practice before AI projects. Standards for quality depend on the use case and should be set case by case.