Data quality can be a hard thing to grasp. Many data warehouse programs are branded as having bad data quality yet these statements can tend to be void of detail, leaving the data warehouse build team wondering how to improve the data quality.
You can't improve what you can't measure. So, we need a means for measuring the quality of our data warehouse. Abstracting quality into a set of agreed data rules and measuring the occurrences of quality violations provides the measurement.
This approach can also help management to understand the importance that the cleanliness of the data that feeds the data warehouse is to overall data quality. "Garbage in, garbage out", to a degree.
So, what can you do with data quality violations as they are picked up for movement into the data warehouse? There are three basic actions:
- "HELD OUT": Record(s) are held out of the main DW tables due to gross rule violation and placed into "holding" tables for manual inspection and action.
- "REPORTED": Data quality violation is reported on but data is loaded and will remain in the table.
- "CHANGE DATA": Transform data to a value in a master set of "good" values (i.e., Texus is changed to Texas).
One tip about changing data for the data warehouse -- bring the "bad" data into the data warehouse as well. Label it "source" data. This way you can trace back to operational data, which is something many
In the next few tips, we'll explore the types of data quality violations and the appropriate actions.
For more information, check out SearchCRM's Best Web Links on Data Quality.
This was first published in May 2002