What is Master Data?

What is Master Data?

In this podcast, an interesting categorization of data is given by Malcom Chisholm.

In brief, there are six types of data:

1. Meta Data (e.g. tables names…)
2. Reference Data (e.g. code tables…)
3. Enterprise Structure Data (hierarchical enterprise organization of data, e.g. product line…)
4. Transaction Structure Data (e.g. customer, product data…)
5. Transaction Activity Data (transactional event data, e.g. orders…)
6. Transaction Audit Data (states changes in transactional event data, e.g. logs…)

Master data are then the aggregation of reference data, enterprise structure data and transaction structure data. This means that there is not only one category of master data. Thus the meaning of the different types of master data must be taken into account in any good MDM tool.

Necessary metadata

In this note, Bill Inmon complains about the endless task of documenting all possible Metadata. Let’s think a bit about this.

If we take the following definition for metadata : data about data, it is clear that documenting all metadata is impossible and endless. Because metadata is also data, metadata can be information about data but can also be information about metadata. And hence the endless loop of documenting the data used for documenting the data…

But, all metadata do not change that much. As an example, see the CWM specification which describes already most of the metadata needed in the data management domain.
Of course, CWM is not exhaustive and cannot be. But maybe, CWM could play the role of the “necessary metadata” searched by Bill Inmon, at least in the domain of data management.

Then data like averages, maximum and all computed data are not really metadata as I understand them. These data are not really a description of data, they are rather data computed from the data. They depend on instances and not on data classes. They are other informations (complementary information) about the data. And I don’t think they should be called metadata. Otherwise everything is a metadata since everything is a data about some other data. Then we could ask ourselves “what is a data?”

Finally, it seems to be obvious that documenting all metadata is impossible. It’s like writing the perfect program: even for a simple program that takes an input and writes it to the output, there is the possibility to write pages of code for handling all the use cases we can think at. Even then, the program will not be able to handle some unexpected cases.