Don’t Create A Data Governance Hairball by John Schmidt.
From the post:
Are you in one of those organizations that wants one version of the truth so badly that you have five of them? If so, you’re not alone. How does this happen? The same way the integration hairball happened; point solutions developed without a master plan in a culture of management by exception (that is, address opportunities as exceptions and deal with them as quickly as possible without consideration for broader enterprise needs). Developing a master plan to avoid a data governance hairball is a better approach – but there is a right way and a wrong way to do it.
…
As you probably can guess, I think John does a great job describing the “data governance hairball,” but not quite such high marks on avoiding the data governance hairball.
Not that I prefer some solution over John’s suggestions but that data governance hairballs are an essential characteristic of shared human knowledge. Human knowledge, can for some semantic locality avoid the data governance hairball, but that is always an accidental property.
An “essential” property is a property a subject must have to be that subject. The semantic differences even within domains, to say nothing of between domains, make it clear that master data governance is only possible within a limited semantic locality. An “accidental” property is a property a subject may or may not have but it is still the same subject.
The essential vs. accidental property distinction is useful in data integration/governance. If we recognize unbounded human knowledge is always subject to the data governance hairball description, then we can begin to look for John’s right level of “granularity.” That is we can create an accidental property that within a particular corporate context that we govern some data quite closely, but other data we don’t attempt to govern at all.
The difference between data we govern and data we don’t, being what ROI can be derived from the data we govern?
If data has no ROI and doesn’t enable ROI from other data, why bother?
Are you governing data with no established ROI?