There are some good points in the DM Review article on MDM by William McKnight. In particular, I like that he clearly highlights some issues that some parts of our industry are still in denial over:

(a) that customer and product data are just a small, though important, part of the master data picture for an enterprise.
(b) in reality, most companies will end up with more than one approach in MDM “style” or architecture, due to sheer complexity of the enterprise application environments out there. I feel that too often IT architects spend time agonising over the most “elegant” approach when in fact the tools are still evolving, and will no doubt go through more than one iteration of architecture over the next decade. Given the savings to be made out there by improved master data management, waiting for a perfect solution may be a costly and missed opportunity.
(c) data quality is a key part of an MDM project. Indeed, as my colleague Dave Waddington said the other day: “if you don’t have any data governance processes, what exactly is the point of doing a data quality project?”. One-off data quality initiatives may have value, but without the processes to fix the disease rather than treat the symptoms, the patient is never going to get truly better.

Although it might seem obvious that “The best place to manage master data is in the operational environment” as it says in the article, I am not sure that in practice this is always right. Some companies with a very centralised approach may have the discipline to really drive home a single master data system across the enterprise (I was speaking to one just the other day) but many do not. For de-centralised companies it may well make sense to consider alternative architectures, as it may simply be impractical or go against the grain of the corporate culture to have a central, harmonised set of operational master data.

Other than that caveat, it seemed like generally sage advice.