Sherlock Holmes’ thoughts on Starting with MDM

I like Sherlock Holmes.
No. I adore Sherlock Holmes, and must have read at least two or three times every story written by Sir Arthur Conan Doyle.

I recently had a discussion with an enterprise architect who was in the middle of a long Master Data Management program.
After listening to him explaining for one hour his impressive achievements, one thing struck my mind.

He had designed and executed his entire program without actually looking at the data. Everything was there, including the structure of the entities and the rules/policies around these entities.
But, as far as I understood, no detailed data analysis had been executed to get a comprehensive knowledge of the data. I was truly amazed and disconcerted.

After the presentation, I talked to him, and confirmed that he had clearly focused the program on the metadata while entirely ignoring the current data content. The effort and results of the project so far were great, but to me, something was utterly wrong in his approach.

Today, I thought of a statement Sir Arthur Conan Doyle put in the mouth of the most famous detective of all times in the “Scandal in Bohemia”:

“‘I have no data yet. It is a capital mistake to theorize before one has data.”

Maybe I (and Sherlock) are wrong, but I still believe that data management projects (and investigations) should start by looking at the data.

What about you?

11 replies
  1. Devashish Bharti
    Devashish Bharti says:

    Great blog in terms of capturing the critical requirement missed during the initial stages and using Mr. Holmes!

    But I guess it also depends on what stage the project is in and how easy it is to access data (public sector is a different ball game as they have to adhere to strict regulations).

    • FX Nicolas
      FX Nicolas says:

      Thanks for your feedback. You are right: Regulations may prevent entirely data access.

      In my case, the data was available. I believe that their initiative is more at a Master *Metadata* Management (MMM) stage, and will evolve to MDM, with some adjustments.

  2. Michael Christopher
    Michael Christopher says:

    Excellent observation, Watson. But your enterprise architect possessed some key information about the data, such as its specific domain. Metadata can provide essentially everything one needs to know, and establishing metadata is the first step in wrapping the mind around the data. Once that is established, whatever happens to the data will have minimal impact on design or operation. I work with specific types of healthcare data, using record-linkage systems to merge silos. I rarely need to look at data because I know the domain (e.g., healthcare providers, plans…), each of which have a particular structure. Still, I had to become familiar with those structures by looking at lots of data back in the beginning.

  3. Joe Palma
    Joe Palma says:

    Totally agree with Mr. Holmes. I am working on gettng 18 ERP’s together and only had 2 weeks to compiling the information…little mistake. We are now having detailed conversations with the 18 SME around the world to understand their data. Thanks for the great posting.

  4. Steve P
    Steve P says:

    Sherlock might also admonish you not to be dogmatic in your approach. I think the project has to dictate the relative importance of metadata and raw data. In general, I think you are right to assess the condition of the raw data, but as one of the previous posters pointed out, sometimes its a luxury or indulgence that you can’t afford.

  5. Tracy P
    Tracy P says:

    I believe I must have read your blog in a dream. We mapped out our first steps to Master Data Management yesterday and it boiled down to discovery. What data do we have, where is it, and what does it look like (schema and metadata)? As the EA in charge, I have been trying to get my head wrapped around this for some time but it makes logical sense to first figure out what you have. 🙂

  6. says:

    In my opinion, data profiling is an important aspect to be considered before beginning on the MDM journey. How good (or bad) is the data? What are the frequency distributions of various entities that you are planning to master? What is the relationship cardinality among various entities (1:1, 1:m, m:n)? Which entities/attributes have better data quality in which source system? And so on and so forth..
    Answers to some of these questions will definitely help avoid potential rework when MDM solution gets deployed in Production and the tyre hits the road!

    • trida
      trida says:

      If he theorizes about its business state(future and/or current), migration and enterprise transformations using models, and therefore uses a common meta model, then congratulations to him and good luck for data governance.

      But how to change yourself without knowing yourself?
      So agree with pankaj, operational, functional and technical aspect of MDM must be considered into the equation, specially if MDM is not only used to bring value but also to accelerate an enterprise based transformation.


Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *