Data Warehouse or Data Mart?
So, here is the age old question. Should I go the Kimball route (data mart first) or the Inmon route (Enterprise Warehouse first)? While both have merits and demerits (we are not going to get into that discussion here as it has been done before). I think, a "functional" approach should be taken to maximize value. Let me explain. If you look at a physical warehouse, let's say, the warehouse of a shipping company, you don't see one warehouse for canned goods and another for toys and another one for perishables. You may see sections carved out for canned goods, toys and maybe a refrigerated section for perishables. But rarely do you see three separate buildings for each type of item. So why should it be different for a Data Warehouse? Basically, if you look at the data warehousing concept, we will see data marts that are "specific" to a line of business and a warehouse who is the closest thing to a physical warehouse, but often times so scattered and/or so complex that the LOB (Line of Business, if you didn't already know that) seldom finds use for it. So are we stuck? Not really!
A Flexible, Extendable model.
What if you can have the best of both worlds? What if you could rapidly deploy a warehouse, knowing that there is going to be more LOBs added later?
What if, we could deploy this warehouse, knowing that tomorrow, "perishables" may be added to the warehouse and the model has to be adept enough to add "refrigeration"?
What if, you could add more information at different levels of granularity to the same model?
And...the kicker....what if, we could correlate information across the enterprise, with the same model and don't have to create LOB specific data marts?
"Hogwash, won't happen, what did you smoke this morning?" you say?
"Functional" Approach
So, if you are working for an organization whose primary business is NOT building software, then, you, my IT colleague, is there to support the "business user". I have yet to see a business user who has come to me and said "Boy, I wish I had a data warehouse". All of them, however, have asked me one thing. "I wish I had good quality information. And I need it today, if possible". So why not give them just that? Rapid Deployment, Ability to add more information later. Can you design such a model? The answer is a resounding YES!
Has it been done before?
Do I have to repeat myself? I did say YES, didn't I? And I added the word "resounding" to give it a dramatic emphasis! Yes we do have such a model. Unfortunately, I can't share the specifics of it just yet without an NDA or holding your first born hostage. Our FlexDimensional model does just that. It allows us to do a few things:
1. Rapidly deploy with limited up front information (business process, granularity, fact etc)
2. Add more information without having to build new star, snowflake, or whatever else schema
3. Correlate enterprise wide information in a single, extendable model.
So, there you have it. There is always a better way of delivering "good quality information" at the point of decision making!
Subscribe to:
Post Comments (Atom)
Great point Kishore! This age old question (DM vs. DW or even EXCEL for that matter), although is a relevant question (for the IT audience), in most cases, it is not being framed up correctly, when the audience happens to be business users type decision makers. Rightfully so, you have reframed the question with a "value driven information insight" mindset.
ReplyDeleteIf I may, I would perhaps augment your re-framing with few more questions (3 strategy questions and 2 structure questions, as structure must follow/align with strategy in the words of Chandler, when we do these types of approach assessments)
1. What is the problem the company is trying to solve with the analytics platform? i.e. Is it used for top line focused growth strategy efforts or bottom line focused operational improvement or both?
2. What is the decision making style of the company? Fact based quantitative culture or intuitive qualitative culture or both? If the answer is both, what is the weight factor emphasis on both of these approaches?
3. Does the company have a need for real time analytics driven insights from structured and unstructured data, within the context of this emerging big data phenomenon?
4. What is the organization structure look like? (i.e. function driven vs. process driven)
5. Which industry structure, the company is part of? (i.e. mature, emerging, declining etc - mainly to assess the impact from regulation and/or to decide on custom analytics vs. COTS or to make build vs. buy decision etc). For example, to your point about healthcare, the whole healthcare solution landscape is changing (depending upon how we frame the debate as you have done in this blog), as outlined, in one of the hack we had proposed at the MIX site (http://www.managementexchange.com/hack/healthcare-reform-%E2%80%93-purpose-innovation%C2%A9-way%E2%80%A6).
I am sure, if we go one level deeper, we can come up with few more granular questions as well. Nevertheless, answering these analytics approach questions along with your expert guidelines, is the way to go, to arrive at the right type of analytics platform i.e. Functional, Process or situational analytics platforms.
On a related note – one of my big data articles within my blog might be of interest to you as well
http://strategywithapurpose.blogspot.com/2012/01/big-data-driven-real-time-valuation.html
On the whole, a great article!
Regards,
Charles