About three months ago, I started a data management and ETL project for a pretty big bank. Time was of the essence and the bank brought me in because I can get results. In this post, I explain why an overemphasis on results can be a really bad thing–and why all matching isn’t created equal.When my client advised me of the number of disparate extracts of financial data I would receive, I quickly looked for potential commonalities. As most would do on this type of project, I started with the obvious: GL account. While not unique in most systems I have seen, they are often at least part of a multi-field index. In other words, I can almost always sum debits and credits by GL account 1000, for example, especially when I bring in a field like company.
Dealing with Suboptimal Data and Other Limitations
Unfortunately, a few of the extracts only contained account descriptions. When I explained this limitation to my boss, we had the following exchange:
Boss: Well, why not join on description instead of account number?
Me: It’s possible, but I never recommend it. In many systems, these descriptions are not standardized. One missing character, misplaced space, or mutant letter will result in missing data downstream and other problems that are worth trying to address from the beginning.
Boss: We have no other choice. IT won’t change the extract from XYZ financial system. Gotta make do….
Me (in Tweety Bird voice): OK…but I got a baaaaaad feeling about this.
Fast forward to the end of the project. Our reports were off–way off. What’s more, this wasn’t a fundamental design issue. So far off that I had to rejigger a bunch of previously working queries and validation routines. This took a few days and caused other problems. We had to break things that previously worked.
Sometimes we don’t have a choice in life and in golf. We hit the ball into the woods, can’t find it, have to take a penalty stroke, and walk away with a snowman on a par 3. It happens.
Although my client wasn’t happy that we had so many problems, he fully understood what I told him on day one: “Try not to join on descriptions.” Of course, the bell didn’t ring quite as loud when I initially said that. After he saw the results first-hand, that bell was pretty loud–and wouldn’t stop.
Understand that there are limitations with certain types of data matching, as many on this blog has pointed. Some types are much better than others. Some are too restrictive; others are far too liberal. Think about it.
As one of my friends put so nicely, you don’t go to pick up your kid at nursery school and just grab any eight-year old brunette girl. You probably want you own kid.
Data’s the same way.
What say you?
Read more at MIKE2.0: The Open Source Standard for Information Management
![Why don’t they [IT] ever learn?](https://www.smartdatacollective.com/wp-content/uploads/2010/04/don-t-ever-learn-330x220.jpg)
