Given a mountain of data that comes from different sources in different flavors, it could be a daunting prospect to get them all in line. Here are seven simple steps that could help you wade through the first phase of implementing a successful BI process that will not only solve the immediate problems, but also grow with your company.
- Trace your steps backwards, begin with the reports that you have to have in order to run your business. Do not start with the data itself. A lot of BI implementations make this mistake, which will soon overwhelms anyone involved and contributes to a slow-start.
- Identify two types of reports: Those that are very important for strategic decision making (annual bank loan reports, ISO certifications, etc.) and those that are important for tactical everyday decisions (daily status report, daily manufacturing job schedule, warehouse inventory, customer returns, etc.)
- Split your attention into the two types of report you identify above. Pick one report from each type, and work on both at the same time. By working on two different sets of data types simultaneously, you will gain more insight on how to transform the existing data to satisfy both types of reports. Failing to do this, will cause you to backtrack a lot, and that means wasted effort.
- Use of your Time Dimension table. This is the lookup table consisting of different facts about one measure of time (usually a day). In a lot of BI projects, people neglected to use this very useful table, opting to create complex SQL queries, which ends up being error prone, difficult to maintain, and a big time waster.
- Be mindful of your database’s idiosyncrasies. Know what NULL values mean in a field, especially the ones that will be part of a calculation. Know what it means to say BETWEEN two dates. And no, Non-SQL databases are not immune to these quirks. They are just as susceptible to unfamiliar and untested usages.
- Prepare two identical databases that contain your BI tables. Setup two instances of your ETL building each database separately. This way you can switch back and forth when you are testing or having to re-run the ETL in the middle of a busy day.
- You will not catch every bug, and there is no perfect the BI system. Test, and re-check your tests, that’s why you have two identical databases. BI is an ongoing process, therefore it is important to come up with a good system and an implementer who understands this.
By following these simple steps and precaution, you will minimize a lot of backtracking in the middle of the project. And more importantly, by starting the project the right way, you will avoid burn out and frustration that will have more adverse impact on the quality of the data at the end of the process.