When not to build a data warehouse
Clients often come to us having already decided they need a data warehouse. Sometimes they are right. Often, though, the warehouse is a solution to a problem that has not yet been clearly stated.
Start with the decisions
Before discussing architecture we ask a simple question: which recurring decisions are currently made with poor information? If the answer is a handful of monthly reports, a well-maintained set of queries against the operational database may be enough for the next two years.
Signs you can wait
- Most of your reporting comes from one or two systems.
- Nobody on staff would own the warehouse after launch.
- Metric definitions are still being debated.
Signs you should not wait
- Analysts spend more time joining exports than analyzing them.
- Reports regularly disagree because they pull from different systems.
- Operational databases slow down when reports run.
Either way, agreeing on definitions comes first. A warehouse built on contested definitions simply makes the disagreement faster.