Now that SAP Predictive Analysis works directly on SAP BW, every statistician’s heart will beat 2 standard deviations faster: all of a sudden, it is that much easier to find facts in the big data warehouse! Before you start a pilot project or a regular implementation, you need to face the following four points to ensure its success:
1. Yes, this is statistics!
However you look at it, you are bound to have to deal with statistics – something that for many people is unsexy, difficult to relate to and abstract. On the surface, it probably looks harmless since you have such a great tool on your hands, but as soon as you scratch the surface a bit, it becomes necessary to understand the underlying statistics – if you want to derive something useful out of the exercise, that is. In other words, there is no way around statistics: you must know how you are supposed to understand, interpret and explain the result. Otherwise, things go wrong. That’s why you must have your theory sorted out and in place.
2. Big Data is not equivalent to Predictive Analytics
It has become unbelievably popular to stick the “Big Data” tag on everything that has something to do with statistical analysis of data. Alrighty then if that is what it takes for you to get your funding, but you do not need truckloads of data to be able to find meaningful contexts and patterns. Several thousand observations can actually do just fine.
Naturally, more is usually better and allows you to test the model’s validity and reliability but is not a requirement to get going.
3. Master Data Management
Yes. This also applies here: you need to sort out your master data. This point is particularly relevant now that you can pull data in a BW. Your master data are naturally already sorted in your BW, but it is a known fact that there are several occurrences of the same business size such as customers, suppliers, business areas, materials, etc. This is what needs to be sorted out first, and this is a problem that can be resolved nicely with an external hierarchy in BW that is used for grouping, e.g. several customer numbers as the hierarchy is transformed into a flat list in Predictive Analytics.
4. Business case
The golden rule also finds application when you make predictions with Predictive Analytics; the one with the money makes the rules and will not let go of it in up to 43% of the cases. So if you want to prevent this from turning into a purely technical/statistical exercise, you must be able to, as a minimum, make it probable to claw back more money by gaining the insight the analysis provides. It does not need to be dashing, it only has to appeal to common sense and make it possible to gain the insight needed to act based on facts rather than assumptions.