Speaking the language of business intelligence with an Australian accent

Tuesday, August 28, 2007

PerformancePoint Monitoring - Scoring Types

Something that irked many people about BSM was the inability to simply display the number that was driving the KPI indicator images in a scorecard. That is, the result of the threshold banding method which is most often set to Band by normalized value of Actual/Target. This number should not be confused with the KPI score (but it often is). The only way to display the banding method figure is to mimic the calculation for it using MDX in another target. This is quite a lot of work to get a number that is already being calculated...

In PPS we get a choice of scoring methods: Raw or Normalized. The scoring type is set on each target in a particular scorecard. Raw is the number that is calculated by the banding method, that elusive number we could never display in BSM. Normalized (the default setting) is the same score calculation mechanism that BSM employed, no difference. If you want to know more about the process of normalized scoring have a look at this post.

The raw score is not the number you want to use in your Balanced, Six Sigma etc. scorecards. You want Normalized score. Raw score will be very useful in displaying the result of that all-important KPI leaf calculation that drives the indicator image, but because the numbers have not been normalized along a linear scale the eventual rollup through various objectives will almost certainly be incorrect.

Take a look at the following examples. I have created 3 KPIs and filled them with simple fixed values. All KPIs have a threshold banding method of Band by normalized value of Actual/Target, the scoring pattern is Increasing is Better, worst value is set to zero. I created two identical scorecards, the only difference between them is that one target uses Raw scoring and the other uses Normalized.

Raw Score
I purposely made the Actual / Target numbers simple to show that this method will indeed bring back the value of the banding method calculation that many expect to see. Take a look at the percentages in the Target column.




Revenue has a score of 92%. Because the Worst value in the threshold has been set at 0 the calculation is (92-0) / (100-0) = 0.92. Similar calculations are made for the other two KPIs. We have our magic number. Brilliant! But what about the rollup? 74.8% hmmm, no normalization here. Sure the business would like 74.8% with a green light at the top of their scorecard, but is it correct? Remember we're trying to roll up a dollar figure, a number and a percentage. Where's the common ground that enables them to be compared? There isn't one, that's the pitfall of Raw score. If we want to roll these numbers up in a consistent manner then we must use a normalized scoring method.

Normalized Score
Below we have the Normalized score version of the same scorecard using exactly the same KPIs. The rollup at the Business Performance objective is 56.8% and yellow. This is an accurate rollup. The normalized percentages displayed in the KPI targets will confuse users, just as score does in BSM (if displayed), but these numbers are accurate for what needs to be done.



Now, this not to say that Raw score should never be used, it is simply a reminder to be wary of how each score type works and why. Quite often users are not really interested in seeing the score (of any kind), so most of the time the score percentage remains hidden from view. Nonetheless, whether this number can be seen or not does not change the fact that it is being used by PPS to determine which target indicator image is displayed for each KPI. Raw score calculations always drive indicator images at the KPI leaf level irrespective of score settings - in both of the above examples all three leaf KPIs displayed the same indicator. Score type affects the rollup of KPIs which is why we saw a different image when a different scoring method was used. Raw score is very handy to have at your fingertips but be careful of using it when rolling up KPI data of varying kinds in a scorecard.

3 comments:

Unknown said...

Hi Nick, useful post.

btw I got a lot of good stuff from your BSM book. Are you and Adrian planning a book on PerformancePoint?

Anonymous said...

Any news on whether PerformancePoint will support SAP BW 7.0.

Lots of documentation on SAP BW 3.5 but nothing confirmed.

Anonymous said...

It's a little disappointing that the only scoring types are raw and normalized. It would have been great to allow other types of distributions such as geometric, Z-score, etc. As it is now, I have to do the scoring in the ETL and have BSM show the raw score.