Six Sigma
Way
Way
BI
Quality
Quality
Copyright © 2007 - 2008 All rights Reserved
BI Capability
Maturity Model
Maturity Model
BI Comprehensive
Improvement Model
Improvement Model
Six Sigma BI Statistics
The Mean and Six Sigma
If you measure the occurrence of something many times, it is going to vary around some average, i.e. mean, value. The mean is the central tendency of the process. If there are enough occurrences of a coin toss, the average will be 50% heads and 50% tails. Anytime you measure the value of a given occurrence or event, it will almost always vary from the mean. Variation is simply a deviation from expectations. Six Sigma is concerned with this variation of the occurrence from the mean. We study the size, trends, causes, effects of the variation and try to control it.
Long term and Short Term Variation
If you measure the variation for a short time versus over a longer period of time, the variance may change. The probability of a defect in the short term does not account for changes that may be occurring in the process/asset base over the long term. Perhaps query response times are being impacted by changes in volumes or new data additions. Whatever performance measure you choose may, and most likely, will become eroded or changed over time. Performance monitoring and analysis must take into account these different variation types.
Roll up of characteristic behaviors
We have defined quality in terms of a single characteristic or component. However, in reality, there are numerous parts to the business intelligence asset base. All the constructs and components must fit together and work in conjunction. Although we may not translate all these components into CTQ’s, which we monitor at the customer level, all will impact the quality results which are seen by the customer. All these components and constructs will impact the performance. This multiplies the risk of defects in the product. In much of the statistical documentation, the concept of compounding defect risk is described with the example of rolling a die vs rolling multiple dice. With a single die with six sides, if you assume one side is a defect, you have a five in six change of rolling a ‘good’ side. However, with each added die, the odds of a successful result decreases. (with two dice - 10 out of 12/69%; three dice - 15 out 18/58%) With 50 to 100 dice, the probability of never getting a defect drops to near non existence, i.e. 1 in millions. This concept of compounding the defect risk is called risk throughput yield. The Business Intelligence Asset Base, which includes the information product, is extremely complex, with myriad parts. That means we need to expect defects. This means in practice that you must establish an extremely high success probability for each characteristic, if you expect the final product or service to be successful and near defect free.
The Six Sigma BI
Initiative
Initiative
Six Sigma
The term Six Sigma refers to a statistical measure measure which indicates how well a critical characteristic performs compared to the specifications. The measurement methodology and approach covers the two key concerns discussed in the previous sections, i.e. variation over time and the roll-up of characteristic behaviors.
Using the standard scale for sigma, the higher the sigma score, the fewer defects and the closer to ‘perfection’ the characteristic being measured. For example, if a characteristic is defective 31% of the time, it is ‘rated’ at two sigma. The compliance for this characteristic is 69%. If a characteristic operates at three sigma, that means that 67% of the time, the variation in the performance of that characteristic exceeds acceptable levels, i.e. the specifications/goals. The tableat the bottom of the page shows some key points on the sigma scale.
Six Sigma
The term Six Sigma refers to a statistical measure measure which indicates how well a critical characteristic performs compared to the specifications. The measurement methodology and approach covers the two key concerns discussed in the previous sections, i.e. variation over time and the roll-up of characteristic behaviors.
Using the standard scale for sigma, the higher the sigma score, the fewer defects and the closer to ‘perfection’ the characteristic being measured. For example, if a characteristic is defective 31% of the time, it is ‘rated’ at two sigma. The compliance for this characteristic is 69%. If a characteristic operates at three sigma, that means that 67% of the time, the variation in the performance of that characteristic exceeds acceptable levels, i.e. the specifications/goals. The table on the left shows some key points on the sigma scale.
six Sigma
Percent Defective
Defects per Million
1
69%
691,462
2
31%
308,638
3
6.7%
66,807
4
0.62%
6,210
5
0.023%
233
6
0.00034%
3.4
7
0.0000019%
0.019
The Six Sigma Scale
The process shift of 1.5 sigma. (long term vs short term variation)
Those of you statistics 'nerds' out there most likely understand already that the Six Sigma tables are based on a little thing called a 'process shift of 1.5 sigma' and there is some doubts about the origins and reality of the six sigma goals of 3.4 per million. Since there appears to be a shift in the quality level for short term vs long term measurements.
The only reason to mention this here is that we want to add a couple of points about using 3.4 million defects per million (i.e. six sigma?) as our goal for Business Intelligence.
• The measurements and structure of the Six Sigma DMAIC process are rigorous and can be very expensive to implement.
We have to ask ourselves, ' What Sigma level to we need to shoot for in order to product a top notch Business Intelligence product.'