Interpreting Span Measurement - P95-P5

P

patkim

Back in 2002-2003 I was working with a company which was one of the vendors to General Electric (GE).
GE with its strong focus on Six Sigma, we were literally bombarded with Six Sigma knowledge. Now I am retracing my steps and I have some queries about a Metric called Span that GE has been using to assess On Time Delivery Performance. Unfortunately very little info is available on the internet about Span. My very little knowledge about Span has only added to my confusion.

Span is a measure of dispersion, Defined as P95-P5. Span removes the extreme outliers of your process by looking only at the difference between the 95th and 5th percentiles.
GE Used (and may still be using) Span to assess on time delivery performance.

If I correctly understand, lower the Span better the performance?
However I guess Span can be totally misleading in case of Normal data.
e.g. all my deliveries are late by 3 days. Now my variation from target is 3 every time so P95 is also 3 & P5 is also 3. Span is 0 but I was late every time resulting into huge customer dissatisfaction.

So how do you really interpret Span? Can someone help? If I have a few early deliveries by -2 days and few late deliveries by +4 days and say Span is 3 what does it practically mean?

In a six sigma project to improve Span, is Span data translated to Sigma Value? Is so how?

Thanks.
 

Miner

Forum Moderator
Leader
Admin
Span is a measure of dispersion (i.e., variation), not of location (i.e., mean).

The concept is based on the theory that customers are impacted to a greater extent by unpredictability than they are by consistently "poor" service. In your case, the customer may not be pleased by your 3 days late delivery, but if they know that you are 3 days late 95% of the time, they can plan for that and the impact is minimal. On the other hand, you might average 0 days late yet vary from 3 days early to 3 days late. Your customer cannot plan with that amount of variation.

With Span, the idea is to minimize the dispersion first (maximize predictability) then work on hitting the target. It appears that you already have a Span that has been minimized, so your next step would be to shift the average closer to zero.

You could convert Span plus mean to a sigma value, but I don't recommend it. It would reduce the usefulness of the metric and divert attention away from what is important.
 
Top Bottom