Eight Stages of Knowledge

Barnes 1997 discusses eight stages of knowledge that a company may go through in understanding their processes and uses a case study on Nabisco biscuits to illustrate them. The stages are

1. Complete Ignoranceshift scheduling software. No knowledge of the variables that affect the response and one consider all variation to be random. May not even know how to measure the response itself.

2. Awareness. Begin to build a list of the factors that one believes could affect the process. (e.g. ingredients, baking time, weather etc)

3. Learn to Measure Key Variables. Begin to measure the variables that one thinks may affect the response. (e.g. time baking using a watch, measure ingredients using cups, count the number of strokes used in mixing etc.)

4. Control the Mean. Knowing how to measure the key variables it follows that one will attempt to control them in order to control the response. E.g. have a countdown timer to tell you when to take the cookies out of the oven, standardise measuring cup size etc.

5. Process Capability and a Recipe. Work on reducing the variation in key variables and document a process that gives reasonable results i.e. write down a recipe and maintain control of all the variables. Cookies should then become more consistent.

6. Process Characterisation. Conduct a series of experiments to discover how certain variables affect the response so that if a customer wants a sweeter or a lower fat cookie one knows which variables to change.

7. Know Why. Develop knowledge of the interactions between input variables and the response as well as the strength and direction of these relationships. Build a model of the process to predict what effect changes in certain variables shift scheduling softwarewill have on the cookie. E.g. you know how much less sugar to put in for the desired reduction in cookie sweetness.

8. Complete Knowledge. Since there are infinite secondary variables it is impossible to have perfect knowledge of a process. It is however practical to say that one has reached stage eight when you have a model which will predict output characteristics to 10% of the tolerance band for changes in inputs across a 2:1 range, including all interactions.

The 8 stages of knowledge development detailed above suggest a stepwise process of improvement in them self. It may theoretically be possible to skip stages 4 and 5 in a rush to know why a process behaves as it does. However without instilling the disciplines involved in these middle stages it is easy for the developed knowledge to be lost, and for the process fall back to stage 3. The total approach described above ensures that improvements that are made to the process are maintained over the long term.