SESSION 11: THE IMPROVEMENT PHASE OF YOUR SIX SIGMA PROJECT

1
SESSION 11 THE IMPROVEMENT PHASE OF YOUR SIX
SIGMA PROJECT

• INTRODUCTION TO
• IMPROVING BUSINESS PERFORMANCE SIX SIGMA, LEVEL
  1
• APRIL 16 - 18, 2007

2
SESSION OBJECTIVE

• Outline steps in the Improvement Phase of Six
  Sigma project implementation

3

• Without continual growth and progress, such
  words as improvement, achievement and success
  have no meaning
• Benjamin Franklin

4
QUESTIONS TO ASK IMPROVE PHASE

• Ask yourself the following questions
• What is the possible root cause of defects?
• How can you eliminate these causes?
• What changes in product/service or process design
  are required to achieve your improvement goal(s)?
• How do you know that the changes will be
  effective?
• What are the next steps toward achieving the
  goal(s)?
• Has Finance been involved in the project to date
  and do they fully understand the cost
  implications of your improvement plan?
• Are you satisfied with the cooperation level and
  support you are getting?
• What other support actions/activities are needed
  to accelerate progress?

5
RECAP FROM THE MEASUREMENT PHASE ON

• Measurement Phase
• You know your key metrics
• You know your data is valid
• Analyze Phase
• Has enabled you to create a set of qualified Xs
  suspected of causing the defects

6
CONSIDERATIONS IMPROVE PHASE

• Fundamentally, the improve phase is about
• good judgment and
• using data to derive solutions
• Base your thinking around
• Y (X)
• making sure you fully understand the
  relationships between the Ys and Xs (proof)

7
So What

• Fundamentally, the heart of the Improvement Phase
  is questions 3 and 4
• What changes are required and
• How do you know they will be effective?

8
Answering the SO WHAT Question

• We use two common techniques to answer the
  questions
• Correlation analysis and
• Experimentation (specifically Design of
  Experiments (DOE))

9
CORRELATION ANALYSIS - 1

• Remember we want to establish input/output
  relationships
• what factors (Xs) (inputs) are affecting our
  output (Ys) (customer satisfaction), the most.
• Simple way is to use graphical method of
  correlation

10
CORRELATION ANALYSIS - 2

• Correlation analysis determines the extent to
  which values of two quantitative variables are
  proportional (linearly related) to each other
• Correlations lie between 1 and 1 with
• 1 representing strong positive linear
  correlation
• - 1 strong negative linear correlation
• and 0 no linear correlation
• The level of Correlation is expressed by the
  Correlation Coefficient (r) and is a measure of
  the strength of the correlation
• The closer to one (either positive or negative)
  the higher the correlation
• 0.80 and above indicates the correlation is
  important
• 0.20 or less means the correlation is not
  significant
• Remember Y (X)
• STRAIGHT LINE EQUATION
• Y MX Constant

11
TYPICAL CORRELATION PATTERNS
12
TYPICAL CORRELATION PATTERNS
13
TYPICAL CORRELATION PATTERNS
14
TYPICAL CORRELATION PATTERNS
15
Apply Pareto Principle to Determine r

• THINK 80 20
• Guidelines for using Pareto to determine
  correlation
• Target is to ensure the oval encompasses 80 of
  the data points
• No more than 3 data points can be outside the
  lower half of the oval
• No more than 3 can be outside the upper half
• Step 1 Draw an oval around the plot of points
• Step 2 Measure maximum diameter A with a scale
• Step 3 Measure minimum diameter B with a scale
• Step 4 Value of r is estimated by (1 (B/A))
  where the sign is a plus if the A diameter slopes
  upward and minus if the A diameter slopes
  downward

16
HOW DOES IT WORK AND HELP US?

• Example Common business issue
• spending on advertising budget and impact on
  sales?
• Consider the next two graphs

17
Advertising versus Average Sales
Advertising Cost
Average Dollar Sales
Advertising Cost
18
Graphical Method for determining Correlation
A
. ..
. ..
. ..
. ..
. ..
. ..
. ..
. ..
. ..
. ..
. ..
. ..
. ..
Y
. ..
. ..
. ..
. ..
. ..
. ..
. ..
. ..
B
(X)
19
Cost of Advertising vs Average sales
Average Sales
B 4.6 cm
A 9.0 cm
To calculate r use the formula (1 (B/A)) r
1 (4.6/9) 0.48
Advertising Cost
20
SO WHAT NEXT? IS THERE A CORRELATION BETWEEN THE
TWO VARIABLES?

• TO INFER FROM THE COEFFICIENT IF THERE IS SOME
  CORRELATION BETWEEN THE TWO VARIABLES WE USE
  DECISION POINTS TABLES
• DECISION POINT TABLES ARE RELATED TO SAMPLE SIZE
• WHEN USING THE TABLES DISREGARD ANY NEGATIVE SIGNS

21
SAMPLE SIZE AND DECISION POINTS
22
Correlation and Decision Points
In the Cost of Advertising example the sample
size is 10, so the decision point is
0.632. r 0.48 Negative No
correlation Positive - 1.0
0.0 0.632 1.0 Decision
Point As the coefficient r is below the
decision point there is no correlation between
cost of advertising and sales
23
DESIGN OF EXPERIMENTS

• USE EXPERIMENTATION AS AN ALTERNATE APPROACH
• In experiments we usually either control inputs
  or vary them according to a plan
• Traditionally, we evaluate a single variable and
  keep all others constant relatively simple but
  drawback is that it does not show what happens
  when two or more variables change at one time- it
  would be possible to do by running every possible
  combination of factors at least once and test the
  effects of the interactions very quickly this
  means we have to do a lot of experiments for
  example five factors would require running the
  experiment 32 times
• To overcome this we use the DESIGN OF EXPERIMENTS
  approach

24
DESIGN OF EXPERIMENTS (DOE)

• DESIGN OF EXPERIMENTS strategy allows us to run
  tests according to a specific structure and with
  specific methodology for analyzing results.
• Step 1Determine settings for each input
  variables (factors) in advance
• Step 2 During the experiment adjust the factors
  to specified settings
• Step 3 Run the process and
• Step 4 Measure and record output variable for
  one or more units of output (think
  products/services delivered)
• Step 5 Analyze data to determine vital few input
  factors
• Step 6 Create a model to estimate Y (X)

25
Overview of Improve Phase using DOE

• Define the problem
• Establish experimental objective
• Select variables and chose levels for input
  variables
• Set experimental design
• Run experiment and collect data
• Analyze data
• Draw conclusions
• Replicate and Validate results

26
Step 1 Define Problem

• Describe problem in practical business terms that
  people can understand in the same way.
• Example Problem Cost impact of advertising,
  media and sales force on seasonal profits is not
  known
• Statement Historical data indicates that
  spending is all over the map with no
  understanding of the return for money spent
  resulting in a tripling of cost to the business.

27
Step 2 Establish Experimental Objective

• Example The business owners wish to be more
  confident in their plan for seasonal promotional
  expenditures.
• Objective Statement
• The experiment should show that our plan will
  reduce cost by 70 with no adverse effects to
  seasonal customer requirements

28
Step 3 Select Variables and Chose Levels for the
Input variables

• Step 1 Select both output (Ys or response)
  variables an and input ( Xs independent
  variables)
• Step 2 Chose the levels for each input variable
  by level we mean a setting. Generally we set
  two low or minimum as one and high or maximum as
  the other. Sometimes we set a third the
  normal setting or the mean.
• Step 3 Where data is available from the Measure
  and Analyze Phases we select the extremes of the
  process for cashiers for example we could use
  the experience considering three months to be low
  and more than 1 year high. The levels should
  reflect a range of reality for each X and it is
  important to fully test the range recognizing
  that this may result in additional defects as a
  direct result of the testing this is to be
  expected and anticipated.
• Step 4 Code each level either 1 (high) or 1
  (low) for simplicity of record keeping

29
Step 4 Select Experimental Design

• Example Note there are many design and complex
  methods beyond the scope of our course to
  discuss.
• Basic Concepts
• Experimental Design is a simple table or matrix
  of possible combinations of factors and levels
  you are studying
• A single combination is called a treatment
  combination - the level of the factors at a given
  condition is the treatment combination that
  results in a given observation that is recorded
  for subsequent analysis

30
Step 4 Select Experimental Design continued

• Example Setting shower temperature 2 variables
  hot (105 degrees) and cold water (50 degrees)
• Step 1 Select 2 levels low pressure (turn knob
  quarter turn counterclockwise) high pressure 2
  turns counter clockwise)
• Experimental design for 2 factors at 2 levels is
  calculated as 2K where 2 is number of levels
  for each factor and k number of factors
• In example we have 2 levels for 2 factors and
  each factor

31
Step 4 Select Experimental Design continued
THE MORE FACTORS THE MORE THE COMBINATIONS
INCREASE
32
Step 5 Run Experiment and Collect Data

• Design Table is your plan for setting factors to
  the level specified in the treatment combinations
  - now run experiment
• To valid results we need to run the treatment
  combinations more than once
• See example chart following for three trials

33
Experimental Results
34
Step 6 Analyze Data

• Once again many ways to analyze data from DOE
  tests
• Purpose here is only to illustrate simple
  principle of DOE
• Graphical plot of X inputs and Y outputs by
  plotting individual values for each variable or
  group in a vertical column making it easy to spot
  trends For example, you can easily see what
  setting you need to get the water as hot as
  possible Cold -1 (low) Hot 1 (high)

35
Step 6 Analyze Data
36
Step 7 Draw Practical Conclusions

• Note you need to understand/balance practical
  significance with statistical significance.
• You can reach practical conclusions that are not
  statistically significant bake a cake imagine
  a cake with 6 different ingredients you find
  that you can reduce the amount of one of the
  ingredients to a point where the difference in
  quality has no statistical significance
• THERE IS NO POINT TO REMOVING THE INGREDIENT AS
  IT DOES NOT ALTER THE TASTE (QUALITY INDICATOR
• HOWEVER THE INGREDIENT MAY BE EXPENSIVE AND
  COULD FOR EXAMPLE REPRESENT 40 OF THE INGREDIENT
  COSTS WHICH YOU CAN REDUCE IF YOU CUT THE AMOUNT
  OF THE INGREDIENT IN HALF

37
Step 8 Replicate/Validate results

• Once you have your DOE conclusions you need to
  validate them by running the desired settings
  validation will be self evident if the results
  are what you expect

38
Step 9 Conduct a Phase Gate Review

• Report your findings to the champions!!!!

39
IMPROVE PHASE DELIVERABLES

• The basic deliverables for the IMPROVE Phase
  include
• A project status form/report
• Metric Graph
• Tools, as applicable for determining
  improvements
• DOE plan, Gage R and R, 3level Pareto charts
• Contingency table, update FEMA
• Solution to question what is Root Cause of
  Defect(s)
• Quantified Improvement Plan(s)/Next Steps
• Complete Project review (Phase Gate)

40
SUMMARY

• The Six Sigma project team begins the Improve
  Phase by selecting the performance characteristic
  that needs to be improved to achieve the goal(s)
• It then diagnoses those characteristics to
  determine/reveal the major sources of variation,
  using correlation and regression analysis
• After this, we apply statistically designed
  experiments (DOE) to identify the key process
  input variables (Xs).
• The team tests the variables that were filtered
  during the Analyze Phase and identified as our
  Vital Few Factors.
• The DOE experiments define the interactions
  between the vital factors and can yield
  interesting facts enabling a rapid movement to
  improve the process(es) in question