Let's not just throw the decision 'ball over the fence'. Instead, let's take the time to define How a non-trivial repeating decision should be made. That will provide us measurable process consistency.

1. REAL PROCESS IMPROVEMENT
& DECISION SUPPORT
Real improvement requires Decision Support.
MarcVandenplas mjvandenplas@gmail.com 415.425.1436 (all rights reserved)

2. Thesis:
• A ‘process’ contains ‘decisions’ – or it is just a series of steps.
• Some decisions are easy to make and don’t need to be prescribed.
• Other decision are not so easy to make and do need to be prescribed.
• If we don’t describe ‘how’ decisions are to be made, we’re not really adding value.
Flip a Coin
Tails
Heads
Make a
Decision
Easy Decision
• What are the inputs required for the decision?
• How are those inputs used to make the decision?
• What are the probable risks and rewards?
Research
whether to
Make & Sell
Product
Make & Sell?
Yes
No
Decision Support Needed

3. At the end of this 15 page presentation, you will:
• Have been ‘stepped-through’ a simple process improvement exercise
• See how important it is to provide decision support
• Know how a Decision Support Model can be absolutely necessary to a process
• Understand what a Decision Support Model is
• (And) Suspect that a Decision Support Model is really just another Process Map
Structure of this presentation:
• Start with a simple process.
• Add a decision – with a decision support model for selecting from alternatives
• Further Improve the process
• Add a more complex decision – with a support model for selecting from alternatives
• Show the an improved process is comprised of (1) the process diagram and (2) the
decision support model

4. Begin
Make & Sell
Product
Costs (Actual)
Unit Revenue
(Actual)
Units Sold
(Actual)
Profit or Loss
End
Actual
Gamble
For example, say that this basic process needs improving (and it really does).
The Process is to:
1. Make & Sell our Product
2. Gamble on these:
• Costs
• Revenue per Unit Sold
• Number of Units Eventually Sold
3. See whether we win or lose.

5. Begin
Research
whether to
Make & Sell
Product
Make & Sell?
Make & Sell
Product
Yes
Costs
(Actual)
Unit Revenue
(Actual)
Unit Sold
(Estimated)
Profit or Loss
End
No
Costs
(Estimated)
Unit Revenue
(Estimated)
Units Sold
(Estimated)
ActualIn-House Estimates
Gamble
One obvious improvement is to get In-House Estimates of Costs, Unit Revenue, and
Units Sold.
• Then decide whether to Make & Sell (or not).

6. Our first decision support model easily corresponds to our improved process diagram.
Research
whether to
Make & Sell
Product
Make & Sell?
Costs
(Estimated)
Unit Revenue
(Estimated)
Units Sold
(Estimated)
In-House Estimates
Profit
or Loss
Units
Sold
Cost
Price
Make &
Sell?
‘Influence Diagram’

7. A ‘Decision Tree’ corresponds to the ‘Influence Diagram’.
Low
Profit_or_Loss
Nominal
Profit_or_Loss
High
Profit_or_Loss
Few
Some
Many
Price
Low
Med
High
Units
Sold
Yes
Cost
No
Make &
Sell?
‘Decision Tree’
Profit
or Loss
Units
Sold
Cost
Price
Make &
Sell?
‘Influence Diagram’

8. The green circles contain the In-House estimated probable values.
Low
Profit_or_Loss
Nominal
Profit_or_Loss
High
Profit_or_Loss
Few
Some
Many
Price
Low
Med
High
Units
Sold
Yes
Cost
No
Make &
Sell?
Few
0.5
Some
6.3
Many
14.2
Units
Sold
Units estimated to be sold
Probabilities

9. The Risk Profile is computed using the In-House estimated probable values. It says that we
will make $26.50, risk-adjusted, if we make and sell our product.
But we have a 52% chance of losing money. Can we improve our chances by further
improving our process?
-80 -60 -40 -20 0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340
Profit or Loss
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
100%
Make & Sell
No
Risk Profile
ProfitLoss
About a 52% chance of
losing money.
Cost
50%
[-65.50]Few
Cost
30%
[54.50]Some
Cost
20%
[214.50]Many
Units Sold
[26.50]Yes
[0.00]No
Make & Sell?
[26.50]

10. So far, we have only our In-House estimates. What if we could purchase some more
information? Would that be cost-effective?
Let’s improve our process by considering buying an outside study on Market Size. We
know that there are many to choose from, so we add a decision to select a study.
Begin
Research
whether to
Make & Sell
Product
Make & Sell?
Make & Sell
Product
Yes
Costs
(Actual)
Price
(Actual)
Market Size
(Estimated)
Profit or Loss
End
No
Costs
(Estimated)
Price
(Estimated)
ActualIn-House Estimates
Consider an
Alternative
Market Size
Studies
Use an Alternative Study?
Pay Cost of
Additional
Study
No
Yes
Gamble
Use Study Result instead of In-house Market Size (Estimated)
Market Size
(Estimated)

11. Precisely ‘how’ the decision of which study, if any, to buy is complicated. We can’t just
‘throw it over the fence’ and hope for consistent, measurable decision-making.
We need a second decision support model.
‘Combine’ Selected Alternative Study Results with Market Size (Estimated)

12. Our second model is more complicated than the first, but under our illustrating
assumptions, it tells us that the best decision is to use Size Study B.
• $26.63 ExpectedValue (EV, the risk-adjusted expected profit or loss)
• We should Not ‘Make & Sell’ if the results of Study B says that we will sell ‘Few’
Make & Sell?
0.00
[26.50]No Study
Study A Results
-12.00
[24.95]use Size Study A
Units Sold
[-2.22]Yes
[-0.50]No
Make & Sell?
37%
[-0.50]Few
Units Sold
[50.28]Yes
[-0.50]No
Make & Sell?
43%
[50.28]Some
Units Sold
[26.00]Yes
[-0.50]No
Make & Sell?
20%
[26.00]Many
Study B Results
-0.50
[26.63]use Size Study B
Use an Alternative Study?
[26.63]

13. By itself, the EV is not very informative: we should examine the risk profiles.
Study B,‘combined’ with the In-House Estimate, has the ‘best’ risk profile.
Note that ‘best’ is a matter of ‘risk tolerance’, which is a matter of ‘policy’, which is
another discussion. For example, risk policy might prefer to use Study A.
-100 -80 -60 -40 -20 0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340
Profit or Loss
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
100%
Use In-House Estimate
use Study A
use Study B
Risk Profile
ProfitLoss
The ‘best’ risk profile.

14. The Decision Support Model is really a ‘decision tree’ that encodes the computation of the
best decision. Hidden in the model are advanced statistics.
Low
Profit_or_Loss
Nominal
Profit_or_Loss
High
Profit_or_Loss
Low
Med
High
Price
Few
Some
Many
Cost
Yes
Units
Sold
No
No Study
Cost_of_Additional_Study
a
Make &
Sell?
Few
Some
Many
a
use Size Study A
Cost_of_Additional_Study
Study A
Results
Few
Some
Many
a
use Size Study B
Cost_of_Additional_Study
Study B
Results
Use
an
Alternative
Study?

15. So the delivered ‘Package’ of Improved Process should have two components:
1. The Improved Process Map, with added decisions
2. Decision Support for all non-trivial decisions
QED
Begin
Research
whether to
Make & Sell
Product
Make & Sell?
Make & Sell
Product
Yes
Costs
(Actual)
Price
(Actual)
Market Size
(Estimated)
Profit or Loss
End
No
Costs
(Estimated)
Price
(Estimated)
ActualIn-House Estimates
Consider an
Alternative
Market Size
Studies
Use an Alternative Study?
Pay Cost of
Additional
Study
No
Yes
Gamble
Use Study Result instead of In-house Market Size (Estimated)
Market Size
(Estimated)
Low
Profit_or_Loss
Nominal
Profit_or_Loss
High
Profit_or_Loss
Low
Med
High
Price
Few
Some
Many
Cost
Yes
Units
Sold
No
No Study
Cost_of_Additional_Study
a
Make &
Sell?
Few
Some
Many
a
use Size Study A
Cost_of_Additional_Study
Study A
Results
Few
Some
Many
a
use Size Study B
Cost_of_Additional_Study
Study B
Results
Use
an
Alternative
Study?