Corporate Decision Analysis:
An Engineering Approach
Victor Tang
Product Lifecycle Team
April 2006
2
Observations from corporate experience about
senior-executive decisions under uncertainty.
Data quality
largely untested
Range of alternatives considered
narrow
Importance of decision variables
largely unaddressed
Uncontrollable variables
largely unaddressed
Impact of uncontrollable variables largely unaddressed
Predictive power
low
3
Decision outcome =
f(what’s controllable & uncontrollable)
4
In situ experiment #3
High-tech manufacturing company (US)
•
$700 M/year global high-tech manufacturer failing to
generate profit.
•
Company de-listed from stock exchange.
•
Board of directors appoints new president.
Wants an assessment of his turnaround strategy, survey
of alternatives and their prospects to generate profit.
5
engagement process
president
working group
president
working group
Problem definition
decision
decisionsituation
situation
goals
goals&&objectives
objectives
Goal & objectives
controllable
controllableand
and
uncontrollable
factors
uncontrollable factors
Data collection &
de-biasing
sample
samplescenarios
scenarios
scenarios’
scenarios’forecasts
forecasts
working group
Generating
alternatives
working group
Analyzing
alternatives
analyze
analyzewhat-if’s
what-if’s
president
working group
Selecting an
alternative
commit
committo
toaction
action
president
Final report
documentation
documentation
what-if
what-ifconstructions
constructions
6
Frame the problem
problem
outcomes
controllable
variables
uncontrollable
variables
Survival
Profitability in 6 Months
1. Sales, general & admin expenses, SG&A
2. Cost of goods sold, COGS
3. Capacity utilization
4. Customer portfolio structure
5. Sales
6. Financing
1. Customer base changes
2. Senior management interaction
3. Banker actions
4. Loss of critical skill
7
Boundaries of the solution and uncertainty space Î
729 alternatives and 54 uncontrollable environments
Controllable
1.
2.
3.
4.
SG&A
COGS
plant capacity
customer
portfolio mix
5. sales
6. financing
BAU
Uncontrollable
level 1
level 2
level 3
+10 %
+2 %
40 %
current mix
$ 54 M
$ 651 M
60 %
dev<10%, a/t<6%,
mfg.<4%
$ 690M
Mexico action,
+ $12 M annualized
-10 %
-2 %
80 %
dev<20%, a/t<12%,
mfg.<8%
+5 %
China action,
+ $25 M annualized
current
best
no change
weak management
unity
no change
no change
net gain >5%GM
strong management
unity
US banks relax terms
gain 1 or 2 skills
-5 %
$10M short
worse
1. cust. base change net loss >5% GM
2. senior executives’
= current
interactions
US banks drop
3. banker actions
lose 3 skills
4. critical skills
8
controllable
factors
2
1
2
3
2
3
1
3
1
2
2
3
1
1
2
3
3
1
2
1
3
1
1
financing
1
1
2
3
2
3
1
1
2
3
3
1
2
3
1
2
2
3
1
1
3
1
3
sales
2
1
2
3
1
2
3
2
3
1
3
1
2
2
3
1
3
1
2
3
1
3
3
portfolio
2
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
1
3
3
2
capacity
2
1
1
1
2
2
2
3
3
3
1
1
1
2
2
2
3
3
3
3
1
1
3
COGS
SG&A
BAU
uncontrollable
factor levels
1
1
2
3
3
1
2
2
3
1
2
3
1
3
1
2
1
2
3
3
3
3
1
level 2
level 2
level 2
level 2
level 1
level 2
level 1
level 1
level 3
level 3
level 3
level 3
current
worst
best
Collect
and
analyze
the data.
9
Debiasing Îdispersions decline and confidence rises
F ORECAS T [ CURRENT] r ound1 & 2
F ORECAS T [ WORS T] r ound 1 & 2
Normal
Normal
25
F ORECAST [ BEST] r ound 1 & 2
Normal
40
80
70
20
30
10
50
Frequency
Frequency
15
40
30
20
10
20
5
10
0
-10.8
-9.0
-7.2
-5.4
PROFIT $ M
-3.6
0
-1.8
forecasts =
stdev È
-16
-14
-12
-10
PROFIT $ M
-8
-6
-4
0
-10
-8
-6
-4
-2
PROFIT $ M
0
2
forecasts È
stdev È
forecasts Ç
stdev È
C O N F D E N C E ( 1 - 5 ) r o u nd 1 & 2
Normal
20
15
round 1
round 2
Frequency
Frequency
60
confidence rises
stdev È
10
5
0
1.5
2.0
2.5
3.0
3.5
PROFIT $ M
4.0
4.5
5.0
10
Unconstrained exploration of what if’s
Contribution of each variable to the outcome
COGS Cportfolio SG&A
72 %
9%
7%
sales
7%
What if ?
BAU
BAU
BAU
BAU
BAU
BAU
BAU
BAU
U
+
+
+
+
+
+
+
business as usual
COGS+
cust. portfolio+
SG&A+
sales+
financing+
plant capacity+
COGS+ + portfolio+
financing
3%
capacity
2%
profit $ M
current
-5.54
-2.04
-3.99
-4.35
-4.43
-4.90
-5.16
-0.40
worst
-9.40
-5.90
-7.73
-8.28
-8.27
-8.17
-8.43
-4.24
best
-2.89
+0.43
-1.15
-1.68
-1.90
-2.41
-2.72
+2.18
11
Manufacturing company results:
Plan versus actual
controllable
factors
SG&A
COGS
plant utilization
portfolio actions
sales
financing
results
level
3
2.5
2
2
1.5
1.5
Plan
values (level)
$54 M-10%
$651 M – 1%
60 %
no change
$690 M -2.5%
shortfall ~$5 M
derived $ -1.13M
actual performance
level values (level)
vs. plan
3
same
=
2.5
same
=
Ç
2.5
70 %
Ç
2.5
improved mix
È
1
$690 M - 5%
È
shortfall ~$10 M
1
reported to SEC $ 1M
derived
$ 0.41 M
• loss of $16 M same quarter last year
• loss of $13 M previous quarter
12
Services company results:
Plan versus actual
Post-BAU plan
3
Project approach
2
Cost contingency
2
Project delivery
3
derived
results
3.6
3.8
4.1
change project
leader &
program mgr.
3 waves. Focus
US & Japan
use some
contingency
meet delivery
date
environment
worst
current
best
delivered
values (level)
level
Project leadership
values (level)
agreed result
level
level
controllable
factors
values (level)
3
=
3
=
2
=
2
=
2
=
2
=
3
=
2
environment
-
2.7
2.9
slip
2-3 months
environment
worst
current
4.1
best
3.2
actual resultxxÎx
3.5
best
13
Executives were enthusiastic about the method
“Let’s take this to our board of directors.”
“Approach will make better decisions.”
“... excellent, rational ...Understand risk with factors
cannot control.”
“Value of this process is in the process not the
conclusions.”
“This process visualizes the decision … instead of
intuition.”
14
Measurement system analysis
15
Forecasts vs. derived estimates give an indication of
an operator’s repeatability across forecasts.
participant 3
participant 2
participant 1
10.0
10.0
10.0
5.0
5.0
5.0
0.0
0.0
131333
133113
222121
313113
0.0
131333
323311
133113
222121
313113
323311
131333 133113 222121 313113 323311
-5.0
-5.0
-5.0
-10.0
-10.0
-10.0
particpant 5
participant 4
10.0
10.0
5.0
5.0
0.0
0.0
131333
133113
222121
313113
131333
323311
-5.0
-5.0
-10.0
-10.0
source: gage r&r calculations.xls
Individual forecasts
133113
222121
313113
323311
Derived estimates using
L18 operator data
16
Individual forecasts of 5 (test) treatments gives us
an indication of reproducibility across “operators”
Profit forecasts $M
10.0
5.0
p1
p2
0.0
p3
131333
133113
222121
313113
323311
p4
p5
-5.0
-10.0
17
Can identify source of low quality data
actual variation part-part
variation over all
treatments
overall
variation in
measurements
(forecasts)
measurem’t
system
variation
Gage R&R
repeatability
variation in forecast by
one operator for a given
treatment
48 %
82 %
3%
49 %
reproducibility
variation in forecasts
of different operators for
a given treatment
11 %
7%
all
w/o op 4
18
New approach to ...
senior-executive decisions under uncertainty.
Data quality
can be improved
Range of alternatives considered
entire solution space
Importance of decision variables
can be determined
Uncontrollable variables
can be determined
Impact of uncontrollable variables
can be determined
Predictive power
higher
19
NEW WAY TO ...
ƒ Analyze corporate decision-making
Controllable variables
Uncontrollable variables
ƒ Explore the entire solution space
Systematically and economically
ƒ Explore outcomes over entire space of uncertainty
Unconstraint range of what-if scenarios
20
Experiment power
power of experiment
1
power
0.8
0.6
0.4
0.2
1.
8
1.
6
1.
4
1.
2
1
0.
8
0.
6
0.
4
0.
2
0
0
difference from BAU profit $M
ƒ Power is the ability to detect a difference when one exists.
ƒ Power is the probability that you will reject a premise when
it should be rejected.
21