Harvard Business School 9-894-004

Rev. December 4, 1997

DO
Decision Analysis

Why do managers usually find making decisions difficult? First, uncertainty is

inherent in most important decisions. A product manager, for example, must decide whether
to market a new product, even though she is uncertain of production costs, consumer
receptivity, and how the competition might react. A second reason is that decision makers
NO
must balance many conflicting objectives. A division manager may decide to invest in a new
product that has the potential to be a blockbuster in ten years, even if the division’s morale
and short term profitability suffers. Third, the large number of alternatives a manager must
evaluate often serves to confuse and even paralyze the decision maker. Finally, a manager
often does not even know the full set of alternatives from which she must choose.

In short, it often becomes too difficult to make a decision based only on intuition:
formal analysis may be necessary. Decision analysis is a logical and systematic approach for
analyzing decision problems. It takes a “divide and conquer” approach to decision making,
T
breaking down the process into a number of steps (summarized in Figure 1):

1. Decision structuring: Identify

a. the set of alternatives from which to choose;

CO
b. the set of critical uncertainties that affect the outcomes of the decisions;
c. criteria for choosing among the competing alternatives.

We represent alternatives and uncertainties in chronological order using a decision

tree.

2. Assessment and information gathering: Collect relevant information,

PY
a. assessing the likelihood of the various uncertain events;
b. assigning values to the various consequences of the decision problem.

George Wu prepared this note as a basis for class discussion.

Copyright © 1993 by the President and Fellows of Harvard College. To order copies or request permission to
reproduce materials, call 1-800-545-7685 or write Harvard Business School Publishing, Boston, MA 02163. No
part of this publication may be reproduced, stored in a retrieval system, used in a spreadsheet, or transmitted in
any form or by any means—electronic, mechanical, photocopying, recording, or otherwise—without the
permission of Harvard Business School.

1
894-004 Decision Analysis

3. Evaluation of the decision problem: Analyze the information provided in the first two
steps to determine which alternatives to undertake.

4. Sensitivity analysis: Determine if the decision is sensitive to changes in

probabilities, valuation of outcomes, or other assumptions you have made.
DO
Figure 1

Assessment
Information Informationof
Evaluation
Decision and
Gathering Gathering Sensitivity
Structuring Decision Analysis
NO
Information Problem
Gathering

Decision Decision Decision Decision

Traps Traps Traps Traps
T
Each step of this four-stage process yields insights into the decision problem. If the
insights gained at any stage are sufficient to allow you to make a decision, then there is no
need to proceed further. On the other hand, the process is often iterative: you might return to
a previous step in the process if an insight leads you to consider another alternative or
question an assumption.
CO
Many believe, falsely, that decision analysis is merely a set of mechanical techniques
that spits out “the best option” if applied properly. This characterization sells short both
decision analysis and the manager as decision maker. While it is true that it is easy to use the
techniques we describe here incorrectly, sound decision making requires much more than a
technical command of decision analysis. The practice of decision making is as much an art as
a science. We emphasize throughout this note that a wise decision requires judgment,
creativity, as well as technical proficiency. Furthermore, following these four steps won’t
PY
ensure a good outcome, or even a good decision. Lurking in the background at every stage in
the process are decision traps, pitfalls that managers face in making sound choices. These
decision traps can severely bias the decision process and lead to poor decisions.

2
Decision Analysis 894-004

A Simple Decision Problem

Consider the following simplified decision problem:

Sarah Chang is the owner of a small electronics company. In six months a proposal
is due for an electronic timing system for the 2000 Olympic Games. For several years,
DO
Chang’s company has been developing a new microprocessor, a critical component in a
timing system that would be superior to any product currently on the market. However,
progress in research and development has been slow, and Chang is unsure about whether
her staff can produce the microprocessor in time. If R&D succeeds in developing the
microprocessor, then there is an excellent chance that Chang’s company will win the
$1,000,000 Olympic contract. If they do not, then there is a small chance that she will still be
able to win the same contract with an alternative inferior timing system that has already
been developed.

If she continues the project, Chang must invest $200,000 in research and
NO
development expenses. In addition, making a proposal requires developing a prototype
timing system at an additional cost of $50,000. Finally, if Chang wins the contract, the
finished product will cost an additional $150,000 to produce. Chang must decide whether
to abandon the project or whether to continue investing in the venture.

Qualitative Structuring
T
Many of us begin attacking a decision problem before we even understand what the
decision problem is all about. We start deciding before we know what we should be
deciding among, or what criterion we will use to choose among the various alternatives.
Asking a few simple questions up front can provide a surprising amount of clarity. In
particular, decision makers should begin by asking themselves the following three
CO
questions:

• What are my alternatives?

• What are the critical uncertainties that affect the outcomes of my decisions?

• What objectives am I trying to achieve in making a decision? What decision

criteria will I use to choose among the competing alternatives?
PY
Alternatives, uncertainties, and decision objectives are three basic components of
virtually every decision problem. We look at each of these three aspects from Chang’s point
of view.

Alternatives

Chang should first identify what options are available to her. It is six months prior
to the proposal date. What can Chang do? First, Chang can abandon the project altogether,
thereby avoiding the risks of failing to develop the microprocessor. On the other hand,

3
894-004 Decision Analysis

Chang can continue to invest in the project. If she continues with the project, at some
juncture she may need to decide whether to make a proposal.

Uncertainties

We often have to make decisions without knowing exactly what will happen
DO
tomorrow. Nevertheless, a wise decision maker must consider the major uncertainties and
make judgments about how likely these uncertain events are to occur. One way to elicit a
list of critical uncertainties is to imagine a clairvoyant, an individual who could look into
the future and indicate perfectly what might happen. Imagine that a clairvoyant will allow
Chang to ask a few questions. What questions can Sarah ask the clairvoyant that might
help her choose wisely? First, she may begin: “Will R&D succeed in developing the
microprocessor?” Second, she might ask, “If R&D succeeds, will we win the contract?”
Chang might conclude by asking: “If R&D fails, will we win the contract?” Of course, there
are many other uncertainties that Chang might wish to have resolved, such as the success of
other company projects or whether a venture capitalist might be interested in investing in
NO
her firm. However, Chang need not consider these factors if they do not bear on her
decision of whether to abandon or continue the Olympic project.

Objectives

We make decisions as means of achieving objectives. Thus, good decision making

requires a clear understanding of our objectives. What are Chang’s objectives? First and
most obvious, she wishes to make money. It is easy to stop here and omit other legitimate
T
objectives Chang may have, such as to attract venture capitalist money, to enhance her
company’s reputation, etc. These objectives should be considered in the analysis if they are
important and the choices she might make impact on these objectives.

It is often appropriate to consider a unitary objective, maximize net cash flow. In

CO
maximizing net cash flow, you should choose the alternative that yields the largest positive
cash flow or smallest negative cash flow. Applying this criterion to a choice between $10
and $20 is straightforward. However, what if the choice is between $25 for sure and a .50
chance at $60? Because a .50 chance at $60 has two possible outcomes, $0 and $60, the
criterion of maximizing net cash flow doesn’t help.

How do we value a .50 chance at $60? Clearly, a .50 chance at $60 is worth less than
$60, the best outcome. It is also worth more than $0, the worst outcome. Since the
probability of receiving $60 is .50, the value of a .50 chance at $60 might be $30 – halfway
PY
between $0 and $60. Similarly, the value of a .20 chance at $60 might be $12, 20 percent of
the way between $0 to $60.

In this example, $30 is the expected monetary value (EMV) of a .50 chance at $60.
EMV is determined by multiplying each outcome by its probability and then summing
these products. As another example, the EMV of a .30 chance at $50 and a .70 chance at
$100 is

.30($50) + .70($100) = $85.

4
Decision Analysis 894-004

It is not always appropriate to use EMV. Some thoughtful individuals will value a .50
chance at $60 at less than its EMV of $30 because they dislike taking risks or are risk averse.
Similarly, a risk prone individual prefers a .50 chance at $60 to $30 for sure.

A corporation with limited capital might reject a positive EMV alternative if the
chance of a loss is too high. Thus, Chang might wish to consider risk explicitly if her
DO
company is perilously close to bankruptcy. Even so, maximizing EMV is a reasonable
criterion for a surprisingly large number of decision problems. It should be used when the
stakes are small compared to the resources of the company or the individual decision maker
and when money is the primary concern. Even when a decision maker wants her decision
to reflect risk aversion or non-monetary objectives such as enhancing company reputation,
EMV can be used as a benchmark. Taking a first cut using EMV might provide a great deal
of insight in the decision problem. The decision maker might then take a more refined cut
that incorporates risk attitudes and non-monetary objectives.

Thus, in this example, we make the initial simplifying assumption that Chang
wishes to maximize EMV.
NO
Decision Trees, Alternatives, and Risks

In the next stage of analysis, Chang should identify the options available to her and
the uncertainties she faces. We represent the chronological sequence of these options and
uncertain events in a decision tree, which can be thought of as a roadmap of the decision
T
problem.

It is six months prior to the proposal date. What can Chang do? As noted earlier,
Chang can abandon the project altogether, thereby avoiding the risks of failing to develop
the microprocessor. On the other hand, Chang can continue to invest in the project. If
CO
these are the two possible options available to Chang, her decision tree begins with the two
branches shown in Figure 2.
Figure 2

Continue
PY

Abandon

Note that a decision is represented in a decision tree by a square or decision node.

5
894-004 Decision Analysis

At this point, Chang should ask herself: “Then what?” If Chang abandons the
investment, then she is done; she does not need to consider this path any further. On the
other hand, what might happen if Chang decides to continue with the project? In the next
six months, either Chang’s engineers will succeed or fail in their effort to develop the
microprocessor. Chang is unsure which of the two possible events will occur. In a decision
tree, we represent an uncertain event with a circle or chance node. Chang’s decision tree so far is
DO
shown in Figure 3.

Figure 3

Succeed

Continue
NO
Fail

Abandon
T
Chang should continue to ask herself, “Then what?”: what could happen in the
event of technological success and in the event of technological failure? First, suppose that
Chang’s engineers develop the microprocessor in time. Chang must then decide whether to
make a proposal. Since it costs $50,000 to produce a prototype, it is not totally obvious
CO
whether or not Chang should submit a proposal. If Chang is unsure of whether an option is
a good one, she should include it in the decision tree: if the alternative proves to be worse
than another option, it will be eliminated naturally at a later stage in the analysis.

If she enters a proposal, she may either win or lose the contract (Figure 4). Of
course, winning or losing the contract is an uncertain event. If she doesn’t submit a
proposal, there is nothing further to continue.
PY

6
Decision Analysis 894-004

Figure 4

Win
Make Proposal
Lose
Succeed
DO
Don't
Make Proposal
Continue

Fail
NO
Abandon

Finally, we can repeat this reasoning for the case of technological failure. Without the new
microprocessor, Chang must decide whether to propose the alternative inferior timing
system. At last, we arrive at the decision tree depicted in Figure 5.
T
Figure 5

Win
Make Proposal
Lose
CO
Succeed

Don't
Make Proposal
Continue Win
Make Proposal
Lose
PY
Fail

Don't
Make Proposal

Abandon

The process of Chang asking “Then what?” has produced a concise description of
her decision problem. The decision tree is a chronological depiction of the alternatives

7
894-004 Decision Analysis

immediately available to Chang, as well as future alternatives and uncertainties. The

decision tree provides immense clarity about the decision problem – perhaps providing
enough insight that it becomes clear to the decision maker which alternative she should
undertake. In this problem, if the chance of technological success is low, the cost of
development high, and the rewards of winning the contract low, then Chang might
conclude that she should abandon the project. Problems that are not so clear cut require a
DO
more complete analysis; the decision maker must assess the likelihood of the uncertain
events and estimate the value of the possible outcomes.

In this example, the set of options and uncertainties is fairly small. In many real
world problems, however, the process of depicting a decision tree requires more creativity.
Even in this simple problem, the analysis could easily be enhanced by considering
additional alternatives: for example, investing $300,000 in development instead of $200,000,
abandoning the effort after three months, etc. It is up to the manager to use her ingenuity
and creativity to ensure that the decision tree captures all the reasonable options available
to her. Creative decision makers can often conceive of alternatives others would overlook.
Often the process of capturing the decision problem as a decision tree will suggest
NO
alternatives that might not have been imagined otherwise.

The Uncertain Future

In almost every important decision, the outcome of the decision depends not only on
the alternatives the manager chooses but on external events that are not under the
manager’s control. For example, an investor’s return depends on which stocks she owns
T
(her decision) and whether these stocks appreciate or depreciate (an external event); a
farmer’s annual income depends on the crops planted (his decision) and the weather and
market price for the crops (external events); and a publisher’s profits depends on which
books she publishes and how much she spends in promotion (her decision) and consumer
demand for books and the state of the economy (external events).
CO
Probabilities measure the likelihood of uncertain events. If the probability of an
event is 0, then that event is impossible. If an event will happen for sure, then the
probability of that event is 1. Probabilities provide us with a precise numerical language for
communicating judgments about the uncertain future. Often you will hear statements like:
“I think that there is a pretty good chance that our sales will be at an all time high this year,”
or “It’s unlikely that our competitor will change his price in the near future,” or even “I’m
quite unsure about how the government will act on this issue.” These phrases attempt to
PY
communicate a judgment about the uncertain future in a language that is imprecise at best.
In contrast, probabilities require the decision maker to be explicit in her pronouncements:
instead of merely saying that there is a “pretty good” chance, she must give her opinion of
how good a chance.

In our example, Chang must estimate the probability of developing the

microprocessor. Naturally, Chang will want to obtain as much knowledge about the
problem as possible, by conducting statistical analysis, talking to her engineers, reviewing
past projects, and perhaps consulting outside experts. This research forms the input for her
probability judgments.

8
Decision Analysis 894-004

Let’s say that Chang believes that the probability of successfully developing the
microprocessor in six months is .40 or 40 percent. Thus the chance of technological failure is
.60 or 60 percent. In addition, Chang indicates that there is .90 probability that she will win
the proposal if she does develop the microprocessor. Without the microprocessor, she
estimates that she has only a .05 chance of winning. These probabilities are reflected in the
decision tree depicted in Figure 6.
DO
Although these judgments are necessarily subjective and hence not “right” or
“wrong,” they are not arbitrary. The probabilities reflect Sarah Chang’s judgments about
the likelihood of various events. But what does it really mean when Chang says that the
probability of technological success is .40? If there is a bag containing 100 balls, 40 of which
are red, then we all understand that the probability of choosing a red ball out of the bag is
.40. Here the probability is “objective”: If we drew a great many balls from the bag,
replacing the drawn ball each time, about 40% of the balls drawn would be red. Although
the probability of technological success does not have such a straightforward physical
interpretation, it does have a meaning. Basically Chang is saying that she believes that
developing the microprocessor is just as likely as drawing a red ball out of bag in which 40
NO
percent of the balls are red.

Figure 6

Win
Make Proposal .90
Lose
Succeed .10
T
.40
Don't
Make Proposal
Continue Win
Make Proposal .05
CO
Lose
Fail .95
.60
Don't
Make Proposal

Abandon
PY

Endpoints and Monetary Consequences

Even with the decision tree in Figure 6, Sarah Chang could not make a decision
unless she knew the value of the various outcomes. For example, how much better off is
she winning the contract than failing to win it? In this stage of the process, Chang must
evaluate each terminal point in the decision tree. For each path in the decision tree or
scenario, a unique set of choices and events has transpired. For example, in one scenario,

9
894-004 Decision Analysis

Chang decides to continue with the project, her engineers are successful in developing the
new microprocessor, and she is awarded the contract. Presumably this is the best scenario
of all. But how good is it?

Recall that Chang wants to maximize the expected monetary cash flow. Thus Chang
must evaluate the net cash flow that results at each end point. In this step, she should
DO
identify all of the relevant costs incurred in the various activities, as well as the relevant
revenues. If she wins the contract, then Chang will expend $200,000 in development costs,
$50,000 in developing the prototype, and $150,000 in producing the final product. Thus the
total costs to her will be $400,000. Since she wins a contract worth $1 million, the net cash
1
flow is $600,000. If she develops the technology, makes a proposal, but is not awarded the
contract, then the net cash flow would be $-250,000 (the development costs plus the cost of
developing the prototype). The other cash flows can be calculated by subtracting the
relevant costs from the relevant revenues. They are shown for each scenario in Figure 7.

Figure 7
NO
Win
$600,000
Make Proposal .90
Lose
$-250,000
Succeed .10
.40
Don't
Make Proposal
$-200,000
Continue Win
$600,000
T
Make Proposal .05
Lose
$-250,000
Fail .95
.60
Don't
CO
Make Proposal
$-200,000

Abandon
PY
What’s Best: Folding Back the Tree

At this point, Chang has structured the problem and used her judgment to assess the
critical uncertainties and to evaluate the endpoints. Now she is ready to combine these

1
A more detailed analysis would reflect the differential timing of the cash flows and the time value of money.
Thus, the various cash flows would be expressed in terms of their net present value. We will not worry about
these details in this note.

10
Decision Analysis 894-004

judgments to determine the best course of action. We will start at the end points of the
decision tree and work our way backwards to the present. Let’s assume that Chang
decided to continue the venture, that her engineers were successful in developing the
microprocessor technology, and that she went ahead with the proposal. Thus she is at node
(E) in the decision tree. At this juncture, either she will win or lose the contract. Recall that
Chang wants to maximize EMV – expected monetary value. If we weight the “Win” and
DO
“Lose” outcomes by their respective probabilities, we find that the EMV for node (E) is
.90($600,000) + .10($-250,000) or $515,000. Thus we replace the uncertain event at node (E)
with its EMV, $515,000 (Figure 8).

Figure 8

Win
$600,000
Make Proposal .90
E
$515,000 Lose
$515,000 $-250,000
Succeed .10
NO
C
.40
Don't
$86,000 Make Proposal
$-200,000
Continue Win
B $600,000
Make Proposal .05
F
$-200,000 Lose
$-207,500 $-250,000
Fail .95
D
A $86,000 .60
T
Don't
Make Proposal
$-200,000

Abandon
CO
At node (C), Chang has a decision of whether or not to make a proposal. If she makes the
proposal then she can “expect” $515,000. However, if Chang does not make the proposal
then she will lose the $200,000 she expended in development costs. Clearly, she should
make the proposal. Thus at node (C), we substitute the EMV of the better of the two
decisions, “Make proposal.” We can “prune” the inferior option, “Don’t Make,” from the
decision tree. A pruned inferior option is denoted by two short parallel lines.
PY
We now have a procedure for analyzing the entire decision tree:

1. start at the endpoints of the decision tree;

• replace each chance node with that chance node’s EMV (multiply the
probability of each event by that event’s EMV and then sum up)

• replace each decision node with the highest EMV of any of the
alternatives;

11
894-004 Decision Analysis

2. work backwards toward the beginning of the tree.

We call this process of simplifying the decision tree folding back the tree or
backwards induction. Chang can decide what to do initially at decision node (A), because
she knows what she would choose if she were at decision node (C) or at decision node (D).

If we continue this process to the beginning of the tree, we find that Chang should
DO
continue investing in research and development and then make a proposal only if her
company succeeds in developing the microprocessor. The EMV of this strategy is $86,000.
Figure 8 depicts the complete analysis.

Consequences Not Contained in the Cash Flows

In most decisions, there are consequences that are not purely monetary. In these
situations the non-monetary consequences should be evaluated and included in the
NO
analysis. In this example, we have simplified the situation by assuming that developing the
microprocessor technology is valuable only because it increases the probability of winning
the contract. However, having the technology in hand will often have considerable value
outside this simple context: the microprocessor technology may open doors for other
projects in the future, may enhance the reputation of Chang’s company, etc. These benefits
can be quite substantial and should be reflected in the analysis.

In general, if the manager believes that the various end points on her decision
diagram leave her with different prospects for the future or achieve different non-monetary
T
objectives, she should assess the value of these future differences and add them to the cash
flows. Thus, the analysis should reflect any differences in customer relations, employee
morale, equipment, inventories, reputation, etc.

In our simple example, continuing to invest is the best initial choice. If winning the
CO
Olympic contract improves Chang’s chances of attracting venture capital or enhances the
company’s reputation, then these objectives should be included in the analysis. However,
here, no additional analysis is necessary: it is easy to see that these factors just strengthen
the argument for continuing investment in R&D.

Sensitivity Analysis
PY
In a previous section, we discussed how Sarah Chang might think about the
judgmental probabilities. She may indicate that .90 is the probability of winning the
contract if she has the microprocessor technology, but this judgment might be based on
relatively little knowledge or experience. Chang might not feel particularly comfortable
about investing such a large sum of money in this project without giving some more
thought to the probability.

If Chang spent more time thinking about the chance of winning, perhaps reflecting
on what her competitors might be up to, then the probability she assigns to winning might

12
Decision Analysis 894-004

change. On the one hand, if, after further reflection, she decided that the probability of
winning was actually higher than .90, then the decision to continue the project would be
unaffected. On the other hand, the probability of winning might drop. How low could the
probability fall before “Continue” becomes worse than “Abandon”? First, Chang might
choose some lower probability of winning, say .40 and fold back the tree with this new
probability. In this case, she would find that the EMV of “Continue” has become negative,
DO
$-84,000. After some trial and error or some simple algebra, Chang would determine that
the EMV of “Continue” is positive when the probability of winning is greater than .647; it is
negative when the probability is less than .647. Chang might not be able to state precisely
the probability of winning, but she might feel confident that the probability is higher than
.647. In this case, it is not worth investing her time refining her original probability
assessment, because further reflection on the probability of winning would not change her
decision to continue with the project.

Chang determined whether her decision to continue the project was sensitive to
different probabilities of winning. The reasoning used in sensitivity analysis can also be
applied to other judgments a manager might make such as the value of company reputation
NO
or employee morale. Recall that we emphasized that decision analysis is an iterative process.
In the first stage, a decision maker may make a tentative assessment or valuation or a
simplifying assumption in order to reach a preliminary conclusion about which alternatives
to undertake. For example, Chang chose to ignore the non-monetary benefits of having the
microprocessor technology. After folding back the tree, she should determine whether her
decision to continue investing is sensitive to any of the tentative judgments she made
earlier. In this case, any benefits of developing the microprocessor technology only
reinforce her decision to continue the project. On the other hand, if Chang finds that her
T
decision is very sensitive to some other judgment, then she may want to invest some
resources refining this judgment. Thus a smart decision maker can use sensitivity analysis
to identify which judgments need to be considered more carefully and which do not.
CO
Decision Traps

At each step in the process, the manager might make an error in judgment. These
errors in decision making, decision traps, are often systematic.2 In the structuring phase, for
example, managers often focus on the “wrong” problem. A manager might frame the
decision as a tactical choice, when it is in fact a “strategic” decision. Or a production-
oriented company might tend to view all of their decisions through an engineering lens,
even though such a lens may not always be appropriate. Looking at the world in such a
PY
way might focus attention exclusively on efficiency, neglecting concerns such as meeting
the needs of the consumer.

Managers also are not particularly good at estimating probabilities. They tend to be
wildly overconfident in their judgments. When a product manager says that there is only a

2 A longer account of decision traps is found in J.Edward Russo and Paul J.H. Schoemaker (1989), Decision Traps:
The Ten Barriers to Brilliant Decision-making and How to Overcome Them, New York: Simon & Schuster. This book
also coined the term decision trap.

13
894-004 Decision Analysis

10% chance that demand might be as low as 100,000 units, the actual probability might be
30% to 50%. In addition, we tend to be biased in how we weigh evidence, tending to
overweight recent and salient evidence and underweight historical information. For
example, Chang’s estimate of R&D’s success in developing the microprocessor might be
largely based upon the most recent R&D report. She might neglect the stack of historical
R&D reports sitting in her file cabinet, however relevant these reports might be.
DO
Epilogue

Sarah Chang’s decision problem was relatively simple. Most managerial decision
problems are more complicated. The sheer size of these problems can often paralyze a
decision maker. However, decision analysis offers a systematic approach to decision
making, breaking down the decision problem into smaller, more easily digestible pieces. At
each step in the process – structuring, assessment and evaluation, analysis, and sensitivity
NO
analysis – the manager can gain insight into her decision problem. The end result will be
better decisions.
T
CO
PY

14