DSME 4020B

Decision Modeling and Analytics

Spring 2022
Lecture 1

Prof. Sean Zhou

Email: [email protected]
Office: 902, Cheng Yu Tung Building
Tel: 39437812

1
 Course Objectives
• Analyze the business problem situation.
• Formulate quantitative model to represent the
problem situation and assist decision making.
• Solve the problem using Excel spreadsheet.
• Interpret the results and perform sensitivity analysis.
• Explain the usefulness and the limitations of the
modelling approach to solving decision problems.

2
 Couse Outline
• Part 1: Decision Analysis

• Part 2: Optimization

• Part 3: Simulation

3
 Class Schedule
Date Topic Required Reading

Jan. 13 Introduction to Modeling and Decision Analysis I Chapters 1-2 (skip 2.3)

Jan. 20 Decision Analysis II Chapters 13

Jan.27 Decision Analysis III Chapters 13

Feb.10 Introduction to Optimization & Linear Optimization I Chapter 9

Feb. 17 Linear Optimization II Chapter 9

Feb.24 Linear Optimization III Chapter 9

Mar.3 Midterm Exam

Mar.10 Network Models Chapter 10

Mar.17 Integer Optimization Chapter 11

Mar. 24 Nonlinear Optimization Chapter 8

Mar. 31 Monte Carlo Simulation Chapter 14

Apr. 14 Monte Carlo Simulation II Chapter 14

Apr.21 Optimization in Simulation Chapter 15

Apr.28 (TBC) Final Exam

4
 Textbook and References
• Textbook:
- Business Analytics: The Art of Modeling with
Spreadsheets (5th edition), by Stephen G. Powell, Kenneth R.
Baker, Wiley
- E-book available via CUHK library website (but with only a
limited number can be accessed at the same time)
• References:
- Introduction to Management Science, by Bernart W.
Taylor III
- Data, Models, and Decisions: The Fundamentals of
Management Science, by Dimitris Bertsimas, Robert M.
Freund

5
 Grading
• Class participation: 10%
• Homework: 25%
– Every two weeks, there will be one assignment.
• Mid-term exam: 30%
• Final exam: 35%

6
 Teaching Assistant

Name: Zhuoluo Zhang

Office: Room 951, Cheng Yu Tung Building
Phone: 3943-1657
E-mail: [email protected]

7
 Simple Rules
• Be punctual.
• One break, 15 minutes.
• Cell phones should be turned off or silent.
• Participate.
• “No question is stupid in this class”.

8
 Lecture Outline
• Introduction to Model

• Introduction to Decision Analysis

9
Introduction to Model

10
 Introduction to Model
• What is model?
“a simplified representation of reality”
- Maps.
- Architectural models.
- Economic models, e.g., d=a-bp.
- Business models, e.g., Brick-and-mortar vs. clicks business
model, platform-based businesses, etc.
- Decision models: perceive, organize the business logic, and
make a business decision, e.g., optimization models

11
 Introduction to Model
• Model takes many different forms: mental,
visual, physical, mathematical, etc.
• Modelling is the process of creating a
simplified representation of reality and
working with this representation to
understand or control some aspect of the
world

12
 Models in Business and Business
Education
• We refer to people who routinely collect and analyze data,
build and analyze formal (decision) models in their
professional lives as business analyst
• Strong modeling/analytics skills are particularly important
for consultants, as well as for financial analysts, marketing
researchers, entrepreneurs, and others who face
challenging business decisions of real economic
consequence
• Most of the models used in education are highly simplified,
or stylized, in order to preserve clarity. Stylized models are
frequently used to provide insights into qualitative
phenomena, not necessarily to calculate precise numerical
results for practical use.

13
 Benefits of Models in Business
Decision Making
• Model can help you
– Improve business intuition
– Gain managerial insights
– Guide your decision-making
– Make inexpensive errors
– Explore the impossible
– Provide information in a timely manner
– Reduce costs
– ……

14
The Real World and The Model World

15
 Excel Spreadsheets
• A vehicle for working with models.
• We translate a problem into a spreadsheet for solution.
• Get yourself familiar with basic functions of Excel by
reading Chapters 3-4 and practicing the examples there.
• Several spreadsheet add-in packages, i.e., Solver, Data
Analysis, Analytic Solver.
• Be careful when using excel spreadsheet
- Erroneous input.
- Incorrect references in spreadsheet.
- Wrong formula.
- ……

16
 Modeling and Analytics in Business
• Dell
– Pioneered the direct sales model allowing customers to fully configure the
PC they buy.
– Changing market dynamics, technological advancements and Dell's global
expansion required that the company cater to diverse customer needs and
purchase behaviors.
– Around 2014, it starts to offer fixed configurations through multiple
channels
– Develop solutions applying analytics to address key challenges, e.g.,
inventories, procurement, transportation, prices, across the value chain
and deliver profitable growth in the new channels
– Benefits:
• Improved online conversion rate, increased ocean shipment and enhanced
customer satisfaction since 2010.
• delivered a margin impact of more than $140 million through reduction in
markdown expenditure.

17
Modeling and Analytics beyond Business
• Dutch Delta Program Commissioner
– 55% of the Netherlands is susceptible to flood risk.
– Government spends roughly €1 billion on protection
by dikes and dunes.
– In total there are 3,500 kilometres of primary dikes in the Netherlands
– The country’s second Delta Committee recommended increasing all
protection standards by at least a factor of ten, a highly costly step
given limited funds
– Collect data and develop models to evaluate the recommendation
– Benefits
• Demonstrated that it is efficient to limit increased standards to only three
critical regions.
• €7.8 billion less investment costs while strengthening the country’s
defence against major disaster.
18
Modeling and Analytics beyond Business
• Mayo Clinic
– First and largest nonprofit medical group practice, located in Rochester,
Minnesota, U.S.
– More than 4,500 physicians and scientists and 57,100 allied health staff.
– Facing low utilization of their operating rooms (OR), combined with fluctuating
empty days and days with overtime to complete scheduled surgeries.
– Conduct descriptive research and develop predictive model that suggests
more efficient schedules.
– Benefits

Measure Before After

Primetime OR utilization 47% 66%
Days with No Surgery 33% 16%
Days with Overtime 38% 28%

19
 Problem-Solving Process
Explore the mess,
search for information

Identify a problem

Search for solution

(analytics is often
Feedback loops involved)

Evaluate solution
(analytics is often
involved)

Implement a solution

20
 Example: Invivo Diagnostics
• Invivo Diagnostics is a $300M pharmaceutical
company.
• It builds on the strength of a single product
that accounts for over 75% of revenues.
• In 18 months, the patent for this product will
expire.
• The CEO wants to explore ways to plug the
expected $100M-$200M revenue gap as
revenues from this product declines.
21
 Explore the mess, search for
information
• What problems do we face?
• What is the gap between the current one and ideal
one?
• What are the causes?
• What actions are available?
• This stage is complete when we
- provide a description of the current situation
- identify and collect relevant data

22
 Identify a problem
• What is our objective?
• Did we deal with a similar one before?
• How to state the problem (narrow it down)
• This stage is complete when we formulate the problem
– In what ways might we slow the decline in revenues from our
patented drug?
– In what ways might we increase the chances of success of R&D on new
products?
– In what ways might we increase market share for our existing
products?
– In what ways might we resize the firm to match declining profits?
– In what ways might we partner with other firms?
– In what ways might we reduce the time to market for the six drugs
currently under development?

23
 Search for solution
• What alternatives do we have?
• What methods do we use?
• What criteria do we use?
• This stage is complete when we produce a list of
potential solutions
– Hire outside firms to conduct clinical trials and develop
applications for Food and Drug Administration (FDA) approvals.
– Invest a higher percentage of the R&D budget in drugs with the
most promise of winning FDA approval.
– Focus the drug portfolio on drugs in the same medical category.

24
 Evaluate solution
• What factors within our control can improve the
outcome?
• What factors outside our control will alter the
outcome?
• This stage is complete when we produce a list of
recommended actions along justifications.
– R&D cost reduction
– Increase in market share
– Months of development time saved
– Increase in probability of FDA approval

25
 Implement a solution
• What are the barriers to successful implementation?
• Will there be support, or resistance?
• This stage is complete when we execute the solution.

26
Introduction to Decision Analysis

27
 Learning Objective
• Decision making without probability
- Three different criteria to determine the optimal
decision.
• Decision making with probability
- The framework is to use a decision analysis tool
called Decision Trees.

28
 Elements of Decision Analysis
• Although decision making under uncertainty occurs in a wide
variety of contexts, all decision problems have four common
elements:

1. A set of decisions (or strategies, alternatives) available,

2. The state of nature (or events/uncertainties),
3. A set of possible outcomes of state, and
4. A value model that prescribes monetary or utility values
for the various decision-outcome combinations.

• Once these elements are known, the decision maker can

systematically find an optimal decision, depending on the
optimality criterion chosen 29
Decision making without probability

30
Case Example: Entrepreneur's Problem
• Stephen Marts has a new product idea.
• Developing the product will cost $7 million.
• In a good market, the product will yield $10 million (revenue),
in a bad market, the product will yield $5 million.
• He could also sell his idea for $0.75 million to company A
(without incurring the developing cost).
• Alternatively, he could also cooperate with company B. In a
good market (10 million of revenue), company B promises him
a share of 40% in the profit; in a bad market, he will get
nothing.

31
Case Example: Entrepreneur's Problem
• Payoff table
- The rows correspond to alternative actions.
- The columns correspond to possible states.

Good Market Bad Market

Develop 3 -2
Sell to Company A 0.75 0.75
Work with Company B 1.2 0

32
Case Example: Entrepreneur's Problem
Good Market Bad Market
Develop 3 -2
Sell to Company A 0.75 0.75
Work with Company B 1.2 0
• If Stephen knows that the market conditions will be
good, what is his preferred strategy?
• What if the market conditions are bad?
• Choose alternative with the best outcome!
• But now he does not know the exact market condition
33
 Maximax Payoff: Optimistic Approach
1. For each decision, list the best possible outcome
across scenarios
2. Choose the decision that offers the best value

Good Market Bad Market Max

Develop 3 -2
Sell to Company A 0.75 0.75
Work with Company B 1.2 0

34
 Maximax Payoff: Optimistic Approach
1. For each decision, list the best possible outcome
across scenarios
2. Choose the decision that offers the best value

Good Market Bad Market Max

Develop 3 -2 3
Sell to Company A 0.75 0.75 0.75
Work with Company B 1.2 0 1.2

35
 Maximax Payoff: Optimistic Approach
1. For each decision, list the best possible outcome
across scenarios
2. Choose the decision that offers the best value

Good Market Bad Market Max

Develop 3 -2 3
Sell to Company A 0.75 0.75 0.75
Work with Company B 1.2 0 1.2

Develop is the best decision.

36
Maximin Payoff: Pessimistic Approach
1. For each decision, list the worst possible outcome
across scenarios
2. Choose the decision that offers the best value

Good Market Bad Market Min

Develop 3 -2
Sell to Company A 0.75 0.75
Work with Company B 1.2 0

37
Maximin Payoff: Pessimistic Approach
1. For each decision, list the worst possible outcome
across scenarios
2. Choose the decision that offers the best value

Good Market Bad Market Min

Develop 3 -2 -2
Sell to Company A 0.75 0.75 0.75
Work with Company B 1.2 0 0

38
Maximin Payoff: Pessimistic Approach
1. For each decision, list the worst possible outcome
across scenarios
2. Choose the decision that offers the best value

Good Market Bad Market Min

Develop 3 -2 -2
Sell to Company A 0.75 0.75 0.75
Work with Company B 1.2 0 0

Sell to company A is the best decision!

39
 Regret
• Example: when the market is good, the best
payoff is 3. Suppose you choose to sell to
company A
– if the market turns out to be good, you may regret
that you did not choose develop the product and
the loss is 3-0.75=2.25
• Regret is the monetary difference between
the best possible payoff for that state and the
payoff for the specific decision.
40
 Minimax Regret
1. For each state, find the best possible outcome.
2. For each decision, calculate its regret for each state.
3. For each decision, list the largest regret across
states.
4. Choose the decision that offers the lowest regret.

41
 Minimax Regret
1. For each state, find the best possible outcome.
2. For each decision, calculate its regret for each state.
3. For each decision, list the largest regret across states.
4. Choose the decision that offers the lowest regret.

Good Market Bad Market

Develop 3 -2
Sell to Company A 0.75 0.75

Work with Company B 1.2 0

42
 Minimax Regret
1. For each state, find the best possible outcome.
2. For each decision, calculate its regret for each state.
3. For each decision, list the largest regret across states.
4. Choose the decision that offers the lowest regret.

Good Market Bad Market

Develop 3-3=0 0.75-(-2)=2.75
Sell to Company A 3-0.75=2.25 0.75-0.75=0

Work with Company B 3-1.2=1.8 0.75-0=0.75

43
 Minimax Regret
1. For each state, find the best possible outcome.
2. For each decision, calculate its regret for each state.
3. For each decision, list the largest regret across states.
4. Choose the decision that offers the lowest regret.

Good Market Bad Market Max Regret

Develop 3-3=0 0.75-(-2)=2.75 2.75
Sell to Company A 3-0.75=2.25 0.75-0.75=0 2.25

Work with Company B 3-1.2=1.8 0.75-0=0.75 1.8

44
 Minimax Regret
1. For each state, find the best possible outcome.
2. For each decision, calculate its regret for each state.
3. For each decision, list the largest regret across states.
4. Choose the decision that offers the lowest (maximum)
regret.
Good Market Bad Market Max Regret
Develop 3-3=0 0.75-(-2)=2.75 2.75
Sell to Company A 3-0.75=2.25 0.75-0.75=0 2.25

Work with Company B 3-1.2=1.8 0.75-0=0.75 1.8

Work with company B is the best decision.

45
 Summary of Decision Making without
Probability
• In practice, there often are more than two
possible actions and two possible outcomes
• The Maximax payoff criterion seeks the largest of
the maximum payoffs (with respect to outcomes)
among the actions.
• The Maximin payoff criterion seeks the largest of
the minimum payoffs (with respect to outcomes)
among the actions.
• The Minimax regret criterion seeks the smallest
of the largest regrets (with respect to outcomes)
among the actions.
46
 Recap of Lecture 1
• Model is a simplified representation of reality.
• Decision-making/problem-solving process
• Decision making without probability
– Maximax payoff criterion: seeks the largest of the
maximum payoff
– Maximin payoff criterion: seeks the largest of the
minimum payoffs
– Minimax regret criterion: seeks the smallest of the
largest possible regrets

47
Decision making with probability

48
Case Example: Entrepreneur's Problem

Good Market Bad Market

Probability 0.4 0.6

• How do you get the probabilities?

– Subjective judgement, predictive analytics (from
data)
• Incorporate information about the likelihoods
of all the possible states of nature.
49
 Expected Payoff Criterion
1. For each decision, compute the expected payoff
2. Choose the alternative with the best expected payoff

Good Market Bad Market Expected Payoff

Probability 0.4 0.6
Develop 3 -2 3*0.4-2*0.6=0
Sell to Company A 0.75 0.75 0.75

Work with Company B 1.2 0 1.2*0.4=0.48

50
 Expected Payoff Criterion
1. For each decision, compute the expected payoff
2. Choose the alternative with the best value

Good Market Bad Market Expected Payoff

Probability 0.4 0.6
Develop 3 -2 3*0.4-2*0.6=0
Sell to Company A 0.75 0.75 0.75

Work with Company B 1.2 0 1.2*0.4=0.48

Sell to company A is the best decision.

51
 Decision Tree
• A practical problem often involves many
alternatives, uncertainties (state of nature or
event, e.g., market condition), and possible
outcomes
• Often times the decisions/uncertain events
are sequential in time
• Without any tool, a direct approach may be
too complex to help decision maker to solve
the problem and find the best decisions

52
 Decision Tree
• A decision tree is a tree-like graphical tool that can
be used to analyze the decisions and their possible
consequences, including chance event outcomes,
resource costs, and utilities.
• Enable decision maker to have a clear view of all
important aspects of the problem:
– decisions available/alternatives,
– uncertain outcomes of events with their probabilities,
– economic consequences, and
– chronological order of events

53
 Case Example: SciTools
• SciTools Inc., a company that specializes in scientific instruments,
has been invited to bid for a government contract.
• The contract calls for a specific number of these instruments to be
delivered during the coming year.
• The bids must be sealed, so that no company knows what others
are bidding, and the lowest bid wins the contract and gets paid the
bid.
• Cost estimates:
– $5,000 to prepare a bid, and
– $95,000 to supply the instruments if it wins the contract.
• SciTools believes that there is a 30% chance that there will be no
competing bids.

54
 Case Example: SciTools
• In case there is competition, based on past contracts of this
type, possible lowest bids from competitors and associated
probabilities are given below:

Lowest Bid from Competitors Probability

Less than $115,000 0.2=P(<115,000|co
mpetition)
Between $115,000 and $120,000 0.4
Between $120,000 and $125,000 0.3
Greater than $125,000 0.1
P(A and B)
Conditional probability: P(A|B)= 𝑃(𝐵)

55
 Four Factors in A Decision Tree
• Decisions i.e. options/strategies
available to decision maker

• Outcomes i.e. externalities outside

control of decision maker

• Probabilities i.e. risk/uncertainty

level of the outcomes

• Payoffs i.e. rewards accruing to

different decisions & outcomes

56
 Decisions (In SciTools Case)
• SciTools only consider the following options:
– Don’t bid
– Bid
• $115,000
• $120,000
• $125,000

57
 Outcomes (In SciTools Case)
• What may happen?
– No bid from competitors
– Lowest bid from competitors < $115,000
– Lowest bid from competitors is between $115,000
and $120,000
– Lowest bid from competitors is between $120,000
and $125,000
– Lowest bid from competitors > $125,000

58
Probabilities Of Outcomes (In SciTools
Case)
• What will happen (possible outcomes)?
– No bid from competitors
0.3
– Low bid from competitors is <$115,000

– Low bid from competitors is btw $115,000 and $120,000

– Low bid from competitors is btw $120,000 and $125,000

– Low bid from competitors is above $125,000

59
 Payoffs (In SciTools Case)
Actual Outcomes

Competitors’ lowest bid (in $’000)

Payoff Table
No bid <115 115~120 120~125 >125

No bid 0 0 0 0 0
Decisions

115 15 -5 15 15 15
SciTools’ Bid
(in $’000) 120 20 -5 -5 20 20

125 25 -5 -5 -5 25

Probability 0.3 0.14 0.28 0.21 0.07

60
 Simplified Payoffs
Summarized Outcomes

Monetary Value ($’000) Probability

that
SciTools SciTools SciTools
Wins Loses Wins
SciTools’ No Bid 0 0 0.00
Bid
Decisions

(in $’000) 115 15 -5 0.86

120 20 -5 0.58

125 25 -5 0.37

61
 Decision Tree Modeling
• For the same problem, different decision trees may
be used to model it, depending on:
– How events (and thus outcomes) are defined?
– How decisions are defined and sequenced?

62
 Decision Tree For SciTools (V1.1)
Outcome 0.3 15k
Decision Probability
0

0.14
-5k

0.28
15k

Decision 0.21
Node 15k
Payoff
Probability
Node 0.07
15k

63
 Evaluating A Decision Tree
• Decision tree is drawn from left to right
(following the sequence of time).
• Evaluation is done from right to left. We refer
this process as rolling back the tree.
• Each node on the decision tree will have an EV
associated with it after evaluation.

64
 EV
• The expected value, or EV, for a decision is a
weighted average of all possible payoffs for this
decision, weighted by probabilities of the outcomes.
• “Playing the averages”: Choosing the decision with
the largest EV
– EV for each probability node is calculated (see following
example)
– EV for decision node = largest EV amongst all decisions

65
 EV of “Bid $115k” Decision
0.3 15k

0.14
12.2k -5k

0.28 EV=0.3*15k+0.14*(-5k)+0.28*15k
15k +0.21*15k+0.07*15k
=12.2k
0.21
15k

0.07

15k

66
 Max Criterion
in Evaluating a Decision Node
Don’t bid 0

12.2k Maximum among all EVs

12.2k Bid 115k of the probability nodes
Bid 120k 9.5k

6.1k
Bid 125k

67
 Complete Decision Tree 15k
0 0.3
0.14 -5k

12.2k Low bid 115k~120k 0.28 15k

0.21 15k 20k

0.3
0.07 0.14 -5k
15k
12.2k
9.5k
Low bid 115k~120k 0.28 -5k

0.21 20k
0.3 25k
0.07
20k
0.14 -5k

6.1k Low bid 115k~120k 0.28 -5k

0.21 -5k
0.07
25k

68
 Optimal Decision (For SciTools)
• The best strategy for SciTools is to bid
$115,000 for the contract.
• The expected return (or profit) is $12,200.

69
 Decision Tree V1.2
0
0.86 15k
Don’t bid Win
12.2k

12.2k Lose 0.14 -5k

Bid 115k
0.58 20k
9.5k Win

Bid 120k Lose 0.42 -5k

Bid 125k 6.1k Win 0.37 25k

Lose 0.63
-5k

Based on summarized payoffs

70
 Decision Tree V1.3
0.3 0.2 -5k
15k
0 No Competition Low bid
<115k
12.2k
Don’t bid Low bid
0.7 115k~120k 0.4
With 15k
12.2k Bid 115k Competition Low bid
120k~125k
12.2k Bid 9.5k
120k 0.3
Bid 15k
Low bid
Bid 6.1k
>125k

125k
0.1
15k

71
 Decision Tree V1.4
0.3 0.2 -5k
15k
0 No Competition
Lose
12.2k
Don’t bid
0.7 Win 0.8
Has 15k
12.2k Bid 115k Competition

12.2k Bid 9.5k

120k
Bid

Bid 6.1k
125k

72
 Procedure for Constructing Trees
1. Determine decisions and uncertainties
(probabilities).
2. Start the tree with a decision node, representing
the first decision.
3. Place probability nodes that follows the first
decision node.
4. Select possible outcomes for the probability
nodes.
5. Continue to expand the tree with additional
decision nodes and probability nodes until the
overall outcome can be evaluated.
73
Rollback Procedure for Analyzing Trees
1. Start from the last sets of nodes (leaves).
2. For each probability node, calculate its expected value
(EV).
3. Replace each probability node by its expected value.
4. For each decision node, find the best decision
(maximum benefit or minimum cost)
5. Replace each decision node by the best value, and
note which choice is best
6. Repeat 2-5 till the first decision node (root of tree)

74