Previously …

Improving our decision making skills,

taking advantage
Decision Models:
of decision models
Optimization
and technology
Simulation Decision Analysis
Technology: (Excel Solver) (SimVoi) (Treeplan)
Case #1: Red Brand Canners Eagle Airlines Freemark Abbey
Case #2: RBC-2 Dynatron Freemark Abbey-2
Business
Case-based project: Model
Report #1 Analysis
Report #2 Decision
Report #3
Problem

Page1
Decision Frame
Information

Objectives

Alternatives Risks
Options Constraints

Intuition

Page2
Decisions

AW

AJ
AP

BW
BJ

BP

Page3
Excel Decision Support Model

Page4
Key Insights
Intuition and Analysis are both key in tackling decisions

Structuring a problem allows dealing with it systematically

Objectives
Alternatives
Risks and Constraints

Identify relevant information for decision making

Ignore overheads
Ignore sunk costs

Tools can help identify the optimal allocation of resources

Page5
Fundamentals of Optimization
─ Session 2 ─
 Nestle (Switzerland)
▪ Production/storage/distribution of its
product line (e.g., chocolate, food).

▪ How many of each products to produce to

maximize chocolate sales?

▪ Decision Factors:
— How much chocolate to source and
from where (e.g. Brazil)? Where should Robert Wang
they be delivered to?
— Which factories should produce
Crunch and Butterfinger, and in what
quantities?
— How and where should the final
products be stored?

Outcome: Saves $1M/project-person

(avg. duration of each project: 3 months)
Page7
Zara (Spain)
▪ How to distribute limited inventory across all the
stores in a fast-fashion retail network?

▪ Objectives: increase time Zara products spend

on display, minimize unsold items, etc.
▪ Things to consider: Very short product life cycles,
shipments from HQ, transshipments, which
factories to produce from, etc.
Felipe Caro
▪ Outcome: increase sales by 3-4% = $275M in
additional revenues in 2007.

Page8
All about (linear) optimization

• Components of an optimization problem

• Linear class of optimization problems (LP) & write

it up

• Feasible sets and shadow prices

• Binary variables / Mixed Integer LP (or MILP)

Page9
General optimization problems
▪ (1) Maximize/minimize an objective function
— E.g., total profit contributions from tomatoes

▪ (2) by changing decision variables

— E.g., how much A/B tomatoes in which products

▪ (3) subject to a set of constraints

— E.g., those in supply, demand, quality

▪ Example of optimization model: Zara

Felipe Caro
Page10
Linear optimization

. An optimization problem is “linear” if the objectives and constraints involve only

additions, subtraction, and scalar multiplications (i.e., no multiplication between
decision variables).

. Why consider linear optimization?

– The optimal solution is guaranteed to be found
– Many classes of problems are or can be approximated as “linear”
– Simple setting makes for clearer understanding useful concepts (e.g.,
feasible sets, shadow prices)

Page11
Decision variables

AW

AJ
AP

BW
BJ

BP

Page12
RBC’s objective

▪ Maximize Contribution = ConWhole + ConJuice + ConPaste,

— ConWhole = $247 * (AW + BW)

— ConJuice = $198 * (AJ + BJ)
— ConPaste = $222 * (AP + BP)

Page13
RBC’s constraints
▪ Supply constraint:
— AW + AJ + AP <= 600
— BW + BJ + BP <= 2,400

▪ Demand constraint:
— AW + BW <= 14,400
— AJ + BJ <= 1,000
— AP + BP <= 2,000

▪ Quality constraint:
— 9*AW + 5*BW >= 8*(AW + BW)
— 9*AJ + 5*BJ >= 6*(AJ + BJ)
— 9*AP + 5*BP >= 5*(AP + BP)

▪ Nonnegative variables:
— AW, AJ, AP, BW, BJ, BP >=0

Page14
RBC’s problem

▪ Maximize $247 * (AW + BW) + $198 * (AJ + BJ) + $222 * (AP + BP)

Subject to:
— AW + AJ + AP <= 600
— BW + BJ + BP <= 2,400
— AW + BW <= 14,400
— AJ + BJ <= 1,000
— AP + BP <= 2,000
— 9 * AW + 5 * BW >= 8 * (AW + BW)
— 9 * AJ + 5 * BJ >= 6 * (AJ + BJ)
— 9 * AP + 5 * BP >= 5 * (AP + BP)
— AW, AJ, AP, BW, BJ, BP >=0

Page15
RBC’s problem with additional tomatoes

▪ Maximize $247*(AW+aW+BW) + $198*(AJ+aJ+BJ) + $222*(AP+aP+BP) - $255*80

Subject to:
— AW + AJ + AP <= 600
— BW + BJ + BP <= 2,400
— aW + aJ + aP <= 80
— AW + aW + BW <= 14,400
— AJ + aW + BJ <= 1,000
— AP + aP + BP <= 2,000
— 9 * (AW + aW) + 5 * BW >= 8 * (AW + aW + BW)
— 9 * (AJ + aJ) + 5 * BJ >= 6 * (AJ + aJ + BJ)
— 9 * (AP + aP) + 5 * BP >= 5 * (AP + aP + BP)
— AW, AJ, AP, aW, aJ, aP, BW, BJ, BP >=0

Page16
Binary/integer variables

▪ We had examined whether or not to purchase the extra 80,000lbs of A grade tomatoes.
— To do so, we had run two models and compared the two outputs. Could we make the
purchase or not decision as part of a single model?
— Yes, by using the binary (switch) variables.

— As an extension, linear optimization can include binary (0 or 1) or integer-valued

variables. These are called Mixed Integer Linear Program (MILP) and are efficiently
solved for “modest” number of binary/integer variables.
Page17
Example: Additional A grade tomato
Trick: Link the
binary
decision to
the constraint

Page18
Feasible sets & Shadow prices

▪ Feasible set: set of decision variable satisfying all the constraints

▪ Shadow prices: marginal improvement in the objective by increasing the feasible set.
— Having one more unit of a given resource (e.g., having one more unit of Grade A
tomato?)
— Having one more unit of demand (e.g., having 1000lbs more in demand for juice
or paste?)

Page19
Sensitivity Analysis

Page20
Sensitivity Analysis

Page21
Red Brand Canners: Initial Purchase Decision

• Suppose today is May 1st

• Everything else is the same (e.g., tomato costs 18 cents per pound), but
there is uncertainty about demand

• How much tomato crops should RBC purchase? (The allocation decision
of Grade A and B tomatoes across tomato products will follow.)

Page22
Demand Scenario

Whole Tomato Juice Tomato Paste

Scenario Probability
Tomato (lbs) (lbs) (lbs)
High 80% 14,400K 1,000K 2,000K
Low 20% 600K 500K 1,000K

Page23
Decision Frame
Information

Objectives

Alternatives Risks
Options Constraints

Intuition
Page
Sequence of Decisions

How many of which

tomato product Profit High
0.8 to produce?

High

How much tomato

crop to purchase? Demand scenario

Low
0.2 How many of which
tomato product Profit Low
to produce?

Page
High Demand Scenario
• Assume we know the demand state will be high

Page26
Low Demand Scenario
• Assume we know the demand state will be low

Page27
High-Low Scenario
• Assume the order for tomato crop is determined assuming the demand state will be high but demand state
turns out to be low

Page28
Low-High Scenario
• Assume order for tomato crop is determined assuming the demand state will be low but demand state
turns out to be high

Page29
Scenario Analysis
•Decision
Assume we know the demand
Tomato state will
Crop be low
Profit High (80%) Profit Low (20%) Average Profit
Optimal for High 3,636,364 $144,636 -$185,345 $78,640

Optimal for Low 1,818,182 $88,485 $72,318 $85,252

Optimal for High Optimal for Low

$200,000
$150,000
$100,000
$50,000
$0
High Low
-$50,000
-$100,000
-$150,000
-$200,000
-$250,000

Page30
Optimal Decision-Maximizing Average Profit
• Alternative that generates highest average profit

Page31
Optimal Decision-Maximizing Average Profit
• Alternative that generates highest average profit

Page32
Optimal Decision-Maximizing Average Profit
Decision Tomato Crop Profit High (80%) Profit Low (20%) Average Profit
Optimal for High 3,636,364 $144,636 -$185,345 $78,640
Optimal For Low 1,818,182 $88,485 $72,318 $85,252
Optimal Maximizing 2,727,272 $132,727 -$31,439 $99,894
Average Profit

Page33
Key Insights
▪ Linear optimization
— Make sure objectives and constraints involve only +,-,*, and / (excluding “*” or “/”
of variables by other variables).
— If so, Excel Solver finds the optimal solution efficiently. If not, you can’t be sure.
▪ Feasible sets
— Set of decisions which satisfies all the given constraints
— Optimal decision is the best choice within the feasible set
— Increase in the feasible options imply an improvement in the objective
▪ Shadow Prices:
— Marginal improvement in the objective function from adding one more unit of a
given resource.

▪ Binary/Integer variables can be added.

▪ Can incorporate simple uncertainty

Page34
Next Week: Workshop

Next session, you will be presented with a practical business problem in the form of a
case. The case accompanies a set of questions that will put what you’ve learned to
the test! You can begin work immediately when it is released.

You will have our assistance (me + TAs) working through your case questions during
the workshops. You will working with randomly assigned groups. This format will
guide your learning.

On following Tuesday, you will be given access to a quiz on Moodle (opens 10am and
closes on Wednesday 10am). The quiz will ask specific (randomized) questions
pertaining to the answers to your workshops. You must complete the quiz individually.

Page