Supply Chain Management

Class 6
Inventory Management: Newsvendor Model

Sajjad Najafi

1
WHAT DID WE SEE LAST TIME?

2
The EOQ Model

• Demand • Excess Demand

• Constant vs. Variable • None
• Known vs. Random • All orders are backordered
• Continuous vs. Discrete • Lost orders
• Lead Time • Substitution
• Instantaneous • Planning Horizon
• Constant vs. Variable • Single period
• Known vs. Random • Finite vs. Infinite horizon
• Review Time • Number of Items
• Continuous vs. Periodic • One vs. Many
• Capacity
• Constrained vs. Unlimited
3
The EOQ Model
• EOQ is a good starting point when setting up an inventory system
• EOQ is driven by cost minimization
• Order size and average inventory are directly related
• Trade-off between ordering and holding costs
• The holding and ordering costs are equal at the EOQ
• The total cost curve is “flat” near the optimum
Cost
• EOQ is robust
Ordering+Holding
Costs

Lowest Holding Cost =

Cost

Ordering Cost =
Q
4 Q*
The EOQ Model

• Demand • Excess Demand

• Constant vs. Variable • None
• Known vs. Random • All orders are backordered
• Continuous vs. Discrete • Lost orders
• Lead Time • Substitution
• Instantaneous • Planning Horizon
• Constant vs. Variable • Single period
• Known vs. Random • Finite vs. Infinite horizon
• Review Time • Number of Items
• Continuous vs. Periodic • One vs. Many
• Capacity
• Constrained vs. Unlimited
5
This Class
• Incorporate demand variability in the analysis

6
This class: Newsvendor Problem

• Demand • Excess Demand

• Constant vs. Variable • None
• Known vs. Random • All orders are backordered
• Continuous vs. Discrete • Lost orders
• Lead Time • Substitution
• Instantaneous • Planning Horizon
• Constant vs. Variable • Single period
• Known vs. Random • Finite vs. Infinite horizon
• Review Time • Number of Items
• Continuous vs. Periodic • One vs. Many
• Capacity
• Constrained vs. Unlimited
7
The Apparel Industry: Classic Business Process

Purchase Retail Discount

Design Manufact. Distr.
Raw Mat Sales Sales

12 months

8
Profitability & product margins in the apparel industry

Typical product margin (gross margin) Average industry profitability

in 2017: 52.32 % in 2017: 4.16 %

Gross margin Net margin

Source: http://csimarket.com 9
Profitability and product margins in the apparel industry

Profitability by Company within Retail Apparel Industry (4Q 2017)

Company Gross Margin Net Margin

Abercrombie & Fitch co. 58.45% 6.33%
Gap, inc. 36.81% 4.23%
Nordstrom, inc. 36.86% 4.29%
Lands End, inc. 38.86% 7.78%
Urban outfitters, inc. 31.28% 0.12%
 Why is profitability in clothing retail so low?

• Lead times are long

• Fashion leads to high uncertainty
• Forecasting is very imprecise

Firms end up with a lot of excess inventory which must be discounted.

At the same time, many customers do not find what they want.

It is a gamble!

11
So, let’s ask the data analyst to forecast the demand!

12
The data analyst’s answer might be this… (discrete distribution)

13
 …or this (continuous distribution)
Probability Density

Demand
Average (Expected)
Demand = 80 units

14
 …or this (continuous distribution)
Probability Density

Area under the curve

gives me probability

D1 D2
Demand
Average (Expected)
Demand = 80 units

15
 The Usual Fashion Gamble
Probability Density
Perfect forecast
Demand = Order
[utopia! Does not exist in real life!]

Assume we ordered
exactly average demand

Demand
Average (Expected)
Demand = 80 units

16
 The Usual Fashion Gamble
Probability Density
Perfect forecast
Demand = Order
[utopia! Does not exist in real life!]

Assume we ordered
exactly average demand

Demand < Order

Excess Stock
Value loss due to discount
sales when actual demand
is lower than expected

Demand
Average (Expected)
Demand = 80 units

17
 The Usual Fashion Gamble
Probability Density
Perfect forecast
Demand = Order
[utopia! Does not exist in real life!]

Assume we ordered
exactly average demand

Demand < Order Demand > Order

Excess Stock Stockouts
Value loss due to discount Value loss due to missed
sales when actual demand sales when actual demand
is lower than expected is higher than expected

Demand
Average (Expected)
Demand = 80 units

18
 The Usual Fashion Gamble
Probability Density
Perfect forecast
Demand = Order
[utopia! Does not exist in real life!]

Assume we ordered
exactly average demand

Demand < Order Demand > Order

Excess Stock Stockouts
Value loss due to discount Value loss due to missed
sales when actual demand sales when actual demand
is lower than expected is higher than expected

Demand
Average (Expected)
Demand

So how much should we order if both are bad?

19
Newsvendor Problem
• Every morning, a newsvendor purchases newspapers to sell
during the day
• Tradeoffs
• If he purchases too few newspapers, he loses profit on unsatisfied
customers
• If he purchases too many newspapers, he loses profit on unsold
newspapers
• Question: How many newspapers should he purchase?
Profits Just right!

Too little Too much

20 Quantity
Main Tradeoff: too many vs too little

Time
What I order

21
Main Tradeoff: too many vs too little

Time
What I order Sale 1 Sale 2 What is left
! !

22
Main Tradeoff: too many vs too little

Need to Salvage!
(e.g., discount)

Time
What I order Sale 1 Sale 2 What is left
! !

Too Many!

23
Main Tradeoff: too many vs too little

Need to Salvage!
(e.g., discount)

Time
What I order Sale 1 Sale 2 What is left
! !

Too Many!

Time
What I order Sale 1 & 2 Sale 3&4 Lost Sale What is left
! ! 5&6&7
"

24
Main Tradeoff: too many vs too little

Need to Salvage!
(e.g., discount)

Time
What I order Sale 1 Sale 2 What is left
! !

Too Many!

What
Could’ve
been sold
more
Time
What I order Sale 1 & 2 Sale 3&4 Lost Sale What is left
! ! 5&6&7
"
Too Few!

25
too many (overage) vs too little (underage) costs

Need to Salvage!
(e.g., discount)

Time
What I order Sale 1 Sale 2 What is left
! !

Co : Overage Cost (cost of having an excess, over-ordering)

What
Could’ve
been sold
more
Time
What I order Sale 1 & 2 Sale 3&4 Lost Sale What is left
! ! 5&6&7
"
Cu : Underage Cost (cost of having a shortage, under-ordering)

26
 You never guess the demand right!

You always have either shortage or excess inventory!

The question is making which mistake is less costly for you!

27
Newsvendor Model: Example

• Rocky stocks newspapers to sell on the

same day. He pays $0.5 for each newspaper,
and sells for $1.5. How many newspapers
should he stock?

What are the trade-offs? “Rocky”

- If he stocks too few newspapers

- If he stocks too many newspapers

28
Newsvendor Model: Example

• Rocky stocks newspapers to sell on the

same day. He pays $0.5 for each newspaper,
and sells for $1.5. How many newspapers
should he stock?

What are the trade-offs? “Rocky”

- If he stocks too few newspapers
He misses a $1 profit on one lost sale. (underage cost)

- If he stocks too many newspapers

He wastes $0.5 per unsold newspaper. (overage cost)
29
Inventory or Capacity as Option

Value Creation Value Destruction

Positive option value: Optimal capacity Negative option value:

additional capacity is value Zero option value additional capacity is value
creating destroying

30
Solving the Newsvendor Problem
• I pay $0.5 for each paper, and sell for $1.5. So, my profit
margin is $1 per paper.
• Assume the daily newspaper demand distribution:

• If I buy 87 papers, profit = 87

$_____

• Should I buy the 88th? Let’s see...

31
Marginal Analysis: 88th Paper
• With probability _____, the 88th paper will not be sold, and it
costs me $____
• With probability _____, the 88th paper will be sold and brings
me profit of $_____
• Cost < > Benefit

• Should I buy the 88th paper?

32
Marginal Analysis: 88th Paper
0.03 the 88th paper will not be sold, and it
• With probability _____,
costs me $____
0.5
0.97 the 88th paper will be sold and brings
• With probability _____,
me profit of $_____
1

• Cost < > Benefit

0.03 x $0.5 < 0.97 x $1
• Should I buy the 88th paper?

33
Marginal Analysis: 88th Paper
0.03 the 88th paper will not be sold, and it
• With probability _____,
costs me $____
0.5
0.97 the 88th paper will be sold and brings
• With probability _____,
me profit of $_____
1

• Cost < > Benefit

0.03 x $0.5 < 0.97 x $1
• Should I buy the 88th paper? Absolutely!

34
Marginal Analysis: 89th Paper
0.10 the 89th paper will not be sold, and it
• With probability _____,
costs me $____
0.5
0.90 the 89th paper will be sold and brings
• With probability _____,
me profit of $_____
1

• Cost < > Benefit

0.10 x $0.5 < 0.90 x $1
• Should I buy the 89th paper? Definitely yes!

35
Marginal Analysis: 90th Paper
0.30 the 90th paper will not be sold, and it
• With probability _____,
costs me $____
0.5
0.70 the 90th paper will be sold and brings
• With probability _____,
me profit of $_____
1

• Cost < > Benefit

0.30 x $0.5 < 0.70 x $1
• Should I buy the 90th paper? I should certainly buy!

36
Marginal Analysis: 91st Paper
0.70 the 91th paper will not be sold, and it
• With probability _____,
costs me $____
0.5
0.30 the 91th paper will be sold and brings
• With probability _____,
me profit of $_____
1

• Cost < > Benefit

0.70 x $0.5 > 0.30 x $1
• Should I buy the 91th paper? Of course not!
Optimal order
90
= _____

37
Trade-offs in a Newsvendor Model

Probability

Newsvendor problem answers the following questions:

1. What is the optimal order quantity?
2. What is the optimal service level? Service level is the probability that you served
all the customers

38
 Marginal Analysis: Generalized

• Ordering one more unit increases the chance of overage …

Expected Marginal Loss on the (Qth + 1) unit = Co x P(D ≤ Q)

39
 Marginal Analysis: Generalized

• Ordering one more unit increases the chance of overage …

Expected Marginal Loss on the (Qth + 1) unit = Co x P(D ≤ Q)
• … but the benefit/gain of ordering one more unit is the reduction in the chance of underage:
Expected Marginal Benefit on the (Qth + 1) unit = Cu x P(D > Q) = Cu x (1- P(D ≤ Q))

40
 Marginal Analysis: Generalized

• Ordering one more unit increases the chance of overage …

Expected Marginal Loss on the (Qth + 1) unit = Co x P(D ≤ Q)
• … but the benefit/gain of ordering one more unit is the reduction in the chance of underage:
Expected Marginal Benefit on the (Qth + 1) unit = Cu x P(D > Q) = Cu x (1- P(D ≤ Q))
of ordering one more unit
Expected gain or loss

As more units are ordered…

Expected gain … the expected gain of ordering
one more unit decreases.
Net benefit … the expected loss of ordering
of Qth unit Expected loss one more unit increases.

0
0 Qth unit ordered
41
Marginal Analysis: Generalized
• If
Expected Marginal Expected Marginal
Loss of selling Qth+1
≥ benefit of selling Qth+1

Co × P(D ≤ Q) ≥ Cu × P(D > Q)

then, do not buy the Qth+1 unit.
• That is, as long as,
Cu
P(D ≤ Q) ≥ ,
Cu + Co
do not buy the Qth+1 unit.

42 42
 How to Compute Q* Given Critical Ratio (Discrete demand)

• With Discrete Demand:

Q* is the smallest quantity at which

Cu
SL ≥
Cu + Co

• With Normal Demand N(μ, σ):

Newsvendor Recipe: Discrete demand

𝐶𝑢
1. Determine SL* =
𝐶𝑢 + 𝐶𝑜
2. Find the cumulative probabilities, i.e.,

Possible Probability of Cumulative

Values of D happening probability
D1 Prob(D=D1) Prob(D ≤ D1)
D2 Prob(D=D1) Prob(D ≤ D2)
… … …
Dn Prob(D=Dn) Prob(D ≤ Dn)

3. Q* is the smallest quantity Q at which the service level (cumulative

probability) exceeds the optimal service level SL*, i.e.,
Cu
P(D ≤ Q) ≥
Co + Cu
44
Newsvendor Recipe: Discrete demand (Rocky Example)
Cu
1. Determine SL* = =0.67
Co + Cu
2. Find the cumulative probabilities, i.e.,
Possible Probability of Cumulative
Values of D happening probability
87 0.03
88 0.07
89 0.2
90 0.4 Service Level 0.7

91 0.2
92 0.07
93 0.03

3. Q*is the smallest quantity Q at which the service level (cumulative

Cu
probability) exceeds the optimal service level SL*, i.e.,P(D ≤ Q) ≥ C + C
o u
45
Newsvendor Problem: Normal Demand Distribution

46
 How to Compute Q* Given Critical Ratio (Normal demand)

• With Discrete Demand:

Q* is the smallest quantity at which

Cu
SL ≥
Cu + Co

• With Normal Demand N(μ, σ):

Q* is the quantity at which

Cu
SL =
Cu + Co

But, how could we find Q*?

Q* with Normal Demand: Method Using Excel
N(μ, σ)

Q* = norm.inv (SL*, μ, σ)

Or
Cu
Q* = norm.inv ( , μ, σ)
Cu + Co
Q* with Normal Demand: Method without Excel (Standard Normal Table)

Normal N(μ, σ) Standard Normal N(0,1)

μ Mean = 0
σ Std. dev. = 1

μ 0
Q* − μ
z=
σ

Cu
• Using Standard Normal Table find the z-value where P(Z ≤ z) =
Cu + Co
Q* − μ
• Then, find Q* from z =
σ
, or equivalently

Q* = μ + zσ 0
Normal distribution (tutorial)
• All normal distributions are characterized by two parameters:
mean = μ and standard deviation = σ .
• All normal distributions can be transformed to the standard normal that has
mean = 0 and standard deviation = 1.

• For example:
– Let Q be the order quantity, and (μ, σ) the parameters of the normal
demand forecast.
– P {demand is Q or lower} = Prob {the outcome of a standard normal is z
or lower}, where
Q−µ
z= or Q = µ + z × σ
σ
– Look up Prob {the outcome of a standard normal is z or lower} in the
Standard Normal Distribution Function Table.

50
 Example 1:
• Demand has a normal distribution with mean 80 and standard
deviation 20. Assume that Cu = $90 and Co = $10. What is the
optimal order quantity.
Cu
• Optimal service level = = 0.9
Cu + Co Cycle service
level

Q* − μ
Q* z=
σ

• At what z we have P(Z ≤ z) = 0.9 ? z = 1.2819

Q* − μ
z= ⟹ Q* = μ + zσ
σ
Q* = 80 + 1.2816 × 20 = 105.63

Using Excel? Q*=norm.inv(0.9,80,20)=105.63

Newsvendor Recipe: Normal Demand Distribution

𝐶𝑢
1. Determine SL* = 𝐶 + 𝐶
𝑢 𝑜
2. If you have access to Excel, then Q* = norm.inv (SL*, μ, σ) .
3. If not, find the corresponding z-value of the optimal service
level in the standard normal table. Note that, z-value might
be negative. Then, Q* = μ + zσ .
Normal Demand Distribution: Safety Stock

Cycle service level

Probability of stockout

zσ

µ
(average demand) Optimal Order Size:
Q* = µ + zσ

Safety Stock

53
Practice Problem
Purchase cost: $30, Salvage Value: $10, Sales price: $110. Demand in Normally
distributed with average 1000 and standard deviation 200.
• Cu = $80 Shortage Cost
• Co = $20 Leftover Cost
𝐶𝑢
Compute the Critical Ratio =
𝐶𝑢 + 𝐶𝑜
Stock to the level where Probability(No Stockout) = Critical Ratio

Customer Stockout
Service Level probability
80/100=0.8

80% 20%
54
Optimal Inventory
Practice Problem Continued …

Co = $20 Markdown Cost

Cu = $80 Stockout Cost

Optimal Service Level 80/100 = 80%

Average demand: 1000

Standard deviation of demand: 200

Optimal Fashion Gamble:

Q* = μ + zσ
Q* = 1000 + 0.84 × 200 = 1168
55
 How to Compute Q* Given Critical Ratio

• With Discrete Demand:

Q* is the smallest quantity at which

Cu
P(D ≤ Q) ≥
Cu + Co

• With Normal Demand N(μ, σ):

Q* is the quantity at which

Cu
P(D ≤ Q) =
Cu + Co

Step 1: Find the z value corresponding to SL* from the z-table

Step 2: Convert z to the order quantity Q*, Q* = μ + zσ Q* = μ + zσ
Newsvendor Summary

When do we have a newsvendor problem?

• Need to commit to capacity before knowing demand.
• Inventory loses value sharply at the end of a selling period.
• Unmet demand will be lost.

Newsvendor logic determines

• Optimal service level
• Optimal stock level (order quantity)
Next Class
Application of Newsvendor Problem to Service Contexts

58
 Supply Chain Management

Class 7
Inventory Management: Newsvendor Model
(Continued)

Sajjad Najafi

1
WHAT DID WE SEE LAST TIME?

2
The EOQ Model

• Demand • Excess Demand

• Constant vs. Variable • None
• Known vs. Random • All orders are backordered
• Continuous vs. Discrete • Lost orders
• Lead Time • Substitution
• Instantaneous • Planning Horizon
• Constant vs. Variable • Single period
• Known vs. Random • Finite vs. Infinite horizon
• Review Time • Number of Items
• Continuous vs. Periodic • One vs. Many
• Capacity
• Constrained vs. Unlimited
3
Incorporate demand variability in the analysis

4
This class: Newsvendor Problem

• Demand • Excess Demand

• Constant vs. Variable • None
• Known vs. Random • All orders are backordered
• Continuous vs. Discrete • Lost orders
• Lead Time • Substitution
• Instantaneous • Planning Horizon
• Constant vs. Variable • Single period
• Known vs. Random • Finite vs. Infinite horizon
• Review Time • Number of Items
• Continuous vs. Periodic • One vs. Many
• Capacity
• Constrained vs. Unlimited
5
A Useful Demand Distribution: Normal Demand Distribution

6
A Useful Demand Distribution: Normal Demand Distribution

Real demand1

7
A Useful Demand Distribution: Normal Demand Distribution

Real demand1 Real demand2

So how much should we order if both are bad?

8
too many (overage) vs too little (underage) costs

Need to Salvage!
(e.g., discount)

Time
What I order Sale 1 Sale 2 What is left
! !

Co : Overage Cost (cost of having an excess, over-ordering)

What
Could’ve
been sold
more
Time
What I order Sale 1 & 2 Sale 3&4 Lost Sale What is left
! ! 5&6&7
"
Cu : Underage Cost (cost of having a shortage, under-ordering)

9
Newsvendor Problem
• Every morning, a newsvendor purchases newspapers to sell
• Too Many - Too Few Tradeoffs (overage - underage tradeoff)
• How many newspapers should he purchase?
• You can never guess demand right!
• You always have either excess or shortage inventory. Which
mistake is less costly?
• You order a quantity to make sure that the more costly
mistake happens with lower probability
Profits Just right!

Too little Too much

10 Quantity
 “Marginal” Approach:

Ordering one more unit: p = Prob (Demand ≤ Q)

• Increases chance of excess:

Expected loss on one more unit = Co x p

11
 “Marginal” Approach:

Ordering one more unit: p = Prob (Demand ≤ Q)

• Increases chance of excess:

Expected loss on one more unit = Co x p

• But reduces chance of shortages:

Expected gain on one more unit = Cu x (1-p)

12
 “Marginal” Approach:

Ordering one more unit: p = Prob (Demand ≤ Q)

• Increases chance of excess:

Expected loss on one more unit = Co x p

• But reduces chance of shortages:

Expected gain on one more unit = Cu x (1-p)

Order until: Co × p ≥ Cu × (1 − p)

expected leftover cost expected shortage cost

Cu Critical Service
Or: p ≥ Fractile
or
Level
C o + Cu
13 13
Newsvendor Recipe: Discrete demand

𝐶𝑢
1. Determine SL* =
𝐶𝑢 + 𝐶𝑜
2. Find the cumulative probabilities, i.e.,

Possible Probability of Cumulative

Values of D happening probability
D1 Prob(D=D1) Prob(D ≤ D1)
D2 Prob(D=D1) Prob(D ≤ D2)
… … …
Dn Prob(D=Dn) Prob(D ≤ Dn)

3. Q* is the smallest quantity Q at which the service level (cumulative

probability) exceeds the optimal service level SL*, i.e.,
Cu
P(D ≤ Q) ≥
Co + Cu
14
Newsvendor Recipe: Normal Demand Distribution

𝐶𝑢
1. Determine SL* = 𝐶 + 𝐶
𝑢 𝑜
2. If you have access to Excel, then Q* = norm.inv (SL*, μ, σ) .
3. If not, find the corresponding z-value of the optimal service
level in the standard normal table. Note that, z-value might
be negative. Then, Q* = μ + zσ .
Newsvendor Summary

When do we have a newsvendor problem?

• Need to commit to capacity before knowing demand.
• Inventory loses value sharply at the end of a selling period.
• Unmet demand will be lost.

Newsvendor logic determines

• Optimal service level
• Optimal stock level (order quantity)
 How to Compute Q* Given Critical Ratio

• With Discrete Demand:

Q* is the smallest quantity at which

Cu
P(D ≤ Q) ≥
Cu + Co

• With Normal Demand N(μ, σ):

Q* is the quantity at which

Cu
P(D ≤ Q) =
Cu + Co

Step 1: Find the z value corresponding to SL* from the z-table

Step 2: Convert z to the order quantity Q*, Q* = μ + zσ Q* = μ + zσ
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

SL*

SL*
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

SL*

SL*
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

SL*

SL*
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)
SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
140 150 160 170 180 190 200 210 220 230 240 250 260

SL*
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)
SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
z = − 0.553
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189

SL*
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)
SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
z = − 0.553
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189
• Example 2: Assume demand is N(850,150)

Cu = $90 Co = $20

SL*
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)
SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
z = − 0.553
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189
• Example 2: Assume demand is N(850,150)

Cu = $90 Co = $20

More costly mistake! Most afraid of this!

SL*

P(D > Q*)

350 425 500 575 650 725 800 875 950 1025 1100 1175 1250
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)
SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
z = − 0.553
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189
• Example 2: Assume demand is N(850,150)

Cu = $90 Co = $20

More costly mistake! Most afraid of this!

$90 SL*
SL* = P(D ≤ Q*) = = 82 %
$90 + $20
z = 0.915 P(D > Q*)
Q* = μ + zσ = 988 350 425 500 575 650 725 800 875 950 1025 1100 1175 1250
Newsvendor Model Performance Measures

For any order quantity we would like to evaluate the following

performance measures:

– Expected shortages / lost sales

The average number of units demand exceeds the order quantity.
– Expected sales
The average number of units sold.
– Expected leftover inventory
The average number of units left over at the end of the season.
– Expected profit

26
Expected Shortage / Lost Sales: Graphical explanation

(Loss = 0) · (Prob D = 120)

0.12 +
…
0.10 +
(Loss = 20) · (Prob D = 140)
0.08
+
0.06
…
0.04 +
0.02 (Loss = 80) · (Prob D = 200)

0.00
0 20 40 60 80 100 120 140 160 180 200 220
Q = 120

∫D
Exp. Lost Sales = (x − Q) ⋅ Pr[D = x] dx = L(z) ⋅ σ

standard normal loss function

Look up in table, or use Excel: L(z) = Normdist(z,0,1,0) – z*(1-Normsdist(z))

Practice Problem (from previous session)
Purchase cost: $30, Salvage Value: $10, Sales price: $110. Demand in Normally
distributed with average 1000 and standard deviation 200.
• Cu = $80 Stockout Cost
• Co = $20 Markdown Cost
𝐶𝑢
Compute the Critical Fractile (Target Cycle Service Level) =
𝐶𝑢 + 𝐶𝑜
Stock to the level where Probability(No Stockout) = Critical Fractile

Customer Stockout
Service Level probability
80/100=0.8

80% 20%
28
Optimal Inventory
Example continued

Co = $20 Markdown Cost

Cu = $80 Stockout Cost

Target Cycle Service Level 80/100 = 80%

Average demand: 1000

Standard deviation of demand: 200

Optimal Fashion Gamble:

Q* = μ + zσ

Q* = 1000 + 0.84 × 200 = 1168

29
 Shortage From table (provided in exam with formula sheet)

Exp. Shortages (Lost Sales)

= L(z) x σ
= 0.1116 x 200 = 22.32

(Do NOT round too much!

These are averages!!)
Calculating the expected profits - Example Continued

C = $30, S = $10, P = $110.

Demand: μ = 1000 and σ = 200.
Then, SL=80%; Q*=1168. What is the expected profit?

Exp. Shortages = L(z) x σ = 22.32

Exp. sales = exp. demand – exp. shortages = 1000 – 22.32 = 977.68
Exp. leftovers = order size – exp. sales = 1168 – 977.68 = 190.32

Then, Exp. profits = (P × exp. sales + S × exp. leftovers) – C × Q*

= 110 * 977.68 + 10 * 190.32 – 30 * 1168 = 74,408
Newsvendor Recipe (with performance measures):
Normal Demand Distribution
Finding the profit maximizing order quantity:
𝐶𝑢
1. Determine SL* =
𝐶𝑢 + 𝐶𝑜

2. If you have access to Excel, then Q* = norm.inv (SL*, μ, σ).

3. If not, find the corresponding z-value to the optimal service level in the standard
normal table. Then, Q* = μ + zσ.

For Performance Measures Calculations:

1. Expected Shortages = L(z) x σ (Find L(z) from L(z) table which is given to you)
2. Expected sales = exp. demand – exp. shortages
3. Expected leftovers = order size – exp. sales
4. Expected profits = (P × exp. sales + S × exp. leftovers) – C × Q*

32
Fashion Products: A good forecast is always a distribution

Demand Probability of happening with Probability of happening with

Price $100 Price $120

600 .3 .6
1200 .4 .3
2400 .3 .1

Expected demand 1380 960

Fashion Products: A good forecast is always a distribution

Demand Probability of happening with Probability of happening with

Price $100 Price $120

600 .3 .6
1200 .4 .3
2400 .3 .1

Expected demand 1380 960

Elasticity of demand or demand function

Fashion Products

35
Expected profit calculations

Purchasing cost = $35/sweater

Salvage value = $15/sweater

1.a) Suppose selling price $100 & Order Size = 1200

If Demand = 600 (happens with 30% chance):

Profit Contribution
=

36
Expected profit calculations

Purchasing cost = $35/sweater

Salvage value = $15/sweater

1.a) Suppose selling price $100 & Order Size = 1200

If Demand = 600 (happens with 30% chance):

Profit Contribution
= Revenues – Variable Costs
= (600 x $100 + 600 x $15) – (1200 x $35)
= $27,000

37
Fashion Products

27K

38
Fashion Products

27K

78K

78K

39
Fashion Products

27K

62700 78K

78K

40
Expected profit calculations

If demand = 1200:
Profit Contribution = 1200 x $100 - 1200 x $35
= $78,000
If demand = 2400: = $78,000

Expected profit contribution (price $100, batch size 1200)

= 0.3 x (profit contribution if demand = 600)
+ 0.4 x (profit contribution if demand = 1200)
+ 0.3 x (profit contribution if demand = 2400)
= $62,700

41
Fashion Products

39K

39000 39K

39K

27K

69300 62700 78K

78K
69300
3K

69300 54K

156K

42
Fashion Products

39K

39000 39K

39K

Is there a faster way? 27K

69300 62700 78K

78K
69300
3K

69300 54K

156K

43
Newsvendor Recipe: Frequency-based (Discrete) demand

Finding the profit maximizing order quantity:

𝐶𝑢
1. Determine Target Cycle Service Level =
𝐶𝑢 + 𝐶𝑜
2. Find the cumulative probabilities
3. Optimal order Q* is the smallest Q whose cumulative probability exceeds
the service level, or
Cu
Pr[D ≤ Q*] ≥
Co + Cu

For Profit Calculations:

1. Expected sales = exp. demand – exp. shortages
2. Expected leftovers = order size – exp. sales
3. Expected profits = (P × exp. sales + S × exp. leftovers) – C × Q*

44
Back to Fashion Buying Example

P = $100
Cu =
C = $35
Co =
S = $15

Demand Prob. Cum.

600 0.3 0.3
1200 0.4 0.7 Cu
=
2400 0.3 1.0 Cu + C o

Cu
Order until (or set to smallest Q so that): Pr[Demand ≤ Q] ≥
Co + Cu
45
Back to Fashion Buying Example

P = $100
Cu = $100 – $35 = $65 Most afraid of this!
C = $35
Co = $35 – $15 = $20
S = $15

Demand Prob. Cum.

600 0.3 0.3
1200 0.4 0.7 Cu 65
= = 0.76
2400 0.3 1.0 Cu + Co 65 + 20

Cu
Order until (or set to smallest Q so that): Pr[Demand ≤ Q] ≥
Co + Cu
46
Newsvendor Model Performance Measures

For any order quantity we would like to evaluate the following

performance measures:

– Expected shortages / lost sales

The average number of units demand exceeds the order quantity.
– Expected sales
The average number of units sold.
– Expected leftover inventory
The average number of units left over at the end of the season.
– Expected profit

47
Performance Measures
Fashion Example: Order Size = 1200 & P=$120

Price $120 Demand Sales Shortages Leftovers

.6 600
.3 1200
.1 2400

Expected
960
Value

48
Performance Measures
Fashion Example: Order Size = 1200 & P=$120

Price $120 Demand Sales Shortages Leftovers

.6 600 600 0 600

.3 1200 1200 0 0
.1 2400 1200 1200 0

Expected
960 840 120 360
Value

Expected Sales = Average Demand – Expected Shortages

Expected Leftovers = Order Size – Expected Sales
49
Performance Measures
Fashion Example: Order Size = 1200 & P=$120

Price $120 Demand Sales Shortages Leftovers

.6 600 600 0 600

.3 1200 1200 0 0
.1 2400 1200 1200 0
Both are Positive!!!
Expected
960 840 120 360
Value

Expected Sales = Average Demand – Expected Shortages

Expected Leftovers = Order Size – Expected Sales
50
Service Level vs Fill Rate
Fill Rate

• Cycle Service Level

probability of NOT stocking out in an order cycle

Cares about the Frequency of Stock-outs

• Expected Fill Rate

percentage of demand filled immediately from inventory
Mean Demand − Exp. Shortages Exp. Sales
Fill Rate = =
Mean Demand Mean Demand

Cares about the Magnitude of Stock-outs

52
Cycle Service Level vs Fill Rate cycle service level reflects the frequency of stockouts the percentage of cycles where
stockouts occur

fill rate reflects the amount stocked out, the percentage of demand not met

Product inventory

53
What is the Fill Rate given a minimum in-stock probability (SL)

• Suppose we wish to find the order quantity for the Sweater 501 while
targeting at least a 99% in-stock probability (SL). Demand is normally
distributed with mean demand 3192, and std. dev 1181. What would
be your expected fill rate given the targeted in-stock probability?

54
What is the Fill Rate given a minimum in-stock probability (SL)

• Suppose we wish to find the order quantity for the Sweater 501 while
targeting at least a 99% in-stock probability (SL). Demand is normally
distributed with mean demand 3192, and std. dev 1181. What would
be your expected fill rate given the targeted in-stock probability?

• Step 1:
– Find the z-statistic that yields the target in-stock probability.
– z = 2.33

• Step 2:
– Convert the z-statistic into an order quantity for the actual demand distribution.
– Q=µ+zxσ
= 3192 + 2.33 x 1181 = 5944
55
Calculating the expected fill rate
Exp. Demand − Exp. Shortages Exp. Sales
Fill Rate = =
Exp. Demand Exp. Demand

Need to know Exp. Shortage to calculate Fill Rate.

• Expected shortages. = L(z) x σ = L(2.33) x 1118 = 0.003389 x 1181 = 4.0

• Expected sales = exp. demand – exp. shortages

= 3192 - 4 = 3188

• Expected fill rate = expected sales / mean demand

= 3188 / 3192
= 99.9 %
56
Next Class
Revenue Management

57
 Supply Chain Management

Class 8
Revenue Management

Sajjad Najafi

1
Projects
Goal: To apply the principles of Operations and Supply Chain Management that you
learn in the class to a real-world business situation:
• Study the current processes;
• Identify 1 or 2 challenges/issues/problems related to Operations or SCM;
• Analyze them;
• Provide improving suggestions using the concepts and tools studied in class.

Report: Your report is your slides, and everything included (pictures, videos, etc.).

Evaluation: Based on the following:

• Description of the company and its industry;
• Problem definition and the description of the current process;
• Analysis of the current process;
• Suggestions improving the current process;
• Presentation of the project and report.
2
Project Proposal Reminder
Proposal: Submit a one-page description, due on 14/3 end of day in France
(in PDF to the Dropbox folder that will be shared), including:
• Names of team members
• Name of company and a short description of company and its industry
• Name and position of the contact person and date of at least one
confirmed meeting time
• Rough project plan (schedule)

3
What did we do last time?
Newsvendor Problem
• Every morning, a newsvendor purchases newspapers to sell
• Too Many - Too Few Tradeoffs (overage - underage tradeoff)
• How many newspapers should he purchase?
• You can never guess demand right!
• You always have either excess or shortage inventory. Which
mistake is less costly?
• You order a quantity to make sure that the more costly
mistake happens with lower probability
Profits Just right!

Too little Too much

5 Quantity
 How to Compute Q* Given Critical Ratio

• With Discrete Demand:

Q* is the smallest quantity at which

Cu
P(D ≤ Q) ≥
Cu + Co

• With Normal Demand N(μ, σ):

Q* is the quantity at which

Cu
P(D ≤ Q) =
Cu + Co

Step 1: Find the z value corresponding to SL* from the z-table

Step 2: Convert z to the order quantity Q*, Q* = μ + zσ Q* = μ + zσ
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

SL*

SL*
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)
SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
z = − 0.553
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189

SL*
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)
SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
z = − 0.553
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189
• Example 2: Assume demand is N(850,150)

Cu = $90 Co = $20

More costly mistake! Most afraid of this!

SL*

P(D > Q*)

350 425 500 575 650 725 800 875 950 1025 1100 1175 1250
Examples
• Example 1: Assume demand is N(200,20), purchase cost = $250, selling price =
$350, salvage value = 0.

Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)
SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
z = − 0.553
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189
• Example 2: Assume demand is N(850,150)

Cu = $90 Co = $20

More costly mistake! Most afraid of this!

$90 SL*
SL* = P(D ≤ Q*) = = 82 %
$90 + $20
z = 0.915 P(D > Q*)
Q* = μ + zσ = 988 350 425 500 575 650 725 800 875 950 1025 1100 1175 1250
Newsvendor Model Performance Measures

For any order quantity we would like to evaluate the following

performance measures:

– Expected shortages / lost sales

The average number of units demand exceeds the order quantity.
– Expected sales
The average number of units sold.
– Expected leftover inventory
The average number of units left over at the end of the season.
– Expected profit

11
Newsvendor Recipe (with performance measures):
Normal Demand Distribution
Finding the profit maximizing order quantity:
𝐶𝑢
1. Determine SL* =
𝐶𝑢 + 𝐶𝑜

2. If you have access to Excel, then Q* = norm.inv (SL*, μ, σ).

3. If not, find the corresponding z-value to the optimal service level in the standard
normal table. Then, Q* = μ + zσ.

For Performance Measures Calculations:

1. Expected Shortages = L(z) x σ (Find L(z) from L(z) table which is given to you)
2. Expected sales = exp. demand – exp. shortages
3. Expected leftovers = order size – exp. sales
4. Expected profits = (P × exp. sales + S × exp. leftovers) – C × Q*

12
Service Level vs Fill Rate
Fill Rate

• Cycle Service Level

probability of NOT stocking out in an order cycle

Cares about the Frequency of Stock-outs

• Expected Fill Rate

percentage of demand filled immediately from inventory
Mean Demand − Exp. Shortages Exp. Sales
Fill Rate = =
Mean Demand Mean Demand

Cares about the Magnitude of Stock-outs

14
Cycle Service Level vs Fill Rate

Product inventory

15
This Class
• Learn what revenue management is and why it is important

• Understand how to balance underage and overage costs in the

context of capacity and revenue management

• Two applications: (1) Booking Limits; (2) Overbooking

16
What is common to airlines and hotels?
Airline Seats

How do you decide who pays what and when?

Airline Seats
Hotel Rooms

How do you decide who pays what and when?

What is Revenue Management
• Price-based RM:
• A strategy that adjusts prices based on product availability, customer
demand, and remaining duration of the sales period.
• Example: Rental cars, airline prices, ride sharing (Uber)

• Capacity-based RM:
• A strategy for allocating limited resources to different customer
segments at different price points.
• Example: Booking classes in flights, early reservations in hotels,
stadium tickets

• Today: two applications of Capacity-based RM:

• Booking Limits
• Overbooking
Revenue Management: Booking Limit
Consider the following situation:
• You manage a hotel with 200 rooms in Paris. The PSG game against
Barcelona is coming up. Sports fans either book a room in advance or
wait until the last minute.
• Advance booking: Bargain rate $200/night
• Late booking: Premium rate $500/night

• Demand is ample – you can easily sell out all rooms offered at the
bargain rate.

• Should you fill up rooms with advance customers? Why? Why not?
• How many rooms should be reserved for customers arriving late
(protection level)?
− How should you think about this problem?
22
Booking Limit and Newsvendor

Newsvendor Problem Room Reservation

Newspapers are “perishable” Hotel rooms are “perishable”
Random newspaper demand Random last-minute demand

Decide how many newspapers to Decide how many rooms to reserve

purchase for last-minute customers

If stock too few newspapers, miss If reserve too few rooms, miss
potential sales potential premium customers

If stock too many newspapers,

If reserve too many rooms, revenue
money wasted on unsold
lost on empty rooms
newspapers

23
Revenue Management
• Booking Limit
• Assume the number of last minute customers is normally distributed
with mean 75 and standard deviation 25
• Advance booking: Bargain rate $200/night
• Late booking: Premium rate $500/night

• How many rooms should be set aside for last-minute customers?

• Co =
• Cu =

24
Revenue Management
• Booking Limit
• Assume the number of last minute customers is normally distributed
with mean 75 and standard deviation 25
• Advance booking: Bargain rate $200/night
• Late booking: Premium rate $500/night

• How many rooms should be set aside for last-minute customers?

• Co = $200 (cost of an excess: missing early arrivals)
• Cu = $500-$200 = $300 (cost of a shortage: missing additional profit from
late arrivals)

25
Revenue Management
• Booking Limit
• Assume the number of last minute customers is normally distributed
with mean 75 and standard deviation 25
• Advance booking: Bargain rate $200/night
• Late booking: Premium rate $500/night

• How many rooms should be set aside for last-minute customers?

• Co = $200 (cost of an excess: missing early arrivals)
• Cu = $500-$200 = $300 (cost of a shortage: missing additional profit from
late arrivals)

Critical ratio = 300 / (200+300) = 0.6,

z(0.6)=0.25
You should reserve 75+0.25x25 = 81.25 rooms for last minute customers
(Round up to 82 rooms)
26
When to Use Revenue Management?
• Booking limit is the maximum capacity to be sold at the lower
price

• Business characteristics that are conducive to the application of

booking limits?
• Distinguishable customer segments with different price sensitivities
• Customers with higher valuation arrive later
• Capacity is fixed
• Capacity is perishable
• Capacity can be sold in advance

• Applications: Hotels, Airlines, Car Rentals, etc.

27
 Booking Limit: Multiple Fare Classes

• The booking limit is the number of airline seats you are willing to sell in a fare
class or lower.
• The protection level is the number of airline seats you reserve for a fare class
or higher.
• Let Q be the protection level for the high fare class.
• Q is in effect while you sell low fare tickets.
• With two fare classes, the booking limit on the low fare class is 100 – Q:
– You will sell no more than 100 – Q low fare seats because you are
protecting (or reserving) Q seats for high fare customers.
0 100

Sell no more than the low Q seats protected for

fare booking limit, 100 - Q high fare customers
28
 Discount Seat Allocation
Three Fare classes: Full, Medium, Low Fare

Nested Control

Low Fare

100 Deep Discount Fare,

seats Medium Medium Fare & Full Fare
Medium Fare
Fare
& Full Fare

No. of seats
Full Fare Protected for
Full fare

Protection levels
29
Booking Limits (Discounted Fares) – Example

An aircraft has 100 seats, and there are two types of fares: full ($500) and
discount ($100). Although there is unlimited demand for the discount fare,
demand for full fare is estimated to be equally likely anywhere between 11
and 30. How many seats should be protected for full-fare passengers
booking at the last minute?

30
Example continued …

Demand of discounted Demand of full

fare customers fare customers

Shortage

Protection Level Flight takes Excess

decision off

Seats reserved for

FULL fare customers

31
Example continued …

Demand of discounted Demand of full

fare customers fare customers Protected too few seats;
loose additional margin
of full fare customers

Shortage Cu = 500 – 100 = 400

Shortage:

Protection Level Flight takes Excess: Co = 100

decision off Protected too many
seats; loose margin
of low fare customers
Seats reserved for
FULL fare customers

32
Example continued …

Demand Probability Cumulative

11 0.05 0.05
12 0.05 0.10
13 0.05 0.15
.
.
25 0.05 0.75
26 0.05 0.80
27 0.05 0.85
28 0.05 0.90
29 0.05 0.95
30 0.05 1.00

33
Example continued …

Cu 400
= = 0.8 Interpretation ?
Cu + C 0 500
Demand Probability Cumulative
11 0.05 0.05
12 0.05 0.10
13 0.05 0.15
.
.
25 0.05 0.75
26 0.05 0.80
27 0.05 0.85
28 0.05 0.90
29 0.05 0.95
30 0.05 1.00

Answer: 26 seats should be protected for full-fare customers.

Or equivalently, a booking limit of 100-26 = 74 seats should be applied for
discount fare reservations.
34
How airlines decide what you pay?

https://youtu.be/-oJlJ5oo5AM?t=87
Revenue management challenges…

36
Revenue management challenges…

• Demand forecasting.
– Wealth of information from reservation systems but there is seasonality, special
events, changing fares, and truncation of demand data.
– Customer behaviour (sell-up, switching between flights, ...)
• Dynamic decisions.
• Substitutable capacity:
– Different seat types
• Group reservations.
• Multi-leg passengers/multi-day reservations for cars and hotels:
– Not all customers using a given piece of capacity (a seat on a flight leg, a room for one
night) are equally valuable.
• How to construct good “fences” to differentiate among customers?
– One-way vs round-trip tickets.
– Saturday-night stay requirement.
– Non-refundability.
– Advanced purchase requirements.

37
Messing with Airlines: Hidden City Flight

38
Messing with Airlines: Hidden City Flight

https://money.cnn.com/2015/05/01/investing/united-airlines-lawsuit-skiplagged/index.html 39
Some Background
• A typical airline operates with 73% of its seats filled but needs to fill
70% of its seats to break even

• On a 100-seat aircraft, the difference between making and losing

money is just a handful of passengers.

40
Overbooking is a common industry practice

41
Overbooking - United Airlines

https://www.youtube.com/watch?v=u_WIB527f-I
42
Overbooking
• You are managing the hotel in Paris (described before) during Tour de
France. Suppose you cannot charge latecomers a different rate
because of a city ordinance that prohibits “price gouging” during the
Final Four weekend.
• You can easily get 200 reservations and fill up your hotel
• But, you are afraid that some people with reservations may not show
up

• In this scenario, how many reservations should you take?

• More than 200?
• How many more?
• What information would you need to make this decision?

− How should you think about this problem?

43
Overbooking and Newsvendor

Newsvendor Problem Room Reservation

Newspapers are “perishable” Hotel rooms are “perishable”
Random newspaper demand Random no-shows

Decide how many newspapers to

Decide how many overbooks
purchase

If stock too few newspapers, miss If too few overbooks, miss

potential sales potential sales

If stock too many newspapers,

If too many overbooks, incur loss of
money wasted on unsold
good-will
newspapers

44
Overbooking
• Assume the number of no-shows is distributed as follows:

• The fixed room rate is $200 /night, non-refundable

• When a customer is denied to stay at the hotel due to overbooking,
you transfer the customer to another hotel which costs you $500.
• How many reservations should you take?
• Co =
• Cu =

45
Overbooking
• Assume the number of no-shows is distributed as follows:

• The fixed room rate is $200 /night, non-refundable

• When a customer is denied to stay at the hotel due to overbooking,
you transfer the customer to another hotel which costs you $500.
• How many reservations should you take?
• Co = $500 - $200 = $300
• Cu = $200

46
Overbooking
• Assume the number of no-shows is distributed as follows:

• The fixed room rate is $200 /night, non-refundable

• When a customer is denied to stay at the hotel due to overbooking,
you transfer the customer to another hotel which costs you $500.
• How many reservations should you take?
• Co = $500 - $200 = $300
• Cu = $200
! Critical ratio = 200 / (200+300) = 0.4
Therefore, you should overbook 4 rooms
47
Practice Problem 1
• CheriFM is a radio station that has 25 thirty-second advertising slots
during each evening. It is early July and the station is selling
advertising for July 15th world cup final game in the evening after the
game.
• The station can sell all slots now for $4000 each. But if France wins the
game on July 15th, there will be an opportunity to sell slots on that
day for $10,000.
• How many slots should ROSS Radio save for the last minute? Last
minute demand is distributed according to the following table:

Slots 9 10 11 12 13 14 15
Prob 0.05 0.10 0.15 0.20 0.30 0.10 0.10

48
Practice Problem 1 - Solution
• Not reserve enough slots for last minute
• Cu=$10,000-$4000=$6000

• Reserve too many slots for last minute

• Co=$4000

• Critical ratio = Cu/(Cu+Co) = 0.6

Slots 9 10 11 12 13 14 15
Prob 0.05 0.10 0.15 0.20 0.30 0.10 0.10

49
Practice Problem 2
• HEC admissions office receives a large number of applications each
year and needs to decide how many offers to make. Since some
students will decide to pursue other opportunities, the office will
admit more than ideal class size 560.
• It is estimated that in the upcoming year, the number of people who
will not accept the offer is normally distributed with mean 40 and
standard deviation 20.

• (a) Suppose 560 were admitted, what is the probability that the
incoming class size will be less than or equal to 500?
• (b) It is 5 times more expensive to have a student in excess of 560
than to have fewer students accept. How many admission offers
would you make?

50
Practice Problem 2 – Solutions
• (a) Pr(the incoming class size will ≤500)= Pr(people not accepting
offer ≥60)
• z= (60-40)/20=1, NORMSDIST(1)=0.8413
• Pr(people not accepting offer ≥60) = 1-0.8413 =0.1587

• (b) Co=5*Cu, therefore Cu/(Cu+5Cu)=1/6=0.1666

• z=NORMSINV(0.1666)= -0.9674
• Q= =40-(0.9674)*20 = 20.65 = 21 (roundup)
• Admission offers = 560 + Q = 581

51
Summary
• Revenue management and overbooking give demand flexibility
where supply flexibility is not possible
• The Newsvendor model can be used:
• Single decision in the face of uncertainty
• Underage and overage penalties
• These are powerful tools to improve revenue:
• American Airlines estimated a benefit of $1.5B over 3 years
• National Car Rental faced liquidation in 1993 but improved via yield
management techniques
• Delta Airlines credits yield management with $300M in additional
revenue annually (about 2% of year 2000 revenue.)
• Marriott, Harrah’s, NCL, NBC …

52
 Supply Chain Management

Class 10
Forecasting & Analytics

Sajjad Najafi

1
LittleField Simulation
- Access to data on March 7th at 8pm (Paris Time) (you just have access to
data to analyze and think about how to better manage the firm, yet you do
not have the control of the firm)

- Starts on March 9th at 8pm (Paris Time) and last for about 7 days (you
take control of the firm and then can change the firm’s policies)

- Report Due on March 30th end of day in France

- A brief introduction at the end of session

- The detailed assignment and an overview will be posted on Blackboard

under the “LittleField Simulation” folder.
2
Mid-Term Exam

• The Mid-term will cover everything until session 9, with the

exception of the NV model – we will use it throughout the
Supply Chain Design part, and test you on this in the final.
Thus, sessions 1-5 and 9 will be included in the midterm exam.

• Open book

• Review sessions and more information next two sessions

3
What did we do last time?
What did we do last time? Newsvendor met EOQ
• Incorporated demand variability into the EOQ model

• Examined the dynamics of continuous and periodic review inventory systems

• Drivers of safety stock (Too Much - Too little Trade-off)?

• Demand variability
• Service Level
• Lead time

5
Fixed order Quantity Recipe (Q, ROP): Normal Distribution

Order Quantity: Find it using EOQ formula.

ROP Calculation:
1. A target service level SL is given to you.
2. Find the corresponding z-value to the target service level in the
standard normal table. Or using z = normsinv(SL) in Excel.
3. Find the average and standard deviation of demand during
Lead Time from the demand distribution as
μL = μd ⋅ L

σL = σd ⋅ L

4. The ROP is ROP = average demand during lead -me + SS = μL + zσL

We do not have access to Paul the Octopus

Source: fifa.com
8
This Class
• Understand the role of forecasting in supply chain management

• Discuss various quantitative forecasting techniques

• Oro-Chimie if we have time

• Littlefield overview

9
Forecasting is very ( very very … ) hard
• Thomas Watson, legendary CEO of IBM, forecasted the demand
for computers. He predicted that the world market for computer
would be Five. Yes, you read correctly, not 5 million, 5 computers.
(in his defence, he made this forecast in the ’50).

• Yahoo, in 1998, had a chance to license an innovative new search

technology created by a pair of Stanford grad students (Google)
for $1 million. As Jeremy Ring (a top sales executive at Yahoo)
writes, “That $1 million price tag was probably the best deal
offered in the history of Silicon Valley, California, the United
States, planet Earth, and the Milky Way Galaxy.” In 2002, Yahoo
had a second chance to buy Google. This time, CEO Terry Semel
offered $3 billion for the company; Google turned him down,
reportedly holding out for $5 billion.
10
Be Careful

11
Be Careful

12
Be Careful

13
Be Careful

14
Impact of demand forecasting on SCM

15
Demand Forecasting

Units sold per day

200

Forecast
100

0 7 14 21 28 35 42 49 Today 63

Day

16
Sales vs Demand
Demand

Inventory Inventory Sales=Inventory

Sales=Demand

Demand

What We Observe What We Observe

How to get an estimate of demand, if demand > then inventory

(i.e., sales limited by availability rather than demand)
• Keep track of orders, customer requests even if you cannot
ship / sell items
• Look at sales by competitors with similar products, who did
not run out (industry comparison)
• Market survey (ask customers about purchase intent)
• Look at the trend in demand when you run out and the time
still available until end of period

17
Time Series Framework
• Method of using past occurrences to model the future
• Assumes some regular and recurring basis over time

• Basic Components Demand Rate

• Trend
• Persistent movement in one direction
• Typically linear but can be exponential, quadratic, etc. Time
• Seasonal variations Demand Rate
• Movement that is periodic to the calendar
• E.g., hourly, daily, weekly, monthly, quarterly, etc.
• Cyclical movements Time
• Periodic movements below and above the trend (not tied to calendar) Demand Rate
• E.g., due to recession, depression or recovery
• Random fluctuations
• Unsystematic, irregular and unpredictable variations, noise Time
• E.g., due to unforeseen events such as union strike or tornado
18
Time Series

Demand

Summer 2018 Summer 2019 Summer 2020

Time

What should be the forecast for next period?

19
Time Series

Demand

Summer 2018 Summer 2019 Summer 2020

Time

What should be the forecast for next period?

20
Time Series

Demand

Seasonal
components

Random
fluctuation
Summer 2018 Summer 2019 Summer 2020
Time

What should be the forecast for next period?

21
Evaluating the Quality of a Forecast
• Forecast error for period t:

Forecast error for t (FEt) = Forecast for t - Actual demand for t

• Simple Averaging of the forecast error is NOT a good idea. Why?

• Forecast errors could be positive (overestimating)
• Forecast errors could be negative (underestimating)
• A negative FE and a positive FE will cancel each other out.

• We use one of the following measures for forecast quality:

N
∑t=1 FE2
Mean Squared Error (MSE) =
N
N
∑t=1 | FE |
Mean Absolute Error (MAE) =
N

22
Time Series

Moving average methods

Time Series Methods

Trend method

1. Moving average methods:

• Simple moving average
• Weighted moving average
• Exponential smoothing

2. Trend method

23
Moving Average Method
• A moving average is a series of arithmetic means
• Effective if there is little or no trend
• Used for smoothing since it provides an overall impression of data
over time

• Formula:

24
Moving Average Method
• You’re manager of a museum store that sells historical replicas. You
want to forecast sales for 2011 using a 3-period moving average.

Year 2006 2007 2008 2009 2010

Demand 4 6 5 3 7

25
Moving Average Example
• Moving average example:

Response Moving Total Moving Average

Time
(Yi) (n=3) (n=3)

2006 4 N/A N/A

2007 6 N/A N/A
2008 5 N/A N/A
2009 3 15 5.00
2010 7 14 4.67
2011 N/A 15 5.00

26
Weighted Moving Average Method
• Weighted moving average method:
• To put more weight on recent data and less weight on older data
• Weights based on intuition
• Often lay between 0 & 1 and sum to 1

• Formula:

27
Actual Demand, Moving Average, Weighted Moving Average
Weighted moving average (0.1,0.3,0.6)
Follows the trend more closely but reacts more to the random changes

30

Actual sales
22,5
Sales Demand

15

7,5 Moving average

Lags the trend but reacts less to the random changes

0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month

Weighted moving average can give more importance to recent days – then it reacts faster to trends but also more to
random changes…
28
Effect of “window-size” n
130 Most oscillations but closest to the trend
Historical
120
n = 3 periods
110 n= 7 periods
100 n= 21 periods

90

80

70

60
Smoothest but lags the trend the most
50

40

30
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47

Period

Larger n reacts slower to trends – but also smoothens out better random changes…
29
Exponential Smoothing Method
Sales
8
Actual Forecast
6
Weighted Moving Average require:
1- to keep sufficient historical data, all past
4
n periods of data are needed; and
2
2- all the corresponding weights to all the
9 9 9 9 9 0 past n periods.
Year

• Is a form of weighted moving average

– Weights decline exponentially
– Most recent data weighted most

Exponential Smoothing Method • Requires only one “weight”: smoothing constant (α)
– Ranges from 0 to 1
– Often subjectively chosen. It should be chosen to

• Advantage:
Involves little record keeping of past data
30
Exponential Smoothing Method
• Ft = Forecast value in period t
• At = Actual demand in period t
• α = Smoothing constant (between 0 and 1)
• Forecast by Exponential Smoothing
Forecast for t = α (Actual demand for t-1) + (1-α) (forecast for t-1)

• Rearranging it gives us

Ft = Ft-1 + α(At-1 - Ft-1)

What it really does:

Ft= αAt - 1 + α(1-α)At - 2 + α(1- α)2At - 3 + α(1- α)3At - 4 + ... +(1- α)tF0
31
Exponential Smoothing Solution

Ft = Ft-1 + 0.1(At-1 - Ft-1)

Forecast, Ft
Quarter Actual
( α = 0.1)

1 180 175.00 (Given)

2 168 175.00 + 0.1(180 - 175.00) = 175.50
3 159 175.50 + 0.1(168 - 175.50) = 174.75
4 175
5 190
6 205

32
Forecast Effects of Smoothing Constant α

Ft = α At - 1 + α(1- α) At - 2 + α(1- α)2At - 3 + ...

Weights
α 2 periods ago 3 periods ago
Prior Period
α α(1 - α) α(1 - α)2

α= 0.10 10% 9% 8.1%

α= 0.50 50% 25% 12.5%

Small alpha: almost like moving average

Large alpha: weighted average with strong weight on last period
33
Impact of α
210

Forecast (α = 0.5)
More oscillations but follows the trend more closely
195
Actual Tonage

180

Actual Forecast (α = 0.1)

Smoother but lags the trends
165

150
1 2 3 4 5 6 7 8 9
Quarter
Larger alpha gives more importance to recent days –
reacts faster to trends but also reacts more to random changes…

Choose an α value that minimizes the forecasting error (MSE or MAE) 34

Disadvantages of Moving Average Methods
• When actual demand has a trend, averaging techniques will either
• overestimate the demand (for downwards trend), or
• underestimate the demand (for upwards trend).
• Seasonality in demand is ignored

Example: Weighted moving average (0.1,0.3,0.6)

30 Overestimation

Actual sales
22,5
Sales Demand

15
Underestimation
7,5
Moving average
0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month 35
More Advanced Methods Needed: Trend Method

• Useful when time series has a clear linear trend with little
variation around trend line

Forecast of period t: yt = a + b t
36
 Estimating Trends
120 N
∑t=1 FE 2
Forecast of period t: yt = a + b t a, b chosen to minimize MSE =
N
100

80

60

40
b: slope

20
a
0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47
Period
37
 Trend Method (how to find the parameters)

∑'!! − *'̅!,!
%=
∑' " − *'̅ "
!! = # + % '

# = !,! − %'̅

— ! = intercept — T = number of periods

— * = slope of the line ∑"
— /̅ = = mean of t values
#
— t = the time period
∑$!
— 3! = forecast for demand — 56 = # = mean of y% values
for period t

These calculation are just included for the curious minds who want to see the blackbox
behind the trend method, and it will not be included in the exams.
Seasonal Fluctuations

120

100

80

60

40

20

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47
Period 39
Calculating the coefficients of seasonality

• Identification of model outside season

a+bt

• Calculation of the coefficients of seasonality for periods t at which we

observe seasonality
St = At – (a + b t) (additive model)

• Calculate average seasonality coefficient

• Or if seasonality is limited to a few data points and known: include dummy in

regression:
a + b t +St

40
Oro-chimie

Delay: 4 months
Segment
Segment Affaire
Special Business
Subcontracting
Segment
Transmission

Réacteur
Productionde
fabrication Segment Marine
Nouvelle
New Generation
génération
Raw Finished
Material Goods Segment Marine
Inventory Inventory Ancienne
Old Generation
génération

41
What do you think of their new approach?

• Figure 2

Ventes mensuelles agrégées (en tonnes)

Monthly consolidated sales (in tons)
30000

25000

20000

15000

10000

5000

0
J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D

Mois
Months

42
What do you think of their new approach?

Main Problems:

• No means to integrate information other than historical data

• No collaboration between forecast analysts, sales, and production and logistics

• Results in problem of data quality (at least in this case)
• Results in problem of acceptance – will it be used??

• Level of Aggregation
• Okay for medium term capacity planning, hiring, subcontracting (overall level;
everything produced in same factory)
• BUT overall average does not help for detailed production planning or placing
orders with subcontractors!
• Mis-match with objective of the forecast!

43
 Monthly
Ventes sales (in
mensuelles (entons) forpour
tonnes) thelesegment
segment
« Marine
"Marine Old Generation
Ancienne génération"»

6000

5000

4000

3000

2000

1000

0
J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D

Months
Mois

“A new generation of engines appeared on the market, which

causes the phasing out of the old generation of engines. The old
generation of engines will disappear from the market by next year.”
44
 Monthly
Ventes sales (in
mensuelles (entons) forpour
tonnes) thelesegment
segment
« Marine
"Marine Old Generation
Ancienne génération"»

6000

5000

4000 - 100.32t + 5153.4

3000

2000

1000

0
J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D

Months
Mois

45
 Monthly sales (in tons) for the segment
«Marine New Generation »

“For us, the new generation of engines is a fantastic opportunity: For now, we are basically
the only company on the market able to produce the additives for this new type of engines.
[…] it means for us a growing market, in my opinion at least for several years.” 46
Monthly sales (in tons) for the segment
«Marine New Generation »

10166,5 + 75,55 t
Seasonality: July :+1000 ,
August : +2000, September :
+1000

47
 Monthly sales (in tons) for the segment
Ventes mensuelles (en tonnes) pour le segment
« Additive Transmission »
"Additifs Transmission"

7000

6217 tons
6000

5000

4000

3000
J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D

Mois
Months
48
Moving Averages

∑ Demand in Previous n Periods

MA =
n

How determine n?
Orochemie:
• “steps correspond essentially to the fact that we become (or not) the main supplier for
certain engine producers”
• know how long existing contracts are valid
• Need additional (exogenous information to predict jumps)

49
 Monthly sales (in tons) for the segment
Ventes mensuelles (en tonnes) pour le segment
« Additive Transmission »
"Additifs Transmission"

7000

6217 tons
6000

5000

4000

3000
J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D

Mois
Months
50
51
What if we lack past data and we have no causal relationship?

• Need to rely on judgemental data

• Combine judgemental forecast with analysis:

• Assemble a dataset of products using a similar forecasting approach
• The basic assumption is that forecast errors verified in the past will happen again
• Can even use the past to “forecast the forecast error”!

52
Example 1: Demand forecasting at Sport Obermeyer

• “Committee” of experts create forecast, taking into account “fashion

trends” in Europe
• Get “independent” # from each member – Why??

Demand
• Mean demand = Average of committee
– Could adjust for bias (optimistic / pessimistic): x
+ average over- or under-estimation of demand
x x x
x x x
• Std. dev. of demand = t. std. dev. of committee forecast
– Agreement about forecast (or lack thereof) provides a good sense Forecast
for the uncertainty in demand
– Use historical data to derive standard deviation of demand from
std. dev. in committee forecasts
– “t” is a coefficient to incorporate more or less uncertainty

53
Example 2: A/F Ratios

• Gather dataset of similar products (+ whose demand forecasts are obtained

using similar forecasting techniques)
• Calculate the ratio between the actual demand observed (A) and the
forecasted demand (F)
• Calculate the mean (μ) and std. dev. (σ) of the ratios
• Use whatever you can to obtain the new forecasted demand (F)

• The characteristics of expected demand will then be:

– Mean: μD = μ x F
– Std. Dev.: σD = σ x F

54
Example 2: A/F Ratios

Actual A/F
Product Forecast
Demand Ratio
A01 90 140 1.56
A02 120 83 0.69
A03 140 143 1.02
B149 170 163 0.96
B320 170 212 1.25
D021 180 175 0.97
D238 180 195 1.08
E3 270 317 1.17

μ 1.09
Forecast for a certain product: F=200 σ 0.25
Then μD = μ x F= 218 and σD = σ x F=50.3
55
 Littlefield Technologies

A Brief Overview
Access to the simulation: op.responsive.net/lt/najafi/entry.html

An FAQ is provided at http://responsive.net/purchaseFAQ.html

Your Job Shop:
 Simulation
• Job shop assembling Digital Satellite System Receivers

• You have 3 stations—Board Stuffing, Testing, Tuning

• Process Flow—Board Stuff—Test—Tune—Test

• At start, you have 3 board stuffing machines, 2 testing machine,

and 1 tuning machine

• You can buy additional board stuffing machines for $90,000, testers
for $80,000 and tuning machines for $100,000. You can sell
machines but you only get $10,000 for them.
 Simulation Timeline
• The factory runs for 50 days without your control

• Student teams control factory from day 50 to day 217

• Factory runs for another 50 days using management policies chosen

by students effective on day 217
 Revenue/Cost at LittleField
• Revenue
• Contracts

• Cost
• Inventory
• Machine
 Choice of Contract
Three contracts to choose from
1) quoted lead time = 7 days,
max lead time = 14 days,
< 24 hours = $1000
price = $750 if delivered within the quoted lead> 72 hours = zero
time
2) quoted lead time = 1 day,
max lead time = 3 days,

* Factory must still

price = $1000 if delivered within the quoted lead time
3) quoted lead time = 12 hours,
max lead time = 24 hours,
purchase inventory
price = $1250 if delivered within the quoted lead for
time orders earning
zero revenue !
• Prices decreases linearly after the quoted LT and reaches zero at the max LT
• Initially, all teams start with offering customers contract #1 but you can change this
in the game.
 Choice of Capacity
• You need to decide whether you have sufficient machines of each
type. < 24 hours = $1000
• Remember buying machines is expensive > and
72youhours = zero
get very little when
you return them so make this decision carefully.

• You can buy additional board stuffing machines

* Factoryfor $90,000, testers
must still
for $80,000 and tuning machines for $100,000. You can sell machines
but you only get $10,000 for them. purchase inventory
for orders earning
zero revenue !
 Factory Process

• Every step has its own process time

• Littlefield measures average daily
utilization rates at each station
• Queues hold waiting jobs
• The factory holds a maximum WIP of 100
orders
lead time = process time + wait time
Arrival Rate × Process Time
Expected Uzlizazon =
# of machincs
 Inventory Decisions
• Orders arrive randomly at the factory
• Each arriving customer order represents 60 receivers
• Each receiver needs one kit of raw materials (thus, 60 kits per each
order)
• Each kit costs $10 ($600 per order)
• Every shipment has a fixed ordering cost = $1,000
• Supplier’s lead time is always 4 days
• Determine (Q, ROP) policy for raw material inventory

• Raw material are ordered as soon as the following Three criteria are
met:
1) Cash on hand is sufficient for the order quantity
2) Inventory on-hand ≤ Reorder Point
3) No orders for raw material kits are open (placed but not received)
 Example: Contract 2

< 24 hours = $1000

> 72 hours = zero

* Factory must still

purchase inventory
for orders earning
zero revenue !
An example using Contract 2
LT ≤ 24 hours = $1000, LT ≥ 72 hours = zero
* The factory uses inventory with every order—even when they earn zero revenue!
 Maximum Profit
Choice of Contract
< 24 hours = $1000
Lead Time Max Revenue Max Profit
Contract
Window
>
per Order
72 hours =
per Order
zero
1 7 - 14 days $ 750 $ 150
* Factory must still
2 1 - 3 days $ 1000 $ 400
purchase inventory
3 0.5 - 1 day $ 1250 for orders
$ 650earning
zero revenue !
$600 is the cost of raw material for each order which will be subtracted from the
revenue.
 Summary: Main decisions

• Purchasing Raw Material

– When to place an order (ROP)
– Size of each order
• Choosing Contracts
– Which type of contract to commit to based on your capacity and your inventory
• Adjusting the Stations’ Utilizations
– How many machines should I buy for each station based on your demand forecast
and the contracts you have chosen
 Playing
• The game starts on day 50.

• You will have control of the factory from day 50 to 217. (For 168
days). You will lose control of the factory starting from day 218 but
the factory will run for another 50 days without your control.

• Each “real day” corresponds to roughly 24 “simulation days”. So the

168 days will last about 7 days.

• The simulation will run the last 50 days from day 218 to 268. On day
268 the factory will stop the operation.

• See Blackboard for the assignment with detailed information and an

overview of the simulation.
 Objective
• Each team start the game with $1,000,000 in cash.
• You earn 10% annual cumulative interest on all cash you
hold.
• You get revenues by satisfying orders (according to
contract you picked) and incur costs for buying machines
and buying raw materials.

• Objective: To end the game with the most cash.

• May the best team win!
 Deliverables
– Simulation in teams
• Agree on teams
• Think of team name + password everyone remembers!

– Access to data on March 7th (you just have access to data to analyze and think
about how to better manage the firm, yet you do not have the control of the firm)

– Starts on March 9th at 8pm (Paris Time) and last for about 7 days (you take
control of the firm and now can change the firm’s policies)

– Deliverables: Submit a three-page (main body) executive summary of the actions

taken by each team during the game including the reasonings and the supporting
analysis for what you did and suggestions of what should have been done (if
different from what you did) only in PDF format. Due on March 30th end of day in
France (name of the file must be the name of your team).
 Causions

1. Procrastination is a common mistake, be sure to log in

during the first day of your assignment.
2. Two heads are better than one, consult your
teammates as a first course of action.
Good Luck!
 Logging Into Your Factory

Enter team name

example
Enter team’s password
••••••

Click OK

Access to the simulation: Click on the following link and choose your section
op.responsive.net/lt/najafi/entry.html
Explore Your Factory
 Download Data

• Click the download button.

• Choose Save and then open with Excel, or Save As and append .xls to the file
name.
 Download Data

• Click the download button.

• Choose Save and then open with Excel, or Save As and append .xls to the file
name.
 Making A Decision:
Additional Capacity

Read message,
enter password,
A popup
and confirm. appears,
2
click edit data.
The Transaction History records each
••••••

A second popup appears,

successive decision with a day index,
To purchaseRead message,
a machine,
enter value then
and click ok.
parameter description, and new
click on a click
Click close.
station
history, icon
to
value.
Note:
double an check.
edit data button only appears if
decisions are enabled for a given parameter.
It is never available when the simulator is
suspended or finished.
Next Class

• One thing to do with these forecasts? — Aggregate Planning

• Read Courtine Case

78
 Supply Chain Management

Class 11
Aggregate Planning

Sajjad Najafi

1
Project Proposal Reminder
Proposal: Upload a one-page description in PDF, due on 14/3 end of day in
France to the following Dropbox links:
EN01: https://www.dropbox.com/request/CyRVTNJfkQbetEOq1qbn
EN03: https://www.dropbox.com/request/jX9tUbIRpSGdBSQYeXRl
EN04: https://www.dropbox.com/request/iv5zSHQLHNhQRg15WkGX
including:
• Names of team members
• Name of company and a short description of company and its industry
• Name and position of the contact person and date of at least one
confirmed meeting time
• Rough project plan (schedule)

2
What did we do last time?
Potential Decision at Apple - MBP 13’’
Potential Decision at Apple - iMacs
Potential Decision at Apple - iMacs

In a longer timeframe (6-18 months) does Apple think about sales for each such
incremental model of its products?
 Potential Decision at Apple

For this longer timeframe (6-18 months) it is easier to come up with an aggregate plan
The Planning Process

Long-range plans (over one year)

Capacity decisions critical to long range plans
Issues:
Research and Development
New product plans
Capital investments
Facility location/expansion
Top
executives Intermediate-range plans (3 to 18 months)
Issues:
Sales and operations planning
Production planning and budgeting
Operations Setting employment, inventory,
managers with sales subcontracting levels
and operations Analyzing operating plans
planning team Short-range plans (up to 3 months)
Scheduling techniques
Issues:
Job assignments
Operations Ordering
managers, Job scheduling
supervisors, Dispatching
foremen Overtime
Part-time help
Responsibility Planning tasks and time horizons

8
What is the problem?

Working
Capacity
hours
Workload

Zone 3
Zone 1

Zone 2

Time

Zone 1 : Workload too high – risk of delivery delays

What is the problem?

Working
Capacity
hours
Workload

Zone 3
Zone 1

Zone 2

Time

Zone 1 : Workload too high – risk of delivery delays

Zone 2 : Insufficient work-load – risk of under-utilization of the capacity
What is the problem?

Working
Capacity
hours
Workload

Zone 3
Zone 1

Zone 2

Time

Zone 1 : Workload too high – risk of delivery delays

Zone 2 : Insufficient work-load – risk of under-utilization of the capacity
Zone 3 : Capacity increase to absorb growing workload
What is the problem?

Workload fluctuations due to Capacity fluctuations due to

- Seasonality - Capacity investments or
retirements of old plants
- Exceptional orders
- Hiring or worker turn-over
- Vacations, extended sick-leaves,
maternity leaves, …
- Scheduled maintenance, new-
product ramp-ups
What is the problem?

Workload fluctuations due to Capacity fluctuations due to

- Seasonality - Capacity investments or
retirements of old plants
- Exceptional orders
- Hiring or worker turn-over
- Vacations, extended sick-leaves,
maternity leaves, …
- Scheduled maintenance, new-
product ramp-ups

We want to match capacity and workload (demand)

over the medium term horizon at the lowest possible cost
Aggregate planning
• Aggregate plan:
• Medium term plan (typically 12 months)
• Examples:
• Manufacturing firm: production plan
• Service firm: staffing plan
• Objective:
• Match supply and demand over the next few months on an aggregate
level

• Output:
• Production levels and resource requirements for one or few product
families using similar resources (labor, equipment)
• Disaggregation breaks the plan down into greater detail resulting in
master production schedule
14
Aggregate Planning Goals

• Meet demand
• Use capacity efficiently
• Meet inventory policy
• Minimize costs:
– Labor
– Inventory
– Plant and Equipment
– Subcontracting
– …

15
Inputs for Aggregate Planning

• Definition of a single overall measure of output and sales (“aggregate

unit”)
• A reasonably accurate aggregate forecast
– In terms of this aggregate unit
– Across all products produced using the same equipment
• Capacity - measured in the same aggregate unit
• Various costs and constraints (see next slides)

16
 How do we get the aggregate demand?

• Versions of the same product? Directly forecast the aggregate demand.

• Different products (Orochemie!)? Define a standard product and calculate
aggregate requirements. Example with standard product AA:

How much time do we need to meet demand for each product?

How many standard units can we make in the same time?

time [hrs. /unit] Demand Demand in standard units

AA 2.0 2000 2.0 * 2000 / 2.0 = 2000
BB 3.0 1000
CC 1.5 2400
DD 1.0 1200
Aggregate requirement

17
 How do we get the aggregate demand?

• Versions of the same product? Directly forecast the aggregate demand.

• Different products (Orochemie!)? Define a standard product and calculate
aggregate requirements. Example with standard product AA:

How much time do we need to meet demand for each product?

How many standard units can we make in the same time?

time [hrs. /unit] Demand Demand in standard units

AA 2.0 2000 2.0 * 2000 / 2.0 = 2000
BB 3.0 1000 3.0 * 1000 / 2.0 = 1500
CC 1.5 2400 1.5 * 2400 / 2.0 = 1800
DD 1.0 1200 1.0 * 1200 / 2.0 = 600
Aggregate requirement 5900

18
Two generic strategies

Aggregate
Demand/ Demand
Output

Time
19
Two generic strategies
• Chase strategy:
• Production quantity equals the aggregate demand for each period

Aggregate
Demand/ Demand
Output

Chase

Time
20
Two generic strategies
• Chase strategy:
• Production quantity equals the aggregate demand for each period

• Level strategy:
• Production quantity equals the average demand over the planning
horizon
Aggregate
Demand/ Demand
Output
Level

Chase

Time
21
Tradeoffs
• Chase Strategy • Level Strategy
• Lower inventory build-up • Constant production rate
• Low holding costs • High resource utilization
• Low obsolescence costs • Lower capacity requirements

• Less backorders/stockouts • Stable workforce levels

• Low hiring/firing costs
• More flexible • Motivation/Union relations
• Responsive to demand
changes • Predictable budgets

• Applicable to service industries • No need for subcontracting

• Quality
22
Basic plan based on pure strategies

Period January February March April

Aggr.
5000 6000 8500 4500
Demand

Chase
Strategy
Level
Strategy

23
Basic plan based on pure strategies

Period January February March April

Aggr.
5000 6000 8500 4500
Demand

Chase
5000 6000 8500 4500
Strategy
Level
Strategy

24
Basic plan based on pure strategies

Period January February March April

Aggr.
5000 6000 8500 4500
Demand

Chase
5000 6000 8500 4500
Strategy
Level
6000 6000 6000 6000
Strategy

Level Strategy:
Average = (5000+6000+8500+4500)/4 = 6000

+ make adjustments based on constraints

25
Example of Adjustments: Capacity Constraint
Chase Strategy:

Period: P1 P2 P3 P4

Capacity 8000 8000 8000 8000

Demand 5000 6000 8500 4500

Basic Plan
(Chase)

End Inventory

Adjusted Plan

End Inventory

26
Example of Adjustments: Capacity Constraint
Chase Strategy:

Period: P1 P2 P3 P4

Capacity 8000 8000 8000 8000

Demand 5000 6000 8500 4500

Basic Plan
5000 6000 8500 4500
(Chase)

End Inventory 0 0 0 0

Adjusted Plan

End Inventory

Capacity constraint violated @ P3. What should we do?

27
 Example of Adjustments: Capacity Constraint
Chase Strategy:

Period: P1 P2 P3 P4

Capacity 8000 8000 8000 8000

Demand 5000 6000 8500 4500

Basic Plan
5000 6000 8500 4500
(Chase)

End Inventory 0 0 0 0

Adjusted Plan 5000 6500 8000 4500

End Inventory 0 500 0 0

We shie 500 units of producgon from P3 to P2 (not to P1 or P4) to produce closer to the desired period,
avoid shortages and reduce costs of inventory.
28
Example of Adjustments: Shortages

Level Strategy

Period: P0 P1 P2 P3 P4

Capacity 8000 8000 8000 8000

Demand 5500 6000 8500 4500

Basic Plan
6000 6000 6000 6000
(Level)

Inventory

Adjusted
Plan

Inventory

Suppose the inigal inventory is 500 units

29
Example of Adjustments: Shortages

Level Strategy Inventoryt = Inventoryt−1 + Productiont − Demandt

Period: P0 P1 P2 P3 P4

Capacity 8000 8000 8000 8000

Demand 5500 6000 8500 4500

Basic Plan
6000 6000 6000 6000
(Level)

Inventory 500 1000 1000 -1500 0

Adjusted
Plan

Inventory

Shortage of 1500 @ P3

30
Example of Adjustments: Shortages

Level Strategy Inventoryt = Inventoryt−1 + Productiont − Demandt

Period: P0 P1 P2 P3 P4

Capacity 8000 8000 8000 8000

Demand 5500 6000 8500 4500

Basic Plan
6000 6000 6000 6000
(Level)

Inventory 500 1000 1000 -1500 0

Adjusted
6500 6500 6500 4500
Plan

Inventory 500 1500 2000 0 0

Increase producgon by 500 in P1, P2, and P3 and reduce by 1500 in P4.
Overall, produce as close as possible to the LEVEL plan to keep capacity need (e.g. overgme) low
31
Tradeoffs
• Chase Strategy • Level Strategy
• Lower inventory build-up • Constant production rate
• Low holding costs • High resource utilization
• Low obsolescence costs • Lower capacity requirements

• Less backorders/stockouts • Stable workforce levels

• Low hiring/firing costs
• More flexible • Motivation/Union relations
• Responsive to demand
changes • Predictable budgets

• Applicable to service industries • No need for subcontracting

• Quality
32
 Case of JC Company
(Operations and Supply Management: The Core,
Jacobs, Chase)
How much to produce in each month? (production requirements)
-
+
+
-
+
+

Safety Stock each month

 Example Pure Plans

Plan 1: Produce to exact monthly production requirements using a regular 8-hour day by
varying workforce size
aka Chase

39
 Example Pure Plans

Plan 1: Produce to exact monthly production requirements using a regular 8-hour day by
varying workforce size
aka Chase

Plan 2: Produce to meet expected demand over the next six months by maintaining
constant workforce.
aka Level

40
 Example Pure Plans

Plan 1: Produce to exact monthly production requirements using a regular 8-hour day by
varying workforce size
aka Chase

Plan 2: Produce to meet expected demand over the next six months by maintaining
constant workforce.
aka Level

Plan 3: Produce to meet minimum expected demand (January) using a constant

workforce. Subcontract to meet additional output requirements.

41
 Plan 1

Plan 1: Produce to exact monthly production requirements using a regular 8-hour day by
varying workforce size

42
 Plan 1

Plan 1: Produce to exact monthly production requirements using a regular 8-hour day by
varying workforce size

capacity

43
 Plan 1

Plan 1: Produce to exact monthly production requirements using a regular 8-hour day by
varying workforce size

+7 +0 +4 +1 -1

44
 Plan 2
Plan 2: Produce to meet expected demand over the next six months by maintaining
constant workforce.

But what level of workforce do we need?

45
 Plan 2
Plan 2: Produce to meet expected demand over the next six months by maintaining
constant workforce.

But what level of workforce do we need?

16840 hrs

1000 hrs

On average we need 17 workers

46
 Plan 2
Plan 2: Produce to meet expected demand over the next six months by maintaining
constant workforce.

Imagine for now that we have 10 workers

47
 Plan 2
Plan 2: Produce to meet expected demand over the next six months by maintaining
constant workforce.

Imagine for now that we have 10 workers

48
 Plan 3
Plan 3: Produce to meet minimum expected demand (January) using a constant
workforce. Subcontract to meet additional output requirements.

Imagine for now that we have 10 workers

49
 Plan 3
Plan 3: Produce to meet minimum expected demand (January) using a constant
workforce. Subcontract to meet additional output requirements.

Imagine for now that we have 10 workers

50
Comparing all three plans

51
Midterm Review I

52
• The way you Manage your Operations and Supply Chain can have
a significant impact on your firm’s performance.
• The objective of a supply chain is to Maximize Value (Value =
Revenue - Cost)

• Four dimensions:
- quality
- delivery
- flexibility
- cost (price)
• Compromise is unavoidable
• The ones you should focus on — depends on your strategy

• Focused strategy has its advantages but also has some risks
Strategic trade-off: High Quality vs. Low Prices

Quality

Low Price

54
Process types

Uniqueness (Variety)

• Project
• Job shop Job Shop
• Batch
• Assembly line Flow Shop
• Continuous flow

Volume
• A company’s competitive and marketing strategies need to be
translated to product attributes; process needs to be designed
accordingly
55
Definitions

• Cycle Time: The average time between the completion of successive

units
• Flow (or Throughput) Time: The length of time a unit spends at a given
stage/process
• Capacity: The rate with which units flow through the stage/process
• Inverse of flow or cycle time
Average Throughput Rate To-do
UTILIZATION = ρ = =
Capacity Can do
Cycle time is NOT necessarily equal to Flow time
Ford F150: Cycle time (~2 minutes) while Flow Time (~20 hours)
56
Throughput Rate and Capacity

• Capacity is the maximum output rate when working at full speed.

• Individual stages have capacities.
• The overall process has a capacity (process capacity) that is
constrained by the bottleneck resource.
• What flows out (Throughput Rate) is limited by what flows in and how
much the process can actually produce. So,
Throughput Rate = minimum (Input, Capacity)

57
Bottleneck Analysis Example: Shopping in
• Determine the capacity of each stage/resource: First determine the
cycle time and then convert to the capacity for each stage:

Cashier Movers Receptionis

10 min/ord
Customer Completed
Order Order
4 min/ord 0.5 min/ord

10 min/ord

Payment Retrieval Reception

Payment Retrieval Reception

(15 orders/hr) (12 orders/hr) (120 orders/hr)

• Retrieval is the bottleneck.

Utilization of the Bottleneck(Retrieval)? What if Darty improves Reception?

Depending on the Input Rate What if Darty improves Retrieval?

58
Process Analysis Example: Capacity of bread-making on Two Lines

Production in batches of 100 loaves at a time

Mix Proof Bake

Raw 45 min/batch 45 min/batch 60 min/batch

WIP FGI
Mat. Pack

40 min/batch

Mix Proof Bake

30 min/batch 45 min/batch 60 min/batch

59
Process Analysis Example: Capacity of bread-making on Two Lines

Production in batches of 100 loaves at a time

Mix Proof Bake

Raw 45 min/batch 45 min/batch 60 min/batch

WIP FGI
Mat. Pack

40 min/batch

Mix Proof Bake

30 min/batch 45 min/batch 60 min/batch

Process capacity = minimum {2, 60/40} = 1.5 batches/hour

60
 Bread Baking Example, cont.
• What is the minimum Throughput (Flow) Time (for a rush order)?

TTMin = 30 min + 45 min + 60 min + 40 min = 175 min

• What is the Throughput Rate

– if the process runs at capacity? 1.5 batches / hour

– if you launch one batch every 2 hours? ½ batch / hour

– if you launch three batches every hour? 1.5 batches/hour

– What happens if input > capacity?

• What is the process utilization

– if the process runs at capacity? 100%

– if you launch one batch every 2 hours? 33%

61
Croissant Baking: An Assembly Operation!

Raw Mat. WIP

Dough Roll & Dough
Mix Proof Fill & 5 min/batch
Cut
Fold
5 min/batch 15 min/batch 5 min/batch

Bake 20 min/batch

Raw Mat. WIP

Filling Mix Filling
Pack 10 min/batch

10 min/batch

FGI
Production in batches of 50 croissants at a time

62
Croissant Baking cont.
• What is the capacity of the process? (batches per hour)

Bottleneck: Baking
Bottleneck capacity = process capacity = 3 batches per hour

• What is the minimum process throughput time?

Dough available after: 5 min + 15 min + 5 min = 25 min
Filling available after : 10 min
TTMin = 25 min + (5 min + 20 min + 10 min) = 60 min = 1 hr

• What is the cycle time of the process?

CT = 1 / 3 batches per hour * 60 min / hr = 20 min / batch

• If you start a batch of croissants every 40 minutes, what is the process

utilization?
% of time busy: 20min/40min=50%
or TR/Capacity: 1 * (60/40) batches per hour / 3 batches per hour = 50%
63
 Supply Chain Management

Class 12
Production Planning / MRP

Sajjad Najafi

1
Project Proposal Reminder
Proposal: Upload a one-page description in PDF, due on 14/3 end of day in
France to the following Dropbox links:
EN01: https://www.dropbox.com/request/CyRVTNJfkQbetEOq1qbn
EN03: https://www.dropbox.com/request/jX9tUbIRpSGdBSQYeXRl
EN04: https://www.dropbox.com/request/iv5zSHQLHNhQRg15WkGX
including:
• Names of team members
• Name of company and a short description of company and its industry
• Name and position of the contact person and date of at least one
confirmed meeting time
• Rough project plan (schedule)

2
What did we do last time?

Demand, Output

Aggregate
demand

Chase

Level

3
Tradeoffs
• Chase Strategy • Level Strategy
• Lower inventory build-up • Constant production rate
• Low holding costs • High resource utilization
• Low obsolescence costs • Lower capacity requirements

• Less backorders/stockouts • Stable workforce levels

• Low hiring/firing costs
• More flexible • Motivation/Union relations
• Responsive to demand
changes • Predictable budgets

• Applicable to service industries • No need for subcontracting

• Quality
4
This Class: MRP (Material Requirements Planning)
• A system for scheduling production (/order) of inventory:
• How much is needed?
• When is it needed
MAJOR AIRCRAFT
Processes PART
Primary,
Secondary material
Supplier . FRONT
FUSELAGE
Processes
Primary,
Secondary material

Processes FINAL ASSEMBLY

Primary,
Secondary material
. REAR
FUSELAGE
Processes
Primary,
.
Secondary material
.

Processes .
Primary,
Secondary material
.
WINGS
Processes
Primary,
Secondary material

= MANUFACTURING = MATERIAL, PART OR COMPONENT = ASSEMBLY 5

Example: Bicycle manufacturing orders for week 26

Demand = 1000
Due date: week 25
1 week lead time
(assembly time)

6
Example: Bicycle manufacturing orders for week 26

Demand = 1000 1 week

lead time
Due date: week 24

Demand = 1000
Due date: week 25
1 week lead time
(assembly time)

7
Example: Bicycle manufacturing orders for week 26

Demand = 1000 1 week

lead time
Due date: week 24

3 weeks
lead time

Demand = 1000
Demand = 1000
Due date: week 22
Due date: week 25
2 weeks 1 week lead time
lead time (assembly time)

Demand = 2000
Due date: week 23

8
Example: Bicycle manufacturing orders for week 26

Demand = 1000 1 week

lead time
Due date: week 24

3 weeks
lead time

3 weeks
lead time Demand = 1000
Demand = 930 Demand = 1000
Due date: week 22
(= 1000 – 70 in stock) Due date: week 25
2 weeks 1 week lead time
Due date: week 19 lead time (assembly time)

Demand = 2000
Due date: week 23

9
When MRP and when continuous/periodic review models?

Independent and Dependent Demand

Independent Demand
End item • Demand of end products
X • Demand is unpredictable
Subassemblies
• Requirements are determined by
inventory control policies
A G
Dependent Demand
BOM B C D D • Demand dependent on
requirements of end product
E (BOM)
• Determine the replenishments
using the net requirements
calculations (MRP)
Components & raw material
• Possibly EOQ to determine batch
size, and some safety stock in
case of uncertain lead-time
Material Requirements Planning (MRP)
• Materials requirements planning (MRP) is a means for
determining the number of parts, components, and materials
needed to produce a product.

• Based on a master production schedule, of end-items

(independent demand) and
• Determines exact unit numbers needed
• Determines the dates when orders for those materials should be
released, based on lead times

11
Inputs and outputs of MRP
• Inputs:
• Master Production Schedule (# of end-items to be produced during
specific periods)
• Bills of Material (BOM)
• Inventory Records

• Outputs:
• Recommended Production Schedule (for the parts you produce)
• Recommended Purchasing Schedule (for the parts you purchase)

12
BOM (or The Product Structure Tree)
• Bill of Materials:
• The demand for an item is related to the demand for its parent item
• Given an end-item’s demand from the MPS (independent demand), the
demand for all its parts and components can be calculated (dependent
demand)

• Example:
A Level 0

B(4) C(2) Level 1

D(2) E(1) D(3) F(2) Level 2

13
Example of MRP Logic and Product Structure Tree

Product Structure Tree (or BOM) for Assembly A Lead Times

A 1 day
A B 2 days
C 1 day
D 3 days
E 4 days
B(4) C(2) F 1 day

Total Unit Demand

Day 10 50 A
D(2) E(1) D(3) F(2)
Day 8 20 B (Spares)
Day 6 15 D (Spares)

• Given the product structure tree for “A” and the lead time and
demand information, provide a material requirements plan that
defines the number of units of each component and when they will
be needed.
14
First, the number of units of “A” are scheduled backwards to allow for their lead
time. So, in the materials requirement plan below, we have to place an order for
50 units of “A” on the 9th day to receive them on day 10.

Day: 1 2 3 4 5 6 7 8 9 10
A Required 50
Order Placement 50

LT = 1 day
 Next, we need to start scheduling the components that make up “A”. In the case
of component “B” we need 4 B’s for each A. Since we need 50 A’s, that means
200 B’s. And again, we back the schedule up for the necessary 2 days of lead
time.

Day: 1 2 3 4 5 6 7 8 9 10
A Required 50
Order Placement 50
B Required 20 200
Order Placement 20 200

LT = 2 Spares

A
4x50=200

B(4) C(2)

Careful:
Look at Order Placement (i.e.,
D(2) E(1) D(3) F(2) prod. Launch or procurement)
NOT at requirements!
Finally, repeating the process for all components, we have the final materials
requirements plan:

Day: 1 2 3 4 5 6 7 8 9 10
A Required 50
LT=1 Order Placement 50
B Required 20 200
LT=2 Order Placement 20 200
C Required 100
LT=1 Order Placement 100
D Required 55 400 300
LT=3 Order Placement 55 400 300
E Required 20 200
LT=4 Order Placement 20 200
F Required 200
LT=1 Order Placement 200

A
Part D: Day 6
B(4) C(2) 2*20 (to produce spares for B) + 15 (spares)

D(2) E(1) D(3) F(2)

Lot-Sizing Decisions

• MRP determines net requirements but order quantity does not have to equal net
requirements
• Many lot sizing rules:
– Lot for lot (L4L): Order = NR
– EOQ (Fixed Order Quantiry): If NR>0 order Q units (or multiples of Q depending
suppliers batching constraints)
– Fixed Period: Order every x periods -- e.g., every Monday
– …
• Best lot sizing strategy is the one with the minimum cost.
• Trade-off?
– Inventory holding costs versus setup / ordering costs !

18
PFSA Case

19
Components and Products

• BOM

• Requirements for the next 16 months

 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Requirement FP1 10 20 10 20 20 10 10 30 20 20 20 30 SA1
Requirement FP2 60 40 80 SA2
Launch FP1 PP3
Launch FP2
Requirement SA1
Requirement SA2 PP1
Requirement PP3 PP2
Inventory SA1 PP3
Launch SA1
Inventory SA2
Launch SA2
Requirement PP1
Requirement PP2 Orderi
Requirement PP3 SA1 : 1
Inventory PP1 SA2 : 3
Procurement PP1
Inventory PP2
Procurement PP2
Inventory PP3
Procurement PP3

The average manufacturing cycles of SAs and of FPs are two weeks.
The average procurement lead times are two weeks for PP1 and PP3 and one week for PP2.
 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Requirement FP1 10 20 10 20 20 10 10 30 20 20 20 30
Requirement FP2 60 40 80
Launch FP1
Launch FP2
Requirement SA1
Requirement SA2
Requirement PP3
Inventory SA1
Launch SA1
Inventory SA2
Launch SA2
Requirement PP1
Requirement PP2
Requirement PP3
Inventory PP1
Procurement PP1
Inventory PP2
Procurement PP2
Inventory PP3
Procurement PP3

The average manufacturing cycles of SAs and of FPs are two weeks.
The average procurement lead times are two weeks for PP1 and PP3 and one week for PP2.
Any potential problem with this approach?

23
Workload Across Time
What can we do?

• Produce earlier => build inventory

• Planned delay => reduce safety stock, expedite
• Change capacity => overtime

• In case of major problems:

Change Master Production Schedule!
– Change timing of production for different products

25
Problem: Work Load > Capacity

– Production order generation ignores capacity constraints (Infinite capacity scheduling)

– After the MRP procedure, the work loads must be calculated for each resource and
one must check that they are compatible with the actual capacities
– If not, adjust the capacities or try to place the orders earlier
– Aggregate plan takes care of capacities for end-times, not necessarily of capacities for
subassemblies!

Work Order x Work Order x

Capacity

Load
Components
Requirements
Date

Before Load smoothing After Load smoothing

26
Adjusted Workload

27
PFSA: Which products to stock?

FP1 FP2 SA1 SA2

SA1 2 2 PP1 2
SA2 3 PP2 2
PP3 4 4 PP3 2 2
Hours 6 4 Hours 4 4

Component Production Production

cost costs time Cost Total
PP1 45 SA1
PP2 15 SA2
PP3 30 FP1
1 hour 30 FP2

28
PFSA: Which products to stock?*

FP1 FP2 SA1 SA2

SA1 2 2 PP1 2
SA2 3 PP2 2
PP3 4 4 PP3 2 2
Hours 6 4 Hours 4 4

Component Production Production

cost costs time Cost Total
PP1 45 SA1 150 4 120 270
PP2 15 SA2 90 4 120 210
PP3 30 FP1 1290 6 180 1470
1 hour 30 FP2 660 4 120 780

29
PFSA: Which products to stock?

FP1 FP2 SA1 SA2

SA1 2 2 PP1 2
SA2 3 PP2 2
PP3 4 4 PP3 2 2
Hours 6 4 Hours 4 4

SA1 SA2 FP1 FP2

Cost 270 210 1470 780
Time 4 hrs 4 hrs 6+8+12 4+8
=26 hrs = 12 hrs
Cost/hr 67.5 €/h 52.5 €/h 56.54 €/h 65 €/h

Other Considerations?
• Obsolescence risk - SA2 only needed for FP1
• Not a problem if just a few weeks, but if for a long time…

30
 Any other problem of varying workload?

800

600

400

200

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Midterm Review II
Supply Chain
• A sequence of operations and organizations involved in producing
and delivering goods/services
• The structure of a typical supply chain:

Retailer

33
Supply Chain Flows

Physical Flow
•Raw materials
•Work in process
•Finished goods

Retailer

Information and Cash Flow

•Orders, contracts
•Payments
34
Croissant Baking: An Assembly Operation!

Raw Mat. WIP

Dough Roll & Dough
Mix Proof Fill & 5 min/batch
Cut
Fold
5 min/batch 15 min/batch 5 min/batch

Bake 20 min/batch

Raw Mat. WIP

Filling Mix Filling
Pack 10 min/batch

10 min/batch

FGI
Production in batches of 50 croissants at a time

35
Croissant Baking cont.
• What is the capacity of the process? (batches per hour)

Bottleneck: Baking
Bottleneck capacity = process capacity = 3 batches per hour

• What is the minimum process throughput time?

Dough available after: 5 min + 15 min + 5 min = 25 min
Filling available after : 10 min
TTMin = 25 min + (5 min + 20 min + 10 min) = 60 min = 1 hr

• What is the cycle time of the process?

CT = 1 / 3 batches per hour * 60 min / hr = 20 min / batch

• If you start a batch of croissants every 40 minutes, what is the process

utilization?
% of time busy: 20min/40min=50%
or TR/Capacity: 1 * (60/40) batches per hour / 3 batches per hour = 50%
36
Three Measures of Process Performance

1. Inventory: the number of units (products/customers) contained in the

process
2. Flow (Throughput) Time: the time it takes for one unit (product/customer)
to get through the process.
3. Throughput Rate (Output Rate): the rate at which the process is delivering
output (finished products / served customers).

Notice: We previously talked about Four measures of Product performance

(price, quality, flexibility, responsiveness)
37
 Little’s Law

Throughput Rate [units/hr]

Inventory [units]
... ... ...

Throughput Time [hrs]

Inventory = throughput Rate x throughput Time

(easier to remember) [ units ] = [ units / time ] x [ time ]

• If you know 2 of them, you can determine 3rd.

• Different interpretation for inventory: Customers, orders, money, applications,
WIP, etc.

Little’s Law is true in steady state

38
An Example Using Little’s Law

Inventory = throughput Rate x throughput Time

A company processes 6000 claims per year (50 Weeks). Processing

time of each application is 2 weeks. What is the average number of
applications in the process?

R = 6000 [claims / year] / 50 [weeks / year] = 120 [claims/ week]

T = 2 [weeks]
I = R x T = 120 [claims/ week] x 2 [weeks] = 240 claims

39
Little’s Law: Summary
• Provides a relationship between the inventory, throughput rate
and flow time of a process

• Always holds on average (e.g., the actual flow time may be

different from the flow time given by Little’s Law)

• Can be applied to any part of the process

40
Little’s Law and Inventory

I = R x T => R = I / T
The same throughput rate can be obtained by large inventory and long flow time, or
small inventory and short flow time.

Which one of the following systems do you prefer?

— System 1: R = 1, I = 10, T = 10
— System 2: R = 1, I = 1000, T = 1000

Given a fixed throughput Rate, Little’s law tells us nothing on whether we would
have large or small inventories (delays).

We looked into the root causes of having queues (inventory) and delays in a system.

41
 Service with no Queues
Gantt Chart illustrating Service
Arrival Inter-Arrival Service
Patient
Time Time Time

1 7:00 0 4

2 7:05 5 4

3 7:10 5 4

4 7:15 5 4

5 7:20 5 4

6 7:25 5 4

7 7:30 5 4

8 7:35 5 4

9 7:40 5 4

10 7:45 5 4

11 7:50 5 4

12 7:55 5 4
7:00 7:10 7:20 7:30 7:40 7:50 8:00
Average Inter-Arrival Time is 5 minutes, Average Service Time is 4 minutes. No Waits!
What is odd about this Service Process?
 A More Realistic Service
Arrival Inter-Arrival Service
Patient Patient 1 P3 P5 P7 P9 P11
Time Time Time

P2 P4 P6 P8 P10 P12
1 7:00 0 5

2 7:07 7 6 Time

3 7:09 2 7
7:00 7:10 7:20 7:30 7:40 7:50 8:00
4 7:12 3 6

5 7:18 6 5
3
3
6 7:22 4 2

7 7:25 3 4
2 2 2 2
2
8 7:30 5 3
Number of cases

9 7:36 6 4 1
1

10 7:45 9 2

11 7:51 6 2 0
2 min. 3 min. 4 min. 5 min. 6 min. 7 min.
12 7:55 4 2
Service times

Average Inter-Arrival Time is 5 minutes, Average Service Time is 4 minutes.

Will the service performance be the same as before?
 Variability Leads to Waiting
Arrival Inter-Arrival Service
Patient
Time Time Time

1 7:00 0 5
2 7:07 7 6
Service time
3 7:09 2 7
4 7:12 3 6
5 7:18 6 5
6 7:22 4 2
7 7:25 3 4
8 7:30 5 3
9 7:36 6 4 Wait time
10 7:45 9 2
11 7:51 6 2
12 7:55 4 2

7:00 7:10 7:20 7:30 7:40 7:50 8:00

Variability is Evil! 4

1
Inventory
(Patients at lab) 0
The View of the Process

So far we looked only at this

Waiting Time Service Time

Arrivals

Flow Time

Where Variability might be Hiding

https://www.youtube.com/watch?v=K3axU2b0dDk
Example: Call Center
Waiting Time Service Time

Arrivals

λ = 10.8 [calls/hr] μ = 12 [calls/hr]

1 1
AIT = = 5.55 [min] τ = = 5 [min]
λ μ
There is one operator, who receives 10.8 calls per hour on average. The operator can handle 12 calls per hour.
Suppose that the CVa=0.99, and CVs=0.7. What is the average wait before service begins (on hold)? What is the
average total time in the system?
τ
ρ= = 0.9
s ⋅ AIT

τ ρ CV a2 + CV s2
Time in the queue: Wq = ⋅ ⋅ ≈ 33 [min]
s 1−ρ 2
Total time in the system: W = Wq + τ
How to Reduce Waiting?
Waiting Time Service Time

Arrivals

Wq

1. Speed up service time τ

2. Add capacity (reduce utilization) s
3. Reduce variability CVa, CVs
4. Resource Pooling
 Benihana

A New Look…

At an Old Classic

http://www.youtube.com/watch?v=epoIPgQrAyQ

Adapted from material by G.Schmidt and R. Ernst

Benihana Restaurant

Average Utilization (no batching) — 42.22%

Average Utilization (with batching) — 56.1%
(4) Resource Pooling

Independent Resources Interarrival time: AIT = 4 min, CVa=1

2x(s=1)
Service time: τ = 3 min, CVs=1

Capacity utilization: ρ = τ / AIT = 0.75 each server

⎛ 0.75 ⎞ ⎛ 1 + 1 ⎞
Wq = 3 * ⎜ ⎟*⎜ ⎟ = 9 min
⎝ 1 − 0.75 ⎠ ⎝ 2 ⎠
Pooled Resources
(s=2) New AIT: AIT = 2 min
Exponential distributions: CV unchanged
New utilization: ρ = τ / (s*AIT)= 3 / (2*2) = 0.75

3 ⎛⎜ 0.75 2 ( 2+1) −1 ⎞⎟ ⎛ 1 + 1 ⎞
Wq= * ⎜ ⎟
*⎜ ⎟ = 3.95 min
2 ⎝ 1 − 0.75 ⎠ ⎝ 2 ⎠

The new time is less than half of the old one!!!

Pooled Systems perform better!

https://www.youtube.com/watch?v=1I2Vs3XwH0w
Inventory Charts

Demand Uniform
Deterministic

Time

Order

Time

Receive Lead Time is 0

Time

Inventory Slope = Demand Rate

Time
51
EOQ: Notation
• Data:
• D = Demand rate (units/yr)
• C = Cost of purchasing or producing a unit (€/unit)
• S = Setup cost or cost per order or cost per production run (€/order)
• H = Annual holding cost per unit of inventory (€/unit/yr)
• Often: H = iC where i = annual percentage holding cost

• Decision:
• Q = Quantity of an order (units)

• Objective:
• Minimize the total cost

52
EOQ: Total Cost
Inventory

Time

Purchasing
Ordering Cost Holding Cost
Cost

Number of orders / yr Average Inventory

Annual Holding
Annual Ordering Cost
Cost
53
EOQ Example: Approach B
Cost

Ordering+Holding
Costs

Lowest Holding Cost =

Cost

Ordering Cost =
Q
Q*
Sensitivity
Analysis w.r.t.
D,S,H?
54
Sensitivity Analysis (if we order Q instead of Q*)

TC(Q)
TC(Q*)

1.25

Q
0.5 1 1.5 Q*

55
EOQ Example
• At a Darty shop:
• D = 1200 GPS units/yr
• C = €200 / unit
• S = €2000 / order
• i = 25% per year

• What is the optimal order size Q for Darty?

• How to determine the optimal Q ?

• A) Calculate the total cost for all possible values of Q and choose the
one which results in the lowest value
• B) Solve analytically

56
EOQ: What if there is a delivery lead time?

Reorder
point

Time
Receive Place Receive
order order order
Reorder point = DxL

Lead Time:
L days

57
Your turn: Auto Repair

AutoRepair is a large automobile repair shop. It installs about 1,250 mufflers per year,
18 percent of which are for imported cars. (Assume a very steady demand rate.) All of
the imported-car mufflers are purchased from a single local supplier at a cost of $18.50
each. The shop uses a holding cost based on a 25 percent cost of capital. The fixed
cost for placing an order is estimated to be $28.
1. Determine the optimal number of imported-car mufflers the shop should purchase
each time an order is placed, and the time between the placement of orders.
EOQ= Sqrt(2*28*1250*0.18/(18.5*0.25)) = 52 mufflers
Cycle time = Q/D = 52/(0.18*1250) = 0.23 years ≅ 12 weeks
2. If the replenishment lead-time is six weeks, how many units are in stock when you
order (reorder point)?

Reorder point = demand during lead time

= [1250*0.18 / 52] * 6 =26 mufflers (round up!)
Adding Uncertainty

Time

59
Adding Uncertainty

Reorder
point Q

Time
Receive Place Receive Place Receive
order order order order order

Lead Time: Lead Time:

L days L days
60
Safety Stock

Reorder
point

Safety Stock

Time
Receive Place Receive Place Receive
order order order order order

Reorder point =
Lead Time: Lead Time: Example:
L days L days Average demand: 120 units/week
Lead time: 3 weeks
Safety stock: 100 units
Re-order point:
61
3*120+ 100 = 460 units
Safety stock
• Safety Stock provides a cushion against demand variability

• Reorder Point (ROP) = Average Demand over Lead Time + Safety

Stock

• A shortage occurs when the actual demand over lead time

exceeds ROP

• What factors influence safety stock?

• Demand variability
• Service Level
• Probability of no stockout during the lead time
• Lead time

62
Fixed order quantity model: How to compute the ROP?
• With discrete demand:

• With continuous, normally distributed demand:

LT Demand

63
How to compute
• Knowns:
• Standard deviation of daily demand:
• Lead time in days:

• Unknown:
• Standard deviation of demand over the entire lead time:

64
Summary of fixed order quantity model
• Order Quantity: EOQ

• Reorder Point: Avg. LT Demand + Safety Stock = μL + z ⋅ σd ⋅ L

• Safety stock is determined from:

• Service level
• Demand variability
• Lead time
• For normal demand:
Prob (demand over lead time ≤ ROP) ≥ Target Cycle SL
65
Our Example

Example:
Average demand: 120 units/week
Lead time: 3 weeks
Std dev of demand: 25 units / week
Service Level: 99%
Re-order point:
3*120 + 2.33 * 25 * Sqrt(3) = 461 units

66
Your turn! HEC Corp – pg. 25 in course package

Continuous review system, operates 50 weeks per year

Product characteristics:

• Demand = 20,000 units/year

• Standard deviation of weekly demand = 130 units
• Ordering cost = $25/order
• Holding Cost to hold one unit for 1 year = $4
• Replenishment lead-time = 2 weeks
• Target cycle service level = 95%

67
Example: Determining Q

• How many units should be ordered each time?

2 DS 2 * 20000 * 25
Q= = = 500
H 4
• Cycle stock?
= 250
• What is the number of orders per year (on average)?
Number of order cycles: 40

68
Example: Calculating ROP

• ROP ?
= mean lead time demand + Z x std. dev over L

= (20000 / 50) * 2 + 1.64 * 130 * √2

= 800 + 302 = 1102
Safety Stock
• Pipeline inventory?
Little’s Law! WIP = TT * TR = 2*20,000/50=800
• Average Inventory
= cycle stock +safety stock+pipeline inventory
= (500/2) + 302 + 800 = 1352 < 1500!
69
 Supply Chain Management

Class 13
Quality Management and Statistical Process
Control

Sajjad Najafi

1
Reminder: LF Simulation Report (Due March 30th)
Deliverables: Your team will write a three-page (main body) executive
summary of the actions taken by the team during the game including the
reasonings and the supporting analysis for what you did and suggestions of
what should have been done to improve the factory performance (if different
from what you did).
The team representative submits the report (only in PDF format) to the
following Dropbox links (the name of the PDF file should only consist of
your Team ID chosen at the registration time):
EN01: https://www.dropbox.com/request/CbAndcXWk0SqPcM7178n
EN03: https://www.dropbox.com/request/rSy7LEjEwqglJ3t24Rh8
EN04: https://www.dropbox.com/request/QjDH3hUAn6ARRBTR04bi

2
Quiz (Quality Management)
Available from March 31st and remains open until Final Exam.

For all quizzes:

Available until Final Exam
Multiple attempts are allowed
Effort based (not grade based): Students will get full grade of a quiz as long
as they take the quiz on Blackboard even once and regardless of the actual
grade displayed.

3
This class
• Understand how to measure and control product/service quality
• Learn the concept of six sigma

4
What is Quality? Two complementary aspects.

• Design Quality (intended quality to offer to customers)

• Target niche of the product in the marketplace ← A strategic decision
for the firm
• Dimensions: Performance, Features, Reliability/durability, Serviceability,
Aesthetics
• Identifies a set of design specifications: targets (ideal dimensions) and
tolerances (acceptable deviations from these ideal dimensions)
• A firm designs a product/service to address the need of a particular
market

• Conformance Quality (actual level of quality of product)

• Degree to which the product/service design specifications are met.

5
What is the root cause of all quality problems?
Without variation either all products are non-defective
or all are defective

Inputs Output

Process Variation in the output

is the root cause of
quality problems
Variations

Whether or not a product is defective depends on

the set of design specifications.

6
Example: LEGO design specifications
• Brick hight
• Target hight (ideal hight) is 9.6mm
• There are natural variations in hight

• Tolerance limits (specification limits) specify how much brick

hight can vary yet still meeting the design quality, e.g., it
specifies that brick hight must fall between 9.5mm and 9.7mm.
• Lower Tolerance Limit (LTL) = 9.5 mm
• Upper Tolerance Limit (UTL) = 9.7 mm

• Anything beyond the tolerance limits is considered a defect

(i.e., inconsistent with design specifications).
7
Probability of Defects
• Any process has natural variations in the outcome it produces. A
defect occurs when the variation is outside of the tolerance
limits.

• Probability of defects depends on:

1- amount of variations in the system (standard deviation)
2- tightness of tolerance limits (UTL and LTL)

8
Who is a better target shooter?

(A) (B)

9
Who is a better target shooter?

(A) (B)
Less variability = Higher quality
Not just the mean is important, but also the variance !
10
Amount of Variations

(A) (B)

11
Amount of Variations
• A process with less variability is more consistent in the outcome
it produces, so has it less probability of producing defects.

Probability Probability Probability Probability

of defect of defect of defect of defect

(A) (B)

12
Tightness of Tolerance Limits

13
Tightness of Tolerance Limits

14
Quality Measure: Sigma Capability
• How many standard deviations the process mean is away from the
closest tolerance limit?

• Higher Sigma Capability index means that the process has a higher
quality of conformance.
15
Sigma Capability example

µ − LTL σ σ σ σ σ σ σ σ UTL − µ
=5 =3
σ σ

LTL µ UTL

This distribution depicts 3-Sigma Capability

16
Sigma capability example

If Sigma Capability z ≥ 2 ⇒
Probability of defect ≤ 5%

If Sigma Capability z ≥ 3 ⇒
Probability of defect ≤ 0.3%

17
Quality measures: LEGO example

• LTL = 9.5 mm, UTL = 9.7 mm, μ = 9.65 mm, σ = 0.025 mm

• Sigma Capability = _______

18
Quality measures: LEGO example

• LTL = 9.5 mm, UTL = 9.7 mm, μ = 9.65 mm, σ = 0.025 mm

2
• Sigma Capability = _______
• What if μ shifts to 9.60 mm?
• Sigma Capability = _______

19
Quality measures: LEGO example

• LTL = 9.5 mm, UTL = 9.7 mm, μ = 9.65 mm, σ = 0.025 mm

2
• Sigma Capability = _______
• What if μ shifts to 9.60 mm?
4
• Sigma Capability = _______
• What is the probability of defects?

20
Quality measures: LEGO example
• What is the probability of defects?

• P(brick is too short) = P(X ≤ 9.5) = 1 × 10 -11

Either convert normal distribution to standard normal distribution and
use the z-table or use excel: P(X ≤ a) = NORMSDIST((a-μ)/σ)
• P(brick is too tall) = P(X ≥ 9.7) = 1 - P(X ≤ 9.7) = 0.02275

• So, the probability of a defect is about 2.275%.

21
Improving process capability: Mean Shift

22
Improving process capability: Variability Reduction

23
Improving process capability: Both Variability Reduction and Mean Shift

24
A CEO Problem
• You are the CEO of a car company that will launch a new car
• The thickness of the fuel tank is critical in ensuring that the tank
does not leak after a crash. Based on crash tests, it has been
found that acceptable thickness should be between 0.086 to
0.094 inches. Thicker than 0.094 inches tanks are less likely to be
punctured, but are more costly to manufacture (not desirable).
• The engineering team told you that they designed a process that
produces tanks with thickness of 0.090 inches on average, with a
standard deviation of 0.001 inches.

• What is the sigma capability and the probability of a defect?

https://www.youtube.com/watch?v=lgOxWPGsJNY
25
A CEO Problem

( )
0.090 − 0.086 0.094 − 0.090
z = min , =4
0.001 0.001

• This is 4-sigma process.

• The probability of no defects is 99.9%.

Do you give a green light for the launch?

26
Isn’t 99.9% Quality of Service Enough?!

• No electricity, water or heat for 8.8 hours each year

• No telephone service or TV transmission for 10 minutes each week

• Two short (or long) landings at DTW each week

• At least 20,000 wrong prescriptions per year

• 4,220 newborn infants dropped by doctors or nurses onto hospital

floors

27
Isn’t 99.9% Quality of Service Enough?!
• Motorola developed the “Six Sigma” quality improvement process in
1986 to improve process so that the probability of defects is
statistically insignificant.

• A six sigma process produces two defects every one billion products
(practically zero defects)
28
Six Sigma
• A philosophy and set of methods to eliminate defects in product/
service processes

29
Defects in a multi-step process

P(no defect) = P(no defect in A)*P(no defect in B)

30
Isn’t 99.9% Quality of Service Enough?!
• In a manufacturing process, the output is defective if even just one
step produces a defect…

• If each step is a “4-sigma process” with no-defect probability 99.9%

• If each step is a “6-sigma process”

31
Why aim for 6 sigma?

Even if process looses (temporarily) its centering, low defect rate

With an average center shift of 1.5 sigma, still ≤ 4 defects / million
Foundation of Six-Sigma: Two views of quality

TRADITIONAL VIEW (Quality Assurance) TAGUCHI’S VIEW

“Good enough” if within range Keep improving: mistakes are costly!

COST
COST

TARGET TARGET
Tolerance Tolerance

33
Implementing Six Sigma (DMAIC)

34
Process Control

• We’ve gotten our process to a place we’re happy with –

will it stay that way?
• There is still variability in the system!

• Why might the process change?

• Machine miscalibration, workers getting lax or rusty,
suppliers changing materials

35
Process Control

• Line 1: write letter R 8 times with your usual writing hand

• Line 2: write letter R 8 times with the opposite hand
• Line 3: write letter R 4 times with right and 4 times with left hand

36
Process Control

Common Cause Variation (low level)

Common Cause Variation (high level)

Assignable Cause Variation

• Need to measure and reduce common cause variation

• Identify assignable cause variation as soon as possible

37
Two Types of Causes for Variation

Common
Causes

Special/
Assignable Causes

38
Process control for LEGO
• Suppose you are the quality inspector of LEGO. How do you know if
your process is behaving normally?
• You check the samples of bricks. If the height of a brick is less than
9.596 mm or more than 9.604 mm, you look for causes and correct if
necessary.
• The control limits are 9.596 (LCL) and 9.604 mm (UCL).

39
Using a control charts
• A control chart plots sample data over time
• Used to monitor process over time to ensure it remains stable and is
operating normally.
• Inspect samples that are outside of control limits.

40
We can have various cases…

UCL UCL UCL

µ µ µ
LCL LCL LCL

UCL UCL UCL

µ µ µ
LCL LCL LCL

41
Patterns in Control Charts

Upper control limit

Target

Lower control limit

Normal behavior. Process is "in
control."

42
Patterns in Control Charts

Upper control limit

Target

Lower control limit

One plot out above (or below).
Investigate for cause. Process is "out
of control."

43
Patterns in Control Charts

Upper control limit

Target

Lower control limit

Trends in either direction, 5 plots.

Investigate for cause of progressive
change.

44
Patterns in Control Charts

Upper control limit

Target

Lower control limit

Two plots very near lower (or
upper) control. Investigate for
cause.

45
Patterns in Control Charts

Upper control limit

Target

Lower control limit

Run of 5 above (or below) central

line. Investigate for cause.

46
Patterns in Control Charts

Upper control limit

Target

Lower control limit

Erratic behavior. Investigate.

47
Control Limits vs. Tolerance Limits
• Control Limits:
• Whether the process is performing predictably
• Tolerance Limits (specification limits):
• Whether the process is performing acceptably (by customers)
• “voice of the process” vs. “voice of the customer”

Tolerance Limits Control Limits

Customers Producers,
Who determines?
Designers Quality Controllers

What happens Actions need to be

Defects
beyond limits? taken

48
How To Choose Control Limits?
• Control Limit must fall within the Tolerance limits (specification limits):

Is tighter control always better?

• Tight controls: More likely to detect problems, but may end spend all your time
chasing problems
• Tradeoff between failure costs and prevention costs

49
Practice Problem
Service quality control at Sigma Auto Wash requires the length of its
service to 30 ± 3 min. A current report indicates that the current average
service time is 31 min with a standard deviation of 1 min.
a) Calculate the sigma capability for this example

b) What is the probability of a defective service?

50
Practice Problem
Service quality control at Sigma Auto Wash requires the length of its
service to 30 ± 3 min. A current report indicates that the current average
service time is 31 min with a standard deviation of 1 min.
a) Calculate the sigma capability for this example

⎧ X − LTL UTL − X ⎫ ⎧ 31 − 27 33 − 31⎫

z = Min ⎨ , ⎬ = Min ⎨ , ⎬ = Min{4, 2} = 2
⎩ σ σ ⎭ ⎩ 1 1 ⎭

b) What is the probability of a defective service?

Probability of defect = P(too short service) + P(too long service) = 0.02278
• P(too short service) = P(X ≤ 27) = P(z ≤ (27-31)/1) = P(z ≤ -4) = 0.00003
Or using Excel: NORMSDIST(-4) = 0.00003
• P(too long service) = P(X ≥ 33) = 1 - P(X ≤ 33) = 1 - P(z ≤ 2) = 0.02275
Or using Excel: NORMSDIST(2) = 0.02275
51
 Supply Chain Management

Class 14
Toyota Production System &
Lean Operations

Sajjad Najafi

1
Reminder: LF Simulation Report (Due March 30)
Deliverables: Your team will write a three-page (main body) executive
summary of the actions taken by the team during the game including the
reasonings and the supporting analysis for what you did and suggestions of
what should have been done to improve the factory performance (if different
from what you did).
The team representative submits the report (only in PDF format) to the
following Dropbox links (the name of the PDF file should only consist of
your Team ID chosen at the registration time):
EN01: https://www.dropbox.com/request/CbAndcXWk0SqPcM7178n
EN03: https://www.dropbox.com/request/rSy7LEjEwqglJ3t24Rh8
EN04: https://www.dropbox.com/request/QjDH3hUAn6ARRBTR04bi

2
Quiz (Toyota Production System)
Available from March 31st and remains open until Final Exam.

For all quizzes:

Available until Final Exam
Multiple attempts are allowed
Effort based (not grade based): Students will get full grade of a quiz as long
as they take the quiz on Blackboard even once and regardless of the actual
grade displayed.

3
What did we learn last time?
Who is a better target shooter?

(A) (B)
Less variability = Higher quality
Not just the mean is important, but also the variance !
5
What is the root cause of all quality problems?
Without variation either all products are non-defective
or all are defective

Inputs Output

Process Variation in the output

is the root cause of
quality problems
Variations

Whether or not a product is defective depends on

the set of design specifications.

6
Quality Measure: Sigma Capability
• How many standard deviations the process mean is away from the
closest tolerance limit?

• Higher Sigma Capability index means that the process has a higher
quality of conformance.
7
Sigma capability example

If Sigma Capability z ≥ 2 ⇒
Probability of defect ≤ 5%

If Sigma Capability z ≥ 3 ⇒
Probability of defect ≤ 0.3%

8
A CEO Problem
• You are the CEO of a car company that will launch a new car
• The thickness of the fuel tank is critical in ensuring that the tank
does not leak after a crash. Based on crash tests, it has been
found that acceptable thickness should be between 0.086 to
0.094 inches. Thicker than 0.094 inches tanks are less likely to be
punctured, but are more costly to manufacture (not desirable).
• The engineering team told you that they designed a process that
produces tanks with thickness of 0.090 inches on average, with a
standard deviation of 0.001 inches.

• What is the sigma capability and the probability of a defect?

https://www.youtube.com/watch?v=lgOxWPGsJNY
9
A CEO Problem

( )
0.090 − 0.086 0.094 − 0.090
z = min , =4
0.001 001

• This is 4-sigma process.

• The probability of no defects is 99.9%.

Do you give a green light for the launch?

10
Isn’t 99.9% Quality of Service Enough?!
• Motorola developed the “Six Sigma” quality improvement process in
1986 to improve process so that the probability of defects is
statistically insignificant.

• A six sigma process produces two defects every one billion products
(practically zero defects)
11
Isn’t 99.9% Quality of Service Enough?!
• In a manufacturing process, the output is defective if even just one
step produces a defect…

• If each step is a “4-sigma process” with no-defect probability 99.9%

• If each step is a “6-sigma process”

12
This class
• Toyota Production System (TPS)
• Lean Operations: A concept that aims to reduce costs by
eliminating all forms of waste

• Just-In-Time (JIT): A concept that aims to produce or procure

the products at the right time at the right quantity ultimately
eliminating inventories

13
The Machine That Changed the World

International Motor Vehicle Program (IMVP)

Toyota Production System (TPS), aka Lean

Toyota Production System (TPS)
What is Lean Operations?

• Eliminate Waste (Muda無駄)

• Anything that does not add value

• Anything that does not help meet

customer requirements

• Anything customers would not be

willing to pay for

Taichi Ohno
The Seven Deadly Wastes (Muda)
1. Inventory (when more than needed)
2. Overproduction
3. Waiting
4. Motion
5. Transportation handling
6. Overprocessing
7. Defect correction

• How can we eliminate waste?

• How can we make operations leaner?

https://www.youtube.com/watch?v=v429BrFiYEk
17
Flow Time (In)efficiency

Industry Process Actual Average Theoretical Flow Time

Flow Time Flow Time Efficiency
Life Insurance New Policy 72 hrs. 7 min. 0.16%
Application
Consumer New Graphic 18 days 2 hrs. 0.14%
Packaging Design
Commercial Consumer Loan 24 hrs. 34 min. 2.36%
Bank
Hospital Patient Billing 10 days 3 hrs. 3.75%
Automobile Financial 11 days 5 hrs 5.60%
Manufacture Closing
• Actual Flow Time: All the time unit spends in the stage or system (includes waiting times)
• Theoretical Flow Time: Minimum time unit spends in the stage or system if no wait
Lean Transformation

Process Metrics

Flow time efficiency = Theoretical flow times/Actual Flow Time

Lean transition raises this ratio

Capacity Utilization = Throughput/Capacity

Lean transition raises this ratio

Variability ratio = Cp = Standard deviation of process time divided by mean

Lean transition lowers variability

19
Being lean: Eliminate all seven types of waste
1. Inventory
2. Overproduction
3. Waiting
4. Motion
5. Transportation handling
6. Overprocessing
7. Defect correction

20
Toyota Production System (TPS)
Being Lean: Eliminate Inventory
The more incidents, the
Expose problems more inventory required to
keep operations afloat...

...but the more inventory,

the less visible the incidents
and underlying causes

Production

Bad
Poor Quality Material Long Lead-
Poor
Time Maintenance
22
Being Lean: Eliminate Inventory
Remove the Lower inventory increases
obstacle problem visibility

Slowly lower the water to

expose the rocks

WIP reduced

Bad
Poor Quality Material Long Lead-
Poor
Time Maintenance
23
Being Lean: Eliminate Inventory
Remove the Lower inventory increases
obstacle again! problem visibility

Slowly lower the water to

expose the rocks

WIP reduced further

Poor Quality Long Lead-

Poor
Time Maintenance
24
Being lean: Eliminate all seven types of waste
1. Inventory
2. Overproduction
3. Waiting
4. Motion
5. Transportation handling
6. Overprocessing
7. Defect correction

25
Toyota Production System (TPS)
Being Lean: Eliminate overproduction
• Pull vs. Push:

• Push: Input availability triggers execution; Production pushes product

Stage 1 Stage 2 Stage 3

Inputs Outputs

• Pull: Output lack of availability triggers execution; Customer demand

pulls product

Stage 1 Stage 2 Stage 3

Inputs Outputs

Information flow
27
Material flow
Being Lean: Eliminate overproduction
• Controlling the flow with a push system:

Traditional approach:
Anticipation

Push system: Each

workstation produces at its
28 maximum capacity
Being Lean: Eliminate overproduction
• Controlling the flow with a pull system:

JIT approach:
Reaction

Pull system: Each workstation is controlled by

the short-term requirements of downstream
flow29
Being Lean: Eliminate Overproduction
• Implementing a pull system: The Kanban system

Full container with Kanban card on it

Kanban card

• The Kanban card is actually the work authorisation form. So,

number of Kanban cards determines the level of WIP

30
Being Lean: Eliminate Overproduction
• Example of a Kanban card
Card ID Card #

Item #

Item ID
Description

Number of parts Vehicle

per container

31
Toyota Production System (TPS)

https://www.youtube.com/watch?v=P-bDlYWuptM
Being lean: Eliminate all seven types of waste
1. Inventory
2. Overproduction
3. Waiting
4. Motion
5. Transportation handling
6. Overprocessing
7. Defect correction

33
Toyota Production System (TPS)
Being Lean: Eliminate Overproduction
• One-Unit-at-a-Time Flow

No Waiting and No Batches Waiting and Batches

Being Lean: Eliminate Waiting
• One-Unit-at-a-Time Flow: Reducing the batch size

Example Process A B C D

1 min/job 1 min/job 1 min/job 1 min/job

Batch Mfg Flow Mfg

4 (Lot size = 4) Space 1 (Lot size = 1) Space
0 0 1
1 1
1
2 2
1
3 4 3
4 4 1

A B 4
C D A B C D
8 8

4
12 12
Time 4 Time
16 mins 36 4 mins
Being Lean: Eliminate Waiting
• Reduce the size of production batches
• Reduce setup times
• Major components of setup cost:
• Machine downtime
• Loss of capacity due to setups

The shorter the setup times (the lower S),

the smaller the production batch can be!

37
Being Lean: Eliminate Waiting
• How to Reduce setup times: Single Minute Exchange of Die (SMED) Method
• Discriminate between “internal” and “external” setup tasks
• Internal setup tasks: can only be executed while the machine is stopped
• External setup tasks: can be done while the machine is still operating
• Transform internal operations into external operations
• Preparation prior to shutdown and changeover

38
Being Lean: Eliminate Waiting
• Mixed-model production (Heijunka)

39
Toyota Production System (TPS)
Being Lean: Eliminate Overproduction
• Takt time (translation of customers’ demand into production rate)
• Production cycle should be synchronized with demand cycle

Orchestra: Follow Tact imposed by conductor.

JIT: Follow Tact imposed by demand.
Being lean: Eliminate all seven types of waste
1. Inventory
2. Overproduction
3. Waiting
4. Motion
5. Transportation handling
6. Overprocessing
7. Defect correction

42
Eliminating unnecessary motion (The Founder, 2016)
Being Lean: Eliminate unnecessary transportation
• Layout re-engineering

44
Being Lean: Eliminate unnecessary transportation
• Layout re-engineering • Requirements
• Large enough volume of each
Workshop organization based
on product routings product (capacity utilization)
• Low-cost tools (duplication)
• High machine reliability

• Advantages
• Streamline flows
• Minimized moving distances
• Reduced work-in-process
• Improved quality
• Improved communication
45
Being lean: Eliminate all seven types of waste
1. Inventory
2. Overproduction
3. Waiting
4. Motion
5. Transportation handling
6. Overprocessing
7. Defect correction

46
Being Lean: Eliminate unnecessary processing
• Unnecessary processes and operations traditionally accepted as
necessary
• Performing unnecessarily stringent inspection when suppliers’
processes ensure defect-free products
• Continuing to train employees in tasks/skills that are no longer needed
• Healthcare examples:
• Gathering unnecessary information from patients
• Nurses and doctors asking the same questions
• Producing hard copies when computer files are sufficient
Being lean: Eliminate all seven types of waste
1. Inventory
2. Overproduction
3. Waiting
4. Motion
5. Transportation handling
6. Processing
7. Defect correction

48
Toyota Production System (TPS)
Being Lean: Eliminate defects
• Quality control
• Fool-proof/Fail-safe design (Poka-Yoke)

• Inspection at the source (Jidoka)

• Line-stopping empowerment (Andon)

50
Being Lean: Eliminate defects
• Poka-Yoke: Fool-proof design

51
Being Lean: Eliminate defects
• Jidoka: Inspection at the source

52
Being Lean: Eliminate defects
• Andon Cord (a form of Jidoka): Line-stopping empowerment

53
Being Lean: Eliminate defects
• Quality at the source
• Control quality when and where problems occur

54
Toyota Production System (TPS)
Even Giants Fail - Toyota

https://www.youtube.com/watch?v=NGe3EOJ-CMY

“…the company has delayed recalls, kept a tight lid on disclosure of potential problems and attempted to
blame human error in cases where owners claimed vehicle defects.”

Jan. 22, 2010: “There is a possibility that certain accelerator-pedal mechanisms may, in rare instances,
mechanically stick in a partially depressed position or return slowly to the idle position,” said Toyota
spokesman Irv Miller. [part made by US supplier, whom Toyota did not name]

“We pursued growth over the speed at which we are able

to develop our people and our organization, and we should
sincerely be mindful of that. I regret that this has resulted
in the safety issues described in the recalls we face today”
Akio Toyoda, Toyota CEO

56
Success stories
• Company experiences with implementation of “lean” principles
• Average inventory reduction of about 50%

• Reduction of throughput time of 50% to 70%

• Reduction in setup times of up to 50% without major investments in

plant or equipment

57
When does JIT not work?
• End-product demand is highly variable
• JIT needs a stable schedule

• JIT requires a cultural change

• Workers need to accept responsibility for making process changes, managers should
be willing to cede control

58
Summary
• Lean operations/TPS aims to reduce all types of waste in an
operational process

• Various tools/concepts from TPS

• Muda
• JIT
• Kaizen
• Heijunka
• Jidoka
• Kanban
• Andon

59
 Supply Chain Management

Class 15
Transportation

Sajjad Najafi

1
What did we learn last time?
The Seven Deadly Wastes (Muda)
1. Inventory
2. Overproduction
3. Waiting
4. Motion
5. Transportation handling
6. Overprocessing
7. Defect correction

• How can we eliminate waste?

• How can we make operations leaner?

3
What’s Left?

SOURCES OF Process
SUPPLY CHAIN DESIGN SUPPLY CHAIN MGMT
ADVANTAGE Execution

INSTRUMENTS JIT/ TPS SC Sourcing SCM Projects/

Structure Review

SESSIONS 13, 14 15,16 17, 18 19-21 22-24

FOCUS ON THE SUPPLY CHAIN

LITTLEFIELD
SIMULATION
Definitions of the SCM
“Supply Chain Management encompasses the planning and
management of all activities involved in sourcing and procurement,
conversion and all Logistics Management activities. Importantly, it also
includes coordination and collaboration with channel partners, which
can be suppliers, intermediaries, third-party service providers, and
customers. In essence, Supply Chain Management integrates supply and
demand management within and across companies”
Source: Council of Supply Chain Management Professionals

“Supply Chain Management deals with the control of materials,

information and financial flows in a network consisting of suppliers,
manufacturers, distributors and customers”
Source: Stanford Supply Chain Forum

“Call it distribution or logistics or supply chain management. By

whatever name, it is the sinuous, gritty and cumbersome process by
which companies move material, parts and products to customers”
5 Source: Fortune Magazine
Traditional View: Supply Chain Management = Logistics

Logistics in the Manufacturing Firm

• Profit 4% Profit
Logistics
• Logistics Cost 21%
Cost

Marketing
• Marketing Cost 27% Cost

Manufacturing
• Manufacturing Cost 48% Cost

6
Importance of Logistics for Economy

2017 US business logistics costs (USBLC) = $1.495 trillion or 7.7 % of

US GDP – more than the GDP of many countries!
– Transportation cost: 64.6%

– Inventory carrying costs: 28.6%

– Other (IT, services) : 6.7%

What is the trend that we see now?

(Source: CSCMP)

7
Transportation Choices

Location of suppliers, plants, warehouses/cross-docks and customers

determine the chain or network structure to be supported by
transportation flows.

8
Transportation Modes

Mode…

9
So, how do you choose transportation mode?

10
 So, how do you choose transportation mode?

What does it influence? Metrics of competitiveness?

11
 So, how do you choose transportation mode?

Mode of transportation influences… Metric of competitiveness?

• Cost • Cost

• Size of individual items & overall capacity • Availability

(Service level)
• Transportation time & reliability
• Frequency of shipment • On-time delivery
• Flexibility
• Flexibility
• Customer lead-time requirements
• Risk
– Damages (Indiana Furniture Industry
created own fleet)
– Delays (Hanjin transport)
– Strikes
–…

12
Transportation Mode Choice Criteria

• Quantities moved
• Nature and physical characteristics of goods
• Lead time customer requirements
• Customer location (distance, concentration)
• Transportation time
• Frequency of transportation
• Reliability (lead time variance)
• Total cost (all costs induced by transportation)

13
 Economic drivers influence transportation costs

• Distance
• Weight
• Density
• Stowability
• Handling
• Frequency of shipment
• Liability
• Risk
• Market
• Location
• Route (China to US > 2* US to
China)
• Transportation Mode…
Source: Transportation Operations, McGraw-Hill (Chapter 9)

14
Road Transport

15
Examples of truck loading

16
Truck loading

European standard pallets: 1200 mm x 800 mm or 1200 mm x 1000 mm

33 pallets

13,50 m x 2,40 m

32 pallets

Depending on type of goods/packaging, possibility of putting

several layers of pallets

17
Road transportation options
• For small parcels:
• Package services
• For pallets:
• LTL (Less than Truck Load) carrier
• TL (Truck Load)

Cost Cost

Weight Weight

Packages LTL / TL
18
Truckload (TL)
• Pros
• Door to door service
• Speed
Long-haul
transportation • Cost (if truck is full)

• Cons
Direct door-to-door
• Cost (if truck is not full)
• Frequency
• High inventory

• Carriers
• Own fleet
19 • Truck rental
 Less Than Truckload (LTL)

Milk run
• Pros
• Transport in small quantities
• High frequency
• Low inventory
Factory

port
• Cons
rans
l t
au
Lon
g-h • Longer lead time due to milk
runs
• High cost
•No handling during
the long-haul
transport • Carriers
•Same truck carries • Mory, Norbert Dentressangle,
out the milk runs
and the long haul Exel, Danzas
20
Packages
• Pros
• Flexibility
Packages
• High frequency
• Low inventory
Grouping
Center

Long-haul
(large truck) • Cons
Distribution
Center
• High cost

Milk Runs
(small truck)
• Carriers
• DHL, Fedex, etc

21
A typical trade-off: Response time versus Cost

Transportation Cost

not worth it “unaffordable”

to decrease to increase
response response time
time further further

Speed
(Response Time)

22
Rail Transportation

• A car: 50 tons
• Full train: 2 000 to 3 000 tons

• Requires two handling

operations (at departure and
arrival stations)

• Often slow

23
Sea Transportation

•Types:
•Container ships

24
Sea Transportation

•Types:
• Bulk carriers: coal, grains, ore, etc.

25
Sea Transportation
• Types
• Specialized vessels: car, liquefied gas, forest product, etc.

Car carrier Liquefied gas carrier

26
How Containerization Shaped the Modern World

https://www.youtube.com/watch?v=Gn7IoT_WSRA&ab_channel=TED-Ed
27
Container Vessel

• Container Vessel (up to 400 meters long) carries between 1500 – 20,000 6
meters long containers
• Container types:
• 20 feet or 40 feet or 15 / 30 tons (depending on goods density)
• normal / refrigerated
• full container / grouping
• Handling (30 per hour)
28
How are containers loaded?

https://www.youtube.com/watch?v=kj7ixi2lqF4

29
Air Transportation

30
Air Transportation

B 747 - 400

747 cargo: 100 to 120 tons

29 pallets of 96” x 125” (10 feet)

31
Cost Components
• Transportation costs
• Loading
• Outbound transportation
• Transshipment
• Main transportation
• Transshipment
• Inbound transportation
• Unloading
• Inventory costs
• Departure point (preparation of shipping load)
• In-transit (depends on transportation lead time)
• Arrival point (depends on transportation lot size)
• Information processing costs
• Ethical and Environmental Issues
32
Choice of Transportation Modes in SC

33
Reminder: Inventory Management

34
Deterministic Demand (EOQ)
Have dxL units while waiting for shipments

35
Uncertain Demand
Is it enough to have dxL units while waiting for shipments?

36
Safety Stock
Safety Stock (SS) protects against stockout during the lead time

37
Inventory Management

• So far only ”Demand” uncertainty

What if “Supply” is also uncertain?

(Lead Time uncertainty)

38
Recipe: Safety Stock with Demand and Supply uncertainty
Demand: Normal Distribution N ∼ (μd, σd)
Lead Time: Normal Distribution N ∼ (μL, σL)
1. A target service level SL is given to you.
2. Find the corresponding z-value to the target service level in
the standard normal table. Or using z = normsinv(SL) in Excel.
3. Safety Stock is

SS = z ⋅ μL ⋅ σ d2 + μd2 ⋅ σ L2

If demand is constant (σd = 0), then

SS = z ⋅ σLD

where σLD = μd ⋅ σL is the standard deviation of demand over the

lead time.
FIXIT Hardware Chain - An Example

Part 1: FIXIT is a hardware chain that operates in the Piqsburg area and receives shipments from its supplier
ACME in Portland. Demand at FIXIT each month is 30,000 packs of bulky widgets. Currently shipments from
ACME to FIXIT are sent by truck, the capacity of a truck is 5000 packs and it costs $3000. The transit rme is 5
days. In addiron, the data shows that the lead-rme has a standard deviaron of 2 days. FIXIT’s accounrng
group has esrmated that the annual holding cost is 20% of the cost of the product. ACME supplies a case of
widgets for a price of $15 per pack. Assume a desired service level of 95%.

Part 2: FIXIT has been provided an offer to move product by RAILCAR, a RAIL brokerage that works closely
with railroads to schedule shipments. The shipment price for one wagon load of widgets from Portland to
Piqsburg is $6000, with a capacity of 15,000 packs. Transit rme is expected to be 15 days with a standard
deviaron of 10 days. If FIXIT were to use rail to ship product, what would be the impact on its costs?

Note: FIXIT’s demand is constant (i.e., standard deviation of demand is zero).

28
FIXIT Hardware Chain - An Example

ACME – supplier
$15 / pack of widgets
FIXIT - hardware chain
monthly demand: 30,000 packs
of widgets
Holding cost rate: 20%/year
41
Target Service Level: 95%
Margin: $2

Capacity: 5,000 packs Capacity: 15,000 packs

Cost: $3000 / truck Cost: $6000 / load
Transit: avg. 5 days, σ=2 days Transit: avg. 15 days, σ=10 days

28
FIXIT Hardware Chain - An Example

ACME – supplier
$15 / pack of widgets
FIXIT - hardware chain
monthly demand: 30,000 packs
of widgets
Holding cost rate: 20%/year
42
Target Service Level: 95%
Margin: $2

Capacity: 5,000 packs Capacity: 15,000 packs

Rail is better?
Cost: $3000 / truck Cost: $6000 / load
Transit: avg. 5 days, σ=2 days Transit: avg. 15 days, σ=10 days

28
Need to Know “Total Logistics Costs”

All costs that depend on the mode of transportation, including costs of

transportation, warehousing, inventory carrying, administration, order
processing, insurance.

43
Inventory Calculations
• Inventory (on hand + on order)
Q
• On-hand (warehouse) inventory (= +z⋅ μL ⋅ σ d2 + μd2 ⋅ σ L2 )
2
Q
• Cycle inventory (= 2
)
2 2 2
• Safety inventory (= z ⋅ μ ⋅
L dσ + μ ⋅
d L) σ

• On-order (pipeline) inventory (= μd ⋅ L, by Little’s Law)

• Why inventory level depends on the choice of transportation?

• Lead Time depends on the choice of transportation. Then, Lead
Time changes the on hand (via safety inventory) and on order
inventory.
• Each mode has a certain Q which impacts the cycle inventory.
44
 Recipe: Choice of Transportation Mode
Demand: Normal Distribution N ∼ (μd, σd)
Lead Time: Normal Distribution N ∼ (L, σL)

1. A target service level SL is given to you. Find the corresponding z-value from
the table.
2. Find transportation cost =
(average demand in each period / capacity of the mode) ⋅ cost per delivery
3. Calculate the total inventory (on-hand + on-order)
• On-hand (warehouse) inventory = Q/2 + z ⋅ μL ⋅ σ d2 + μd2 ⋅ σ L2

• On-order (pipeline) inventory = μd ⋅ L (by Little’s Law)

3. Find the total inventory holding cost = (Q/2 + z ⋅ μL ⋅ σ d2 + μd2 ⋅ σ L2 + μd ⋅ L) ⋅ H
(H is per unit holding cost per period of time over which transportation cost is
computed.
4. Find total cost = transportation cost + total inventory holding cost. Then, pick
the mode with the minimum cost.
 FIXIT - Continued

Capacity: 5,000 packs 15,000 packs

Cost: $3000 / truck $6000 / truck
Transit: 5 days, σ=2 days 15 days, σ=10 days
Total costs: $20,697 /month $ 21,737 /month

46
 FIXIT - Continued

Capacity: 5,000 packs 15,000 packs 15,000 packs

Cost: $3000 / truck $6000 / truck $6000 / truck
Transit: 5 days, σ=2 days 15 days, σ=10 days 10 days, σ=3 days
Total costs: $20,697 /month $ 21,737 /month $17,609 /month

There is NO direct relationship between the response time and cost, if you
consider the TOTAL LOGISTICS COSTS!

For FIXIT calculations, see Excel!

47
 Supply Chain Management

Class 16
Distribution Networks and Risk Pooling

Sajjad Najafi

1
What did we do last time?
 Recipe: Choice of Transportation Mode
Demand: Normal Distribution N ∼ (μd, σd)
Lead Time: Normal Distribution N ∼ (μL, σL)

1. A target service level SL is given to you. Find the corresponding z-value from
the table.
2. Find transportation cost =
(average demand in each period / capacity of the mode) ⋅ cost per delivery
3. Calculate the total inventory (on-hand + on-order)
• On-hand (warehouse) inventory = Q/2 + z ⋅ μL ⋅ σ d2 + μd2 ⋅ σ L2

• On-order (pipeline) inventory = μd ⋅ L (by Little’s Law)

3. Find the total inventory holding cost = (Q/2 + z ⋅ μL ⋅ σ d2 + μd2 ⋅ σ L2 + μd ⋅ L) ⋅ H
(H is per unit holding cost per period of time over which transportation cost is
computed.
4. Find total cost = transportation cost + total inventory holding cost. Then, pick
the mode with the minimum cost.
FIXIT Hardware Chain - An Example

ACME – supplier
$15 / pack of widgets
FIXIT - hardware chain
monthly demand: 30,000 packs
of widgets
Holding cost rate: 20%/year
4
Target Service Level: 95%
Margin: $2

Capacity: 5,000 packs Capacity: 15,000 packs

Cost: $3000 / truck Cost: $6000 / load
Transit: avg. 5 days, σ=2 days Transit: avg. 15 days, σ=10 days

28
 FIXIT - Continued

Capacity: 5,000 packs 15,000 packs

Cost: $3000 / truck $6000 / truck
Transit: 5 days, σ=2 days 15 days, σ=10 days
Total costs: $20,697 /month $ 21,737 /month

5
 FIXIT - Continued

Capacity: 5,000 packs 15,000 packs 15,000 packs

Cost: $3000 / truck $6000 / truck $6000 / truck
Transit: 5 days, σ=2 days 15 days, σ=10 days 10 days, σ=3 days
Total costs: $20,697 /month $ 21,737 /month $17,609 /month

There is NO direct relationship between the response time and cost, if you
consider the TOTAL LOGISTICS COSTS!

For FIXIT calculations, see Excel!

6
This class
• Reactive capacity
• Risk pooling

7
Quiz (Risk Pooling)
Available from April 7th and remains open until Final Exam.

For all quizzes:

Available until Final Exam
Multiple attempts are allowed
Effort based (not grade based): Students will get full grade of a quiz as long
as they take the quiz on Blackboard even once and regardless of the actual
grade displayed.

8
Reactive Capacity: Combining Transportation Modes
An example of combining several transportation modes

Quantities
FedEx
3

Contracted
2 Fleet

1 Own Fleet

time

Advantage?
10
Another example of combining several transportation modes

Quantities
FedEx
3

Contracted
2 Fleet

1 Own Fleet

time
Advantage:
• Better customer service
• High transportation costs, only IF demand turns out to be very high

11
Would you ever consider both?

+ ?

(10 days)

12
Reminder: Newsvendor Model
Newsvendor Problem
• Every morning, a newsvendor purchases newspapers to sell
• Too Many - Too Few Tradeoffs (overage - underage tradeoff)
• How many newspapers should he purchase?
• You can never guess demand right!
• You always have either excess or shortage inventory. Which
mistake is less costly?
• You order a quantity to make sure that the more costly
mistake happens with lower probability
Profits Just right!

Too little Too much

14 Quantity
 How to Compute Q* Given Critical Ratio

• With Normal Demand N(μ, σ):

Q* is the quantity at which

Cu
P(D ≤ Q) =
Cu + Co

Step 1: Find the z value corresponding to SL* from the z-table

Step 2: Convert z to the order quantity Q*, Q* = μ + zσ Q* = μ + zσ
Examples
• Example 1: Assume demand is N(200,20)
Cu = $100 Co = $250

SL*

SL*
SL*
Examples
• Example 1: Assume demand is N(200,20)
Cu = $100 Co = $250

More costly mistake! Most afraid of this!

SL*

SL*
SL*
Examples
• Example 1: Assume demand is N(200,20)
Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)

SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
Q* = 189
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189

SL*
Examples
• Example 1: Assume demand is N(200,20)
Cu = $100 Co = $250

More costly mistake! Most afraid of this!

$100 P(D > Q*)

SL* = P(D ≤ Q*) = = 29 % SL*
$100 + $250
Q* = 189
140 150 160 170 180 190 200 210 220 230 240 250 260

Q* = μ + zσ = 189
• Example 2: Assume demand is N(850,150)
Cu = $90 Co = $20

More costly mistake! Most afraid of this!

$90 SL*
SL* = P(D ≤ Q*) = = 82 %
$90 + $20
Q* = 988 P(D > Q*)
350 425 500 575 650 725 800 875 950 1025 1100 1175 1250

Q* = μ + zσ = 988
Newsvendor Recipe: Normal Demand Distribution (μ, σ)

Finding the profit maximizing order quantity:

𝐶𝑢
1. Determine SL* =
𝐶𝑢 + 𝐶𝑜

2. If you have access to Excel, then Q* = norm.inv (SL*, μ, σ).

3. If not, find the corresponding z-value to the optimal service level in the standard
normal table. Then, Q* = μ + zσ.

For Performance Measures Calculations:

1. Expected Shortages = L(z) x σ (Find L(z) from L(z) table which is given to you)
2. Expected sales = exp. demand – exp. shortages
3. Expected leftovers = order size – exp. sales
4. Expected profits = (P × exp. sales + S × exp. leftovers) – C × Q*

20
Choosing Transportation Mode (variability in demand)

Train: 10 days (no variability) FedEx: 1 day (no variability)

Cost of transport: $0.3 per unit Cost of transport: $10 per unit

Demand: N(mean = 100, sigma = 10) per month

Margin: $1
No customers should be left without the product!

21
Choosing Transportation Mode (variability in demand)

Train: 10 days (no variability) FedEx: 1 day (no variability)

Cost of transport: $0.3 per unit Cost of transport: $10 per unit

Demand: N(mean = 100, sigma = 10) per month

Margin: $1
No customers should be left without the product!

How much to order by train?

How many on average will be covered (ordered) by FedEx?

22
A Useful Demand Distribution: Normal Demand Distribution
Service Level = Critical Ratio = Prob of meeting all demand
= P(D≤Q*) = Cu / (Cu+Co)

Optimal TRAIN order

size:
Q= µ + zσ

23
A Useful Demand Distribution: Normal Demand Distribution
Service Level = Critical Ratio = Prob of meeting all demand
= P(D≤Q*) = Cu / (Cu+Co)

Real demand1

Optimal TRAIN order

size:
Q= µ + zσ

24
A Useful Demand Distribution: Normal Demand Distribution
Service Level = Critical Ratio = Prob of meeting all demand
= P(D≤Q*) = Cu / (Cu+Co)

Real demand1 Real demand2

Optimal TRAIN order

size:
Q= µ + zσ

Ordering more by train - less chance that will order expensively by

Fedex, but higher probability of overage

What are Cu and Co? 25

Choosing Transportation Mode (variability in demand)

Train: 10 days (no variability) FedEx: 1 day (no variability)

Cost of transport: $0.3 per unit Cost of transport: $10 per unit

Demand: N(mean = 100, sigma = 10) per month

Margin: $1

Cu =
Co =

26
Choosing Transportation Mode (variability in demand)

Train: 10 days (no variability) FedEx: 1 day (no variability)

Cost of transport: $0.3 per unit Cost of transport: $10 per unit

Demand: N(mean = 100, sigma = 10) per month

Margin: $1

Cu = FedEx cost - Train cost = $9.7

Co = Train cost - Salvage value = $0.3 - $0 = $0.3

CF = 9.7/(9.7+0.3) = 0.97 What does it mean?

27
Choosing Transportation Mode (variability in demand)

Train: 10 days (no variability) FedEx: 1 day (no variability)

Cost of transport: $0.3 per unit Cost of transport: $10 per unit

Demand: N(mean = 100, sigma = 10) per month

Margin: $1

Cu = FedEx cost - Train cost = $9.7

Co = Train cost - Salvage value = $0.3 - $0 = $0.3

CF = 9.7/(9.7+0.3) = 0.97 What does it mean?

z = 1.88 and L(z) = 0.012

Qtrain = mean + z*sigma = 100 + 1.88*10 = 119 units

28
Choosing Transportation Mode (variability in demand)

Train: 10 days (no variability) FedEx: 1 day (no variability)

Cost of transport: $0.3 per unit Cost of transport: $10 per unit

Demand: N(mean = 100, sigma = 10) per month

Margin: $1

Cu = FedEx cost - Train cost = $9.7

Co = Train cost - Salvage value = $0.3 - $0 = $0.3
What if FedEx is
CF = 9.7/(9.7+0.3) = 0.97 What does it mean?
cheaper?
z = 1.88 and L(z) = 0.012

Qtrain = mean + z*sigma = 100 + 1.88*10 = 119 units

29
How much will be delivered by FedEx on average?

Train: 10 days (no variability) FedEx: 1 day (no variability)

Cost of transport: $0.3 per unit Cost of transport: $10 per unit

Demand: N(mean = 100, sigma = 10) per month

Margin: $1
z = 1.88 and L(z) = 0.012
Qtrain = mean + z*sigma = 100 + 1.88*10 = 119 units

Qtrain Real demand E Shortage = L(z)*sigma = 0.012*10 < 1

Shortage = FedEx

30
What about other measures?
For Profit Calculations:
1. Expected Shortages = L(z) x std. deviation
2. Expected sales = exp. demand – exp. shortages
3. Expected leftovers = order size – exp. sales
4. Expected profits = (p × exp. sales + s × exp. leftovers) – c × Q*

31
What about other measures?
For Profit Calculations:
1. Expected Shortages = L(z) x std. deviation
2. Expected sales = exp. demand – exp. shortages
3. Expected leftovers = order size – exp. sales
4. Expected profits = (p × exp. sales + s × exp. leftovers) – c × Q*
Unit of Demand

Satisfied by NOT Satisfied by

train capacity train capacity

Sales (train) Shortage -> Fedex

Sales (fedex)

32
What about other measures?
For Profit Calculations:
1. Expected Shortages = L(z) x std. deviation
2. Expected sales = exp. demand – exp. shortages
3. Expected leftovers = order size – exp. sales
4. Expected profits = (p × exp. sales + s × exp. leftovers) – c × Q*
Unit of Demand Unit of Product ordered by Train

Satisfied by NOT Satisfied by

Sold Not sold
train capacity train capacity

Sales (train) Shortage -> Fedex Sales (train) Leftover

Sales (fedex)

33
What about other measures?
For Profit Calculations:
1. Expected Shortages = L(z) x std. deviation
2. Expected sales = exp. demand – exp. shortages
3. Expected leftovers = order size – exp. sales
4. Expected profits = (p × exp. sales + s × exp. leftovers) – c × Q*
Unit of Demand Unit of Product ordered by Train

Satisfied by NOT Satisfied by

Sold Not sold
train capacity train capacity

Sales (train) Shortage -> Fedex Sales (train) Leftover

Sales (fedex)

E Shortages (total) = 0
E Sales (Fedex) = L(z)*σ Shortage covered by Fedex
E Sales (train) = μ - L(z)*σ
E Sales (total) = E Sales (train) + E Sales (Fedex) = μ - L(z)*σ + L(z)*σ = μ
E Leftover = Q*train - E Sales (train) = Q*train - (μ - L(z)*σ)
E Profits = μ*$1 - $0.3 Q*train - $10 QFedex 34
Risk Pooling
Strategies for reducing safety stocks
• Risk Pooling
• Idea: Supplying independent demand streams from one location rather
than multiple locations requires less safety stock

• Tool: Adding independent and identically distributed random variables

• Mean:

• Std. Dev.:

• Applications:
• Component commonality in product design
• Portfolio effects in finance/risk pooling
• Storage centralization

36
Component Commonality

37
Problem: Risk Pooling - Component Commonality
• The daily demand for zipper A from apparel style 1 is normally
distributed with mean of 100 units and std. dev. of 30 units.
Similarly, the daily demand for zipper B from apparel style 2 is
normally distributed with mean of 300 units and a std. dev. of 40
units.

1. Assuming an order lead time of 1 day, how much safety stock should be
kept for each zipper to achieve a 98% service level (Hint: Use z=2 for
98% service level)? What would be the total safety stock level?
2. What would be the total safety stock level if the same type of zippers is
used for each style?

38
Problem: Risk Pooling - Component Commonality
• The daily demand for zipper A from apparel style 1 is normally
distributed with mean of 100 units and std. dev. of 30 units.
Similarly, the daily demand for zipper B from apparel style 2 is
normally distributed with mean of 300 units and a std. dev. of 40
units.

1. Assuming an order lead time of 1 day, how much safety stock should be
kept for each zipper to achieve a 98% service level (Hint: Use z=2 for 98%
service level)? What would be the total safety stock level?
2. What would be the total safety stock level if the same type of zippers is
used for each style?

SS = z ⋅ σd ⋅ L SS1 = 2 ⋅ 30 SS2 = 2 ⋅ 40 Total(SS) = 140

Combined Demand for A + B ∼ N(μ = 300 + 100,σ = 302 + 402 ) = N(400,50)
SScombined = 2 * 50 = 100 < 140

39
Benefits of risk pooling/centralization
• Centralization reduces safety stocks (pooling effect)
• Offers better service for the same total inventory investment or the
same service for a smaller total inventory investment

• Different methods to achieve pooling efficiencies:

• Physical centralization
• Information centralization
• Commonality
• Postponement/Late customization

• Cost savings are proportional to the square root of the number of

locations pooled

40
Inventory (Risk) Pooling - Centralisation
You keep Safety Stock to meet the demand with a desired service level
(with a corresponding z-score)

Decentralized Distribution Centralized Distribution

zσ zσ
zσ zσ
zσ
zσ

Total SS = k stores × zσ Total SS = z × std = z × k stores × σ

Remember Square Root Law: Total safety stock required to provide a specified service
level increases by square root of the number of locations in which it is held.
41
Centralization / Decentralization: Effect on safety stocks

Warehouse 1

Warehouse 2
Source
Warehouse 3

Warehouse 4

Demands

Decentralized Solution
SS = N ⋅ (zSLσL)

42
Centralization / Decentralization: Effect on safety stocks

Warehouse 1

Warehouse 2
Source Central
Source
Warehouse 3 Warehouse

Warehouse 4 Demand

Demands

Decentralized Solution Centralized Solution

SS = N ⋅ (zSLσL)

43
Centralization / Decentralization: Effect on safety stocks

Warehouse 1

Warehouse 2
Source Central
Source
Warehouse 3 Warehouse

Warehouse 4 Demand

Demands

Decentralized Solution Centralized Solution

SS = N ⋅ (zSLσL) > SS = zSL ⋅ N ⋅ σL

One centralizes the demands at the central warehouse

44
 Benefit of location pooling

1
safety stock at central depot 1
=
• Location pooling reduces demand 0.7 safety stock at N local depots N
uncertainty
• Reduced demand uncertainty reduces 0.5
the inventory needed
• Declining marginal returns to risk
pooling! 0.33

1 2 4 9 N

Most of the benefit can be captured by pooling only a few territories!

45
Amazon’s distribution network across time:

https://edition.cnn.com/interactive/2018/10/business/amazon-distribution-map/index.html

46
Risk pooling via location pooling

Pros, cons and alternatives

• Pros: Reduces demand uncertainty, which allows a firm to reduce inventory, increase
service, expand the product line or a combination of all three

• Cons: Location pooling moves inventory away from customers

• Creates inconvenience for sales reps / stores (longer lead time)
• May create costs to ship product to customers (more costly transport to customers)

• Alternatives: Virtual pooling

• Each rep / store keeps own inventory but shares inventory with nearby reps / stores
if needed

47
Summary: Factors Influencing Distribution Strategy

• Distribution network impacts Supply chain costs :

– Inventories (cycle stock, pipeline inventory, safety stock)
– Transportation
– Facilities and handling
– …

• Distribution network influences customer service :

– Response time
– Product availability
– Customer experience
– After sales service
– Returnability
– …

Not a decision to be taken by “logistics” alone!

48
Key Lessons about distribution

• When considering impact of new distribution system, need to consider impact on all
players of supply chain, i.e., consider impact on customers and suppliers in addition to
cost impact on logistics.

49
 Supply Chain Management

Class 17
Supplier Selection and Management

Sajjad Najafi

1
What did we do last time?
Choosing Transportation Mode (variability in demand)

Train: 10 days (no variability) FedEx: 1 day (no variability)

Cost of transport: $0.3 per unit Cost of transport: $10 per unit

Demand: N(mean = 100, sigma = 10) per month

Margin: $1

Cu = FedEx cost - Train cost = $9.7

Co = Train cost - Salvage value = $0.3 - $0 = $0.3

CF = 9.7/(9.7+0.3) = 0.97 What does it mean?

z = 1.88 and L(z) = 0.012

Qtrain = mean + z*sigma = 100 + 1.88*10 = 119 units

3
Centralization / Decentralization: Effect on safety stocks

Warehouse 1

Warehouse 2
Source Central
Source
Warehouse 3 Warehouse

Warehouse 4 Demand

Demands

Decentralized Solution Centralized Solution

SS = N ⋅ (zSLσL) > SS = zSL ⋅ N ⋅ σL

One centralizes the demands at the central warehouse

4
What’s Left?

SOURCES OF Process
SUPPLY CHAIN DESIGN SUPPLY CHAIN MGMT
ADVANTAGE Execution

INSTRUMENTS JIT/ TPS SC Sourcing SCM Projects/

Structure Review

SESSIONS 13, 14 15,16 17, 18 19-22 22-24

FOCUS ON THE SUPPLY CHAIN

LITTLEFIELD
SIMULATION
 April 19, end of the day in France
Deadline for project submissions (only in PDF)

• Submit to the following Dropbox link:

• EN01: https://www.dropbox.com/request/o983Yksqs4V7t7dsKOQy
• EN03: https://www.dropbox.com/request/mOJ1s6VWtYBurydGSNhY
• EN04: https://www.dropbox.com/request/4RNo1vgLE9pJrPTVbq4q
This Class
• The importance of sourcing

• Supplier selection and negotiation

• Long- vs short-term relationships

7
Sourcing

• Sourcing is the process of determining what to source, whom to

source it from, and how to manage risks and leverage
opportunities to support overall business strategy

Specs Inputs Transformation Demand Logistics

New Product Marketing

Sourcing Operations Distribution
Development & Sales

8
Historical Context of Sourcing

• Traditional: Large, vertically integrated production

▪ Henry Ford's Rouge plant: Iron ore, glass and rubber come in one end,
cars leave from the other end
▪ Purchasing manager only had clerical and scheduling duties

The Ford Rouge complex,

oil painting in 1949

9
Historical Context of Sourcing

• Then, many changes:

▪ Political – globalization of trade (GATT, NAFTA, EEC)
▪ Cheaper transportation – easy and quick to ship
▪ Cheaper communication – easy and quick to coordinate and
communicate half a world away

• Today: Outsourcing and globalized supply chains

▪ Sourcing manager – market analysis, spend analysis, negotiation, risk
management and strategy duties

10
What is the cost of raw materials?
Why is purchasing important?

Common cost breakdown

Materials Direct Labor
Other Costs

30 %

60 %
10 %

Mode Materials Direct Labor Other Costs

Percentage 60% 10% 30%

12
Why is purchasing important?

Example: cost breakdown of automobile companies

13
 Benefits of reduced spending
• Suppose you purchase materials for $6, which account for 60% of
the total cost, and you sell for $12. Demand = 100 units.

• A 50% increase in demand yields an extra $_____ of profit.

• Your marketing department would be proud!

• A 50% reduction in cost of material yields an extra $_____ of profit.

• Procurement costs are a crucial profitability lever

14
Main purchasing segments
• Raw materials
• Strategic components
• Commodity products
• Indirect materials
• Maintenance, repair and operation (MRO)
• Production support items
• Services
• Capital equipment
• Transportation

15
Commodity products Services

16
Example: Commodity products
• Characteristics
• Strongly standardized product
• Intense international competition

• Targets
• Identify suppliers capable of continuous improvement
• Make suppliers compete to reduce prices
• Work with the suppliers to update their catalog offerings
• Ensure suppliers take advantage of innovations
• Reduce lead time
• Manage risks

17
Example: Services
• Characteristics
• Not standardized
• Multi-dimensional
• Subject to subjective evaluation

• Targets
• Develop measure for service quality and value
• Make suppliers compete in various dimensions
• Take advantage of innovations
• Manage non-performance risks

18
https://www.youtube.com/watch?v=R2IUwElbCnc

https://www.youtube.com/watch?v=Z7Yblt0Trzw
BOSE Corporation
• Better Sound through Research!

• High Quality, innovative audio products

• High Quality of each sub-component and raw materials is essential to good sound.

• Continuous Innovation, many new products developed and introduced.

• Products are radical and break with industry convention.

• Such a setup presents some unique sourcing challenges

• Ensuring very high quality of components is vital.
• Many components are not standard but must be custom-made.
• Getting suppliers to invest in producing new sub-components, before demand is proven is
challenging. These investments are often Bose-specific.
• Proprietary information may have to be shared with suppliers.
• Good coordination is necessary to keep inventory levels low.

21
How can BOSE deal with the challenges?
• Make (Vertical Integration)
• Produce inputs in house
• A captive production facility
• Examples: Oil Refineries with dedicated oil fields; Pit-head Thermal Power Plants

• Buy from Spot Market (Short Term Contracts)

• Source product from a spot market.
• Compare, evaluate and select among alternate suppliers on a continual basis.
• E-procurement, Procurement Auctions, Spot market purchases
• Internet enabled price discovery mechanisms have made this much more common
• Common in US Manufacturing

• Buy (Long term relationships)

• Designate a supplier. Source from the same supplier again and again.
• Contract terms may vary with time.
• Relationship specific investments are often made.
• Cross ownerships, integrated ERP systems, embedded employees.
• Common in Japanese Manufacturing.

22
Make approach
This is a vertical integration approach

• Easier to ensure highest quality within organizational boundaries (+)

• Timely and reliable supplies are easier to ensure (+)
• Market forecasts and other Information flows are seamless (+)
• Both the “supplying” and “buying” unit are working towards the same goal (+)
• The supplying unit has the full incentive to invest in capacity for new products with unproven
demand (+).
• The supplying unit has incentives to invest in producing custom parts (+)
• There is less potential for information leakage to competitors. (+)

• There are lower economies of scale: These increase prices/costs and decrease investments in
R&D (for the component), ability to learn.
• Benefits of Focus are lost. Firms tend to become very diffused and unfocussed in their
activities.

23
 Buy — short-term contracts
In a one-shot interaction, a supplier has the incentives to opportunistically cheat the buyer on
all items that are not covered in their contract or are legally enforceable. Thus,

• Quality verification is hard for most products. For commoditized or standardized products, it may be
possible (-)
• Timely delivery may be hard to ensure (-)
• Market forecasts and information sharing is often not possible (-)
• Incentives are not aligned. Each firm, the buyer and the supplier want to maximize their own profits
at the expense of the other (-)
• Suppliers have no incentives to invest in capacity for parts for new products with as yet unproven
demand (-)
• Suppliers have no incentive to invest in capacity for custom parts (-)
• Proprietary Information may leak from the supplier to other competitors (-)

• Full benefits of economies of scale. In cost and innovation for supplied components (+)
• Each supplier can be a highly efficient focused firm (+)
• In each period, one can buy at the cheapest price (+)
• Reputational concerns limit the harm to the buyer (+)

In a one-shot interaction, if a supplier can gain by harming the buyer, he would do so!
24
 Buy — short-term contracts (spot market)
In a one-shot interaction, a supplier has the incentives to opportunistically cheat the buyer on
all items that are not covered in their contract or are legally enforceable. Thus,

• Quality verification is hard for most products. For commoditized or standardized products, it may be
possible (-)
• Timely delivery may be hard to ensure (-)
• Market forecasts and information sharing is often not possible (-)
• Incentives are not aligned. Each firm, the buyer and the supplier want to maximize their own profits
Incentives are misaligned!
at the expense of the other (-)
• Suppliers have no incentives to invest in capacity for parts for new products with as yet unproven
demand (-)
• Suppliers have no incentive to invest in capacity for custom parts (-)
• Proprietary Information may leak from the supplier to other competitors (-)

• Full benefits of economies of scale. In cost and innovation for supplied components (+)
• Each supplier can be a highly efficient focused firm (+)
• In each period, one can buy at the cheapest price (+)
• Reputational concerns limit the harm to the buyer (+)

In an one-shot interaction, if a supplier can gain by harming the buyer, he would do so!
25
 Benefits of Make approach: betting against uncertainty

Capacity Investment in a critical component for a new product

MAKE BUY

0 125 250 375 500 0 125 250 375 500

Capacity
Capacity New Product New Product
Investment
Investment Demand Demand

• Bose’s margin on the product is $300. • Supplier’s margin on the product is $50
• If capacity to produce the component falls short, • If capacity to produce the component falls short,
Bose looses 300$. supplier loses $50
• A unit of capacity costs $30 (rate) • A unit of capacity costs $30 (rate)

How much will supplier invest in capacity?

26
 Benefits of Make approach: betting against uncertainty

Capacity Investment in a critical component for a new product

MAKE BUY

0 125 250 375 500 0 125 250 375 500

Capacity
Capacity New Product New Product
Investment
Investment Demand Demand

• Bose’s margin on the product is $300. • Supplier’s margin on the product is $50
• If capacity to produce the component falls short, • If capacity to produce the component falls short,
Bose looses 300$. supplier loses $50
• A unit of capacity costs $30 (rate) • A unit of capacity costs $30 (rate)

Supplier will systematically under-invest in capacity!

27
Buy — long-term Relationships
• If both the supplier and buyer believe they have a profitable long-term relationships, then
they are inclined to not cheat one another; else either could break off the relationship and
the other would lose all future benefits.
• A short-term contract is like a prisoner’s dilemma; whereas a long-term contract is a repeated
prisoner’s dilemma (where one can induce full cooperation)
• Now one can expect for most reasonable demands to be met:

• Easier to ensure higher quality (as incentives to cheat are smaller) (+)
• Timely and reliable supplies are ensured (+)
• Market forecasts and other Information flows can be shared without fear of leakage (+)
• The supplying unit has higher incentives to invest in capacity for new products with unproven
demand (+).
• The supplying unit has incentives to invest in producing custom parts (+)
• There are the full benefits of the economies of scale (+)
• Each supplier can be a highly efficient focused firm (+)
• One is stuck with a supplier (-)

28
When do long-term relationships work?
• These contracts can achieve the best of both worlds! But there are certain conditions to make
this all work.
• Both parties should believe the relationship is going to last for the long-term.
• Both parties should find the collaboration sufficiently rewarding.
• Both parties should be interested in maximizing long-term gains.
• Requiring both parties to make relationship-specific up-front investments is a good way to
ensure these

• One problem with these relationships: Suppose both firms believe it is a long-term
relationship and make relation-specific investments. Now both firms are stuck with each
other (holdup). They could demand unreasonable terms which the other party will find
preferable to meet than have to incur the cost of setting up a new long-term relationship
(Renegotiation). This can be prevented as long as both firms have made enough comparable
relation-specific investments.

• Long Term Relationship if executed correctly, can get all the benefits of buying
and mitigate most of the downsides.

29
Business Model Innovation: Li & Fung Limited

120

90
Sales (HK$ B)

60

30

0
6

08
'9

'9

'0

'0

'0

'0
30
 A Hard Choice: Flexibility versus Commitment
Short Term Relationships Long Term Relationships

• Flexibility to choose best party, each period • Relation Specific investments

• No relation specific investments • Stuck with one party
HOW TO BREAK THE TRADE-OFF?

Intermediaries relieve the trade-off by allowing for dynamic buyer-supplier matching while keep all
relationships intact. The benefits of relationships flow through a 3 tier system and there is additional
flexibility
Li & Fung Limited: What Does it Do?
✓ Buyers outsource sourcing to Li & Fung
✓ Has a global platform of suppliers
✓ Finds suppliers to source from
✓ Ensures good behavior
(Quality, information, etc.)

Global Platform of Suppliers

32
Rolls Royce: Power-By-The-Hour
• A typical relationship: Time & Materials Contract (pay
per repair).
• Incentive problems:
• The supplier (Rolls Royce) wants more repairs.
• The customer (airline) faces all the risks associated with
engine breakdowns and wants fewer repairs.
• The customer’s costs are unpredictable.
• The customer is forced to buy spare parts while what it
wants is the working engine.

• The alternative: Power-By-The-Hour contract.

• Pay per flying hour.
• Ensure predictable costs.
• Both parties have the same incentives.
• Rolls-Royce takes over the risk of break-downs but it is
better positioned to handle it!
Servicization

THE INNOVATION
• Rather than sell the product, Sell the service that the customer cares about…
• Offer “Power by the hour” Style contracts; lease instead of sale
• Increases Revenues and Increases Risk Exposure, Risk is borne by Party that can best manage it.

WORKS BEST WHEN

• Customers do not want to own assets, but only want service
• Customers don’t want to bear risk of asset productivity.
• You can bear the risk better– Risk Pooling, Private Information, Aligned Incentives
Strategic Sourcing Decisions
• Supplier number and diversity
• Many or a few suppliers?

• For how long to source?

• Level of vertical integration

• To build or to buy?
• To what extent and how?

• Auction or face to face negotiation?

• Or both?

• Organizational issues and degree of centralization

35
Multi-sourcing
• Should we have several suppliers for a purchased product?
• Advantages
• Maintain cost competition
• Stock-outs due to production or delivery problems are less likely
• Lower risk of supplier bankruptcy
• Combine local and global sourcing

• Disadvantages
• Lower quantity per supplier (less important customer; loose quantity discounts)
• Quality differences between suppliers (quality control harder)
• More coordination work for purchasing/receiving
• Higher delivery costs (more shipments)

• Where to locate suppliers? How to choose them?

36
Many suppliers

• Commonly used for commodity

products
• Purchasing is typically based on
price
• Suppliers compete with one
another
• Supplier is responsible for
technology, expertise, forecasting,
cost, quality, and delivery

37
Few Suppliers

• Buyer forms longer term

relationships with fewer suppliers
• Create value through economies of
scale and learning curve
improvements
• Suppliers more willing to participate
in JIT programs and contribute
design and technological expertise
• Cost of changing suppliers is huge
• Trade secrets and other alliances
may be at risk

38
Vertical Integration

• Developing the ability to

produce goods or services
previously purchased
• Integration may be forward,
towards the customer, or
backward, towards suppliers
• Can improve cost, quality,
delivery, and inventory but
requires capital, managerial
skills, and demand
• Risky in industries with rapid
technological change

39
Make or buy and outsourcing decisions

• Make-or-buy decisions
• Choosing between obtaining
products and services
externally as opposed to
producing them internally
• Outsourcing
• Transfer traditional internal
activities and resources to
outside vendors
• Efficiency in specialization
• Focus on core competencies

40
Supplier selection criteria
• Cost:
• Price, conditions of payment, transport, induced costs

• Quality:
• Certified products, after-sales service

• Delivery performance:
• Short delivery lead times, reliable deliveries, flexibility

• Services:
• Packaging, delivered on site, safety stocks

• Security:
• Financial position, part of a conglomerate, reputation
41
Supplier Selection Analysis
• Many factors play a role
• Factor-weighting technique considers multiple criteria
• Each factor is assigned a weight and a score
• Choose the supplier with the best weighted score
FABER PAINT SMITH DYE
SCORE (1-5) WEIGHT SCORE (1-5) WEIGHT
CRITERION WEIGHT
(5 HIGHEST) x SCORE (5 HIGHEST) x SCORE
Engineering/
.20 5 1.0 5 1.0
innovation skills
Production process
.15 4 0.6 5 0.75
capability

Distribution capability .05 4 0.2 3 0.15

Quality performance .10 2 0.2 3 0.3

Facilities/location .05 2 0.1 3 0.15

Financial strength .15 4 0.6 5 0.75

Information systems .10 2 0.2 5 0.5

Integrity .20 5 1.0 3 0.6

Total 1.00 3.9 4.2

42
 Traditional sourcing methods
• Catalogues
• Example: Office supplies

• Request for quotes (RFQ)

• One buyer, many sellers, standardized requirements
• Example: Specifiable machined parts

• Request for proposals (RFP)

• One buyer, many sellers, non-standard items
• Example: Construction Projects

More complex

43
Methods of negotiation
• Face-to-face:
• Traditional

• Can take several weeks

• Iterative communication

• Can adjust an imperfect initial RFQ

• Spec changes, alternatives, etc.

• With one supplier, the outcome depends on buyer-supplier relationship

44
Methods of negotiation: Auctions
Methods of negotiation
• Procurement Auctions:
• Formally defined as:

A mechanism of bid submission, together with rules assigning

payments and contracts based upon submitted bids

• Newly enabled by internet

• Typically completed in a swift manner

• Communication over quote details is front-loaded

• Leverages competition

46
Face to Face Negotiation vs. Auction

Face-to-Face Negotiation Procurement Auction

Parties 1 buyer, 1 supplier 1 buyer, many bidders

No upper bound, can take Bounded, typically a few

Time
several weeks hours

Iterative, revealed step-by-

Communication Bid details specified upfront
step

Limited, Can adjust imperfect

Competition Leverage buyer power
initial RFQ

Depend on buyer-supplier
Outcomes Economically efficient
power

47
E-Sourcing Success Stories
• From interview with a large auto OEM:
1. Initial state: Incumbent charges $1/part
2. Buying agent proposes auction
3. Incumbent tries to pre-empt, cuts price to 80¢
4. Buyer holds competitive auction anyway
5. Incumbent wins contract at 60¢

• Microsoft saved 6% off indirect computer purchases by using

electronic catalogue in which Dell, HP, and Toshiba could see each
other’s price and offerings.

48
When to use e-auctions?
• Items can be clearly specified and translated into prices a supplier
will commit to charge the buyer

• There is a strong likelihood that the current price is sufficiently

higher than the market price

• Switching costs are acceptable

• A sufficient number of qualified, competitive suppliers exists in

the marketplace

• Qualified suppliers are willing to participate in an e-auction

Source: The Role of49

Reverse Auctions in Strategic Sourcing, Beall et al. 2003
Pitfalls of auctions
• Is the lowest price supplier really the best supplier?

• What is the true total cost of the contract?

• Quality
• Warranty
• Service
• Switching costs

50
Centralized purchasing
• Pros:
• Bargaining power over suppliers
• Organizational coordination on purchasing policies
• Standardized purchasing procedures
• Transparency and tight control of organizational spending
• Single interface to suppliers

• Cons:
• Dissatisfied internal customers
• Inflexible purchasing procedures
• Bureaucratic inefficiency

51
Summary
• Purchasing is important; it is used increasingly as a strategy for
competitive advantage

• Auctions are a powerful part of a sourcing manager’s toolkit

• Must consider supplier perspectives

• Beware of eroding supplier trust

52
 Supply Chain Management

Class 18
Littlefield Debriefing

Sajjad Najafi

1
What’s Left?

SOURCES OF Process
SUPPLY CHAIN DESIGN SUPPLY CHAIN MGMT
ADVANTAGE Execution

INSTRUMENTS JIT/ TPS SC Sourcing SCM Projects/

Structure Review

SESSIONS 13, 14 15,16 17, 18 19-21 22-24

FOCUS ON THE SUPPLY CHAIN

LITTLEFIELD
SIMULATION

Submitting your project presentations (PDF):

Deadline April 19 (end of day in France)

Project Presentations:
Session 22 (April 27)
Session 23 (May 11)
Review:
Session 24 (May 16)
SCM Student Project - logistics

What you need to prepare

• ~12 minute presentation (powerpoint/keynote -> pdf) during Sessions 22-23
• All groups will present - no escape from this :-)
• Your report is you slides and everything included (pictures, videos, …)
– Include hidden slides with additional information, e.g., mention all sources of
information (internet pages, company documents, interviews, articles, etc.) – or
provide information in notes pages.
• DEADLINE: April 19, end of day in France
• Submit to the following Dropbox link:
• EN01: https://www.dropbox.com/request/o983Yksqs4V7t7dsKOQy
• EN03: https://www.dropbox.com/request/mOJ1s6VWtYBurydGSNhY
• EN04: https://www.dropbox.com/request/4RNo1vgLE9pJrPTVbq4q
Quiz (Newsvendor)
Available from April 14th and remains open until Final Exam.

For all quizzes:

Available until Final Exam
Multiple attempts are allowed
Effort based (not grade based): Students will get full grade of a quiz as long
as they take the quiz on Blackboard even once and regardless of the actual
grade displayed.

4
What did we do last time?
Sourcing

• Sourcing is the process of determining what to source, whom to

source it from, and how to manage risks and leverage
opportunities to support overall business strategy

Specs Inputs Transformation Demand Logistics

New Product Marketing

Sourcing Operations Distribution
Development & Sales

6
Historical Context of Sourcing

• Traditional: Large, vertically integrated production

▪ Henry Ford's Rouge plant: Iron ore, glass and rubber come in one end,
cars leave from the other end
▪ Purchasing manager only had clerical and scheduling duties

The Ford Rouge complex,

oil painting in 1949

7
Commodity products Services

8
https://www.youtube.com/watch?v=R2IUwElbCnc

https://www.youtube.com/watch?v=Z7Yblt0Trzw
How can BOSE deal with the challenges?
• Make (Vertical Integration)
• Produce inputs in house
• A captive production facility
• Examples: Oil Refineries with dedicated oil fields; Pit-head Thermal Power Plants

• Buy from Spot Market (Short Term Contracts)

• Source product from a spot market.
• Compare, evaluate and select among alternate suppliers on a continual basis.
• E-procurement, Procurement Auctions, Spot market purchases
• Internet enabled price discovery mechanisms have made this much more common
• Common in US Manufacturing

• Buy (Long term relationships)

• Designate a supplier. Source from the same supplier again and again.
• Contract terms may vary with time.
• Relationship specific investments are often made.
• Cross ownerships, integrated ERP systems, embedded employees.
• Common in Japanese Manufacturing.

11
 Benefits of Make approach: betting against uncertainty

Capacity Investment in a critical component for a new product

MAKE BUY

0 125 250 375 500 0 125 250 375 500

Capacity
Capacity New Product New Product
Investment
Investment Demand Demand

• Bose’s margin on the product is $300. • Supplier’s margin on the product is $50
• If capacity to produce the component falls short, • If capacity to produce the component falls short,
Bose looses 300$. supplier loses $50
• A unit of capacity costs $30 (rate) • A unit of capacity costs $30 (rate)

Supplier will systematically under-invest in capacity!

12
 A Hard Choice: Flexibility versus Commitment
Short Term Relationships Long Term Relationships

• Flexibility to choose best party, each period • Relation Specific investments

• No relation specific investments • Stuck with one party
HOW TO BREAK THE TRADE-OFF?

Intermediaries relieve the trade-off by allowing for dynamic buyer-supplier matching while keep all
relationships intact. The benefits of relationships flow through a 3 tier system and there is additional
flexibility
Li & Fung Limited: What Does it Do?
✓ Buyers outsource sourcing to Li & Fung
✓ Has a global platform of suppliers
✓ Finds suppliers to source from
✓ Ensures good behavior
(Quality, information, etc.)

Global Platform of Suppliers

14
Rolls Royce: Power-By-The-Hour
• A typical relationship: Time & Materials Contract (pay
per repair).
• Incentive problems:
• The supplier (Rolls Royce) wants more repairs.
• The customer (airline) faces all the risks associated with
engine breakdowns and wants fewer repairs.
• The customer’s costs are unpredictable.
• The customer is forced to buy spare parts while what it
wants is the working engine.

• The alternative: Power-By-The-Hour contract.

• Pay per flying hour.
• Ensure predictable costs.
• Both parties have the same incentives.
• Rolls-Royce takes over the risk of break-downs but it is
better positioned to handle it!
Servicization

THE INNOVATION
• Rather than sell the product, Sell the service that the customer cares about…
• Offer “Power by the hour” Style contracts; lease instead of sale
• Increases Revenues and Increases Risk Exposure, Risk is borne by Party that can best manage it.

WORKS BEST WHEN

• Customers do not want to own assets, but only want service
• Customers don’t want to bear risk of asset productivity.
• You can bear the risk better– Risk Pooling, Private Information, Aligned Incentives
Littlefield Technologies
Group EN01
Group EN01

Maria Pla
Gloria Fachinger
Gianmarco Milani David Canosa
Lucia Tonolo Moritz Thier
Emma Gugliotta Sofia Dupuis
Lodovico Benvenuti Jeena Miria George
Wenqian Wang
Agnès Moukarzel
Group EN03
Group EN03

Romain Batlle
Zineb Alami
Edgar van OS Hamza Marrakchi
Jade Chen Ines Achour
Jayjay Srijeeta
Philip Mantz Joshua McFadden
Jan Philipp Girgott
Raili Engler
Group EN04
Group EN04

Xinhe Melanie Fan

Jad Hajj
Kwanho Ryu Pia Subiros
Jongwoo Kim Tomás Tedim
Yeonoo Kim Oscar Franz
Ying Hon Ng Elisabeth Ljabach
Axel Svensson
Lucia Maria Schluter
Tell me about your experience with Littlefield
What did you learn?
 Littlefield Mechanics
▪ Factory runs for 50 days without student control

▪ Student teams control factory from day 50 to day 217

▪ Factory runs for another 50 days using management policies chosen

by students effective on day 217
 Factory Performance?
How was the factory performing when you took control?
 Station Utilization
100% 100%

75% 75%

50% 50%
Stuffer Tester
25% 25%
Avg U = 82.6% Avg U = 88.7%
0% 0%
0 13 25 38 50 0 13 25 38 50

100%

75%

50%

25%
Tuner
Avg U = 89.2%
0%
0 13 25 38 50
3600

2700

Stuffer queue
Tester queue
1800
Tuner queue

900

0
0 13 25 38 50
3600

2700

Stuffer queue
Tester queue
1800
Tuner queue
Total

900

0
0 13 25 38 50
 RM kits
9,000

6,750

4,500

2,250

0
0 12.5 25 37.5 50

3600

2700

Stuffer queue
Tester queue
1800
Tuner queue
Total

900

0
0 13 25 38 50
 Job Arrivals
30
Average = 14.5 jobs/day
23

15

0
0 13 25 38 50

Flow time Revenue per job

6 1000

4.5 750

3 500

1.5 250

0 0
0 13 25 38 50 0 13 25 38 50
 Factory Performance?
How was the factory performing when you took control?
▪ Raw material stockout every order cycle
▪ Inventory buildup
▪ High utilization at all stations
▪ Spiky demand
▪ Long lead times
 Managing Littlefield Technologies

▪Revenue:
▪choice of contract
▪lead times

▪Costs:
▪machine
▪inventory
 Choice of Contract
Current actual flow time (days)
6

4.5

Avg ~ 2.97 days

3

1.5

0
0 13 25 38 50
 Choice of Contract
Given current performance, what contract(s) can we
realistically support?
▪ Contract #1: Price $750, LT window 7-14 days ~ Fine
▪ Contract #2: Price $1000, LT window 1-3 days Bankruptcy
▪ Contract #3: Price $1250, LT window 0.5-1 day Bankruptcy

Since potential profit for contract #3 is

$1250-600 = $650, compared with
$750-600 =$150 for contract #1, it makes
sense to see if we can make it work.
Can we effectively serve Contract 2 or 3?
Avg. processing ▪ Theoretical flow time = 9.2 hours
Step
time per job < 0.5 days
1 4.4 hrs ▪ Can serve contract 2 or 3 in
theory, but ...
2 1.1 hrs
Actual flow time (days)
3 1.6 hrs 6

4 2.1 hrs 4.5

Total 9.2 hrs 3

Avg ~ 2.97

1.5

0
0 13 25 38 50
 Why actual flow time is so long?

Jobs spend most of time waiting

Why are they waiting?

Demand exceeds Frequent RM stockout

station capacity
Why frequent RM stockout?

ROP is too Low Order too frequently

Add
Capacity

Revise Revise Order

ROP Quantity
How can we get to a 0.5 day lead time?

39
Revising Inventory Management Policy

RM kits
9,000

6,750

4,500

2,250

0
0 12.5 25 37.5 50

Continuous Review Model (ROP, Q)

 Demand Data
First 50 days Jobs Kits
mean daily demand 14.5 870
SD (daily) 4.1 246
Annual Rate 5293 317,580
Job Arrivals
30
Average = 14.5 jobs/day
23

15

0
0 13 25 38 50

“No significant trend in the average demand” Moving average for forecasting
 Revising the Order Quantity Q

2DS
Q* =
H
D= Avg. Annual demand (units) = 14.5 x 365 = 5293
S= Cost per order ($) = 1000
H= Holding cost ($) = 0.1 x 600 = 60 /order/year

Q* = 420 jobs (25,202 units)

42
 Why actual flow time is so long?

Jobs spend most of time waiting

Why are they waiting?

Demand exceeds Frequent RM stockout

station capacity
Why frequent RM stockout?

ROP is too Low Order too frequently

Add
Capacity

Revise Revise Order

ROP Quantity
 Revising the ROP

• Cycle length = Q/D = 420 / 5293 = 28.96 days

• Cost of overage: (0.1*600)/365 * 28.96 = 4.76
• Cost of shortage (contract3): 1250 - 600 - 4.76 = 645.24
• Critical fractile (optimal service level) = 99.2%
• Z0.992=2.41

ROP = 4 * 14.5 + 2.41 * 4.1 * 4 = 78 jobs

44
 Why actual flow time is so long?

Jobs spend most of time waiting

Why are they waiting?

Demand exceeds Frequent RM stockout

station capacity
Why frequent RM stockout?

ROP is too Low Order too frequently

Add
Capacity

Revise Revise Order

ROP Quantity
Remember queueing?
Waiting time
(or inventory)

High Returns

Low Returns

0 0.2 0.4 0.6 0.8 1

Utilization

• 3 strategies to reduce waiting:

– Decrease service time (speed-up operations)
– Decrease utilization (i.e., increase capacity) The only available to us option!
– Decrease variability (stabilize processes and demand)

Decreasing Utilization (adding more capacity) is the most efficient when

utilization is high — search for the bottleneck!
 Adding Capacity
100% 100%

75% 75%

50% 50%
Stuffer Tester
25% 25%
Avg U = 82.6% Avg U = 88.7%
0% 0%
0 13 25 38 50 0 13 25 38 50

100%

75%

50%

25%
Tuner
Avg U = 89.2%
0%
0 13 25 38 50
Where do we add capacity?

High Returns

Low Returns

0 0.2 0.4 0.6 0.8 1

Station 1 Station 2 Station 3

Utilization 0.826 0.478 0.892
0.895 0.887 0.904
Exploit the Bottleneck!

Bottleneck is station 3

Keep station 3 always busy – should not wait for input.

• Set Priority at Station 2 to Product Coming from Station 1, i.e., first work on
anything going to station 3
• Unless the queue there is already very long there; then priority to finish and
ship products coming from station 3. (Could adjust dynamically in real life…)

49
We need (4 - 3 - 2) machines to have low enough avg. Wq to
move to Contract 3

0.30 3 machines (starting point)

0.25

0.20
Wq , days

4 machines
0.15
0.10
5 machines

0.05

0.00
0.0 0.2 0.4 0.6 0.8 1.0
utilization, ρ
50
 Why actual flow time is so long?

Jobs spend most of time waiting

Why are they waiting?

Demand exceeds Frequent RM stockout

station capacity
Why frequent RM stockout?

ROP is too Low Order too frequently

Add
Capacity

Revise Revise Order

ROP Quantity
 End of the Game Strategy:

Should we continue to use regular ROP policy?

No! Inventory leftover on day 268 is worthless.
How can we end the game with the minimum amount of inventory?
OM tool: Newsvendor model

• One-shot game: balances cost of overstocking vs. opportunity cost of

missed sales
• Determine optimal quantity based on critical ratio (cost of being stuck with
inventory vs. foregone profit if a shortage occur)

Target SL = Cu/(Cu+Co)

• Relevance: all industries where capacity commitments must be made in

advance and demand in uncertain
– All perishable products!
– Fashion industry
– Electricity, airline seats
– Products with short shelf life or short life cycle

– Any product at the end of its life cycle!

OM tool: Newsboy model

• Consider cost of shortages vs. inventory (Contract 3):

– Target SL = Cu/(Cu+Co)
– Co = cost of kit – salvage value = 600 (possibly: +holding cost)
– Cu = selling price – cost of kit = 650 (possibly: -holding cost)
– SL* = 650/1250 = 0.52 (optimal probability of NOT stocking out)

• Mean demand for 50 days = 50(14.5) = 725 jobs on average during

last 50 days
• Ordering up to about 725 jobs on day 217 is a reasonable strategy
orders (Order a little more than mean demand, as stock-outs are more
costly than excess inventory)

Careful: Do not forget inventory on hand (and on order) when placing

last order!
 Supply Chain Management

Class 19
Zara, Fast Fashion, Postponement, Delayed
Differentiation

Sajjad Najafi

1
What’s Left?

SOURCES OF Process
SUPPLY CHAIN DESIGN SUPPLY CHAIN MGMT
ADVANTAGE Execution

INSTRUMENTS JIT/ TPS SC Sourcing SCM Projects/

Structure Review

SESSIONS 13, 14 15,16 17, 18 19-21 22-24

FOCUS ON THE SUPPLY CHAIN

LITTLEFIELD
SIMULATION

Submitting your project presentations (PDF):

Deadline April 19 (end of day in France)

Project Presentations:
Session 22 (April 27)
Session 23 (May 11)
Review:
Session 24 (May 16)
 April 19, end of the day in France
Deadline for project submissions (only in PDF)

• Submit to the following Dropbox link:

• EN01: https://www.dropbox.com/request/o983Yksqs4V7t7dsKOQy
• EN03: https://www.dropbox.com/request/mOJ1s6VWtYBurydGSNhY
• EN04: https://www.dropbox.com/request/4RNo1vgLE9pJrPTVbq4q
Supply Chain Management
Let’s talk about Fashion Industry
THE FASHION INDUSTRY

Design Sourcing Mfg. Distribution Sales Discount

Example

Retail Price: $65

Cost: $15 (production, shipping)

Margin: 77%†
Firm Profitability: 5%!!!!
Grease Jacket from Costume Gallery’s 2008
Catalogue

† Margins are indicative of the fashion garment industry, but bear no resemblance to Costume Gallery’s actual margins

6
THE FASHION INDUSTRY

Design Sourcing Mfg. Distribution Sales Discount

Example

Retail Price: $65

Cost: $15 (production, shipping)

Margin: 77%†
Firm Profitability: 5%!!!!
Grease Jacket from Costume Gallery’s 2008
Catalogue

While Margins are high for fashion goods, but firm profitability is Very Low! Why?

• Demand for such products is driven by fickle trends and is notoriously hard to predict
• Consequently, firms end up with a lot of excess inventory which must be discounted. At the same time, many customers
don’t find what they want.

† Margins are indicative of the fashion garment industry, but bear no resemblance to Costume Gallery’s actual margins

7
ZARA’S FAST FASHION BUSINESS MODEL

Amancio Ortega

Zara NyTimes Magazine Piece

8
ZARA FACTS AND FIGURES

• Part of Inditex (64.8% of sales in 2011)

• Inditex founded in 1963
• Inditex group has several brands (Zara, Pull &
Bear, Massimo Dutti, Bershka, Stradivarius,
Oysho, Zara Home and Uterqüe)
• First Zara shop in 1975
• Present in 96 countries (2016)
• 2259 stores (2019)
• 11 factories (in 2010)
• 500,000 m2 Logistics Center in La Coruna
(in 2010), all logistics centers in Spain
• Revenue: $18.9 billion annually (2018)

Among fastest growing apparel companies in the world!

9
 IS IT PRODUCT INNOVATION?

10
 IS IT PRODUCT INNOVATION?

• Fast copying of leading styles

• Fast delivery in own stores

• Limited Editions

11
 IS IT QUALITY?

12
 IS IT QUALITY?

• Raw material: medium

• Knit: poor (though, became better over time)

• Look: grand!

• Customer satisfaction: Fashion at Low Price!

13
 IS IT VARIETY?

14
 IS IT VARIETY?

• Limited customer variety: only what is on display and in limited

choices (colour, size)

• But every customer is participating in the process: helps determining

the next batch

15
 IS IT MARKETING?

16
 IS IT MARKETING?

• Rely on word of mouth

• Initially no advertising (and still relatively little)

• Store fronts as advertisement

17
 IS IT LOCATION?

18
 IS IT LOCATION?
La Coruna is in a far corner of Europe!

An unlikely place for fast growing company!

19
 IS IT LOW MANUFACTURING COST?

Production close to headquarters… rather than low cost labor locations in

Asia or Eastern Europe
20
 WHAT ABOUT DISTRIBUTION COSTS?

A fancy, huge, expensive warehouse!

But it is utilized only at 50%!
21
 AND TO INCREASE ASSET UTILIZATION…

They’ve built a new expensive huge

warehouse (press release May 8, 2003)

$90 million
123,000 m2
80,000 garments/hr

Are they crazy?

22
 STOCK MARKET THOUGHT SO…

INDITEX

23
SO HOW DO THEY DO IT?
 LET’S COMPARE:

VS

P.S. new logo as of 2019

25
MARKS & SPENCER: CUSTOMER PROPOSITION AND BUSINESS PROCESS

Business Process
• Design Team defines detailed cloth specifications one year before store delivery.
• Merchandisers decide prices and quantities well before start of selling season.
• Stylists focus on quality improvements of traditional styles.
• Quality: High conformance to specification, quality of stitching, etc.
• Many Distributed Warehouses
• Seven Weeks of Inventory

Customer Proposition
• High “Quality” garments, traditional styles, plus sizes.

Clock-Speed
16 month cycle

Design Sourcing Mfg. Distribution Sales Discount

26
ZARA: CUSTOMER PROPOSITION & BUSINESS PROCESS

Business Process
► Design Team develops platform models but holds off finalizing detailed design
► Single production. Production often falls short. New styles pick up unsatisfied demand.
► Internal Raw Material finishing capabilities, allowing for last minute changes
► Production is consolidated in Zara Industrial Areas (close to markets- Mexico, Spain)
► All new products every month
► Word of Mouth Advertising. Location, Location and Location!

Customer Proposition
► Trendy Stylish products, very affordable prices. Buy new styles often, high turnover.

Clock-Speed
21-29 day cycle

Design Sourcing Mfg. Distribution Sales Discount

27
ZARA V/S M&S: KEY BUSINESS PROCESS DIFFERENCES

Marks & Spencer Zara

Quantity Decision made 1 year before sales Quantity Decisions made days before sales

Multiple, Distributed Warehouses Centralized Warehouse

Platform designs are frozen and flexible capacity is

Detailed Designs are frozen well in advance booked. Detailed Product differentiation decision is
postponed.

Demand is satisfied , leftover inventory is heavily Full Demand is often unsatisfied, Customers
discounted substitute other products.

Distant Production Location, Long lead times Close Production Location, Short lead times

7 weeks of Inventory Few days of inventory

28
ZARA BUSINESS PROCESS IN FULL: 5 DAYS LEAD TIME!

Step 1:
Scan Fashion Shows

Step 2: Step 3:
Purchase Raw
Simplify “hits” and Final design of the next Manufacturing
Material
make library of designs batch

Distribution
ZARA BUSINESS PROCESS IN FULL: 5 DAYS LEAD TIME!

Step 1: Step 5:
Scan Fashion Shows Designers “pull” next raw
material batch

Step 2: Step 3:
Purchase Raw
Simplify “hits” and Final design of the next Manufacturing
Material
make library of designs batch

Distribution
Step 4:
Shoppers (and store managers)
“pull” next designs (shapes) &
designers “adapt”

Shopping
Experience
 HOW CAN THEY DO IT IN 5 DAYS?

• Focus on getting it approximately correct

• Define a fast process (including local production)

• Solve the material constraint

• Constrain designers

• Optimise the offer

What risk does Zara still face?

• Risk of buying wrong raw material…

31
ILLUSTRATIVE EXAMPLE: VALUE OF PLATFORM DESIGNS
Early Differentiation Process†
Dyed Yarns Finished Sweaters

Dyeing Knitting

32
ILLUSTRATIVE EXAMPLE: VALUE OF PLATFORM DESIGNS (DELAYED DIFFERENTIATION/POSTPONEMENT)

Early Differentiation Process†

Dyed Yarns Finished Sweaters

Dyeing Knitting

Delayed Differentiation Process (Postponement)

White Garments Finished Sweaters

Knitting Dyeing

► In an early differentiation process, dyeing and knitting quantities have to be decided on the basis of individual demand for
colors.
► In a delayed differentiation process, knitting quantities can be decided on basis of total demand.
► Total Demand has lower variance than the sum of variances of individual demand. This will reduce total mismatch costs and
increase profits

Delayed Differentiation also leads to pooled demand, and reduces mismatch costs!

Anand K. S. and K. Girotra, “The Strategic Perils of Delayed Differentiation,” Management Science, 53:5, May 2007, pp. 697-712.

33
DELAYED DIFFERENTIATION: OTHER EXAMPLES
Paint Industry

Early Differentiation Process

color pigments, retail sales

paint mixing,
packaging

Delayed Differentiation Process (Postponement)

color pigments retail sales,

paint mixing,
packaging

34
HOW DOES DELAYING HELP? VALUE OF TIME (INFORMATION)
GRAPHS INDICATE FORECAST ACCURACY, CLOSENESS TO 45O LINE INDICATES HIGHER ACCURACY

A Year Before Demand 3 Months Before Demand 2 Weeks before Demand

0.02 0.02
0.02

0.015 0.015 0.015

0.01 0.01 0.01

0.005 0.005 0.005

0 0 0
0 50 100 150 200 0 40 80 120 160
0 30 60 90 120

Demand Forecasts become better with time as more information becomes available

Firm 1 orders 1 year ahead of Demand.

It’s forecast is D ~ N(Mean, Variance1). Compute Newsvendor Quantity and Profit
Firm 2 orders 1 month ahead of Demand
It’s forecast is D~ N(Mean, Variance2), where Variance2 < Variance1. Compute NV Qty & Profit.
Mismatch Costs will be higher for Firm 1, Newsvendor profits for Firm 2 will be higher!

Delaying Ordering Decisions reduces demand-supply mismatch, and increases profits

35
AGGREGATING DEMAND FROM TWO MARKETS

Uncertainty for common part lower than that of all individual

products (risk pooling)

Safety stock: Safety stock:

N⋅z⋅σ N⋅z⋅σ

Lower uncertainty (standard deviation) results in

• Lower inventory (safety stock) => hence less leftovers
Q =μ+z⋅σ

• Lower lost sales

E Lost Sales = σ ⋅ L(z)

36
EXAMPLE PROBLEM

• You are selling a particular T-shirt in two colors: red and black. Your demand forecast for
both is N(µ=100,σ=20) and demands are independent. How many T-shirts of each type
should you buy, if you want to provide a service level of 99% (z=2.33)?

• You bought a “Benetton machine” and now can color T-shirts in small batches in your
store. How many “white fabric” T-shirts should buy if you want to provide the same
service level?

37
EXAMPLE PROBLEM

• You are selling a particular T-shirt in two colors: red and black. Your demand forecast for
both is N(µ=100,σ=20) and demands are independent. How many T-shirts of each type
should you buy, if you want to provide a service level of 99% (z=2.33)?

Q=100+2.33*20 = 147 of each type (or 294 in total)

• You bought a “Benetton machine” and now can color T-shirts in small batches in your
store. How many “white fabric” T-shirts should buy if you want to provide the same
service level?

New distribution: µ=2*100=200, σ=Sqrt(2)*20=28.28

Q=200+2.33*28.28 = 266 white fabric ones

38
AGGREGATING DEMAND FROM TWO PRODUCTS
Product 1 Product 2

DEMAND ~ NORMAL (μ, σ)

DEMAND ~ NORMAL (μ, σ)
Combined demand for the two products ?
► The mean demand will be 2*μ, but the combined standard deviation depends on the correlation between the two variables
► Correlation refers to how one random variable’s outcome tends to be related to another random variable’s outcome.

20 20
20
18 18 18
16 16 16
14 14 14
12 12 12
10 10 10
8 8 8
6 6 6
4 4 4
2 2 2
0 0 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
NEGATIVE CORRELATION
NO CORRELATION / INDEPENDENCE POSITIVE CORRELATION

Normal (2μ, 2(1 + correla'on) σ)

Random demand for two products (x-axis is product 1, y-axis is product 2). In scenario 1 (left graph) the correlation is –0.9, in scenario 2 (middle graph) the correlation is -0 and in scenario 3 (the right
graph) the correlation is 0.90. In all scenarios demand is Normally distributed for each product with mean 10 and standard deviation 3.

39
WHEN DOES DELAYED DIFFERENTIATION MAKE SENSE?

40
WHEN DOES DELAYED DIFFERENTIATION MAKE SENSE?

• Customers demand variety

• High demand uncertainty + less uncertainty with total demand than demand
for individual versions.
⇒ Combine items with low or negative correlations.
⇒ In case of perfect, positive correlation there is NO benefit for delayed differentiation!
• Changes not too expensive to implement
– Variety is created late in the production process.
– Variety can be added quickly and cheaply.
– Components needed for variety are inexpensive relative to the generic component.
• Implement at the beginning of the product life cycle!

Decision needs to involve every function in organization!

41
ILLUSTRATIVE EXAMPLE: VALUE OF CENTRALIZED WAREHOUSE (LOCATION POOLING)
Consider two supply chain designs

Decentralized Warehouses
Demand from Territory 1: Normal (µ, σ)
This is the effective demand for W1. Compute NV Profits
W1 Territory 1

Demand from Territory 2: Normal (µ, σ)

W2 Territory 2 This is the effective demand for W2. Compute NV Profits

Centralized Warehouses
Demand for Warehouse is sum of Demand from Territory 1 and Territory 2.
Territory 1 Effective Demand: Normal (2µ, σ (2(1+ρ))1/2), where ρ is the correlation
W between the two markets.
Territory 2
Compute NV Profit for Warehouse.

The Variance in Pooled Demand is less than 2x the variance in original demands!

Having a centralized warehouse, reduces effective demand variance and increases Profits

42
MORE EXAMPLES OF POOLING DEMAND

► Campbell Soup:
► Manufacturers brand name and private label soup (same soup)
► Problem: many different private labels (Giant, Kroger, A&P, etc)
► Solution: Hold inventory in cans without labels, add label only when demand is realized.

► Black and Decker:

► Sell the same drill to different retailers that want different packaging.
► Store drills and package only when demand is realized.

► Nokia:
► Customers want different color phones.
► Design the product so that color plates can be added quickly and locally.

► Hewlett Packard
► Sells printers in different countries. Different markets have different power supply conventions
and require different product labeling.
► Designs and produces generic printers with special power supply modules that can be plugged
in after demand is realized.

43
FASHION INDUSTRY: SUBSTITUTABILITY, DISCOUNTS & CONSUMER BEHAVIOR

Excess stock and unmet

demand are avoided by
stopping production when
market saturates

Next model Next model

Small
batches

Always Positive Always Positive

Option Value Option Value

Customers substitute. Lower cost of lost sales (cost of underage)!

44
FASHION INDUSTRY: SUBSTITUTABILITY, DISCOUNTS & CONSUMER BEHAVIOR

• Consider the underage and overage costs of the newsvendor model (In all previous examples,
we considered effects of the demand distribution)

• In fashion goods, customers can be encouraged to substitute. Thus, if one product is out of
stock, customers may purchase another product. Zara recognized this and modified its
Newsvendor logic to consider a smaller cost of lost sales (cost of underage). Thus, it stocks
less.

• Stocking less reduces the amount of inventory that has to be discounted. Further, in
principle, Zara has very limited sales

• Thus further modifies customer strategic behavior to buy early rather than wait for the
sale. Customers can also be fashion forward this way.

• A typical Zara customer visits the store 17 times/year (compared to 4-5 for Gap)

• All this increases Zara’s total profits!

Overage and underage costs can be modified by changing customer behavior

45
ZARA’S SUCCESS

► Zara Continues to be recognized as the

leader in “Fast Fashion”

THE CHALLENGE FOR ZARA TODAY: COMPETITORS ARE CATCHING UP

► The Dilemma of game changing innovators.

► Zara continues to relentlessly drive further efficiencies
from its newsvendor game.

46
REACTION TO COMPETITION

• Zara hired “designers”

• Zara started advertising
• Zara started global sourcing

• Which products should Zara source in Asia?

– Asian capacity = speculative capacity = best for low risk products
– Local capacity = reactive capacity = best for high risk products

47
KEY LESSONS

• Delaying decisions allows one to use better forecasts and reduces information risk and inefficiency
(lower mismatch costs and higher newsvendor profits)

• Pooling Demand from various locations (Centralized Warehouse) reduces variance in demand (lower
mismatch costs and higher newsvendor profits)

• Delaying Differentiation of products (platform designs) allows one to bet on joint demand for
different product variants, which has lower variance (lower mismatch costs and higher newsvendor
profits)

• Fewer discounts can change customer behavior and encourage immediate purchases and
substitution.

From Cost Focus to Risk Focus– Reduce Information Risks

48
NEXT CLASS

• Mass-customization
• Dual Sourcing

• Check out Timbuk2‘s webpage!

49
 Fresh Connection Registration

Registration deadline: April 20 at 6pm

Access to the interface: April 20 at 6pm
First Round Deadline: April 24 at 6pm
Registration

Go to http://my.inchainge.com
Registration
Registration
Registration

xxx@xxx
Registration
Registration
Registration

Course codes (Do NOT enter the wrong code)

[EN01 class] : GQaR-K9Fc
Registration

Course codes (Do NOT enter the wrong code)

[EN03 class] : ajkA-wRTC
Registration

Course codes (Do NOT enter the wrong code)

[EN04 class] : Ac8x-JARJ
Registration

Operations and Supply Chain management - TFC Nov 2018 - Woonam Hwang
Registration

Operations and Supply Chain management - TFC Nov 2018 - Woonam Hwang

Patrick Willems
Woonam Hwang
Registration

Operations and Supply Chain management - TFC Nov 2018 - Woonam Hwang

Patrick Willems
Woonam Hwang
Registration

Operations and Supply Chain management -

TFC Nov 2018 - Woonam Hwang
How to enter the simulation

Operations and Supply Chain management - TFC Nov 2018 - Woonam Hwang

XXX, XXX
 Supply Chain Management

Class 20
Mass Customization and Reactive Capacity

Sajjad Najafi

1
What’s Left?

SOURCES OF Process
SUPPLY CHAIN DESIGN SUPPLY CHAIN MGMT
ADVANTAGE Execution

INSTRUMENTS JIT/ TPS SC Sourcing SCM Projects/

Structure Review

SESSIONS 13, 14 15,16 17, 18 19-21 22-24

FOCUS ON THE SUPPLY CHAIN

LITTLEFIELD
SIMULATION

Submitting your project presentations (PDF):

Deadline Extended to April 23 (end of day in France)

Project Presentations:
Session 22 (April 27)
Session 23 (May 11)
Review:
Session 24 (May 16)
 April 23, end of the day in France
Deadline for project submissions (only in PDF)

• Submit to the following Dropbox link:

• EN01: https://www.dropbox.com/request/o983Yksqs4V7t7dsKOQy
• EN03: https://www.dropbox.com/request/mOJ1s6VWtYBurydGSNhY
• EN04: https://www.dropbox.com/request/4RNo1vgLE9pJrPTVbq4q
 Fresh Connection
Deadline to register: April 20 at 6pm
Access to the interface: April 20 at 6pm
First round deadline: April 24 at 6pm
Mass Customization and Reactive Capacity
Who is this man?

https://www.youtube.com/watch?v=iOi4b7XeSMg&t=1s 6
Who is this man?

Michael Dell

https://www.youtube.com/watch?v=iOi4b7XeSMg 7
How Michael Dell Revolutionized the PC Industry

HP: Build To Forecast (BTF) Dell: Build To Order (BTO)

Manufacture Demand Demand Manufacture

8
Should we have Finished Goods Inventory or Not ?

HP: Build To Forecast (BTF) Dell: Build To Order (BTO)

Final Product Demand Demand Final Product

Ex. Ready to wear clothes Ex. Airplanes

Fast Food Tailored Suits
Buffet Table Gourmet Restaurant
Factories Dell Computers
Goal A La Carte
Services

9
 To Have Finished Goods Inventory Or Not?

Build To Forecast (BTF)

• Systems that have Finished Goods Inventory (FGI) are called Build To Forecast (or Make to
Stock) systems.
• In such systems we produce well in advance of customer arrival based on our forecast of
demand (Think: Buffet Table). When a customer shows up, we can just supply him from the
Finished goods inventory, and the customer doesn’t have to wait. The downside is that we
have wasted inventory, and we have limited ability to customize the product to individual
taste.

Build To Order (BTO)

• Systems where no finished goods inventory is allowed are called Build To Order (or Make to
Order) Systems.
• In such systems, production is only initiated after a customer arrives, so she has to wait for as
long as it takes to produce the product (including waiting time). (Think: A la Carte). The main
upsides are no wasted inventory, possibility of customization.

10
Build To Order V/S Build To Forecast

Wait Product Variety Waste

BTF (FGI) Shorter (Good) Small (Bad) High (Bad)

BTO (no FGI) Longer (Bad) More (Good) Low (Good)

What does the market dictate?

Customers prefer more and more customized products.

Increased Variety is a very important competitive weapon.

One needs accurate forecasts for build to stock.

Lost Sales in BTF are also a kind of waste (Mismatch between Supply & Demand)

11
 BTO V/S BTF. When Does BTO Make Sense?

• Build To Order is ideal when …

• Customers are patient (willing to wait to receive their order)

• Assembly must be fast, easy and cheap:
• Most likely with a modular product design.
• Customers have a strong preference for variety.
• Inventory is expensive to hold
• High carrying costs, short product life cycles, component price deflation.
• Cannot be too expensive to deliver individual units
• Market forecasts are very inaccurate

12
 The Computer Industry: Catching up with Dell’s Pioneering BTO Model
Inventory turns=COGS/Inventory
100,0

► Dell
75,0
Inventory turns

►Gateway
50,0

►IBM
25,0

►Compaq
0,0
1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004
Year

Dell had a huge advantage but the benefits eroded

13
Timbuk2

Messenger
Bag
Laptop
Sleeve

www.Timbuk2.com
14
Timbuk2

www.Timbuk2.com
15
Dell , Timbuk2 and Original Spin

16
 The Benefits of Customization
• Customization allows a customer to more closely match his wants to the available products. This can
be used to extract a premium.

• Customers don’t care about all dimensions equally. A technique to identify which dimensions are most
important to customers is called conjoint analysis

• Conjoint Analysis surveys people using appropriately designed choices to find out what they really
care about.

The “Most Wanted” painting in America, according to Conjoint Analysis conducted by Komar and Melamid.
See Komar and Melamid’s website on paintings based on attribute surveys: http://www.diacenter.org/km
17
France’s Most and Least Wanted Paintings

Most Wanted Least Wanted

18
Too many dimensions…

Variety of Messenger Bags?

5 sizes *
2 fabrics *
26 colors *26 *26 *
5 liners *
12 logos *
2 (strap)*
2 (insert) *
2 (center divider) *
2 (reflectors) *
2 (left/right) *
2 (pad) *
2 (accessory) *
2 (cell phone) *
2 (holster)

= 5.4 billion combinations

19
How to Deliver on the Options you are Promising?

► Offer Variety on Modular dimensions, such that

modules with limited number of interfaces can be
mixed and matched.

► With modular products you will Assemble-to-Order

rather than building to order. Building to Forecast is
generally only possible for components rather than
final goods.

20
 Spackling at Timbuk2 using Flexible Capacity
Average Demand for Custom + Std
200
Production Volume

150
Standard products (BTF)
100
Demand for Custom

50
Custom products (BTO)
0
0 5 10 15 20 25 30

Day

A combination of Customized and Standard products smooths production levels

Cattani, K., E. Dahan, and G. Schmidt. 2007. Spackling: Smoothing Built-to-Order Production of Mass-Customized Products with Build-To-Forecast production of Standard Items..

21
Dual Sourcing Example:

Assume that all of Timbuk2’s bags are similar, i.e. they all offer Timbuk2 the same margin of
24% of the wholesale price (offered to retailers), if they are produced in China, and a margin of
14% of the wholesale price, if produced in SanFrancisco. You salvage all left-over bags by selling
them in bulk to an outlet store, with the salvage value of 68% of the wholesale price.

How many bags should you source from China?

(a) If you buy bags for the retail channel only from China;

(b) If you source bags from China, but use San Francisco as a back-up, in case
you run out of stock. Do you have enough capacity in San Francisco?

22
Part (a) you buy only from China:

Assume that all of Timbuk2’s bags are similar, i.e. they all offer Timbuk2 the same margin of
24% of the wholesale price (offered to retailers), if they are produced in China, and a margin of
14% of the wholesale price, if produced in SanFrancisco. You salvage all left-over bags by selling
them in bulk to an outlet store, with the salvage value of 68% of the wholesale price.

N(μ, σ 2)
c p
China Timbuk2

Consumer Demand
s
Salvaging

23
Part (a) you buy only from China:

Assume that all of Timbuk2’s bags are similar, i.e. they all offer Timbuk2 the same margin of
24% of the wholesale price (offered to retailers), if they are produced in China, and a margin of
14% of the wholesale price, if produced in SanFrancisco. You salvage all left-over bags by selling
them in bulk to an outlet store, with the salvage value of 68% of the wholesale price.

N(μ, σ 2)
c p
China Timbuk2

Consumer Demand
s
Salvaging

Cost of having one unit less than needed: Cu = p − c = margin = 0.24 ⋅ p

Cost of having one unit more than needed: Co = c − s = 0.08 ⋅ p

Cu 0.24p 0.24
CF = = = = 0.75 → z = 0.674
Cu + Co 0.24p + 0.08p 0.32
24
Part (a) you buy only from China:

Assume that demand is normal with an average 1017 and standard deviation 388 and p=$40.

N(μ, σ 2)
c p
China Timbuk2

Consumer Demand
s
z = 0.674
Salvaging

Q*
China
= μ + z ⋅ σ = 1017 + 388 * 0.674 = 1279

25
Relationship between leftover, demand, order quantity, shortage
Unit of Demand

Is Satisfied → Sales Is NOT Satisfied → Shortages

Expected sales = exp. demand – exp. shortages

Unit of Inventory Ordered

Is Bought → Sales Is NOT Bought → Leftover

Expected leftovers = order size – exp. sales

26
Part (a) you buy only from China:

Assume that demand is normal with an average 1017 and standard deviation 388 and p=$40.

N(μ, σ 2)
c p
China Timbuk2

Consumer Demand
s
z = 0.674
Salvaging L(z) = 0.149
Q*
China
= 1279

Expected Shortages: σ ⋅ L(z) = 0.149 ⋅ 388 = 57.8

Expected Sales: μ − 57.8 = 1017 − 57.8 = 959.2

Expected Leftover: Q* − 959.2 = 1279 − 959.2 = 318.8

Expected Profit: p ⋅ Sales + s ⋅ Leftover − Q* ⋅ c = $8157.76

c = 0.76 ⋅ p
s = 0.68 ⋅ p 27
Part (b) you buy from China but use SF as a back-up:

SF, if needed

cSF
N(μ, σ 2)
cCh p
China Timbuk2

Consumer Demand
s
Salvaging

28
Part (b) you buy from China but use SF as a back-up:

SF, if needed

cSF
N(μ, σ 2)
cCh p
China Timbuk2

Consumer Demand
s
Salvaging

Cost of having one unit less than needed: Cu = 0.24 ⋅ p − 0.14 ⋅ p = 0.1 ⋅ p

Cost of having one unit more than needed: Co = cCh − s = 0.08 ⋅ p

29
Part (b) you buy from China but use SF as a back-up:

SF, if needed

cSF
N(μ, σ 2)
cCh p
China Timbuk2

Consumer Demand
s
Salvaging

Cu 0.1 ⋅ p 0.1
CF = = = = 0.56 → z = 0.151
Cu + Co 0.1 ⋅ p + 0.08 ⋅ p 0.18

Q*
China
= μ + z ⋅ σ = 1017 + 388 * 0.151 = 1076

Q*
SF
=?
30
 Part (b) you buy from China but use SF as a back-up:
L(z) = 0.328

Expected Shortages (not enough inventory bought from China): σ ⋅ L(z) = 0.328 ⋅ 388 = 127.26

Expected Sales (using Chinese inventory): μ − 127.26 = 1017 − 127.26 = 889.74

Expected Leftover (of Chinese inventory): Q*

China
− 889.74 = 1076 − 889.74 = 186.26

31
 Part (b) you buy from China but use SF as a back-up:
L(z) = 0.328

Expected Shortages (not enough inventory bought from China): σ ⋅ L(z) = 0.328 ⋅ 388 = 127.26
Observed (total) Shortages: 0
Expected Sales (using Chinese inventory): μ − 127.26 = 1017 − 127.26 = 889.74
Observed (total) Sales: mu = 1017

Expected Leftover (of Chinese inventory): Q*

China
− 889.74 = 1076 − 889.74 = 186.26
Observed (total) Leftover: 186.26

cCh = 0.76 ⋅ p p − cSF = 0.14 ⋅ p cSF = (1 − 0.14) ⋅ p = 0.86 ⋅ p s = 0.68 ⋅ p

Expected Profit:

32
 Part (b) you buy from China but use SF as a back-up:
L(z) = 0.328

Expected Shortages (not enough inventory bought from China): σ ⋅ L(z) = 0.328 ⋅ 388 = 127.26
Observed (total) Shortages: 0
Expected Sales (using Chinese inventory): μ − 127.26 = 1017 − 127.26 = 889.74
Observed (total) Sales: mu = 1017

Expected Leftover (of Chinese inventory): Q*

China
− 889.74 = 1076 − 889.74 = 186.26
Observed (total) Leftover: 186.26

cCh = 0.76 ⋅ p cSF = (1 − 0.14) ⋅ p = 0.86 ⋅ p s = 0.68 ⋅ p

Expected Profit:

p ⋅ Total Sales + s ⋅ Leftover − Q*

China
⋅ cCh − QSF ⋅ cSF = $8658.128

8658 − 8157
ΔΠ = = 6%
8157 33
What happened at Timbuk2?
• Decided for production in China.

• Going from BTO to BTF (production in China) introduced a lot of complexity into Timbuk2’s
business: hundreds of SKUs popped up - they were still thinking in terms of large product
variety!!

• Small risk in stocking raw materials became high risk in stocking FGI (finished good
inventory), multiple channels had separate stockpiles of inventories, forecasting was a mess.

• Even though sales skyrocketed to $20M in 2006, the company lost $3.5M in the same year
due to supply chain inefficiencies such as write-offs, inventory, material costs and low
margins. It almost went bankrupt.

• New management took over in 2007 and have done a lot to introduce order and reduce
complexity. They cut down from about 800 to about 200 SKUs in wholesale channels,
consolidated inventories and introduced order into forecasting. Now Timbuk2 dual-
produces: 80% of the wholesale volume is in China and the rest in SFO, in case demand
exceeds forecast.

34
The Common Thread?

35
 The Common Thread: Resequencing When Decisions are Made

Decide After Information

TRADITIONAL TIME-LINE

Assortment, Pricing Production Sales Markdowns Receivables

NEW TIME-LINE

Component Sales Production Markdowns Receivables

Assortment, Pricing

THE INNOVATION WORKS BEST WHEN

Change sequence of events– Bets (Production, Hiring) and High Mismatch Costs (Demand Uncertainty)
Information (Demand Information) Modular Product, Desire for Variety
Changes Exposure to Sales Uncertainty Risks, Increase Costs Limited benefits of commitment

36
Summary

Product
Wait Waste
Variety
BTF (FGI) Shorter (Good) Small (Bad) High (Bad)

BTO (no FGI) Longer (Bad) More (Good) Low (Good)

Re-sequencing Processes is a key way to improve Business Processes: Production after Demand
is realized (BTO/BTF)– Works best when high desire for variety and modular production +
background products

37
 Supply Chain Management

Class 21
Supply chain coordination and contracting

Sajjad Najafi

1
What’s Left?

SOURCES OF Process
SUPPLY CHAIN DESIGN SUPPLY CHAIN MGMT
ADVANTAGE Execution

INSTRUMENTS JIT/ TPS SC Sourcing SCM Projects/

Structure Review

SESSIONS 13, 14 15,16 17, 18 19-21 22-24

FOCUS ON THE SUPPLY CHAIN

LITTLEFIELD FRESH CONNECTION

SIMULATION SIMULATION

Project Presentations:
Session 22 (April 27)
Session 23 (May 11)
Review:
Session 24 (May 16)
 Fresh Connection
The deadline for playing in the next 5 rounds of the simulation as posted on
Blackboard before (you need to implement your desired changes before the date/
time below):

2nd round: April 26th at 6:00 pm;

3rd round: April 28th at 6:00 pm;
4th round: May 1st at 6:00 pm;
5th round: May 9th at 6:00 pm;
6th round: May 11th at 6:00 pm.
What did we do last time?
Should we have Finished Goods Inventory or Not ?

HP: Build To Forecast (BTF) Dell: Build To Order (BTO)

Final Product Demand Demand Final Product

Ex. Ready to wear clothes Ex. Airplanes

Fast Food Tailored Suits
Buffet Table Gourmet Restaurant
Factories Dell Computers
Goal A La Carte
Services

5
The Common Thread?

6
 The Common Thread: Resequencing When Decisions are Made

Decide After Information

TRADITIONAL TIME-LINE

Assortment, Pricing Production Sales Markdowns Receivables

NEW TIME-LINE

Component Sales Production Markdowns Receivables

Assortment, Pricing

THE INNOVATION WORKS BEST WHEN

Change sequence of events– Bets (Production, Hiring) and High Mismatch Costs (Demand Uncertainty)
Information (Demand Information) Modular Product, Desire for Variety
Changes Exposure to Sales Uncertainty Risks, Increase Costs Limited benefits of commitment

7
Summary

Product
Wait Waste
Variety
BTF (FGI) Shorter (Good) Small (Bad) High (Bad)

BTO (no FGI) Longer (Bad) More (Good) Low (Good)

Re-sequencing Processes is a key way to improve Business Processes: Production after Demand
is realized (BTO/BTF)– Works best when high desire for variety and modular production +
background products

8
Dual Sourcing Example:

Assume that all of Timbuk2’s bags are similar, i.e. they all offer Timbuk2 the same margin of
24% of the wholesale price (offered to retailers), if they are produced in China, and a margin of
14% of the wholesale price, if produced in SanFrancisco. You salvage all left-over bags by selling
them in bulk to an outlet store, with the salvage value of 68% of the wholesale price.

How many bags should you source from China?

(a) If you buy bags for the retail channel only from China;

(b) If you source bags from China, but use San Francisco as a back-up, in case
you run out of stock. Do you have enough capacity in San Francisco?

9
Part (a) you buy only from China:

Assume that all of Timbuk2’s bags are similar, i.e. they all offer Timbuk2 the same margin of
24% of the wholesale price (offered to retailers), if they are produced in China, and a margin of
14% of the wholesale price, if produced in SanFrancisco. You salvage all left-over bags by selling
them in bulk to an outlet store, with the salvage value of 68% of the wholesale price.

N(μ, σ 2)
c p
China Timbuk2

Consumer Demand
s
Salvaging

Cost of having one unit less than needed: Cu = p − c = margin = 0.24 ⋅ p

Cost of having one unit more than needed: Co = c − s = 0.08 ⋅ p

Cu 0.24p 0.24
CF = = = = 0.75 → z = 0.674
Cu + Co 0.24p + 0.08p 0.32
10
Part (a) you buy only from China:

Assume that demand is normal with an average 1017 and standard deviation 388 and p=$40.

N(μ, σ 2)
c p
China Timbuk2

Consumer Demand
s
z = 0.674
Salvaging

Q*
China
= μ + z ⋅ σ = 1017 + 388 * 0.674 = 1279

11
Relationship between leftover, demand, order quantity, shortage
Unit of Demand

Is Satisfied → Sales Is NOT Satisfied → Shortages

Expected sales = exp. demand – exp. shortages

Unit of Inventory Ordered

Is Bought → Sales Is NOT Bought → Leftover

Expected leftovers = order size – exp. sales

12
Part (a) you buy only from China:

Assume that demand is normal with an average 1017 and standard deviation 388 and p=$40.

N(μ, σ 2)
c p
China Timbuk2

Consumer Demand
s
z = 0.674
Salvaging L(z) = 0.149
Q*
China
= 1279

Expected Shortages: σ ⋅ L(z) = 0.149 ⋅ 388 = 57.8

Expected Sales: μ − 57.8 = 1017 − 57.8 = 959.2

Expected Leftover: Q* − 959.2 = 1279 − 959.2 = 318.8

Expected Profit: p ⋅ Sales + s ⋅ Leftover − Q* ⋅ c = $8157.76

c = 0.76 ⋅ p
s = 0.68 ⋅ p 13
Part (b) you buy from China but use SF as a back-up:

SF, if needed

cSF
N(μ, σ 2)
cCh p
China Timbuk2

Consumer Demand
s
Salvaging

Cost of having one unit less than needed: Cu = 0.24 ⋅ p − 0.14 ⋅ p = 0.1 ⋅ p

Cost of having one unit more than needed: Co = cCh − s = 0.08 ⋅ p

14
Part (b) you buy from China but use SF as a back-up:

SF, if needed

cSF
N(μ, σ 2)
cCh p
China Timbuk2

Consumer Demand
s
Salvaging

Cu 0.1 ⋅ p 0.1
CF = = = = 0.56 → z = 0.151
Cu + Co 0.1 ⋅ p + 0.08 ⋅ p 0.18

Q*
China
= μ + z ⋅ σ = 1017 + 388 * 0.151 = 1076

Q*
SF
=?
15
 Part (b) you buy from China but use SF as a back-up:
L(z) = 0.328

Expected Shortages (not enough inventory bought from China): σ ⋅ L(z) = 0.328 ⋅ 388 = 127.26

Expected Sales (using Chinese inventory): μ − 127.26 = 1017 − 127.26 = 889.74

Expected Leftover (of Chinese inventory): Q*

China
− 889.74 = 1076 − 889.74 = 186.26

16
 Part (b) you buy from China but use SF as a back-up:
L(z) = 0.328

Expected Shortages (not enough inventory bought from China): σ ⋅ L(z) = 0.328 ⋅ 388 = 127.26
Observed (total) Shortages: 0
Expected Sales (using Chinese inventory): μ − 127.26 = 1017 − 127.26 = 889.74
Observed (total) Sales: mu = 1017

Expected Leftover (of Chinese inventory): Q*

China
− 889.74 = 1076 − 889.74 = 186.26
Observed (total) Leftover: 186.26

cCh = 0.76 ⋅ p p − cSF = 0.14 ⋅ p cSF = (1 − 0.14) ⋅ p = 0.86 ⋅ p s = 0.68 ⋅ p

Expected Profit:

17
 Part (b) you buy from China but use SF as a back-up:
L(z) = 0.328

Expected Shortages (not enough inventory bought from China): σ ⋅ L(z) = 0.328 ⋅ 388 = 127.26
Observed (total) Shortages: 0
Expected Sales (using Chinese inventory): μ − 127.26 = 1017 − 127.26 = 889.74
Observed (total) Sales: mu = 1017

Expected Leftover (of Chinese inventory): Q*

China
− 889.74 = 1076 − 889.74 = 186.26
Observed (total) Leftover: 186.26

cCh = 0.76 ⋅ p cSF = (1 − 0.14) ⋅ p = 0.86 ⋅ p s = 0.68 ⋅ p

Expected Profit:

p ⋅ Total Sales + s ⋅ Leftover − Q*

China
⋅ cCh − QSF ⋅ cSF = $8658.128

8658 − 8157
ΔΠ = = 6%
8157 18
This Class
• Bullwhip Effect

• Double marginalization

• How to coordinate the supply chain

• Information sharing
• Contracts
A chain of independent players
• Local information
• Decentralized decision making

Retailer

20
The Bullwhip Effect
Large swings at
the tip
Delivery

Order
Delivery Small
perturbations
at the handle

Order
Delivery

Order

21
The Bullwhip Effect in the auto industry

Auto parts
80%
Autos
60%

40%

20%
% change in demand

0%
GDP = solid line
-20%

-40%

-60%

-80%
Source:Anderson, Fine and Parker (1996)
Bullwhip effect in the US PC supply chain

Changes in
demand
80%

60%

Semiconductor
40% Equipment

20% PC

0%

-20% Semiconductor

-40%

1995 1996 1997 1998 1999 2000 2001

Annual percentage changes in demand (in $s) at three levels of the semiconductor
supply chain: personal computers, semiconductors and semiconductor
manufacturing equipment.
The Bullwhip Effect
• What is it?
The variance of orders is greater than that of sales and the
distortion increases as one moves upstream

• Consequences
• Inefficient production or excessive inventory
• Low utilization of the distribution channel
• Necessity to have capacity far exceeding average demand
• Poor customer service due to stock-outs

Source: The Bullwhip Effect in Supply

24
Chains, Sloan Management Review, Lee et al. 1997
The Bullwhip Effect
• What causes it?
• Order batching

• Pathological incentives
• Trade promotions and forward buying
• Shortage gaming

• Reactive and overreactive ordering

25
Order batching
• Retailers may be required to order
in integer multiples of some batch 70
size, e.g., case quantities, pallet
60
quantities, full truck load, etc.
50
• The graph shows simulated daily 40

Units
consumer demand (solid line) and
30
supplier demand (squares) when
20
retailers order in batches of 15
units, i.e., every 15th demand a 10

retailer orders one batch from the 0

Time (each period equals one day)
supplier that contains 15 units.
Trade promotions and forward buying
• Supplier gives retailer a temporary discount, called a trade promotion.
• Retailer purchases enough to satisfy demand until the next trade
promotion.
• Example: Campbell’s Chicken Noodle Soup over a one year period:
• This product is promoted in January and June, and demand is high in winter.
Total shipments and consumption One retailer’s buy

7000

6000
Shipments
5000

4000
Cases

Cases

3000
Consumption
2000

1000

Jul
Jan

Feb

May

Jun

Aug

Sep

Nov
Oct
Mar

Apr
Dec

Time (weeks)
Shortage gaming
• Setting:
• Retailers submit orders for delivery in a future period.
• Supplier produces.
• If supplier production is less than orders, orders are rationed, i.e., retailers
are “put on allocation”.
• … to secure a better allocation, the retailers inflate their orders, i.e.,
order more than they need…
• … So retailer orders do not convey good information about true
demand …
• This can be a big problem for the supplier, especially if retailers are
later able to cancel a portion of the order:
• Orders that have been submitted that are likely be canceled are called
phantom orders.
Reactive and over-reactive ordering
• How should a firm respond to a “high” demand observation?
• Is this a signal of higher future demand or just random variation in
current demand?
• Hedge by assuming this signals higher future demand, i.e. order more
than usual.
• Rational reactions at one level propagate up the supply chain.
• Unfortunately, it is human to over react, thereby further
increasing the bullwhip effect.
Battling the bullwhip effect
• Vendor Managed Inventory (VMI)

Vendor (supplier) receives inventory and POS data from the

retailer and calculates how much to ship to the retailer. Vendor
manages the inventory of the retailer directly.

30
Vendor Managed Inventory
• Vendors take control of inventory management at the retailer

• VMI Projects
• Dillard department stores, JC Penney and Wal-Mart have shown sales
increases of 20 to 25% and improvement of inventory turnover up to
30%

• VMI Success Story:

• But VMI is not always a success....

31
Vendor Managed Inventory
• VMI Failure:
• Spartan Stores (now SpartanNash), a grocery chain, shut down its VMI
effort about one year after its inception

• Buyers:
• Didn’t trust the suppliers enough
• Carefully monitor inventories and intervene at the slightest hint of trouble

• Suppliers:
• Didn’t do much to allay buyer’s fears
• Didn’t do as effective a job as buyers

32
The Bullwhip Effect
• Sharing information
• Collaborative Planning, Forecasting and Replenishment (CPFR)
• Smoothing the flow of a product
• Coordinate with retailers to spread deliveries evenly.
• Reduce minimum batch sizes.
• Smaller and more frequent replenishments by sourcing from a distributor rather
than directly sourcing from suppliers.
• Eliminating pathological incentives
• Every day low price
• Restrict returns and order cancellations
• Order allocation based on past sales in case of shortages
• Using Vendor-Managed Inventory
Double marginalization and contracts

34
Supply chain coordination via contracts
• Selling to the Newsvendor

$0.15/newspaper $1/newspaper
$0.90/newspaper

Publisher Newsvendor

35
Supply chain coordination via contracts

N(μ, σ 2)
cp = $0.15 cn = $0.9 p = $1

Consumer Demand
Publisher Newsvendor

No salvaging

What are the incentives of each player?

36
Supply chain incentives
• How many newspapers will the newsvendor purchase?
• The newsvendor calculates that:
• Co = ......
• Cu = ......
• The newsvendor wants to maximize his own profit, so he
purchases X newspapers such that:
• P(D≤X) = cu / (cu + co ) = ......

37
Supply chain incentives
• How many newspapers will the newsvendor purchase?
• The newsvendor calculates that:
• Co = cn = $0.9
• Cu = p − cn = 1 − 0.9 = $0.1
• The newsvendor wants to maximize his own profit, so he
purchases X newspapers such that:
• P(D≤X) = cu / (cu + co ) = 0.1

38
Supply chain incentives
• How many newspapers will the newsvendor purchase?
• The newsvendor calculates that:
• Co = cn = $0.9
• Cu = p − cn = 1 − 0.9 = $0.1
• The newsvendor wants to maximize his own profit, so he
purchases X newspapers such that:
• P(D≤X) = cu / (cu + co ) = 0.1
• What if the newsvendor and the publisher decide to maximize
their total profit?
• Co = ......
• Cu = ......
• They should produce X* newspapers such that:
• P(D≤X*) = cu / (cu + co ) = ...... 39
Supply chain coordination via contracts

Our New Black Box

N(μ, σ 2)
cp = $0.15 p = $1

Consumer Demand
Publisher Newsvendor

No salvaging

What are the new incentives of each player?

40
Supply chain incentives
• How many newspapers will the newsvendor purchase?
• The newsvendor calculates that:
• Co = cn = $0.9
• Cu = p − cn = 1 − 0.9 = $0.1
• The newsvendor wants to maximize his own profit, so he
purchases X newspapers such that:
• P(D≤X) = cu / (cu + co ) = 0.1
• What if the newsvendor and the publisher decide to maximize
their total profit?
• Co = ......
• Cu = ......
• They should produce X* newspapers such that:
• P(D≤X*) = cu / (cu + co ) = ...... 41
Supply chain incentives
• How many newspapers will the newsvendor purchase?
• The newsvendor calculates that:
• Co = cn = $0.9
• Cu = p − cn = 1 − 0.9 = $0.1
• The newsvendor wants to maximize his own profit, so he
purchases X newspapers such that:
• P(D≤X) = cu / (cu + co ) = 0.1
• What if the newsvendor and the publisher decide to maximize
their total profit?
• Co = cp = $0.15
• Cu = p − cp = 1 − 0.15 = $0.85
• They should produce X* newspapers such that:
• P(D≤X*) = cu / (cu + co ) = 0.85 42
Who will order more and why?

N(μ, σ 2)
cp = $0.15 cn = $0.9 p = $1
CF = 0.1
Consumer Demand
Publisher Newsvendor

Our New Black Box

N(μ, σ 2)
cp = $0.15 p = $1
CF = 0.85
Consumer Demand
Publisher Newsvendor

43
No coordination vs. Centralization

Double marginalization
Profits

Optimal SC Profit

Wholesaler’s Profit

Optimal Newsvendor
Profit
Supply Chain Profit

Retailer’s Profit

Quantity

44
Calculating Performance Measures

Our New Black Box

N(μ, σ 2)
cp = $0.15 p = $1

Consumer Demand
Publisher Newsvendor

No salvaging

E Shortages = σ ⋅ L(z)
E Sales = μ − E Shortages
E Le.over = Q*
t − E Sales
E ProfitSC = p ⋅ E Sales − cp ⋅ Q*
t

What are the profits of publisher and newsvendor

separately?
45
Coordinating the supply chain
• Publisher wants to induce the newsvendor to purchase more units
• Publisher agrees to buy-back unsold units at $0.8/newspaper
• How many units will the newsvendor purchase?

• Newsvendor calculates that:

• Co = ......
• Cu = ......
• The newsvendor wants to maximize his own profit, so he
purchases X newspapers such that:
• P(D≤X) = cu / (cu + co ) = ......

• Can the publisher induce the newsvendor to purchase X=X*?

46
Supply chain coordination via contracts

N(μ, σ 2)
cp = $0.15 cn = $0.9 p = $1

Consumer Demand
Publisher Newsvendor
s = $0.8

Salvaging through buy-back contract

47
Coordinating the supply chain
• Publisher wants to induce the newsvendor to purchase more units
• Publisher agrees to buy-back unsold units at $0.8/newspaper
• How many units will the newsvendor purchase?

• Newsvendor calculates that:

• Co = cn − s = 0.9 − 0.8 = $0.1
• Cu = p − cn = 1 − 0.9 = $0.1
• The newsvendor wants to maximize his own profit, so he
purchases X newspapers such that:
• P(D≤X) = cu / (cu + co ) = 0.5

• Can the publisher induce the newsvendor to purchase X=X*?

48
How do we calculate how much is bought-back?

N(μ, σ 2)
cp = $0.15 cn = $0.9 p = $1

Consumer Demand
Publisher Newsvendor
s = $0.8

Salvaging through buy-back contract

E Shortages = σ ⋅ L(z)
E Sales = μ − E Shortages
E Le.over = Q*
n − E Sales

49
Buyback contracts
• Advantages:
• Reduce the retailer’s overage costs (via risk sharing)

• Limitations:
• Administrative costs
• Shipping costs
• Sales efforts

50
Buyback contracts
• Why do publishers give refunds for unsold books?
• Publishers give full refunds for unsold books
• 20 million books are returned to the Time Warner Book Group every year

• Publishing Industry:
• Short life cycle: 6-8 weeks
• Highly risky business: 95% of all published books loose money or barely break
even

51
• Rents designer dresses for
10-15% of the dress’ retail price
• Keep the dress for 4 or 8 days
• Return for free. RTR takes care
of dry cleaning.
Characteristics of Rent the Runway
• With a standard wholesale price contract, RTR cannot earn a
profit in the initial weeks of buying a dress:
• RTR’s cost per dress = $750, revenue per rental = $90, variable
cost per rental = $31.
• ...So profit per dress is $90-31=$59.
• What is the required number of rentals to justify purchasing a
dress?

• A dress is unlikely to rent 12+ times in the first week or two.

53
Standard Wholesale price contract

• How many dresses will RTR buy?

• Is this the best decision for the54supply chain?
Characteristics of Rent the Runway
• Lack of inventory is not due to poor forecasting or bad inventory
management, it is due to the economics imposed on RTR.

• Even if each firm in the supply chain chooses optimal actions,

overall supply chain performance may not be optimal.

55
Profit-sharing: A big pie

57
Revenue-sharing contracts
• Blockbuster:
• In the summer of 1997, movie fans flocked to their local Blockbuster eager to
rent The English Patient and Jerry Maguire only to find out that all ten copies
or so had already been checked out

• At $60 a copy, Blockbuster couldn’t afford to stock the number of tapes it

needed
• At $3 per rental, Blockbuster had to rent a tape at least 20 times

• Its suppliers, the movie studios, had to charge $60 to earn enough revenue
themselves

• No one (the suppliers, the retailer and the customers) was happy!

Source: Turning the Supply Chain into a Revenue Chain,

58
Harvard Business Review, Cachon and Lariviere 2001
Revenue-sharing contracts
• In 1998, Blockbuster agreed to give the studios 50% of the rental
fees in return for $9 on tapes

• What does this affect?

• Blockbuster’s overage cost

• How does the studio make up for lost margin?

• Taking a share of revenue

• Who wins?
• Blockbuster and Studios: More revenue for both
• Customers: More availability

59
Revenue-sharing contracts
• When do they work?
• In industries characterized by low marginal cost of production
compared to the price of the product

• Low cost of administering a revenue-sharing program

• Satisfied in industries with high use of IT, like POS scanners etc.

60
Next class

Presentations - Please be in class on time!