Service Levels and Lead Times in Supply Chains:

The Order-up-to Inventory Model

Chapter 13
Medtronics InSync pacemaker
supply chain and objectives
Supply chain:
One distribution center (DC) in Mounds
View, MN.
About 500 sales territories throughout
the country.
Consider Susan Magnottos territory
in Madison, Wisconsin.

Objective:
Because the gross margins are high,
develop a system to minimize inventory
investment while maintaining a very high
service target, e.g., a 99.9% in-stock
probability or a 99.9% fill rate.
Timing in the order up-to model
Time is divided into periods of equal length, e.g., one hour, one month.
During a period the following sequence of events occurs:
A replenishment order can be submitted.
Inventory is received.
Random demand occurs.
Lead times:
An order is received after a fixed number of periods, called the lead time.
Let l represent the length of the lead time.

Fig 13.4:
An example with l = 1
Order up-to model vs. newsvendor model
Both models have uncertain future demand, but there are
differences











Newsvendor applies to short life cycle products with uncertain
demand and the order up-to applies to long life cycle products with
uncertain, but stable, demand.
Newsvendor Order up-to
Inventory
obsolescence
After one period Never
Number of
replenishments
One (maybe two or
three with some
reactive capacity)
Unlimited
Demand occurs
during replenishment
No Yes
InSync demand and inventory at the DC
Average monthly demand = 349 units
Standard deviation of monthly demand
= 122.28
Average weekly demand = 349/4.33 =
80.6
Standard deviation of weekly demand =

(The evaluations for weekly demand
assume 4.33 weeks per month and
demand is independent across weeks.)
81 . 58 33 . 4 / 38 . 122 =
0
100
200
300
400
500
600
700
J
a
n
F
e
b
M
a
r
A
p
r
M
a
y
J
u
n
J
u
l
A
u
g
S
e
p
O
c
t
N
o
v
D
e
c
Month
U
n
i
t
s
Fig 13.2 Monthly implants (columns)
and end of month inventory (line)
InSync demand and inventory
in Susans territory
Total annual demand = 75
units
Average daily demand = 0.29
units (75/260), assuming 5
days per week.
Poisson demand distribution
works better for slow moving
items
0
2
4
6
8
10
12
14
16
J
a
n
F
e
b
M
a
r
A
p
r
M
a
y
J
u
n
J
u
l
A
u
g
S
e
p
O
c
t
N
o
v
D
e
c
Month
U
n
i
t
s
Fig 13.3 Monthly implants (columns) and
end of month inventory (line)
Order up-to model definitions
On-order inventory / pipeline inventory = the number of units that have
been ordered but have not been received.

On-hand inventory = the number of units physically in inventory ready to
serve demand.

Backorder = the total amount of demand that has has not been satisfied:
All backordered demand is eventually filled, i.e., there are no lost
sales.

Inventory level = On-hand inventory - Backorder.

Inventory position = On-order inventory + Inventory level.

Order up-to level, S
the maximum inventory position we allow.
sometimes called the base stock level.
Order up-to model implementation
Each periods order quantity = S Inventory position
Suppose S = 4.
If a period begins with an inventory position = 1, then three units are ordered.
(4 1 = 3 )
If a period begins with an inventory position = -3, then seven units are ordered
(4 (-3) = 7)

A periods order quantity = the previous periods demand:
Suppose S = 4.
If demand were 10 in period 1, then the inventory position at the start of
period 2 is 4 10 = -6, which means 10 units are ordered in period 2.
The order up-to model is a pull system because inventory is ordered in
response to demand.
The order up-to model is sometimes referred to as a 1-for-1 ordering
policy.
What determines the inventory level?
Short answer:
Inventory level at the end of a period = S minus demand over l +1 periods.
Explanation via an example with S = 6, l = 3, and 2 units on-hand at the start of period
1
D
1
D
2
D
3
D
4
?
Period 1
Time
Inventory level at the end
of period four
= 6 - D
1
D
2
D
3
D
4

Keep in mind:
At the start of a period the
Inventory level + On-order equals S.
All inventory on-order at the start of
period 1 arrives before the end of
period 4
Nothing ordered in periods 2-4
arrives by the end of period 4
All demand is satisfied so there are
no lost sales.
Period 2 Period 3 Period 4
Expected on-hand inventory and backorder
This is like a Newsvendor model in which the order quantity is
S and the demand distribution is demand over l +1 periods.
So
Expected on-hand inventory at the end of a period can be evaluated
like Expected left over inventory in the Newsvendor model with Q = S.
Expected backorder at the end of a period can be evaluated like
Expected lost sales in the Newsvendor model with Q = S.
S

Period 1
Time

Period 4
D

S D > 0, so there is
on-hand inventory
S D < 0, so there
are backorders
D = demand
over l +1
periods

Stockout and in-stock probabilities, on-order inventory and fill rate
The stockout probability is the probability at least one unit is
backordered in a period:


The in-stock probability is the probability all demand is filled in a
period:


Expected on-order inventory = Expected demand over one period x
lead time
This comes from Littles Law. Note that it equals the expected
demand over l periods, not l +1 periods.

The fill rate is the fraction of demand within a period that is NOT
backordered:


Expected backorder
Fill rate 1-
Expected demand in one period
=
( ) { }
( ) { } S periods 1 l over Demand Prob
S periods 1 l over Demand Prob y probabilit Stockout
s + =
> + =
1
( ) { } S periods 1 l over Demand Prob
y probabilit Stockout - 1 y probabilit stock - In
s + =
=
Demand over l+1 periods
DC:
The period length is one week, the replenishment lead time is three
weeks, l = 3
Assume demand is normally distributed:
Mean weekly demand is 80.6 (from demand data)
Standard deviation of weekly demand is 58.81 (from demand data)
Expected demand over l +1 weeks is (3 + 1) x 80.6 = 322.4
Standard deviation of demand over l +1 weeks is

Susans territory:
The period length is one day,
the replenishment lead time is one day, l =1
Assume demand is Poisson distributed:
Mean daily demand is 0.29 (from demand data)
Expected demand over l+1 days is 2 x 0.29 = 0.58
Recall, the Poisson is completely defined by its mean (and the standard
deviation is always the square root of the mean)

6 . 117 81 . 58 1 3 = +
DCs Expected backorder assuming S = 625
Expected backorder is analogous to the Expected lost sales in the
Newsvendor model:
Suppose S = 625 at the DC
Normalize the order up-to level:


Lookup L(z) in the Standard Normal Loss Function Table:

Convert expected lost sales, L(z), for the standard normal into the
expected backorder with the actual normal distribution that represents
demand over l+1 periods:



Therefore, if S = 625, then on average there are 0.19 backorders at the end
of any period at the DC.
Other DC performance measures with S = 625
Fill rate




On-hand inventory




Units on order






Performance measures in Susans territory
Look up in the Poisson Loss Function Table expected backorders
for a Poisson distribution with a mean equal to expected
demand over l+1 periods:






Suppose S = 3:
Expected backorder
In-stock
Fill rate
Expected on-hand
Expected on-order inventory
S F(S) L (S ) S F(S) L (S )
0 0.74826 0.29000 0 0.55990 0.58000
1 0.96526 0.03826 1 0.88464 0.13990
2 0.99672 0.00352 2 0.97881 0.02454
3 0.99977 0.00025 3 0.99702 0.00335
4 0.99999 0.00001 4 0.99966 0.00037
5 1.00000 0.00000 5 0.99997 0.00004
F(S ) = Prob {Demand is less than or equal to S}
L (S ) = loss function = expected backorder = expected
amount demand exceeds S
Mean demand = 0.29 Mean demand = 0.58
Choosing S
Target fill rate w/normal demand
Target in-stock w/normal demand
Target fill rate w/Poisson demand
Target in-stock w/Poisson demand

Justifying a service level via cost minimization
Let h equal the holding cost per unit per period
e.g. if p is the retail price, the gross margin is 75%, the annual holding
cost is 35% and there are 260 days per year, then h = p x (1 -0.75) x
0.35 / 260 = 0.000337 x p

Let b equal the penalty per unit backordered
e.g., let the penalty equal the 75% gross margin, then b = 0.75 x p

Too much-too little challenge:
If S is too high, then there are holding costs, C
o
= h
If S is too low, then there are backorders, C
u
= b

Cost minimizing order up-to level satisfies
The optimal in-stock probability is usually quite high
Suppose the annual holding cost is 35%, the backorder
penalty cost equals the gross margin and inventory is
reviewed daily. (Table 13.3 Graphed)
88%
90%
92%
94%
96%
98%
100%
0% 20% 40% 60% 80% 100%
Gross margin %
O
p
t
i
m
a
l

i
n
-
s
t
o
c
k

p
r
o
b
a
b
i
l
i
t
y





Impact of the period length
Increasing the period length leads to larger and less frequent
orders:
The average order quantity = expected demand in a single period.
The frequency of orders approximately equals 1/length of period.

Suppose there is a cost to hold inventory and a cost to submit
each order (independent of the quantity ordered)

then there is a tradeoff between carrying little inventory
(short period lengths) and reducing ordering costs (long
period lengths)
Example with mean demand per week = 100
and standard deviation of weekly demand = 75.
Period lengths of 1, 2, 4 and 8 weeks result in average
inventory of 597, 677, 832 and 1130 respectively (Fig 13.7)
0
200
400
600
800
1000
1200
1400
1600
0 2 4 6 8 10 12 14 16
0
200
400
600
800
1000
1200
1400
1600
0 2 4 6 8 10 12 14 16
0
200
400
600
800
1000
1200
1400
1600
0 2 4 6 8 10 12 14 16
0
200
400
600
800
1000
1200
1400
1600
0 2 4 6 8 10 12 14 16
Tradeoff between inventory holding costs and ordering costs
Costs:
Ordering costs = $275 per
order
Holding costs = 25% per year
Unit cost = $50
Holding cost per unit per year =
25% x $50 = 12.5

Period length of 4 weeks
minimizes costs:
This implies the average order
quantity is 4 x 100 = 400 units

EOQ model:
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
22000
0 1 2 3 4 5 6 7 8 9
Period length (in weeks)
C
o
s
t
478
5 . 12
5200 275 2 2
=

=

=
h
R K
Q
Ordering costs
Inventory holding
costs
Total costs
Fig 13.8
Better service requires more inventory
at an increasing rate
More inventory is
needed as demand
uncertainty
increases for any
fixed fill rate.

The required
inventory is more
sensitive to the fill
rate level as
demand
uncertainty
increases
Fig 13.9: The tradeoff between inventory and fill rate with Normally distributed demand
and a mean of 100 over (l+1) periods. The curves differ in the standard deviation of
demand over (l+1) periods: 60, 50, 40, 30, 20, 10 from top to bottom.
Shorten lead times and you will reduce
inventory
Reducing the lead
time reduces
expected
inventory,
especially as the
target fill rate
increases
0
100
200
300
400
500
600
0 5 10 15 20
Lead time
E
x
p
e
c
t
e
d

i
n
v
e
n
t
o
r
y
Fig 13.10: The impact of lead time on expected inventory for four fill rate targets,
99.9%, 99.5%, 99.0% and 98%, top curve to bottom curve respectively. Demand
in one period is Normally distributed with mean 100 and standard deviation 60.
Dont forget about pipeline inventory
Reducing the lead
time reduces
expected inventory
and pipeline
inventory

The impact on
pipeline inventory
can be even more
dramatic that the
impact on expected
inventory
0
500
1000
1500
2000
2500
3000
0 5 10 15 20
Lead time
I
n
v
e
n
t
o
r
y
Fig 13.11: Expected inventory (diamonds) and total inventory (squares), which is
expected inventory plus pipeline inventory, with a 99.9% fill rate requirement and
demand in one period is Normally distributed with mean 100 and standard deviation 60
Risk-Pooling Strategies to
Reduce and Hedge Uncertainty

Chapter 14
Risk pooling strategies
Redesign of the:
supply chain
production process
product

to reduce uncertainty or to hedge uncertainty

Four versions of risking pooling:
location
product
lead time
delayed differentiation
consolidated distribution
capacity
Location pooling at Medtronic
Current operations:
Each sales representative has her own inventory to
serve demand in her own territory.
Lead time is 1 day from Mounds View DC
e.g., 3 territories, 3 stockpiles of inventory

The location pooling strategy:
A single location stores inventory used by several
sales reps.
Sales reps no longer hold their own inventory, they
must pull inventory from the pooled location.
Inventory is automatically replenished at the
pooled location as depleted by demand.
Lead time to pooled location is still 1 day from
Mounds View DC.
e.g., 3 pooled territories, 1 stockpile of inventory
DC
Territory 1
Territory 2
Territory 3
DC
Territory 1
Territory 2
Territory 3
The impact of location pooling on inventory
Suppose each territorys expected daily demand is 0.29, the required in-
stock probability is 99.9% and the lead time is 1 day with individual
territories or pooled territories.











Pooling 8 territories reduces expected inventory from 11.7 days-of-
demand down to 3.6.
But pooling has no impact on pipeline inventory.
Number of
territories
pooled
Pooled territory's
expected demand
per day
(a)
S units
(b)
days-of-
demand
(b/a)
units
(c)
days-of-
demand
(c/a)
1 0.29 4 3.4 11.7 0.29 1.0
2 0.58 6 4.8 8.3 0.58 1.0
3 0.87 7 5.3 6.1 0.87 1.0
4 1.16 8 5.7 4.9 1.16 1.0
5 1.45 9 6.1 4.2 1.45 1.0
6 1.74 10 6.5 3.7 1.74 1.0
7 2.03 12 7.9 3.9 2.03 1.0
8 2.32 13 8.4 3.6 2.32 1.0
Expected inventory Pipeline inventory

Location pooling & the inventory-service curve
Location pooling shifts the inventory-
service tradeoff curve

For a single product, location pooling
decreases inventory with service
constant
increases service with holding cost
constant
a combination of inventory reduction
and service increase.

Location pooling can be used to
broaden the product line.
0
2
4
6
8
10
12
14
16
0.96 0.97 0.98 0.99 1
In-stock probability
E
x
p
e
c
t
e
d

i
n
v
e
n
t
o
r
y

(
d
a
y
s

o
f

d
e
m
a
n
d
)
Fig 14.2: Inventory-service tradeoff curve for different levels
of location pooling. The curves represent, from highest to
lowest, individual territories, two pooled territories, four
pooled territories, and eight pooled territories. Daily demand
in each territory is Poisson with mean 0.29 and the lead time is
one day.
Why does location pooling work?
Location pooling reduces demand
uncertainty (measured by CV)

Reduced demand uncertainty
reduces the inventory (with
constant service level)

But there are declining marginal
returns to risk pooling!
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
0 1 2 3 4 5 6 7 8
Number of territories pooled
E
x
p
e
c
t
e
d

i
n
v
e
n
t
o
r
y
i
n

d
a
y
s

o
f

d
e
m
a
n
d























0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
C
o
e
f
f
i
c
i
e
n
t

o
f

v
a
r
i
a
t
i
o
n









Fig 14.1: The relationship between expected inventory (diamonds) and the
coefficient of variation (squares) as territories are pooled. Daily demand in
each territory is Poisson with mean 0.29 units, the target in-stock
probability is 99% and the lead time is one day.
Location pooling pros, cons and alternatives
Pros:
Reduces demand uncertainty

Cons:
Location pooling moves inventory away from customers:

Alternatives:
Virtual pooling:
Drop shipping:
Product pooling universal design
ONeill sells two Hammer 3/2 wetsuits that are identical except for
the logo silk screened on the chest.









Instead of having two Hammer 3/2 suits, ONeill could consolidate
its product line into a single Hammer 3/2 suit, i.e., a universal
design, which we will call the Universal Hammer.
Surf Hammer 3/2 logo Dive Hammer 3/2 logo
Product pooling analysis assumptions
Demand for the Surf Hammer is Normally distributed with mean
3192 and standard deviation 1181.

Demand for the Dive Hammer has the same distribution as the Surf
Hammer.

Surf and Dive demands are independent
then the Universal Hammers demand has mean 2 x 3192 = 6384 and std
deviation = sqrt(2) x 1181 = 1670.

Price, cost and salvage value for the Universal Hammer are the
same as for the other two:
Hence, C
o
is 110 90 = 20, C
u
= 180-110 = 70
Same critical ratio = 70 /(20 + 70) = 0.7778
Same optimal z statistic, 0.77
Product pooling analysis results
Performance of the two suits (Surf and Dive)
Total order quantity
Total profit

Universal Hammer
Order quantity:

Profit:




Reduces inventory investment by
Increase profit by
The profit increase of
Demand correlation
0
2
4
6
8
10
12
14
16
18
20
0 5 10 15 20
0
2
4
6
8
10
12
14
16
18
20
0 5 10 15 20
0
2
4
6
8
10
12
14
16
18
20
0 5 10 15 20

Fig Random demand for two
products (x-axis is product 1,
y-axis is product 2).
Correlations are -0.9, 0.0, and
+0.9
In all scenarios demand is
~N(10,3).
Key driver of product pooling
Most effective if the
CV
Universal
< CV
individual

( )
|
|
.
|

\
|
+ =

o
n Correlatio 1
2
1
demand pooled CV
380000
390000
400000
410000
420000
430000
440000
450000
-1 -0.5 0 0.5 1
Correlation
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
C
o
e
f
f
i
c
i
e
n
t

o
f

v
a
r
i
a
t
i
o
n
E
x
p
e
c
t
e
d

p
r
o
f
i
t
The correlation between surf and dive demand for the Hammer 3/2 and the
expected profit of the universal Hammer wetsuit (decreasing curve) and the
coefficient of variation of total demand (increasing curve)
CV
individual
= 1181/2192 = 0.37
CV
Universal
= 1670/6384 = 0.26
Negative correlation in demand for
the individual products is best for
reducing CV
Limitations of product pooling/universal design
May not provide key functionality to consumers with special needs

May be more expensive to produce because additional functionality may
require additional components.

May be less expensive to produce/procure because each component is
needed in a larger volume.

May eliminate brand/price segmentation opportunities
Lead time pooling consolidated distribution
Consider the following two systems from Fig 14.9:
In each case weekly demand at each store is ~P(0.5) and the target in-
stock probability at each store is 99.5%
DC demand is ~N(50,15)
If demands were
independent across stores,
then DC demand would
have a standard deviation of
sqrt(50) = 7.07
Consolidated distribution results





Table 14.4

Consolidated distribution
reduces retail inventory by more than 50%!
not as effectively as location pooling
but consolidated distribution keeps inventory near demand,
reduces inventory even though the total lead time increases from 8 to 9
weeks!
Direct
delivery
supply chain
Centralized
inventory
supply chain
Location
pooling
Expected total inventory at the stores 650 300 0
Expected inventory at the DC 0 116 116
Pipeline inventory between
the DC and the stores 0 50 0
Total 650 466 116
Consolidated distribution summary
Reduces inventory in a supply chain via lead time risk pooling
Decide the total quantity to ship from the supplier, not a total quantity
and its allocation across locations.
Most effective if demands are negatively correlated across locations.
Most effective if the supplier lead time is long and the DC to store lead
time is short.
Increases distance traveled & lead time from supplier to stores.

Other benefits of consolidated distribution:
Easier to obtain quantity discounts in purchasing.
Easier to obtain economies of scale in transportation:
Lead time risk pooling delayed differentiation
A Universal Hammer 3/2 increases ONeills profit, but does not allow
ONeill to differentiate between the Surf and the Dive markets.

Delayed differentiation is an alternative to the Universal Hammer:
ONeill stocks generic Hammers with no logo.
When demand occurs ONeill quickly silk screens on the appropriate logo,
This generates the same profit as the Universal Hammer! Wheee!

When does delayed differentiation make sense?
Customers demand variety.
There is less uncertainty with total demand than demand for individual versions.
Variety is created late in the production process.
Variety can be added quickly and cheaply.
Component needed for variety is inexpensive relative to the generic component.
Other examples of delayed differentiation
Retail paint
HP printer
Private label canned goods manufacturer
Black and Decker
Nokia
Capacity pooling with flexible manufacturing
Consider the following stylized situation faced by GM
10 production facilities
10 vehicle models
Each plant is capable of producing 100 units/day
Demand for each product is ~N(100,40)
Each plant can be configured to produce up to 10 products
But flexibility is expensive.
GM must decide which plants can produce which products before demand
is realized.
After demand is realized, GM can allocate its capacity to satisfy demand.
If demand exceeds capacity, sales are lost.
Four possible capacity configurations: no
flexibility to total flexibility
Figs 14.12 & 14.13 The more links, the more flexibility constructed
In the 16 link configuration plant 4 is flexible enough to produce 4 products but
plant 5 has no flexibility (it produces a single product).
How is flexibility used
Flexibility allows production shifts to avoid lost sales.
Two plant, two product example





If demand turns out to be 75 for product A, 115 for product B then..
Plant
2
1
Vehicle
B
A
Plant
2
1
Vehicle
B
A
No flexibility Total flexibility
Product Demand Plant 1 Plant 2 Sales
A 75 75 0 75
B 115 0 100 100
Total Sales 175
Plant Utilization 88%
Production
With no flexibility
Product Demand Plant 1 Plant 2 Sales
A 75 75 0 75
B 115 15 100 115
Total Sales 190
Plant Utilization 95%
Production
With total flexibility
The value of flexibility
Adding flexibility increases capacity utilization and expected sales:















Fig 14.14: Note that 20 links can provide nearly the same performance as total flexibility!
800
850
900
950
1000
80 85 90 95 100
Expected capacity utilization, %
E
x
p
e
c
t
e
d

s
a
l
e
s
,

u
n
i
t
s
No flexibility
Total flexibility
20 links
11 links
12 links
These data are
collected via
simulation
Chaining: how to add flexibility
A chain is a group of plants and products connected via links.
Flexibility is most effective if it is added to create long chains.
A configuration with 20 links can produce nearly the results of total flexibility as
long as it constructs one large chain:
Hence, a little bit of flexibility is very useful as long as it is designed correctly
When is flexibility valuable?
Flexibility is most valuable when
capacity approximately equals
expected demand.

Flexibility is least valuable when
capacity is very high or very low.

A 20 link (1 chain) configuration
with 1000 units of capacity
produces the same expected sales
as 1250 units of capacity with no
flexibility.
If flexibility is cheap relative to
capacity, add flexibility.
But if flexibility is expensive
relative to capacity, add
capacity.
400
500
600
700
800
900
1000
60 70 80 90 100
Expected capacity utilization, %
E
x
p
e
c
t
e
d

s
a
l
e
s
,

u
n
i
t
s
No flexibility
20 links, 1
chain
C=500
C=750
C=880
C=1000
C=1130
C=1250
C=1500

Fig 14.16 C = total capacity of all ten plants
One way to make money with capacity pooling:
contract manufacturing
A fast growing industry:









But one with low margins:
Total revenue of six leading
contract manufacturers by fiscal
year: Solectron Corp, Flextronics
International Ltd, Sanmina-SCI,
Jabil Circuit Inc, Celestica Inc and
Plexus Corp.
0
10000
20000
30000
40000
50000
60000
70000
1
9
9
0
1
9
9
1
1
9
9
2
1
9
9
3
1
9
9
4
1
9
9
5
1
9
9
6
1
9
9
7
1
9
9
8
1
9
9
9
2
0
0
0
2
0
0
1
2
0
0
2
2
0
0
3
Fiscal year
R
e
v
e
n
u
e

(
i
n

m
i
l
l
i
o
n

$
s
)
2003 data:
Firm Revenue* Cost of goods* Gross Margin
Flextronics 14,530 13,705 5.7%
Solectron 11,014 10,432 5.3%
Sanmina-SCI 10,361 9,899 4.5%
Celestica 6,735 6,474 3.9%
Jabil Circuit 4,729 4,294 9.2%
Plexus 807 755 6.4%
* in millions of $s