Recipe 1: Process Improvement

Step 1: Draw a process flow diagram

Example 1 : Review Sheet 1, Problem 1. The workers are the resources, not the activities. So
we have three resources:

Worker 1 Worker 2 Worker 3

Input Output
Activity 1 Activities 2+3 Activities 4+5

Step 2: Find the capacity of each resource

Capacity is the number of units of flow that a given resource can process per unit of time.

1
Capacity of a resource = .
average time spent per item by the resource

Example 2 : If it takes 6 min (=0.1 hours) to process an item, capacity is 1/6 units per min, or
10 units per hour.

Step 3: Find the bottleneck

“A chain is only as strong as its weakest link, its bottleneck”

• If in-flow = out-flow for all resources (i.e, there is no rework, no loss of yield, etc.), the
bottleneck is the resource with the lowest capacity.

• If in-flow ̸= out-flow for at least one resource, the bottleneck is the resource with the
highest utilization.

Input rate to the resource

Utilization of a resource(ρ) = .
Capacity of the resource

Step 4: Find the system capacity (utilize bottleneck at 100%)

Start with the bottleneck and make its inflow equal to its capacity. Then

• go backwards to find the inflow for the entire system.

• go forward to find the output rate which is called system capacity.

System capacity: System output per unit time when the bottleneck is fully (100%) utilized.

1
 Recipe 2: Managing Variability
Step 1: Write down all the parameters
List of parameters: λ, τ , s, CVa , CVs

Customer parameters

• Arrival rate, λ: number of customers arriving per unit time (e.g., 10 [customers/hour])

• Average interarrival time (IAT): average time between two consecutive arrivals, IAT = 1
λ

• Coefficient of variation for arrivals, CVa : extent of variability of arrivals in relation to

their mean,
standard deviation of IAT
CVa = .
mean of IAT

Example: Poisson arrivals of mean λ have exponential distributed interarrival times with stan-
dard deviation λ1 and mean λ1 , hence CVa = 1/λ
1/λ = 1.

Service parameters

• Service rate (capacity), µ: number of customers served per unit time (e.g., 20 [cus-
tomers/hour])

• Average service time, τ : average time it takes to serve one customer, τ = 1

µ (0.05[h])

• Coefficient of variation for service, CVs : extent of variability of services offered in relation
to their mean,
standard deviation of service time standard deviation of service time
CVs = = .
average service time τ

Example: Service times that are distributed exponentially with parameter µ have a standard
deviation of µ1 and mean µ1 , thus here CVs = 1/µ
1/µ = 1.

Performance measures

• Lq : average number of customers waiting in the queue

• L: average number of customers waiting in the system overall (queued + on service)

• Wq : average waiting time in the queue

• W : average total waiting time in the system (queue+service)

W = Wq + τ

• Little’s Law

Lq = λ · Wq
L = λ · W.

2
Step 2: Find the number of servers s

Step 3: Find the utilization of a server ρ

λ λ·τ
ρ= = .
µ·s s

Step 4: Calculate the average waiting time in the queue

Calculations here depend on the number of servers s:
• Case 1: If s = 1, use the following formula for Wq :
 2
CVa + CV2s

ρ
Wq = τ · · .
1−ρ 2

• Case 2: If s > 1 (there are multiple servers), you can either (1) find Lq from the table
at the end of the review sheet for your ρ1 . After, you can use Little’s law to calculate
L
Wq = λq ; or (2) use the waiting time formula for multiple servers
 2  √
τ ρ CVa + CV2s
Wq = · · · ρ 2(s+1)−2 .
s 1−ρ 2

Step 5: Calculate the average total waiting time in the system

Average total waiting time in the system = queueing time + service time:

W = Wq + τ.

Step 6: Calculate the average number of customers

If you haven’t yet found Lq in step 4, then calculate it using Little’s law:

Lq = λ · Wq .

For the total number of customers in the system, you can also use Little’s law:

L = λ · W.

1
Note that, sometimes, the exact value of ρ for which you would like to find Lq is not in the table. In this case,
you need to use a linear approximation. First, find the closest value to ρ from below. Denote it by ρ and find the
corresponding Lq . Secondly, find the closest value to ρ from above. Denote it by ρ and find the corresponding
L̄q . Then, the value of ρ for ρ < ρ < ρ is approximated in the following way:

Lq − Lq
Lq (ρ) = Lq + · (ρ − ρ)
ρ−ρ

3
Recipe 3: Managing Risks (The Newsvendor Model)
Step 1: Collect the inputs c, r, s, µ, σ

Demand parameters
• How is the demand distributed? (in most of the cases, it will be normal denoted N (µ, σ 2 )).
• µ : mean (average).
• σ: standard deviation (or variance, which is σ 2 ).

Product parameters
• c: unit manufacturing (acquisition) cost (the cost of baking a pie is c = 10$. This includes
labor cost of your grandmother, cost of ingredients, etc.)
• r: selling price per unit (price at which your grandmother sells cooked pies to her neigh-
bours, e.g. r = 12$)
• s: salvage value per unit (your grandmother can give the pie to you at s = 1$ if no one
buys it).
Note: usually s < c < r. When the good is perishable, usually s = 0.

Step 2: Calculate underage and overage costs Cu , Co

• Cu = r − c: consequence of producing (stocking) not enough items (opportunity cost).
• Co = c − s: consequence of producing (stocking) too many items.

Step 3: Calculate the critical ratio

Cu
critical ratio = .
Cu + Co

Step 4: Find order quantity Q∗ which maximized expected profit

• Case 1: If demand if normal N (µ, σ 2 ).
– Find z-value from Table A (standard normal distribution Φ(z)), i.e. find z ∗ such
that:
Cu
Φ(z ∗ ) = .
Cu + Co
Note: basically you need to find your critical ratio in Table A and find the corre-
sponding z-value. If the critical ratio is not there, use approximation – take the
closest z from above.

4
 – Calculate Q∗ using the formula

Q∗ = µ + z ∗ · σ.

• Case 2: If demand follows some discrete distribution F (·).

– Find quantity Q∗ such that

Cu
F (Q∗ ) = .
Cu + Co
Note: Choose the closest Q∗ from above.

Step 5: Calculate performance measures

This step can be done for any arbitrary order quantity Q.

1. Normalize the order quantity to get the z-statistic

Q−µ
z=
σ

2. Look up in the Standard Normal Loss Function Table (Table B) the expected lost sales
L(z) for a standard normal distribution with that z-statistic.

3. Evaluate lost sales (or unmet demand)

E[Lost sales] = σ · L(z).

4. Compute the other quantities as follows

E[Sales] = E[Demand] − E[Lost sales] = µ − E[Lost sales]

E[Leftover inventory] = Order quantity − E[Sales] = Q − E[Sales]
E[Profit] = (r − c) · µ − Cu · E[Lost sales] − Co · E[Leftover inventory]

Additionally, we can calculate the in-stock and stock-out probability as follows.

• In-stock probability = F (Q) : the probability that the firm will end up selling season
without shortages. In other words, it is the probability that the good will be available (in
stock) for every customer. Basically, it is the probability that the realized demand will
be less than Q∗ that you stock.

• Stock-out probability = 1 − F (Q) : the probability that there will be shortages during
the selling period.

5
Table A: Standardized Normal Probability Table

6
Table B: Standardized Normal Loss Table

7
 Table C: Lq Values for the Multi-server Queue

Value of Lq for s servers, with mean utilization rate ρ, assuming Poisson arrivals and exponential
service times (known as a M/M/s queue).

Utilization Number of servers (s)

rate ρ 1 2 3 4 5
.10 .0111 .0020 .0004 .0001 .0000
.20 .0500 .0167 .0062 .0024 .0010
.30 .1286 .0593 .0300 .0159 .0086
.35 .1885 .0977 .0552 .0325 .0196
.40 .2667 .1524 .0941 .0605 .0398
.45 .3682 .2285 .1522 .1052 .0743
.50 .5000 .3333 .2368 .1739 .1304
.55 .6722 .4771 .3583 .2772 .2185
.60 .9000 .6750 .5321 .4306 .3542
.62 1.0116 .7743 .6213 .5109 .4269
.64 1.1378 .8880 .7246 .6051 .5130
.66 1.2812 1.0188 .8446 .7158 .6152
.68 1.4450 1.1698 .9847 .8461 .7367
.70 1.6333 1.3451 1.1488 1.0002 .8816
.72 1.8514 1.5500 1.3423 1.1834 1.0553
.74 2.1062 1.7914 1.5721 1.4025 1.2646
.76 2.4067 2.0785 1.8472 1.6668 1.5187
.78 2.7655 2.4237 2.1803 1.9887 1.8302
.80 3.2000 2.8444 2.5888 2.3857 2.2165
.82 3.7356 3.3661 3.0979 2.8832 2.7029
.84 4.4100 4.0265 3.7456 3.5190 3.3273
.86 5.2829 4.8852 4.5914 4.3526 4.1493
.88 6.4533 6.0414 5.7345 5.3834 5.2682
.90 8.1000 7.6737 7.3535 7.0898 6.8624
.92 10.5800 10.1392 9.8056 9.5290 9.2893
.94 14.7267 14.2712 13.9240 13.6344 13.3821
.96 23.0400 22.5698 22.2088 21.9060 21.6408
.98 48.0200 47.5350 47.1602 46.8439 46.5656
.99 98.0101 97.5176 97.1357 96.8127 96.4274

When one is faced with a value of ρ that lies between two values in the table, one may have
to interpolate between the two values of Lq associated with the two successive values of ρ. For
example, consider s = 2 with ρ = 0.75. This utilization rate lies in between the two table values
of 0.74 and 0.76. The Lq for this M/M/2 queue can be approximated by the following linear
interpolation:

(2.0785 − 1.7914)
Lq = 1.7914 + (0.75 − 0.74) = 1.7914 + 0.1435 = 1.9349
(0.76 − 0.74)