A Simulation Model of Starbucks

Starbucks- A Simulation
By Spandana Pothuri, M12366835
MS-Business Analytics, 2017
University of Cincinnati
Location: Steger student life center Starbucks coffee center
Tool: Arena Simulation Software

Introduction
Starbucks is one of America’s top coffee shops which has cafes all over the world. At the University of
Cincinnati, there are three such coffee shops. The one at the Steger Student Life Center is always
bustling with students in between classes, there are often queues which extend outside of the entrance.
As a part of the business analytics course ‘Simulation Modelling and Methods’, this café is simulated in
this project, to find a plausible solution to that.
Problem Statement
The Starbucks at the Steger Student Life Center is right across the student recreation center and on the
way to the school of business, engineering research center, two libraries, among other things. Being at
such important crossroads, this Starbucks has a lot of rush during peak hours. It is also completely empty
at times during the slack period. Students have a window of about 5-10 minutes to get a drink and leave,
hence having an optimal process flow is given highest priority at Starbucks. The average wait times are
already low, however, when someone orders food or more than 1 beverage, the wait times are higher. It
also depends on the rush, wait time can cross 10 minutes on rare occasions. This project attempts to
create an approximate simulation of Starbucks to identify what changes can be made to decrease the
wait time for students and staff.
Assumptions
The exact system can never be simulated as there is always variability in real life. Nevertheless, for
optimizing the process, an apt replica has been simulated under the following assumptions (which hold
true when the model is running): -
1. The café is open from 7AM – 8PM (that is 13 hours) For this simulation, it runs through the week
(to collect more data). However, the real Starbucks only runs Monday through Friday
2. There are 3 work shifts - Morning shift from 7AM – 2PM, Afternoon Shift from 10AM – 4PM and
the Evening Shift from 2PM – 8PM
3. Workers take breaks only when there is free time. (which they usually have during class hours)
Otherwise for this model they can work throughout their respective shift times
4. Every customer who enters, buys something. The seating area of Starbucks is not considered for
this model since it is not contributing in any way to the purpose of this simulation
5. The data for the peak hour has been collected manually at Starbucks in between 11AM and
12.30PM. The data for the slack hours has been collected from the manager who gave fair
approximations
6. For the various service times, data is pooled from all hours, both peak and slack hours to create
a general approximate distribution, which is used in the model
7. The customer waits at the self-service counter which contains sugars, straws, cup holders,
tissues etc. There is no specific time allocated for the customer at the self-service counter, since
there is generally always excess supplies and this process takes the customer very little time.
Usually during the wait time customers collect their supplies.
8. When a customer collects his/her order, he or she leaves the system.

9. The peak hours are morning 11AM – 2PM and in the evening around 5.30PM – 6.30PM. There is
also a peak when there are holiday deals, like buy one get one drink free. However, that data
has not been collected and hence an average day without any deals is considered for this project
10. There is always one resource at the cashier. It could be any of the employees, but generally
there is a designated person. At the real Starbucks, baristas take turns at the cashier. In this
simulation, a constant fixed resource is allocated to the cashier as it essentially outputs the
same result
11. All the baristas are cross trained, both at the real Starbucks and in the simulated Starbucks
Data Collection
Data was collected at the location during peak hours, during 11AM – 12.30PM. Around 61 customers
have been recorded during this time. Their interarrival times, orders, order times, food processing time
and beverage processing time have all been recorded. Based on observed data and the approximations
provided by the Manager of the café, around 80% of the customers opted for only beverages. 15%
opted for snacks and beverages between classes and about 5% opted for only food. This is a rounded
percentage.
Data distributions
To input the data into Arena, the data values have been fitted to the best fit distributions using the Input
Analyzer tool. To do this, the raw data values (all in minutes) of interarrival times and service times have
been noted in separate .txt files. These files were analyzed in the Input Analyzer, which gave the best fit
distributions for the data. These distributions were then used directly in the Arena system.
Fitted data of inter-arrival times of customers during peak hours (when data was collected): -

Taking 1/1.49 as the rate of arrival of customers during peak hours, approximately 40 customers arrive
per hour. On confirmation with the manager, this figure tallies with their average number of customers
during peak hour as well. During slack hours, approximately 8 customers per hour have been taken into
consideration. This figure was given by the Manager. Some hours have no customers, and some have
more than 20, hence the slack hours balance out by taking 8 customers per hour during a period of one
day. Based on this data, the customer arrival schedule was formed.

Fitted data for Beverage process service time is as follows:
Beverage Service time has been obtained for 42 customers and it follows an exponential distribution
with a mean of 1.07 minutes

Fitted data for food service time is as follows: -
Food service time was collected for 14 customers and this data follows a beta distribution.

Fitted data for the time taken to order is as follows: -
The order time was recorded for 61 customers and this data follows a beta distribution.
Modelling in Arena
The general process that goes on at Starbucks is as follows:
1. The customer enters the store
2. The customer waits in Queue to order
3. The cashier takes the order and the customer pays for his order
4. The order is passed onto the Baristas who divide the work amongst themselves
5. Meanwhile the customer waits in Queue to receive his full order
6. Once the customer collects beverages and drinks, he/she leaves the system
The model consists of various modules, queues, paths and parts. The basic parameters in the system
are: -

Parameter Count Type Action
Customer 1 Entity Part
Cashier 1 Fixed Resource Seize Delay Release
Barista 6 Scheduled Resource Seize Delay Release
Oven 1 Fixed Resource Seize Delay Release
Beverage Machine 2 Fixed Resource Seize Delay Release
Self Service Table 1 Fixed Resource No action
Order 1 Attribute
3 values, belongs to Entity -
Customer
Order Queue 1 Queue FIFO
Beverage Queue 1 Queue FIFO
Food Queue 1 Queue FIFO
Customer Time 1 Variable/Output Time Interval
There is a total of 7 resources that work at Starbucks in a day. One is the cashier and the others are
Baristas. There is one Oven, 2 Beverage machines for both hot and cold beverages. An attribute called
order is defined on every customer. This has 3 values, 1 – only beverage, 2 – beverage and food, 3 - only
food. There are three processes, ordering food, making food, making beverages. Each of these have
their own queues. Customer time in the system is also recorded to use for calculations.
The following is a screenshot of the model: -
A screenshot of the animation: -

Overview of model logic, parameters and steps
1) Run Setup
The simulation is done on a 24-hour clock, which starts at 12AM. 10 Days are replicated to
collect more data
2) Customer arrival
Customers enter through a create module. Customers arrive by the arrival schedule based on
the data collected.

Like in the figure, the arrival times are from 7AM to 8PM and have two peak times from 11PM -
2PM and one from 5.30PM-6.30PM
As soon as the customer arrives, his/her arrival time Is captured by accessing the inbuilt TNOW
function.
3) Order Queue

The customer waits in a queue, if there is any, to place an order.
4) Placing the order
The time taken to place an order and pay is given by the expression derived from the input
analyzer. The resource Cashier is capture by the customer and the cashier is released once the
order has been placed. To achieve this a process module with a ‘Seize Delay Release’ action has
been used as shown: -
To capture the orders of customers, an attribute called ‘Order’ is assigned to each customer.
This is created using an attribute module. Orders of type 1, 2 or 3 are assigned based on
probabilities using the inbuilt DISC() function as follows: -

5) Order path
Based on the order assigned, the process for that customer goes in 3 possible paths. In order to
create this, an N-Way by condition decision module has been used as follows: -
6) Path 1 – Only Beverages
When the order has value = 1, the control goes to the beverage making process which is a
process module with the service time defined by the equation derived from the input analyzer.
It works as follows: -

This process captures one Beverage Machine and one Barista from the available set of Baristas.
7) Path 2 – Food only process
When the order has value = 3, the control goes to the food making process which is a process
module with the service time defined by the equation derived from the input analyzer. It works
as follows: -

This process captures one Oven to heat the cold food and one Barista from the available set of
Baristas.
8) Path 3 – Both food and Beverage
When the order has value = 2, the control goes to the separate module. Here a duplicate of the
customer is created. The original goes to the beverage making process while the duplicate goes
to the food making process. Once both are done, the Batch module unites the two into one
based on their Serial Number. Every customer has a unique serial number.

The separate module is called ‘Parallel Processing’, creates one duplicate of the customer entity.
The Batch module unites these two entities after the last one finished the process on Serial
Number
To duplicate the processes, two Seize and two Delay Release modules have been used. This has
been done to add the respective queues to the original beverage and food process queues. The
resources are shared.

A Simulation Model of Starbucks

In this manner, the processes have been duplicated and parallel processed to simulate the
scenario that takes place at Starbucks.
9) Customer total time –
In order to calculate the total time the customer spent in the system, a record module is used
which calculates the time interval between arrival time which was recorded when the customer
arrived and TNOW () which is the time at which the customer is leaving the system.
10) Customer Type –
Order types are stored using a record module, which writes into a set for the total number of
orders of each type. This was achieved in the following manner: -

11) Customer Exit -
Customer exits the system once the purchases are collected and exits using a dispose module in
the following manner: -
Resources –
1. The following resources were created in the system –

2. The Baristas Schedule –
The baristas follow the following schedule: -
7AM 8AM 9AM 10AM 11AM 12PM 1PM 2PM 3PM 4PM 5PM 6PM 7PM
Resource
Schedule
Three schedules were created to achieve this. Morning Barista Schedule, Afternoon Barista Schedule
and Evening Barista Schedule. Two Baristas work during each shift; hence the capacity is given as 2.
Outside the shift, the capacity is given as 0.

Three resources were created, Morning Barista, Afternoon Barista and Evening Barista. These resources
were given their respective schedules. These three resources have been entered into a Set, from which
processes pick up baristas based on ‘Smallest number busy’.
Queues –
The following process queues were created in the system: -

Results and Interpretation
Upon simulating the model, the following results were seen in the category overview window of Arena: -
An inbuilt KPI called number out records the number of customers who left the system. For this model,
the number of customers to successfully leave the system for a 10-day simulation are 3004 customers.
This is all customers served. This is because the system considers that if a customer comes in the last
minute, he will be served by the resources before they close, and the resources come back the next day
on time. Since we are considering a 24-hour day, all the customers will be served. The customer entity
outputs are as follows: -
Entity Outputs
Customer WIP gives the number of customers in the system at a given time.

Average wait time for a customer is 0.7346 minutes and average service time is 2.1553 minutes.
Therefore, the total average time of customer spent in the system is 2.4632 minutes. What needs to be
reduced here is the maximum wait time which is coming to be 34.8487 minutes.
Queue Times
The maximum wait time is observed when customers place an order of both food and beverage. It is on
average 0.9489 minutes, goes up to a maximum of 16.3970 minutes. This can be optimized further.
Resource Utilizations:
To view how much the resources have been used, the output displays a bar chart as follows: -

The resources can be utilized much more. The morning baristas are used the most since the peak hours
are longer during that time. The least used is the oven which is expected since the number of food
orders are lesser.
Customer total time in system –
The average total time in the system is about 2.4632 minutes. Which is a decently good time. However,
the maximum time is 19.7377 minutes. This can be optimized further.
Statistics on the types of orders: -

As defined, most number of orders are Beverages, about 1900.
Improvements
The maximum total time and maximum wait time in the system are high and can be reduced. The total
time in the system can be reduced as well, to make this more efficient. As the resources are fully utilized
only during the peak hours, the number of baristas could be reduced during slack hours to save money.
However, Starbucks baristas are paid at $8 per hour, which does not cause much difference in company
expenditure by reducing employees.
To determine which resources can be increased to decrease total time and maximum total time in the
system, the process analyzer was used. The process analyzer provides the comparison of statistics across
scenarios.
Comparing scenarios to reduce maximum total time in the system: -
From the figure, adding a beverage machine would reduce the maximum total time in the system to
11.339 from 12.794 minutes.
Comparing scenarios to reduce average total time in the system: -

Adding a cashier could decrease the average total time of customer in system to 2.224 from 2.463
minutes.
Created two models, one with an additional cashier and one with an additional beverage machine.
On comparing the outputs for the new models with the original model, the following outputs were seen.
This was achieved using output analyzer, which performs a hypothesis test to determine if means are
equal and uses a significance level of 0.05
On adding a cashier: -

The average total time of the customer decreased to 2.2242 from 2.4632 and the maximum total time
also decreased to 16.0696 minutes from 19.3777 minutes. It can also be seen that the Food Average
time does not change as the null hypothesis cannot be rejected.
On adding a Beverage Machine: -

Average total time of the customer decreased to 2.4421 from 2.4632 and the maximum total time
reduced to 13.3509 from 19.3777 minutes. The null hypothesis for all three average process times is
significantly different for this model.
Thus far, the best model was obtained on adding a beverage machine to the existing model. The half
width gives the 95% confidence interval for all values in Arena.
Conclusion
The average time spent by a customer at the Starbucks at the Steger Student Life Center is much less
than five minutes. However, during rush hours that duration increases with customers waiting in
queues. To optimize the process and reduce the customer wait and total time, an extra beverage
machine can be added. There are many changes that can be done and in combinations to find the
absolute best model. However, for this project only two such models have been compared. Another
suggestion would be to pre-heat food before rush hours, to save time on heating in the oven.
References: -
Information on Arena - Simulation wit Arena 6/e – W. David Kelton – University of Cincinnati, Randall P.
Sadowski, Nancy B. Zupick, Rockwell Automation
Data –Starbuck Steger Student Life Center, University of Ohio
Image - https://www.usmagazine.com/celebrity-news/news/starbucks-debuts-13-new-red-holiday-
cups-for-2016-season-watch-w449657/

A Simulation Model of Starbucks

More Related Content

What's hot (20)

Similar to A Simulation Model of Starbucks (20)

Recently uploaded (20)

A Simulation Model of Starbucks