Starbucks Deployment Tool to Optimally Assign Employees

IND E 495 By: Carter Herron, Eric Kammers, Dimitri Sastropranoto, Mohammed Said, Yuchao Fan, Weihuan Fu Class of 2017

Project Introduction Resource Allocation Algorithm Validation Platform Model Verification

Problem Statement
Increase Starbucks store manager’s capability by creating a staffing Comparing Output of Algorithm and Simio Model
tool that level-loads tasks across a given amount of employees and Objective Objective
Create a model of a Starbucks store to test different assignment plays and Six different scenarios were run and tested for both the algorithm and
minimizes customer wait time. Use inter-arrival rates, average service times, and product distribution to optimally
verify/validate the output of the assignment algorithm. the Simio model
assign partners to roles in order to reduce time in system for a customer.
• Number of partners: 5-7 • Play format = {POS, BAR, FOOD, SUPP}
Our project will aim to develop an improved and more dynamic
*Simulation model is constructed based on actual store data including number of • Store number: 302 • TIS = customer time-in-systm
version of the current staffing tool by forming a new backend Process • Play format: {POS, BAR, FOOD, SUPP}
algorithm that can provide store specific, and time specific outputs. Pull Data from Starbucks Database arrivals per half an hour, items ordered per half an hour and each station’s average
working time. Comparison Table
• Store number
Example Question: • Time of Day (1/2 hour increments) Time Simio Play Algorithm Play Simio TIS Algorithm TIS
Manager: What tasks should my 10 Starbucks employees be • Product mix (Blended, Brew, Espresso, Warm Food, Ambient Food, Tea)
Simio Model 8:30 am { 2, 2, 1, 2} {3, 2, 0, 2 } 2 min 49 sec 2 min 28 sec
Objects, Stations, and Workers
assigned to during the 9:00am shift at the University Village • Order Channel 10:30 am {1, 2, 1, 3} {3, 2, 0, 2} 3 min 1 sec 2 min 37 sec
location so customer waiting time is minimized? • 7 working stations including 1 POS and 6 order prep stations (Espresso,
12:30 pm {1, 2, 1, 2} {2, 2, 0, 2} 2 min 56 sec 2 min 22 sec
Brew, Oven, Tea, Blended, and GrabAndGo)
• 4 types of workers: POS, BAR, FOOD, CUST SUPP 2:30 pm {1, 2, 1, 2} {3, 2, 0, 1} 2 min 58 sec 2 min 8 sec
Product Mix for Store 111 @ 7:30 AM
50
4:30 pm {1, 2, 1, 1} {2, 2, 0, 1} 3 min 00 sec 2 min 14 sec
45

5:30 pm {1, 2, 1, 1} {2, 2, 0, 1} 3 min 15 sec 2 min 28 sec

Current State vs. Opportunity
40

35

30

25

Current State Opportunity 20

15

Keeping inputs constant, it Treat each store uniquely and 10

5
produces the same output for make allocations based on store 0
TOT_BLENDED TOT_BREW TOT_ESPRESSO TOT_FOOD TOT_TEA
every store specific historical data

Allocate workers by comparing

Algorithm Flow Chart
Allocates workers using an excel the average customer time-in-
based V Lookup Table system between allocations and Allocate In-store Verification
Allocate
level-loading utilization Remaining No Assumptions Went in and observed store 302 (U village) from 10:30am - 11:00am
Assign Partners until
Partners Until TIS Print • Travel distances and speeds are assumed as 1 meter between each on a Thursday
Inputs Inputs Tasks To each Role’s
Minimum Less? output station and 1.4 meters per second. • Observed customer TIS: 5 min 24 sec; {1, 1, 1, 2}
•Ask for an estimate of the MOP •Ask for a time range to allocate Roles Utilization is
System Time is
transactions per ½ hour based off of historical data Less than 100% • No defective products and rework process. • Simio model output: 4 min 40 sec; {1, 1, 1, 2}
Achieved
•Does not ask for number of •Ask for an estimate of how • The workers at oven stations are allowed to help with other stations • Algorithm output: 4 min 1 sec; {1, 2, 0, 2}
Yes
employees available at that time many employees to allocate while the food is warming.
Notes about algorithm • The processing time for Grab-And-Go station is zero.
• Service times for stations weighted off of inter-arrival rates to that station • Exclude outliers such as extreme huge or small quantity orders.
Output
•“What’s possible” - # of
Output
•Expected time in system for
Data
• Same data as assignment algorithm
Results
transactions per ½ hour customer and partner utilization • Time in system is calculated from the initialized assignment vector with 1 Model Logic
•Primary and secondary •Visual aids • OptQuest Plugin
Findings
partner on each role
responsibilities •Minimum # of partners so as to • Experiments are conducted based on any possible combination for ƒ Changing the number of workers
• Time in system has to be at least two seconds lower in order to exit loop
•Service Standards not be understaffed number of workers at each station within certain range. has a significant affect on TIS
• Roles: POS, BAR, SUPP1, SUPP2
• After running all scenarios, the best can be chosen which has the lowest • From 9 to 7 partners the
• Tasks: Process Order, Tea, Brew, Espresso, Blended, Ambient Food, Warm
value of time in the system. customer TIS will increase by
Food
• The utilization of each type of worker is calculated by its busy time 110%
Current Playbook Input Display
divided by the total time running the model. ƒ Current tool output vs. new tool
Example of Output (8 person play @ 10:30 on a Sunday)
Example of Output output
Role: POS Role: SUPP1 • Example: Current TIS = 4.12
Tasks Assigned: Process Order, Tea Tasks Assigned: Warmed Food, Blended min; New TIS = 3.05 min
Partners Assigned: 2 Partners Assigned: 2 • Customer TIS reduced by 26%
Role: BAR Role: SUPP2
Tasks Assigned: Espresso Tasks Assigned: Brew, Ambient Food
Partners Assigned: 3 Partners Assigned: 1 ƒ Changing the number of
workers has significant affect on
Average Time in System: 5 Minutes 29 Seconds
POS role
Average Partner Utilization: 82%
• From 2 to 1 POS partner,
utilization increases 74%
ƒ Target utilization value ranges
from 70% to 80%
• BAR role typically

Concept Design of Deployment Tool

experiences high utilization
due to high product flow

1 2 3 4
Impact on Starbucks
• Developed the backend of a staffing tool that level-loads tasks
across partners and minimizes average time-in-system
• Dynamic: treats every store uniquely
• Effective: Outputs time-in-system and utilization values to
help with manager’s decision making process
• Economic Impact of Current Tool vs. New Tool (6 employee play)
• Current tool TIS = 4.12 min; New Tool TIS = 3.05 min
• Assuming avg. customer transaction = $5.50
Allocation • Inter-arrival rate: 49 customers per ½ hour
Login Input Data Graphs • Potential to increase sales by approx. $440 per day