1 of 31

WP Carey Team & Cohort Building Process Optimization

-Lean Six Sigma

2 of 31

WP Carey School of Business

3 of 31

W. P. Carey Graduate School of Business Today…

15+ Programs

1000+ Students

Avg no of Cohorts per program: 4

​

4 of 31

From January 2018 till date, at the onset of each Fall and Spring semester, cohorts and student teams created by Academic supervisors take as long as 5 weeks resulting in a 10.4% reduction in employee utilization rate in one Academic year.

Problem Statement

Goal Statement

Reduce the time taken to create teams and cohorts by 70% (from 37 days to 11 days) thereby increasing the employee utilization rate by 8.32% (from 31 weeks in an academic year to 34 weeks)

​

5 of 31

ASU’s Academic Process Optimization

Business Problem:

From January 2018 till date, at the onset of each Fall and Spring semester, cohorts and student teams created by Academic supervisors take as long as 5 weeks resulting in a 10.4% reduction in employee utilization rate in one Academic year.

Executive Sponsor

N/A

​

Black Belt Champion

N/A

Objective and Scope

Reduce the time taken to create teams and cohorts by 70% (from 37 days to 11 days) thereby increasing the employee utilization rate by 8.32% (from 31 weeks in an academic year to 34 weeks / 65% utilization to ~ 75% utilization)

​

Scope:

  • Process Analysis, Data Analysis, Stakeholder Interview, Model Building and Process Optimization, Training and Guidance, Performance Metrics and Monitoring, Report and Recommendation

​

Team Members

  • Aniket Srivastava
  • Ganesh
  • Kshitij Waghdhare
  • Nana Amma Debrah-Apomah
  • Priyanka Sharma

​

​

​

Process Owner: Celetia Liang

Black Belt: Brett Duarte

​

​

Investment:

3 months

​

RISKS

Model Deterioration

​

DMAIC

Start / End

Operational Metric

Baseline

Target

Define

03/19 - 03/25

  • Team and Cohort Creation Time: The time taken from collecting data to creating teams and cohorts.
  • Defect: Whenever the creation time is more than 11 days.

​

37 Days

11 Days

Measure

04/10 - 04/14

Analyze

03/25 - 03/26

Expected Benefits

Projected Savings

Improve

03/26 - 04/28

Hard Benefits: Increase Employee Utilization rate by 8.3%

Soft Benefits: Reduced team and cohort creation time frees up the custodian to perform other pressing duties

Strategic Benefits: W.P. Carey School of Business will be known as the faculty supporting ASU’s innovation agenda

26 Days

Control

04/28 - 05/16

6 of 31

What does the data tell us?

7 of 31

Years of Work Experience

8 of 31

MSBA Population

9 of 31

Current Process Flow (with Pain Points)

1 Day

1 Week

3 Weeks

2 Days

1 Day

1 Day

1 Week

2 Days

1 Month

Cycle Time from Data Collection to Student Update

Wait Time

Rush Time

10 of 31

Root Cause Analysis…

11 of 31

5 Whys Analysis

12 of 31

How did we address the issue?

We built a Machine Learning Model to help

13 of 31

Process for FALL

14 of 31

Student Pool

Team 5

Team 1

Team 2

Team 3

Team 4

Cohort 1

Cohort 2

Team Formation

Cohort Formation

Number of Teams are initialised based on the team size as specified by the user

Number of students per Cohort is initialised based on cohort size as specified by the user. The teams are randomly grouped together to fill a cohort

Bottom-Up approach to create diverse Teams and Cohorts

15 of 31

Imagine this to be a team of 3 looking to add the next potential member

Dots → Current Team members

Cross → Potential Team members

In this iteration we add one student in the Cross to the team to increase the team diversity

In this step we calculate distance of each potential new member with respect to the each existing team member

Finally we pick the corresponding maximum distance pairs from each potential new member. Out of these pairs we pick the pair with the maximum distance ensuring diversity

E - B has the maximum distance compared to other pairs. Hence, E is chosen as the potential new member

Algorithm to enhance team diversity

16 of 31

Process for SPRING

17 of 31

Team 1

Team 2

Team 3

Team 4

Team 5

Team 6

Team 7

Team 8

Segregate 9/16

9 month Students

  • Same approach as fall but the student pool is broken down by track(9/16) and further by specialisation

Team Formation - 9 month … we use bottom up approach

18 of 31

Team 1

Team 2

Team 3

Team 4

Team 5

Team 6

Team 7

Team 8

16 month Students

Track 1

Track 2

Track 3

Team Formation - 16 month

​

19 of 31

Team 1

Team 2

Team 3

Team 4

Team 5

Team 6

Team 7

Team 8

COHORT ‘N’

  • Randomly select the Teams to create Cohorts of size ‘m’

Cohort Formation

20 of 31

What is left for the Academic Advisor?

Handling Leftover Students:

  • A small number of students may not be assigned to any team due to even distribution constraints. These students will require manual placement into teams of the user’s choice.

Ensuring Specialization Balance:

  • While the algorithm aims for even distribution, it may occasionally place students of different specializations together. It is necessary to review and adjust these teams to maintain balance within specializations.

Model Performance Evaluation:

  • The performance of the current diversity algorithm should be measured using the Silhouette score. A more negative score indicates greater diversity within the teams and cohorts, which is the desired outcome.

​

21 of 31

Process Map After Improvement

22 of 31

Updated Process Flow

1 Day

1 Week

2 Days

2 Days

1 Day

1 Day

2 Days

2 Days

9 Days

Cycle Time from Data Collection to Student Update

23 of 31

Projected Improvements

Team & Cohort Creation

Time: 11 days

70%

Employee Utilization

Rate: +8.3%

24 of 31

Process Flow Feasibility

25 of 31

To minimize the error margin and ensure that only essential columns are included in the model, we have created templates for input files with fixed columns.

​

Additionally, we developed a standard operating procedure (SOP) that can be easily used by both new and regular users.

Template

SOP

26 of 31

Risks And Future Work

27 of 31

Risks Associated with the Model

03

Over Reliance on Quantitative Measures

02

The model for example, treats every undergrad Major uniquely irrespective of their similarities in the study discipline

​

  • Hence its important to review the models output rather than pure reliance on the quantified measures.
  • These changes can be further implemented to tune the model

Insufficient Data

01

  • We used the median of the Official Work Experience using the median of the data to handle the model calculations
  • If the future data contains more missing data now the model performance might deteriorate.

Mitigation: Performance can be monitored using the Silhouette score

Model Deterioration

Decline in model performance over time Caused by changes in data patterns. Requires periodic updates to maintain accuracy.

​

Mitigation:

  • Implement regular model retraining.
  • Monitor model performance continuously.
  • Update data sources as needed.

28 of 31

Future Enhancements:

​

Refine Data Handling and Preprocessing

  • Data Quality Assurance
  • Feature Engineering
  • Handling Missing Values
    • imputation based on clustering
    • predictive models

​

​

Optimize Machine Learning Model

  • Model validation
  • Hyperparameter Tuning

​

Enhance Automation and Integration

  • Integration with Existing Systems
  • User Interface Improvements

NB: For every enhancement, handlers need to be retrained to ensure continuity

29 of 31

Demo

30 of 31

Meet The Team

Nana Debrah-Apomah

https://linktr.ee/nanadapm

Aniket Srivastava

https://linktr.ee/aniksri

Ganesh Apparaju

Priyanka Sharma

https://linktr.ee/SharmaPriyanka

Kshitij Waghdhare

https://linktr.ee/kshitij1210

https://tr.ee/I2xORXNhJj

31 of 31

Thank You!

The Floor Is Now Open For Questions!!!