Top Banner
April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis
26
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

April 29 - May 1, 2015

Better Donor Engagement Through Cluster Analysis

Page 2: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Agenda

• Introduction/Challenges

• What is the problem we are trying to solve?

• Our Starting Point

• Our Approach

• Alliances

• Data

• Technique

• Our Learnings

• Next Steps

• Recommendations

Page 3: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

The United Way of Greater Cincinnati

• Serve communities in 10 counties in Southwest Ohio, Northern Kentucky and Southeast Indiana

• Over 100,000 donors, advocates and volunteers

• Annually over $50.3M invested in the community

Page 4: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Challenges

• Flat Annual Workplace Campaign

• Aging donor base

Opportunity: Better Engagement With Our Constituents

Page 5: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Our Starting Point

• Lots of data – little actionable information

• Minimal data scrubbing expertise

• Minimal statistical expertise

• Minimal tool set for analytics or visualization

• Existing Workplace Analyses based on macro level metrics

- % Raised over Goal

- 3 – 5 year giving trends

- $/donor

- % Participation

What Is Driving These Results?

Page 6: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Desired Engagement Approach

• Understand underlying constituent behavior

- Provides enlightenment on why the macro level results occur

• Interact with constituents by targeting unique behavioral characteristics

- Improves

Donations

Advocacy

Volunteerism

Page 7: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Our Approach

• Find a partner with expertise willing to teach us

• University of Cincinnati – Department of Operations, Business Analytics and Information Systems

• ‘Sponsored’ 2 Graduate Level Students – Master Thesis

• Began Fall 2013

Problem statement: To use Descriptive Analytics for UWGC’s Individual Constituents to describe our current state numerically and visually, further develop segmentation, correlate variables, and build the basis for predictive experiments with a focus on Long-term Retention

Page 8: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Our Data

• Individual level data for 2006 – 2012

- No Personally Identifiable Information (PII)

- Demographic

- Pledge

- Volunteer Activity

- Recognitions

- Affinity Groups

- Event Attendance

• Frequent Meeting with students and professors

• Lots of data scrubbing!

Page 9: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Donor Cluster Analysis

Used a Hierarchical Clustering Technique

• Based on a set of independent variables

• Does not force the user to specify the number of resulting clusters up front

• Run the clustering model, look at the results, refine parameters, repeat until results make practical sense

Page 10: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Cluster Analysis

Agglomerative Hierarchical clustering algorithm starts with each point as a cluster and recursively joins together nearest clusters based on the least distance measure until there is only one cluster.

We divide the resultant tree formed by this recursive agglomeration based on statistical measures and look for homogenous clusters and their properties.

Dataset Clustering Output

Page 11: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Initial Variable Creation

• Began with a data pool of 14 variables

• Examples:

• Acquisition Rate

• Volunteer Participation Rate

• Average Contribution

• Average Event Attendance

Page 12: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Initial Correlation Analysis

CORRELATION churn_rate acquisition_rate

vol_rate aff_rate reco_rate influencers avg_contrib avg_pledge avg_freq avg_evrgd avg_evses avg_evatt avg_evppl avg_evnumregd

churn_rate 1 0.668 -0.048 0.325 -0.155 -0.063 -0.08 -0.08 0.15 -0.131 -0.11 -0.195 -0.141 -0.107

acquisition_rate 0.668 1 -0.037 0.581 -0.149 -0.059 -0.11 -0.11 0.028 -0.127 -0.107 -0.184 -0.133 -0.095

vol_rate -0.048 -0.037 1 0.04 0.288 0.38 0.16 0.16 0.085 0.439 0.429 0.517 0.441 0.363

aff_rate 0.325 0.581 0.04 1 0.033 0.068 -0.04 -0.04 -0.082 0.072 0.07 0.064 0.073 0.036

reco_rate -0.155 -0.149 0.288 0.033 1 0.203 0.55 0.55 0.133 0.619 0.606 0.657 0.6 0.32

influencers -0.063 -0.059 0.38 0.068 0.203 1 0.09 0.09 0.077 0.271 0.271 0.296 0.282 0.182

avg_contrib -0.078 -0.111 0.157 -0.035 0.553 0.09 1 1 0.117 0.278 0.274 0.295 0.251 0.134

avg_pledge -0.078 -0.111 0.157 -0.035 0.553 0.09 1 1 0.117 0.278 0.274 0.295 0.251 0.134

avg_freq 0.15 0.028 0.085 -0.082 0.133 0.077 0.12 0.12 1 0.082 0.074 0.102 0.08 0.048

avg_evrgd -0.131 -0.127 0.439 0.072 0.619 0.271 0.28 0.28 0.082 1 0.995 0.883 0.988 0.564

avg_evses -0.11 -0.107 0.429 0.07 0.606 0.271 0.27 0.27 0.074 0.995 1 0.852 0.981 0.552

avg_evatt -0.195 -0.184 0.517 0.064 0.657 0.296 0.29 0.29 0.102 0.883 0.852 1 0.861 0.553

avg_evppl -0.141 -0.133 0.441 0.073 0.6 0.282 0.25 0.25 0.08 0.988 0.981 0.861 1 0.57

avg_evnumregd -0.107 -0.095 0.363 0.036 0.32 0.182 0.13 0.13 0.048 0.564 0.552 0.553 0.57 1

Page 13: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Final Correlation Analysis

CORRELATION churn_rate

vol_rate aff_rate influencers

avg_contrib

avg_freq

churn_rate 1 -0.048 0.325 -0.063 -0.08 0.15

vol_rate -0.048 1 0.04 0.38 0.16 0.085

aff_rate 0.325 0.04 1 0.068 -0.04 -0.082

influencers -0.063 0.38 0.068 1 0.09 0.077

avg_contrib -0.078 0.157 -0.035 0.09 1 0.117

avg_freq 0.15 0.085 -0.082 0.077 0.12 1

After variable reduction to reduce multi-collinearity 6 variables remain

Page 14: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Final Variables

• Correlation analysis determined there were 4 independent variables that could be used in the clustering model

- Volunteer Participation

- Churn Rate

- Influencers (Active Contributors Who Registered for 10 or More Events)

- Average Contribution

Page 15: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Refinement Of The Cluster Analysis

• UC Students provided their R code to us used to perform variable correlation and cluster analysis

- R is an open source statistical programming language

- Analogous to SAS or SPSS

• We modified the data input to include only individuals from our top 200 accounts

- Accounts that our Resource Development Professionals focus on

• Reran the clustering analysis process

Page 16: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Initial Cluster Analysis Output

Page 17: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Number of Clusters - Visual Approach

Page 18: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Final Cluster Analysis Output

Page 19: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Cluster Analysis – Numerical Output

Cluster Avg Churn Rate Avg Volunteer Rate Average Influencers Avg Contribution # of Companies1 0.44 0.02 0.01 321.47$ 252 0.24 0.01 0.12 326.83$ 553 0.09 0.02 0.06 439.47$ 634 0.24 0.19 0.29 713.70$ 105 0.16 0.03 2.96 724.35$ 126 0.11 0.02 0.27 1,471.71$ 307 0.14 0.21 1.17 4,647.90$ 4

Average 0.20 0.07 0.70 1,235.06$ 199

Page 20: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Final Cluster Analysis – Informing Strategy

Final Output Generated 7 Clusters – 3 of Which Had Low Average Contribution Rates

•25 Companies With Very High Churn Rate, Minimal Influence And Lowest Average Contribution

•55 Companies Low Average Contribution, Average Churn, Minimal Influencers and Low Volunteerism

•63 Companies With Low Churn Rate, Minimal Influencers, Low Volunteerism and Low Average Contribution

Page 21: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Final Cluster Analysis – Informing Strategy

Characteristics of the Other 4 Clusters

•10 Companies High Volunteer, Average Churn and Average Contribution

•12 Companies With Strong Mix of Influencers and Average Contribution (Largest Overall Workplace Campaigns)

•30 Companies With Low Volunteer Rate and Very Low Churn Rate and High Average Contribution

•4 Companies with Low Churn, Highest Volunteer Rate, Strong Influencers and Highest Average Contribution

Page 22: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

How UWGC Is Using The Results

• Now we have a characterization of individual behaviors at our Top 200 Accounts

• Using that characterization to formulate account specific engagement plans for 2015 campaign

- Capitalize on strengths

- Address areas of opportunities

• Assess results after the 2015 campaign

Page 23: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Concurrent Activities

• Tool selection

- R for analytics

- Tableau for visualization

• Training

• Local Meetups

- Business Intelligence

- Data Analytics

- R

Page 24: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Recommendations For You

• Unless you have strong statistical modeling, data analytics or business intelligence capabilities in house

- Corporate alliance

- Academic alliance

- Other United Ways (Contact Me)

• Tools – Choose Wisely

- Strongly consider tools used by your alliance

- What are local companies using?

• Training

- Meetups (www.meetup.com)

- On-line courses (Coursera www.coursera.org)

- Swirl (swirlstats.com)

Page 25: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.

Contact Information

[email protected]

513-762-7102

Page 26: April 29 - May 1, 2015 Better Donor Engagement Through Cluster Analysis.