Post on 06-Aug-2015
Special Thanks To: Master of Science in Marketing Analysis (UGent) students: Ruben Billiet Sam Beelprez Yasir Ekinci Koen Gommers Jingbo Wei Brian Weidenbaum Assistants: Jeroen D’Haen & Andrey Volkov www.mma.UGent.be
LOYALTY CAN BE PREDICTED! PROF. DR. DIRK VAN DEN POEL
event Jan. 29, 2013
• Brand loyalty is a consumer's preference for a particular brand and a commitment to repeatedly purchase that brand in the face of other choices (Dick & Basu, 1994)
• Behavioral loyalty versus Attitudinal loyalty (Jacoby & Chestnut, 1978) • E.g. monopoly
A non-profit organization that commits itself to the preservation of nature and biodiversity in Flanders
www.natuurpunt.be
• Business objectives • Decrease number of members stopping their membership • More efficient and effective marketing
• Project goals • A better understanding of the data • Create a churn model
• Predict if member will stop in the next year • Provide insights and recommendations to stop customer attrition
• Did member stop membership in past 11 years? • How long did they stay member before they stopped?
• Analyze categories that impact relationship
• Long-term perspective
• Does involvement have an effect on how long a membership lasts?
• How does having or not having a standing order affect memberships in the long term?
• Are some recruitment types more effective than others?
• Does the province where you live have an effect on how long you stay a member?
10.33%
19.95%
9.28% 7.60% 7.07% 5.99% 5.68% 5.61% 5.61% 6.85%
100% 90%
74% 67%
62% 58% 55% 52% 49% 46%
43%
0%
20%
40%
60%
80%
100%
0 1 2 3 4 5 6 7 8 9 10
Surv
ival
Pro
babi
lity
/ Haz
ard
rate
Length of Membership Relationship (in years)
Life-Table Survival Curve
Hazard Function Estimate at Midpoint Survival Distribution Function Estimate
Total # of members: 99.744 Based on data after 2001
• Does involvement have an effect on how long a membership lasts?
• How does having or not having a standing order affect memberships in the long term?
• Are some recruitment types more effective than others?
• Does the province where you live have an effect on how long you stay a member?
Based on data after 2001
Member + donor N = 8.789 Member + volunteer N = 698 Member + volunteer + donor N = 357 Member only = 89.900
• Does involvement have an effect on how long a membership lasts?
• How does having or not having a standing order affect memberships in the long term?
• Are some recruitment types more effective than others?
• Does the province where you live have an effect on how long you stay a member?
Based on data after 2001
100.0%
42.8%
31.9% 24.9%
20.0%
11.5% 9.2% 7.6% 6.9% 5.7% 5.7%
0%
20%
40%
60%
80%
100%
0 1 2 3 4 5 6 7 8 9 10
Survival Probability
Length of Membership RelaEonship aFer stopped standing order (in years)
Survival Distribu-on Func-on Es-mate
Average duration after stopped order 1 year 6 months 11 days
• Does involvement have an effect on how long a membership lasts?
• How does having or not having a standing order affect memberships in the long term?
• Are some recruitment types more effective than others?
• Does the province where you live have an effect on how long you stay a member?
• Does involvement have an effect on how long a membership lasts?
• How does having or not having a standing order affect memberships in the long term?
• Are some recruitment types more effective than others?
• Does the province where you live have an effect on how long you stay a member?
Total # of contacts: 211.731
29.96%
22.56%
18.29%
16.73%
9.32%
1.36% 1.19%
0.58%
Contacts by province
Antwerp
East-Flanders
Flemish Brabant
West-Flanders
Limburg
Wallonia
Others
Brussels
• Predict if a member stops membership within next 12 months
• Gives score between 0 (safe) and 1 (at risk) to each member • Based on historical data • Indicates likelihood of stopping membership
• A search engine for members who are in danger of stopping
INDEPENDENT GAP DEPENDENT
31DEC01 31MAR11 01APR11 01APR12
Model Building
INDEPENDENT GAP DEPENDENT
31DEC01 31MAR12 01APR12 01APR13
Purpose Model
1. Create analysis basetable • Get statistics on members • Create new predictive models
2. Give “Churn Score” to every active member • Churn probability between 0 (safe) and 1 (at risk) • Find members in danger of stopping membership
• Encourage good behavior • Standing orders • Subscriptions
• Encourage involvement • Get more people on the website • Promote donating and volunteering
• Focus on successful campaigns • “Member get member” actions
• Historical data: Missing dates
• More reliable gender data
• Consistent data input • Recruitment type categories
• Fix the missing link with website users • Automatic account creation
• Investigate reasons of stopping • Exit questions • Why do members stop standing orders?
• Volunteering data • Start & end dates • Reason of volunteering
• Gather birthdates (online form) • Membership card + scanners
• Member visits to parks • Natuurpunt shop purchases
• Qualitative Market Research on churn impact factors
• Churn-prevention efforts • Target top x% • Mailings, discount offers, …
• Traffic lights system • Color codes for probability of stopping • More actions for members with red light • Evolution of probability
We are able to predict customer churn/loyalty Predictive Analytics makes churn prevention actionable
• VERHAERT G. & VAN DEN POEL D. (2012), The role of seed money and threshold size in optimizing fundraising campaigns: Past behavior matters!, Expert Systems with Application, 39 (18), 13075-13084.
• VERHAERT G. & VAN DEN POEL D. (2011), Empathy as Added Value in Predicting Donation Behavior, Journal of Business Research, 64 (12), 1288-1295.
• VERHAERT G. & VAN DEN POEL D. (2011), Improving campaign success rate by tailoring donation requests along the donor lifecycle, Journal of Interactive Marketing, 25 (1), 51-63.
• JONKER J.J., PIERSMA N. & VAN DEN POEL D. (2004), Joint Optimization of Customer Segmentation and Marketing Policy to Maximize Long-Term Profitability, Expert Systems with Applications, 27 (2), 159-168.
• A 30' video featuring Dr. Griet A. Verhaert is available on this page.