Donor Retention in Online Crowdfunding Communities A CASE STUDY OF DONORSC HOOSE.ORG Tim Althoff and Jure Leskovec
2 Crowdfunding Revolutionary way to collect funding Crowd collectively funds projects through many small donations Projects rely on contributions from large number of individuals Similar to non-profit organizations & charities
What factors predict donor return? 3
Donor Retention Donors making donations year after year Significant impact on effectiveness of fundraising campaigns More cost-effective to maintain existing donor relationships than to recruit new donors 10% improvement in retention may yield up to 200% more donations (Sargeant 2008) Present retention rates very low (~25%) Challenge for traditional non-profit organizations and online crowdfunding 4
What do we know? Current knowledge (Sargeant 2008) Anecdotal evidence from professionals Small lab experiments Example anecdotes Regularly [ ] show your donors loyalty that you care beyond just making the ask. Thank donors well and promptly How to quantify these factors? Are donor subgroups affected differently? 5
6 This Work Observational study of donor retention factors U.S. platform for public school teachers to request donations of classroom materials Used by >60% of U.S. public schools Data: full trace of donor and project activity (2000-2014) 3.9M donations by 1.5M donors to 638k projects for a total of $282M
7 Related Work Online crowdfunding project dynamics (Mollick 14) predicting project success (Greenberg et al. 14) completing donation bias (Wash 13) Offline charities importance of donor retention (Barber & Lewis 13) retention factors (Sargeant 08) Contributor retention newsgroups (Arguello et al. 06) forums (Lampe & Johnston 05) Q&A Sites (Yang et al. 10) Wikipedia (Halfaker et al. 12)
8
9 Donor Retention in DC.org Fraction of donors 1.00 0.10 0.01 1 2 3 4 5 Number of donations by donor
What factors predict donor return? 10
Project Success = Retention? Project success = 100% funded Trust known driver of loyalty (Sargeant, 2008) 0.30 Donor return 0.25 0.20 Donor teacher referred site donor teacher referred 0.15 No Yes First project successful 11
Four Questions about Donor How Where When How Much do they donate? 12
Tracking How Donors Joined Donor return 0.4 0.3 0.2 0.1 Number of donors 20000 40000 60000 Donor teacher referred site donor teacher referred [0,1) [1,2) [2,3) [3,4) [4,5) [5,10) [10,20) [20,50) [50,100) #Future projects by teacher à Possible explanation: correlated with personal motivation and involvement [100,200) [200,500) 13
14 Four Questions about Donor Site donors more loyal than teacher-referred donors Where When How Much
15 Distance Involvement? Distance between classroom and donor 0.50 Donor return 0.45 Donor teacher referred site donor teacher referred 0.40 10 25 50 100 200 500 1000 2000 5000 Distance between donor and school in km
16 Four Questions about Donor Teacher-referred donors less likely to return Local donors are loyal donors When How Much
17 Donor s Role within Project Donors assume different roles based on when they donate to project Starters: Make initial donation Closers: Finish off projects close to completion
18 0.2 0.3 0.4 0.5 2 4 6 8 N th donation to project (successful) Donor return Project size in total number of donations 1 2 3 4 5 6 7 8 Donor s Role within Project
19 Four Questions about Donor Teacher-referred donors less likely to return Local donors are loyal donors Starters and Closers are particularly loyal How Much
Donation Amount Donors can support DC.org w 15% of donation Donor return 0.6 0.5 0.4 0.3 0.2 Number of donors 25000 50000 75000 100000 125000 Including optional DC.org support False True (0,1] (1,5] (5,10] (10,25] (25,50] (50,100] Donation amount (100,200] (200,500] (500,1000] 20
21 Four Questions about Donor Teacher-referred donors less likely to return Local donors are loyal donors Starters and Closers are particularly loyal Large donations demonstrate commitment
22 Teacher Perspective Acknowledge Support Communicate Impact
23 Donor return to same teacher Acknowledging support Effect of timeliness of thank you note on return to same teacher 0.16 0.12 0.08 0.04 Number of donors 10000 20000 30000 40000 50000 Donor teacher referred site donor teacher referred [0,1) [1,3) [3,6) [6,12) [12,24) [24,72) Teacher response time in hours after fully funded (confirmation letter) [72,168) [168,Inf)
24 Teacher Perspective Thank within first hours! Communicate Impact
25 Communicating Impact Effect of timeliness of teacher impact letter on return to same teacher Donor return to same teacher 0.20 0.15 0.10 0.05 Number of donors 10000 20000 30000 40000 50000 Donor teacher referred site donor teacher referred [0,7) [7,30) [30,60) [60,90) [90,120) [120,150) [150,180) [180,Inf) NA Teacher response time in days after fully funded (impact letter)
26 Teacher Perspective Thank within first hours! Teacher-ref. sensitive to communication
27 Can we predict donor return? Data: 470k first-time donors (26% return rate) Features: split into four groups Time Project Donor Teacher Model: Logistic Regression (SVM & RandomForest gave similar results) Metric: Area under receiver operating characteristic curve (ROC AUC) (10-fold CV)
28 Prediction Results 0.8 0.72 0.73 0.74 0.7 ROC AUC 0.6 0.5 0.5 0.53 0.54 0.55 0.56 0.4
29 Count How likely are donors to return? Calibrated Model? Yes. à Predicted Probability \Impact estimate: of Return Increasing retention by 10% on DC.org would lead to an over 60% increase in donations ($15M) 18000 16000 14000 12000 10000 8000 6000 4000 Logistic Regression (all features) 2000 0 0.0 0.2 0.4 0.6 0.8 1.0 Predicted Probability
Recommendations Platform Teacher Make sure first-timers experience success Recommend local projects to local givers Use donations by anywhere givers wisely Acknowledge support within a few hours Communicate impact within a month Real-world impact: DC.org now recommends smaller and morelikely-to-succeed projects to first-time donors DC.org is rethinking the teacher communication workflow 30
Conclusion Challenge for traditional NPO & crowdfunding Observational study of donor retention factors Predicted donor return Implications for online & offline fundraising Inform crowdfunding communities and NPOs Encourage them to start collecting information 31
32 Thank you! Data available at data.donorschoose.org Paper and slides available at cs.stanford.edu/~althoff Thanks to Vlad Dubovskiy and Thomas Vo at DonorsChoose.org for facilitating the research!
Backup Slides
Project Success (cont.) Potential confounders Successful projects are smaller Donations towards them are larger Almost exact pairwise matching On donation amount, project cost, etc. Still observe 19% difference 34
Project Cost Small project: greater sense of impact Donor return 0.30 0.25 0.20 Number of donations 25000 50000 75000 Project fully funded No Yes [0,200) [200,300) [300,400) [400,500) [500,600) [600,700) [700,800) [800,900) Project cost in $ [900,1000) [1000,1500) [1500,2000) 35
Teacher Perspective Donor relationship management is a skill Expect that teachers get better over time Donor return within one year 0.25 0.20 0.15 0 5 10 15 20 N th project by teacher (all successfull) Donor teacher referred site donor teacher referred Number of donors 2000 3000 4000 5000 36
Future Work Use of prediction models in fundraising campaigns Content analysis of essays, messages, photos Online field experiments to test causal hypotheses 37
Example Thank You Note 38
Example Impact letter 39