Predictive Modeling. RFM. Personas. These are all terms you’ve likely heard before in marketing. These are all processes that organizations use to select who they’re going to contact in any given marketing campaign. In other words, they are all part ofsegmentation strategy.
RFM(or, recency, frequency, and monetary) selection has been around for a long time. It buckets donors into rough groups based on their last gift date, total number of gifts and either last or largest gift amount.
Personasare usually developed based on demographic, psychographic, and behavioral commonalities, and they can be used layered on top of other segmentation (such as RFM) or as a standalone selection tool. They provide a fictional representation of an idealized supporter, such as “White Collar Walter” or “Mid-Level Mindy.” They are often generated through unsupervised learning to group supporters in novel ways. Personas can be specific to an organization, or generic. Personas can be used to target more elusive groups of supporters or potential supporters.
Predictive Modelingis a finely honed combination of science (machine learning and artificial intelligence) plus art (knowledge of client’s program, goals, and history). Predictive modeling combines demographic, psychographic, and behavioral data at the individual level, using it to predict future performance. Models can predict the likelihood of response or propensity to convert (response modeling), or average gift amount, or revenue per donor contacted (value modeling).
At One & All we’ve used modeling to support our client initiatives for over a decade. OurModeling via Purpose(MVP) uses an organization’s CRM data—plus other outside data, as available—to identify the ideal constituents to market. All our models are customized for your organization. We don’t believe in a one-size-fits all approach to modeling. Our customized approach further increases results by 3 to 5% over a generic model. While each model is different, some of the techniques we use are random forest, gradient boosting, logistic regression, and SVM.
OurModeling via Purposecan be used to segment:
And much more!
And, the results speak for themselves. A few highlights from our Modeling via Purpose:
4-8% increase in net revenuewhen compared with RFM or other rough segmentation techniques
65% increase in sustainer conversionwhen compared with the control group
54% increase in annualized donor valuefor sustainers when compared with the control group
Predictive modeling works best when there’s a clear purpose (e. g. identify sustainers, or increase net revenue per donor mailed), appetite for testing and refining, and organizational buy-in. If you’re looking for a partner to help build a customized supporter selection model,let’s connect.