When we refer to donor segmentation in the context of the Constituent Intelligence Hub
(aka CI Hub), we are referring to the segmentation modeling used to define what will
typically be our three primary segments of Top Segment Donors, Middle Segment Donors,
and Bottom Segment Donors.
During the implementation of a customer’s Constituent Intelligence Hub, a specialized
software tool is used to perform statistical donor segmentation analysis on annual giving
data. It uses multiple clustering algorithms to identify natural groupings in donor giving
patterns and determines the optimal segmentation thresholds for donor management and
fundraising strategy.
The clustering methods compared include K-Means Clustering options with Log and
Double Log transformations, Gaussian Mixture Model with Log and Double Log
transformations. We also test versions of the models above, whereby we exclude outlier
gifts on either end of the spectrum.
For example, very low single-donor annual giving totals (e.g., less than $50) are usually
relatively insignificant in terms of their impact on overall fundraising performance year over
year, so we do not want to skew fundraising behavior toward these donors. In fact, in many
cases these are things like event ticket purchases classified as gifts in the organization’s
CRM.
In addition, if an organization has one or a small number of very high outlier donors (like a
founding family or other atypical donor support we will want to exclude them. We would
also look to exclude extraordinary one-time gifts (i.e. “McKenzie Scott type gifts”).
Management and stewardship of those donors will typically look very different from those
of the other donors the advancement team focuses on, so excluding their giving behavior
makes sense.
Once we have run all models and variations, we select the one with the best statistical fit
based on the Silhouette Score, then use a geometric mean to close gaps between
segments to $0.01, so no donor's annual giving falls into a gap between segments.
IMPORTANT: These segments are designed to organize donors into segments that
advancement or development team can use to better manage outreach and donor
management. A segment (group) of donors who have behaved similarly in the past is likely
to behave similarly as a group in the future. These segments are in no way used for things
like donor recognition programs and donors should never know which of these segments
they are assigned to. Donor recognition segments are defined by additional criteria. This
type of segmentation analysis may be considered, but you would not use these exact
segment breakpoints to set thresholds for the various types of donor-facing recognition
programs (Diamond, Gold, Silver levels, for example). Those programs will often have more
than three core segments, and best practice is to orient thresholds around “round
numbers” rather than statistically determined, random-looking thresholds.