In the first two installments (scroll down to see previous posts) of this “trilogy” on segmentation in direct mail fundraising I discussed the basic objectives underlying segmentation, listed some of the challenges and limitations created by typical (often overly complicated) segmentation strategies, and suggested a few very basic principles for effective segmentation.
Now I would like to suggest an alternative, highly data-driven approach: segmentation by giving likelihood. The essence of predictive modeling for fundraising is to identify the constituents in your database with the highest statistical likelihood of giving to your organization, and this information can be used to create segments.
Let’s look again at what I called out in part one as being major objectives of annual giving/direct mail fundraising efforts:
Raise as much money as possible.
- Minimize solicitation costs.
- Analyze results and refine tactics accordingly.
Now let’s consider how using a custom predictive model to drive segmentation can help achieve improve results in these areas. Data-driven fundraising strategy primarily comes down to targeting efforts, time, resources, people, and dollars toward those prospects who are the most likely to give significantly to your organization.
Within an organization’s overall direct mail fundraising efforts, treating various segments differently can involve more than simply varying the solicitation content. For example, on the basis of likelihood segmentation an organization can reduce the number of mailings per year to low-likelihood constituents, reducing solicitation costs. At the same time the most likely prospects can be targeted more specifically by allocating more resources toward increasing the amount of personal contact they receive in order to increase their giving.
These are just starting points. As an organization becomes increasingly data driven, analytics can be used to inform decision making in other areas such as major giving/prospect management, planned giving, and even data management.
The model itself can be the basis of a powerful test effort, by simply sending a modest-sized general solicitation to equal numbers of constituents across your likelihood segments. You can then analyze giving by segment and will see markedly better results from your high likelihood segment, in terms of response rate, average gift size, and total dollars. A test of this sort can be a great confidence builder to help your organization get more comfortable using analytics to drive decisions such as those described above.
Beyond this simple test, following a likelihood-based segmentation strategy would make it easier to test, analyze, and refine regular solicitation efforts (again, testing variables such as content, channel, timing, etc.) since the segments themselves would be simple and stable. Using likelihood as your fundamental basis, you can certainly also segment further in ways that you plan to align with different messages, as long as doing so doesn’t lead you back into the over-complicated and unstable segments we started with.