Focus On: Database Analysis: A Closer Look at Models

The past few years have proven to be challenging for direct marketing fundraising. When you consider the 2001 terror attacks and the ensuing questions about dispersement of funds contributed as a result, Anthrax scares, the war on terrorism, corporate distrust, the recent Catholic Church opprobrium and the troubled economy, it’s no surprise that fundraisers have felt left out in the cold.

A closer look within the fundraising sector finds that the size of the American donor population is shrinking at an alarming rate, as well, with traditional donor segments being impacted the most. Donors cite “trust” (or lack thereof) as the primary reason for their change of heart, and confidence in nonprofits has fallen to an alarming low. As a result, many donors have scaled back the number of charities that they support.

As outside forces make it increasingly difficult to balance success benchmarks (i.e., response rate, average gift, etc.) with volume goals, fundraisers are recognizing that achieving gross revenue objectives at the expense of profitability is a losing battle. Past success has conditioned many of us to increase mail volume to reach our targeted goals, but today’s fundraising environment no longer supports such a practice.

What we need is something to help us better align marketing investments with potential donor value to maximize profitability. As Ginny Renehan, associate development director for the Zoological Society of San Diego, has experienced, “Adding analytic modeling … strategies has proven to be very effective both in cost savings and results.”

Letting go of the past
Nonprofits have begun to recognize that sophisticated analytics provide much of the intelligence necessary to drive targeted, donor-centric management. However, the vast majority of nonprofits continue to embrace the more familiar RFA (recency, frequency, gift amount) approach to donor segmentation.

RFA is borrowed from the catalog industry, which was credited with pioneering the use of RFM (recency, frequency, monetary value) prior to the introduction of the computer. Generally speaking, RFA is used to segment donors based on a few key components of past behavior (often direct mail behavior) to predict future activity. Although simple and often subjective, RFA is an effective segmentation model that combines several pieces of information to form relatively homogeneous groups of donors that can be ranked from best to worst.

Unlike predictive models that rank each donor relative to one another, RFA segmentation models rank groups of donors relative to other groups. No distinction is made between donors within each group. For instance, John Sample and Jane Doe share the same RFA ranking even though Mr. Sample upgrades every year, participates in special events and recently volunteered. By definition, RFA ignores a considerable amount of important information regarding a donor’s relationship with the organization. The resulting lack of precision can translate into thousands of misdirected dollars per direct mail campaign.

RFA has endured in large part because it works well at discerning the best and worst candidates for a mailing. As one client stated, “I spend 20 percent of my time selecting 80 percent of my audience and 80 percent of my time selecting the other 20 percent.” In reality, most fundraising files contain a substantial number of names that fall somewhere between best and worst. As marketers find themselves venturing deeper into their files to achieve gross revenue objectives, they are realizing that RFA becomes a progressively less effective tool for differentiating opportunities. More information about our donors leads to more informed marketing decisions, and nowhere is this more evident than within the not-so-obvious middle ground.

In the past, inherent shortcomings of RFA could be overlooked due to a lack of affordable alternatives, but technology and competition have lowered many of the barriers that have prevented fundraisers from updating their marketing tools.

Embracing the future
Sophisticated analytic techniques have been a staple in the commercial world for many years, and there’s a movement among fundraising organizations to adopt similar best practices. Yet, predictive modeling has not been widely accepted by the nonprofit community primarily because many fundraisers do not have a background in statistics and find it difficult to understand the underlying mathematics.

Although building an effective predictive model requires a trained statistician, the concept is simple:

• Determine your objective (i.e., increase renewal rates).

• Select responders and non-responders from one or more recent renewal campaigns.

• Using available data (behavioral data is sufficient, but third-party demographics can be used to enhance results), build a profile of characteristics that differentiate responders from non-responders.

This combination of characteristics forms an equation, and each variable in the equation is weighted based on its relative importance. The weighted equation is then used to score and rank each individual donor on your database.

Not surprisingly, the backbone of most renewal models includes some variation of recency, frequency and gift amount. However, leveraging a donor’s complete relationship with an organization provides a broader, more stable foundation on which to make objective marketing decisions. In addition to recency, frequency and gift amount, a predictive model is able to consider hundreds of other potential drivers of response and gift size, such as consistency of giving, lifetime value, upgrade/downgrade, preferred month of giving, breadth of support, involvement, tenure, direct mail responsiveness and demographics to create a more meaningful and actionable profile of each donor. Nonprofit organizations have been capturing data about their donors for years, but the real value of that data is not realized until it has been transformed into actionable information.

Modeling for renewal
Organizations that have adopted modeling techniques during these trying times have realized significant gains relative to their peers. As many organizations were mailing deeper to preserve gross revenue at the expense of their net, one of the nation’s premier health-related fundraising organizations was able to cut renewal mail volume 20 percent and increase gross and net revenue 12 percent with the use of predictive modeling. A logistic regression model was developed to predict a donor’s likelihood to respond, and a linear regression model was built to predict gift amount. The two models combined resulted in a score that measures expected value and served as the basis for determining the best targets for renewal.

The model solution enabled the organization to replace a substantial quantity of under-performing names with a more profitable audience that went previously undetected by traditional RFA. The new approach generated a 37 percent increase in average gift and a 19 percent reduction in cost per dollar raised (CPDR). The money saved by mailing fewer names funded two additional mailings during the year that net more than $2 million additional dollars.

Modeling for reactivation
Some seasoned nonprofit veterans use a hybrid approach of RFA, predictive modeling and marketing savvy to move the needle. Another nationally recognized fundraising organization recently turned to predictive modeling to enhance its selection strategy and help offset faltering acquisition results by mining its growing universe of lapsed donors more effectively. Its standard RFA selection strategy was used to select the primary target audience. Then it incorporated the popular hit/no hit strategy — that is, any lapsed records on the house file that hit against outside lists during the merge/purge process were considered secondary candidates for reactivation.

Unfortunately, the combination of both techniques yielded less than half the desired mail volume. To select the remaining 50 percent, the organization tested a response model against RFA. The model-build process considered more than 100 behavioral characteristics to differentiate responders from non-responders, but the final equation only contained the 10 most significant variables, including upgrade/downgrade, direct mail responsiveness, tenure on the file prior to lapsing and consecutive years of giving. The utility of the model was measured relative to an audience of names selected using RFA. Not only did the model deliver superior results relative to the RFA control group, but it also outperformed the primary RFA and secondary hit/no hit audiences across all three performance metrics - response rate (+168 percent), gross revenue (+81 percent) and CPDR (-11 percent). These highly responsive and profitable names were overlooked because the reach of RFA did not extend deep enough into the file to effectively differentiate opportunities.

Final thoughts
Some passionate supporters of RFA will dismiss the mounting case studies supporting the use of predictive models, arguing that RFA delivers adequate results with minimal investment. However, we’re living in a time when adequate results are not keeping pace with escalating costs and attrition rates. Fortunately, small improvements in response or average gift can have a profound impact on a fundraising program. And the cost of building, maintaining and implementing predictive models turns out to be comparable to — if not less than — the cost of traditional RFA.

The movement to bring together critical donor information that has resided in disparate locations is well under way. Consequently, as the pieces of the puzzle fall into place, we’re getting a clearer picture of who our best donors are. By making more efficient use of all available information, fundraisers are in a better position to capitalize on latent opportunities that have previously gone unnoticed. When you consider the amount of money sacrificed chasing the wrong donors from campaign to campaign, fundraisers cannot afford to ignore the benefits associated with more sophisticated marketing techniques.

Andy Wilder is the senior director of analytics at Epsilon, a relationship marketing company that helps clients build customer relationships through integrated marketing services. Epsilon maintains offices in Boston, Washington, D.C., St. Louis and Dallas.