Have you ever looked at a list of prospects and thought: “Gosh, I wish there was a way to segment these prospects by their relationship to my organization?” Well, you’re not alone! Many organizations address this challenge using a measure of engagement and connection called an affinity score. The components of an affinity score vary based on the institution and the analysis, but generally include certain “opt-in” activities a prospect performs that describe their relationship to your organization, like volunteer positions, amount of giving or attending events. The more of these activities that a prospect performs, the higher their affinity.

The Need for Affinity Scoring

An affinity score is a great tool that can augment many of the scores you already use. A wealth screening score by itself is fairly one-dimensional — it says, “X person can give Y amount.” However, it does not indicate whether the person is likely to give, or how involved the person already is. This is where an affinity score can be a useful tool, if used in conjunction with wealth and other indicators.

The UNC Health Foundation has a number of research tools at our fingers that we have used effectively for a few years. But none of them really got to the heart of why a prospect may give. While describing intention is difficult, recognizing and measuring the entire relationship that a prospect may have with the institution can be done with a bespoke affinity score, and one which in our case is centered around a prospect’s patient connection.

Here at UNC Chapel Hill, we have a great internal affinity score that was developed a few years ago, which considers many useful elements describing a prospect’s relationship with the school (e.g. alum status, recency and frequency of giving, etc.) and applies the score on a scale of 0-100. While these factors speak well to alumni involved with an institution like UNC as a school, it does not fully describe the relationship that a patient would have with their health provider. As the interim assistant vice president of prospect development at the UNC Health Foundation, I know patient affinity is much different than alumni affinity, therefore we decided to develop an internal, UNC Health Affinity Score.

Developing Datapoints

So, where did we begin and how did we decide who to model and run the analysis on? Considering the UNC Health Foundation had just wrapped up a $1.5 billion campaign, we decided to pull a list of all campaign donors, which came out to around 40,000 individuals. We used this group to identify the data points most common or unique to this group, since we already knew our campaign donors were highly connected to us. We began with the datapoints in the UNC Affinity Score, and although this specific affinity score did not apply heavily to our patient population, it was still helpful as a starting point. The first thing we did was break down the main UNC Affinity Score by its criteria and see if any of those data points “pop” with our donors. For example, the main UNC Affinity Score contains criteria about season-ticket holdings for UNC’s athletics. While this could be a high affinity factor for alumni and other donors, it did not resonate in the data of our campaign donors, so we did not use it.

On the opposite end, we tested a number of built-in constituency tags, such as alumni, friend, faculty and parent of a student. When testing these record types, we noticed parents of current or former students appeared significantly more among donors who made a major gift in the campaign ($100K and above) than those who made a lower-level gift. Therefore, we knew we had to have this as one of our affinity elements and score it highly.

The patient information was a tricky piece to figure out. We researched which of our campaign donors, or their spouses, had been patients of the UNC hospital. This provided us with specialty and provider data. While we do receive certain appointment information, it is HIPAA compliant and very high level in terms of why the patient came in for an appointment. Looking at this data, we learned that the specialty area information we receive had to be cleaned and combined into broader, more helpful buckets. For example, there are 14 subspecialties of cardiology that needed to be consolidated under simply “cardiology.”

Secondly, we had to see if there was anything significant about the physicians treating our patient-donors. This turned out to be a difficult task, as running analysis on several hundred thousand appointments took significant computer time to perform. However, running some basic regression analyses in R showed a pattern — many of the physicians identified as significant factors were already working with the development staff. This is one of those moments where the art and the science of fundraising align. The UNC Health Foundation already has a list of physicians that our development officers interact with the most, so we used a simple “Yes/No” to signify whether a treating physician was already on our internal list.

Finalizing Criteria and Score Testing

When all was said and done, we had 10 criteria that turned out to be significant descriptors of our major donors. These criteria were sorted into three categories: constituent data, patient data and revenue data. Consulting with senior staff, we assigned scores for each data point. We tried to utilize binary choices as much as possible (e.g. “Yes/No”), but there were a few points where some nuance had to be applied — especially for our revenue category, where we had to score principal and major giving amounts higher than lower amounts. We also applied weightings to a few categories, indicating that these categories were more important than others.

How did we know that we had the right categories overall? Using some basic predictive modeling in R software, we tested the scores on a selection of donors known to us and were able to see that the categories we chose created very high predictive values. This was very encouraging and allowed us to move forward.

With our scores and weights, we could put it all together! An Excel sheet was created to house data for the 10 categories and populate scores using nested if/then formulas to match the data with a scoring sheet. Once we had a numeric score, the last piece was to create easily interpretable labels. We decided to create five engagement segments that correspond with different percentile ranges, as shown below:

Percentile Range	Score	Engagement Segment
<25	<20	Disconnected
25	20	Highly Disengaged
50	27	Disengaged
75	38	Engaged
90	48	Highly Engaged

This categorization allows us to communicate a person’s score with greater description than a simple number would. This score and these segments are what worked best for our organization. Your organization might be different, and that’s okay! The important thing is to recognize that you have the power to create something like this to better describe your prospects affinity and better segment your data.

Using these scores, we have been able to focus development officers on the right prospects and apply greater segmentation to larger prospect lists. Furthermore, those prospects who do score highly may be more likely to respond to outreach from development officers than a traditionally cold lead would be. While these warmer leads with higher affinity points may not have the highest net worth, they are great prospects to cultivate. Using this affinity score with other data like a wealth screening or predictive model can highlight the right people, which ultimately contributes to more dollars raised.

Chris Copsey will be presenting "Going from Good to Great: How to Best Leverage a Consultant Relationship" at Plug In to Technology, taking place May 21-22. Learn more about Copsey's session and Plug In on the Apra website.