It’s Thanksgiving! Time to think about pie—or at least about ways to slice your constituents into meaningful chunks.
Going wonky in this post with some of the detailed ingredients and cooking directions for segmentation.
But first, a bit about the danger of segmenting audiences, which involves grouping and dividing people. A warning:
“Dividing” and the Body of Christ don’t go together particularly well.
Our default posture should be one of unity.
In the church (and in church planting), the “homogeneous unit principle” acknowledges that growth tends to happen more quickly among groups whose members share a common language, ethnicity and socio-economic class.
This principle has been used wisely to develop indigenous churches that show the incarnational nature of the Good News in understandable forms—relevant and transformative. The principle has also been used unwisely to segregate and isolate people within the Body of Christ on the basis of race, class, age, etc., preventing the Church from experiencing unity in the midst of God-ordained diversity.
The challenge in applying the concept wisely is well described in the very first Lausanne Occasional Paper from 1978.
In local congregations, intentionally segmenting people can be fraught with difficulty. Christianity Today’s Andy Crouch addressed this issue eloquently in his 1999 essay For People Like Me from re:generation quarterly, a magazine whose demise I still mourn a decade after it folded. I encourage you to click through and read the whole piece, but here’s the punchline:
For surely one of the scandalous things about the gospel—indicated by Jesus’ own practices of welcoming sinners and eating with them, calling tax collectors along with fishermen to be his disciples, and praying for the forgiveness of his executioners—is that it does not fit the marketer’s (or the Pharisee’s) formula “for people like me.” It is in fact for people not like me—unless they are “a wretch like me,” and wretchedness was never the basis of a successful marketing campaign. Christianity is not a product that can be added seamlessly into the lives of consumers like one more lifestyle-enhancing appliance. It is instead a call to a completely different way of viewing the world, one in which the one who looks least like me is at a minimum my “neighbor” (Luke 10:29-37) and could well be Jesus himself (Matt. 25).
So, before undertaking the task of segmenting an audience, be sure to check your conscience. Segmentation can acknowledge God-given variation in giftedness and experience. That is the beauty of the Myers-Briggs® types—there are no better or worse personalities—each has natural strengths as well as potential blind spots. It can also validate multiple approaches for doing a task, blunting the arguments of those who are fixed on their method as the “one best way” to do a task.
With that addressed, on to wonkiness.
When communicating with large numbers of people (donors, staff, readers/listeners, etc.) segmentation is a strategy that reflects a middle ground between uniformity (one size fits all) and customization (every one unique). Mass communication is often ineffective; individual customization is usually inefficient. In segmentation, an approach is developed for each segment, but within a segment everyone is treated similarly.
Criteria for developing segments include:
- Meaningful subgroups really exist—there is a valid basis for segmenting an audience.
- The subgroups are identifiable—there is a reasonable way to segment an audience.
- The subgroups are actionable—there is a practical use for segmenting an audience.
The Missio Nexus CEO survey that we recently helped with was commissioned, in part, because Missio Nexus frequently heard CEOs asking how other CEOs were dealing with various challenges. The idea was to document and share experiences among the CEO community. The question CEOs were asking presupposes likeness among the peer group.
We thought it could be helpful to see if meaningful subgroups existed which would help focus the question or expand on it. Profiling CEO segments could help CEOs better understand themselves and their peers. Their question could become, “How are other CEOs like me dealing with this issue?” or “Why are CEOs dealing with this issue in different ways?”
Here are the general steps in segmentation, using some of our recent projects as examples.
Step 1: Select a basis on which to explore/generate segments.
You can segment audiences in many ways—some of the most common consider the needs, values, aspirations or behaviors of their audiences. Consider behaviors. If people behave according to certain patterns, those can dictate the communication channels used to reach them. Child sponsorship agencies, for example, use several methods to sign up new sponsors that are behavior based: church partnerships, online ads, concert sponsorship, direct mail. These are real meaningful, actionable segments.
In some of the recent segmentation work that we’ve done, the basis for segmentation has been as follows:
- Church Planters: Behaviors (frequency of “fruitful practice” activities)
- Mission Agency Website Visitors: Needs (information sought)
- Mission Internship Prospects: Motivations (for considering a 1-to-3 year term of service)
In the Missio Nexus CEO survey, one objective was to look at recent progress and current or near-future challenges. So, we developed segments based on relative priorities for organizational, staff and personal-effectiveness (combined). We didn’t worry about why CEOs prioritized one area over another. The shared need to address certain areas was enough.
Our hope is that priority-based segments would be actionable for Missio Nexus as an association that provides regular programming for executives such as C-Suite Webinars. Priority segments give them a guide for how to plan relevant content that provides a balance for each type of CEO. At an event, the CEO audience might not be large enough to justify separate tracks—but breakout sessions could be scheduled in a way that each group is likely to find something of interest.
Step 2: Select a method for creating segments.
If one quantitative measure is extremely important, such as expected lifetime donor value or likelihood of serving with your agency, segments can be driven by their impact on that variable. This situation calls for decision-tree analysis. Many statistical packages include such a method—CHAID and CART are traditional examples. The analyst feeds in a number of predictor variables—often combining different variable types—along with the known outcome of the key variable from a sample. The software will identify a sequence of if-then steps involving the variables that best divide people into groups on the basis of the key measure.
This allows donor or staff prospects to be quickly qualified; responses can vary accordingly. Those with a lower likelihood of giving or joining should not be ignored, but follow-up communication might be done a bit more frugally or infrequently.
Often segmentation will be driven not by a single measure but several measures with a common theme. In our Agency Web Review we used a set of 16 types of information that visitors to mission agency websites might be interested in. In that case, cluster analysis can be a great way to identify segments. K-mean clustering is a well-regarded tool in which the analyst specifies the number of segments (clusters), assuming statistical significance. Most analysts run and compare several variations using different numbers of clusters, selecting the one that seems most practical or intuitive.
Step 3: Create and label the segments
In our mission agency website visitor study, we chose five statistically valid segments (via cluster analysis) that also made intuitive sense to us. Looking to name the segments, we noticed each segment’s various interests. Among the influential variables were short-term opportunities and long-term opportunities.
One segment demonstrated relatively low interest in both kinds of opportunities. That seemed strange—all of those responding had been screened for interest in serving cross-culturally. We dug deeper, examining the group by its demographics. It turned out that many people in the segment were underclass collegians (seniors and new grads were more likely to be in other segments). Aha! Now it made more sense. Their low interest in service opportunities reflected the fact that they were years away from applying for career service. They valued learning about agencies generally and exploring mission-oriented resources (perhaps for use in coursework or for their campus fellowship). This group also included a fair number of non-students whose primary role was mobilizing others to go. Therefore, we named the segment Scouts. They were scouting out info for another time—or for other people. We named the other segments through a similar process.
Step 4: Describe the segments in detail
Saving segment membership to your data set opens up a world of descriptive possibilities by cross-tabbing segment with other variables. In our church planter segmentation, women were especially likely to be in one segment—even though none of the input variables were gender related. The church planters were working among resistant peoples, often in cultures where women are closely protected and limited in their social mobility. It came as no surprise, then, that women made up a large portion of the segment that emphasized prayer and judicious (not bold) sharing. For many, that was the type of ministry that was available without severely violating cultural norms.
One way to see the relationships among segments is through segment maps. These can be created quickly through a bit of reverse engineering. (Warning: statistical terms coming—in case of dizziness, skip down two paragraphs.) We use the variables from the cluster analysis as predictor variables in a discriminant-function analysis. Cluster membership is the variable to be predicted. We save the function coefficients as variables, and then we use the first two sets of coefficients as X-Y coordinates in a scatterplot. When we color code by segment, results look something like this (taken from our Mission Internship Study):
Each point represents a respondent. The segments naturally group together, and the X and Y dimensions distinguish segments from one other. These dimensions, which reflect weighted combinations of the input variables, should be labeled to show how the segments relate to one another.
In the example above, three groups emerged with differing motivations for considering a cross-cultural internship of one to three years. The map showed that the groups can be considered on dimensions related to the purpose of the internship (My Fit vs. Their Blessing) and their level of commitment to long-term mission (Committed vs. Exploring):
- The Where segment is mostly committed to long-term service and want to test their fit in a particular setting or a particular agency;
- The Whether segment is uncertain about long-term service and want to test if they should continue serving after the internship concludes.
- The Whatever segment isn’t concerned about their long-term direction. They simply want to meet people’s needs through the internship without considering their future career path.
Step 5: Develop a scoring model for classifying others
It isn’t easy to get everyone to take a survey, so the segments of constituents who don’t respond—and those emerging in the future—cannot be classified. This limits the value of the segmentation.
The answer is to create a scoring model—either using non-survey data or developing a mini-survey that makes it easy to collect information, such as through a registration form. Here is where we get to the quizzes that let people discover their “personality.”
With decision-tree segments, we simply use quiz questions based on the logic of the tree. For cluster analysis-based segments, we use discriminant analysis. The setup uses the same variables as in the mapping step above, but this time we use a stepwise procedure (to limit the number of variables) and select the option for “Fisher coordinates.” This yields one equation for each segment. When someone takes the quiz, we cross-multiply their responses with the Fisher coordinates and then compare the totals: the largest value is the “predicted” segment—which is shown to the quiz taker and/or added to the constituent database.
Quiz results usually include a thorough description of the predicted segment (and sometimes other segments as well). Discussion questions can be added to help people think about how to maximize the strengths of their personality and to minimize or overcome the weaknesses.
This step is important because marketing research ethics statements usually indicate that participation in a survey should not influence the way the respondent is treated (compared to non-respondents). Therefore, making an effort to classify non-respondents ensures ethical compliance.
Step 6: Develop and carry out a strategy for each segment
With segment membership assigned to constituents, it is time to put the segments into practice. Should we emphasize some segments over others? Should we communicate differently to each segment? Should we develop offerings based on the needs of certain segments? Should we organize staff responsibilities by segment?
The applications for using segmentation are many and far reaching. Segmentation is usually strategic rather than tactical. Since it involves high-level thinking, the segmentation process should have involvement and buy-in of senior leadership from its early stages. In commercial research, I have seen segmentation projects aborted or shelved more frequently than any other kind of research. It should not be undertaken lightly.
That’s the recipe for segmentation. Wonky, yes—but underlying the fun, What kind of ____ are you?” quizzes is real science. If you are thinking about segmentation and would like to have some help in your analysis kitchen, feel free to <a href=”mailto:[email protected]”>email</a> or give us a call. We’re glad to join in the process of delivering information that supports Spirit-led decisions.