Segmentation Overview

Segmentation is the process of splitting a large business market into sub-groups based on some form of shared characteristics. Clients seeking segmentation analysis from Gradient expect a detailed examination of their current or prospective client base to find potential target markets and identify distinct segments/profiles to help them tailor their marketing strategies.

With a segmentation analysis, our clients can:

• Develop new products that meet the needs of various market segments.
• Strategically position themselves against competitors when entering a market or introducing a product.
• Understand preferences and trends that will help them better target their audience with the appropriate marketing messages.

We previously used a demographic segmentation analysis, which segmented business markets based on demographic factors such as age, gender, marital status, family size, occupation, education level, income, race, religion, and so on. This solution has the issue of being overly ambiguous and failing to fully profile our segments. We can’t figure out what shared psychographic or behavioural theme exists in our segments, and these are usually the features required to create a complete segment profile.

To tackle this issue, We use mindset segmentation analysis to develop our segments and only use the demographic factors to add further information during synthesis. During data collection/survey development, we ask questions that have a direct impact on the segmentation goal. Participants in the survey are asked to select the option with which they most identify (these options are on a Likert scale).

A typical mindset segmentation format is presented below.

Dictionary: The names of the segmentation columns and the questions they pertain to are included in a tibble to form a dictionary.

Scree Plot: Used to determine the optimal number of segments/factors. It plots both the component and factors in an elbow plot.

Factor Analysis: This reduces the dimensionality of the segmentation data and displays the top N variables or columns for each factor.

NMF: This is the segmentation model used. More details about this can be found here.

Dispersion: This measures the reproducibility of the clusters obtained from NMF. Ranges between 0 and 1, the higher the better.

Silhouette: Silhouette score is a metric used to calculate the goodness of a clustering technique. It ranges between -1 and 1 where:

• 1: Means clusters are well apart from each other and clearly distinguished.

• 0: Means clusters are indifferent, or we can say that the distance between clusters is not significant

• -1: Means clusters are assigned in the wrong way.

Sparseness: Sparseness quantifies how much energy of a vector (or matrix) is packed into only few components. It ranges from 0 and 1

Breakdown: Displays a tibble of respondent assignments into segments for various K solutions

Consensus map: A plot that displays the clusters/segments from the segmentation model.

Silhouette comparison: This contrasts the NMF model’s Silhouette score with that of other well-known segmentation models, including hierarchical, Kmeans, pam, and Clara.

Typing tool: It indicates how effectively we can predict segment assignment using a subset of the original input variables and is also used to confirm with the client whether the typing tool’s accuracy is enough given the solution.

Indices: The index value is defined as the ratio of “the segment that agrees” to “the entire population that agrees.” Compared to the entire sample, this value shows how much more likely a segment is to agree with the statement.

Deciding on the number of segments is more of an art than science. Segments must be large enough to make targeting them worthwhile. Having too many segments will also make managing your tactics difficult. If you use too few, you won’t get the granularity you require. The elbow plot can help decide the optimal number of segments but you as the analyst would have the final decision on the ideal number of segments.

Typically, a segmentation should have atleast ~1.000 responses. Given that most segmentations end up at roughly 5 segments, that would leave us with on average 250 respondents per segment. In order to draw meaningful conclusions and differences, that is the minimum sample size needed.

In order to identify potential target markets and specific segments/profiles that would assist the client make better business decisions, a segmentation analysis has to be performed.

Typing tools are created very early in the segmentation workflow because it assesses two things:

• How well can we predict segment assignment with a subset of the original input variables?
• We can confirm with the client if the accuracy of the typing tool is good enough given the solution. If this is not the case, we need to go back to the segmentation process and adjust it.

A typing tool for segmentations are typically used by the client to group respondents in future surveys, but without having to use the very large questionnaire. A practical example would be, if they wanted to run a new study, they can simply ask the subset of questions and assign the associated segment to the data. Sometimes they will even implement the typing tool in their CRM, so they can tag all of their customers with the segment for various marketing implementations.

These indexes are values that can help interpret the segmentation. They don’t drive the segmentation, as the indexes themselves are not inputs to the segmentation. In order to understand the segmentation solution, these tables are created. In the simplest form, this is what’s done:

For each variable we calculate;

• The overall proportion of respondents that agree with the statement
  • Note that in a 5-point likert scale, this would be the combined “Somewhat agree” and “Very much agree” percentage
• The segment proportion of respondents that agree with the statement
• The index value, which is segment_agree / population_agree. This value represents how much more likely the segment agrees with the statement than the overall sample.

With this, it’s easier to interpret how a segment is different from the overall population, based on how much they agree with a statement, in contrary to the overall population.

1. Recognize the business problem.
2. Create research questions.
3. Choose survey methodology.
4. Get Data
5. Extract Segmentation Data
6. Preliminary Assessments
7. Run NMF
8. Run diagnostics
9. Find optimal segmentation solution
10. Interpret and synthesize the segments