What it is
Segmentation groups people into clusters that are internally similar and meaningfully different from one another on variables relevant to the business problem.
Overview notes
Common variants
Segmentation is a family of workflows rather than a single algorithm. K-means and hierarchical clustering are two common ways to create candidate solutions.
Decision guide
When to use it
- When one average view hides important differences between people
- When the business needs a targeting or messaging framework
- When you have a robust variable set that captures meaningful variation
When not to use it
- When the sample is too small for stable grouping
- When variables are poorly chosen or inconsistent
- When the business cannot realistically act on multiple groups
Inputs required
- Variables tied to the segmentation objective
- A sample large enough for stable clusters
- A profiling plan for interpretation
Typical outputs
- Cluster assignments
- Segment pen portraits
- Sizing and profiling tables
Simple example
Group category users based on needs, attitudes, and buying behaviors to create four actionable audience types for marketing and product planning.
Strengths
- Helps teams move beyond averages
- Supports targeting and communication strategy
- Can integrate needs behaviors and attitudes
Limitations
- Segment stories can become overfitted or over-romanticized
- Results are sensitive to variable selection and preprocessing
- Actionability can fail if the business cannot use the segments
Common mistakes
- Using every available variable instead of the right variables
- Naming segments too early
- Ignoring stability checks and practical usability
How I use it in practice
I use segmentation when the client needs a strategic framework that can survive beyond one debrief. The process matters as much as the final clusters: variable selection, standardization, and profiling determine whether the output is actually usable.
What is outputted
- Segment IDs
- Summary profiles and comparisons
- Guidance on who each segment is and how to act on it
How to interpret the output
- Start with separation quality and business meaning before naming clusters
- Use profiling variables to explain the segments not to create them after the fact
- Check that each segment can support an action
How to communicate to clients
- Explain how segments were built and what variables drove the solution
- Avoid treating segments as immutable truths
- Anchor recommendations in activation targeting or product implications
Displayr / Q implementation notes
- Freeze the exact variable set before clustering so versions stay comparable
- Standardize inputs consistently
- Export segment assignments with clear labels for downstream tabulation
Mini demo
Segment profile explorer placeholder
A future mini demo can show how changing the number of clusters affects profile clarity and segment size.
This method is marked as a good candidate for a future teaching demo, but v1 keeps the site lightweight for GitHub Pages.