Segments you can defend
Lesson 9 of 24, level 01Index
Support says users are confused. Which users? The founder thinks students, sales thinks freelancers. Both are guessing, and the roadmap depends on the answer.
A segment is only real if it behaves differently
Split by behaviour
- Group A: Completes the job
- Group B: Abandons midway
- Check: Different constraints?
The test for a segment is not that you can describe it. It is that it behaves measurably differently from the rest, and that the difference matters for what you would build.
If two groups convert at the same rate, retain the same, and complain about the same things, they are one group with two labels on it.
Segmentation becomes useful when it changes a product decision. Build groups around a relevant difference in behavior or constraints, then check whether the available evidence supports the distinction.
Use your chosen course project throughout. The additional examples below are fictional practice cases; transfer the method to your own evidence.
Define the split before the story
Pick an observable attribute: first-time versus returning, one account versus several, or individual versus team approval. Specify how a person enters each group and whether groups overlap. Do not name a group “power users” until you can explain exactly how that label is assigned.
Compare comparable populations
Use the same observation window and outcome definition for both groups. Show counts alongside percentages. Five successes out of six and five hundred out of six hundred have the same rate but very different uncertainty. A tiny group can inspire a research question without justifying a broad rollout.
Explain what changes
A segment should lead to a different interview sample, experience, or priority. If both groups need exactly the same fix, keep the simpler design. When differences appear, investigate acquisition source, tenure, and permissions before attributing everything to the segment label. Correlation is a clue for discovery, not a causal mechanism.
A segment must earn its name
- Membership: Observable rule
- Comparison: Same outcome window
- Decision: Different next action
Doing it with AI, and where it breaks
Cross tabulate your coded signal log by user attribute and problem label, and ask a model to find which attribute best separates the complaint pattern.
With small samples it will confidently present a dominant theme that is four people. Always show the count next to the theme, and never report a percentage on fewer than 100 rows.
Your AI workbench
Start with your own notes or clearly labelled practice data. Remove private details before sharing. Replace the placeholders, run the prompt in your chosen AI tool, and keep the output beside its source.
Given this anonymised table [paste], propose at most three behavior-based segments relevant to [decision]. Define membership and show numerator, denominator, time window, and missing values. List confounders and a disconfirming observation for each. Do not invent statistical significance or infer protected attributes.
Before you use the output
- Group membership is reproducible
- Rates include counts and windows
- The split changes a specific decision
Stuck? Try this next
Ask AI to show the rows behind one aggregate, then reproduce the calculation manually. If the sample is small, use it to recruit follow-up interviews rather than to claim a market size.
Keep a brief AI log: input used, useful output, what you checked, and what you rejected. The decision remains yours.
Build it
Three evidenced segments with a behavioural difference named for each.
Google Sheets
Checkpoint
If you did the build, these take two minutes. If you cannot answer one of them, that is the part to go back to.
Moving on marks this lesson complete. Finish the build first, it is the part that counts.