Skip to content
PMcademy
Build75 minutesUnit 3 of 8 · Lesson 3 of 3

Segments you can defend

Unit 3 of 8 · Lesson 3 of 3Index
PriyaWed 11:30

Support says users are confused. Which users? Maya thinks students, sales thinks freelancers. Both are guessing, and next quarter's roadmap depends on the answer.

A segment is real only if it behaves differently

  1. Two guesses, no numbers

    01 Two guesses, no numbers

    Maya says students. Sales says freelancers. Each has a good story, and neither has looked at what those people actually do.

  2. A label is not a segment

    02 A label is not a segment

    A group becomes a segment when it behaves differently: it signs up, stays or gets stuck in a way the others do not.

  3. One group, two names

    03 One group, two names

    If two groups sign up the same, stay the same and complain about the same things, they are one group. Two labels do not make two segments.

  4. Watch the small groups

    04 Watch the small groups

    A loud theme can turn out to be four people. Put the count next to every theme, every time.

How Netflix handled it

NE

Netflix

Public commentary. No affiliation.
The situation

Recommendation systems often start by grouping people by country, age and gender, because that is what signup forms collect.

What they did

Netflix has said publicly that it found those details predicted far less than viewing behaviour, and that it groups members into a large number of taste communities that cut across countries and ages.

Why it matters to you

What people do made better groups than who they are. Treat descriptions as a starting guess and behaviour as the test.

Doing it with AI, and where it breaks

The move

Cross tabulate your coded signal log by user attribute and problem label, and ask a model which attribute best separates the complaint patterns.

The trap

With small samples it confidently reports a dominant theme that turns out to be four people. Always show the count next to the theme, and do not report a percentage on a handful of rows.

Prompt worth stealing

Here is a table of users with attributes and coded problem labels. For each attribute, show the counts of each problem label per group. Flag any group with fewer than 30 rows as TOO SMALL. Do not describe a theme without its count.

Build it

The artefact

Three evidenced segments, each with one behaviour that differs and the number behind it.

Google Sheets

0/4

The worked solution

Stuck, or want to compare?

There is a worked solution. Try the build first. Reading it before you have attempted anything is the fastest way to learn nothing.

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.

Go deeper

Moving on marks this lesson complete. Finish the build first, it is the part that counts.

Previous lessonThree conversations you have not had yet