Segment, cohort, dimension: terms people mix up

A handful of analytics terms get confused so often that the confusion is worth its own page. Segment and cohort. Event and feature. Attribute and segment. Each pair sounds similar and describes something genuinely different, and mixing them up leads to the wrong measurement or the wrong decision. This article takes the pairs one at a time. The definitions are tool-neutral: they describe the underlying concept, not any one product’s labels.

The single terms below are defined in the vocabulary of feature usage analytics; here the focus is on what separates each pair.

Segment vs. cohort

Both are groups of users. The difference is what defines the group.

A segment is defined by a shared attribute: enterprise customers, admin users, users in the EU.
A cohort is defined by a shared event in time: users who signed up in January, users who first tried a feature in Q1.

The test: if the group is defined by who someone is (their plan, role, region), it is a segment. If it is defined by when they did something, it is a cohort. Segments are for comparing kinds of users. Cohorts are for retention — you follow the January cohort week by week and compare it to February.

This is the distinction tools blur most. Some products let you build a cohort from an attribute with no time element at all, then use it exactly as a segment. The concept still holds; the labels do not.

Event vs. feature

An event is one recorded action: a single click, a single export.
A feature is a product capability, which groups several such actions.

Opening the export dialog, choosing a format, and confirming are three events. They belong to one feature: export. Where those actions get grouped into a feature depends on the tool. General event analytics logs each action as its own event, then reconstructs the feature from a combination of them afterward — so you have to be careful not to count three clicks as three uses of the export feature. UsageLens moves that grouping into your application: it decides when the feature was actually used and sends one usage event, so one event already means one feature use.

Attribute vs. segment

These sit next to each other and get swapped constantly.

An attribute (or property) is data attached to an event or user: subscription_tier, role, region.
A segment is the group of users who share a particular attribute value.

So subscription_tier is the attribute, enterprise is the value, and the enterprise customers are the segment. Calling the attribute itself a “segment” is a common shorthand, and some product UIs use it that way, but strictly the attribute is the data and the segment is the group. When you “break down by subscription_tier”, you are using the attribute as a dimension to produce several segments at once.

Active users vs. usage count

Two ways to count the same activity, and they can point in opposite directions.

Active users (or unique users) counts distinct people.
Usage count counts actions, ignoring who performed them.

One hundred uses of a feature could be 100 people using it once each, or 1 person using it 100 times. The usage count is identical; the meaning is opposite — broad shallow reach versus a single dependent user. Judge a feature on usage count alone and you cannot tell the two apart.

Adoption vs. engagement

Adoption asks whether users started using a feature.
Engagement asks how much they use it after adopting.

High adoption with low engagement means everyone tried it once and few came back. The two call for different responses: low adoption is usually a discovery or onboarding problem, while low engagement is usually a usefulness problem. Treating one as the other sends you to fix the wrong thing.

User vs. account

A user is one person.
An account (or customer) is the organization that person belongs to.

In business software many users belong to one account, and the level you measure at changes the answer. “80% adoption” across users can coincide with only a handful of accounts touching a feature, if a few accounts have many users each. For roadmap decisions, account-level adoption often matters more — losing one enterprise account is not the same as losing one free user. Decide which level a metric is measured at before you read it.