Segments and cohorts

The groups of users that Otis compares, how the four built-in segments are decided, and how to define segments and cohorts of your own.

Otis compares groups of users. A segment is a group defined by what users have done recently, and its members change as their behavior changes. A cohort is a group of users who first arrived at about the same time, and its members are decided once and never change.

Every project gets four built-in segments: power user, believer, satisfied quiet and churning. You can define further segments and cohorts by asking Otis. This page explains how membership is decided, so that you know what a segment name or a cohort count refers to.

Purpose of segments and cohorts

A segment answers a question about the present, such as which users are getting value this month. Its counts change from day to day as users move in and out.

A cohort answers a question about users who started together, such as how the users who arrived in March were doing after 30 days. Every member is measured at the same age. That removes the effect of how long each user has had the product, which a comparison between segments can't do.

Both are definitions. A definition is a rule over a fixed set of measures, and Otis applies it by counting. No AI model decides who is a member, so the same activity always gives the same result.

The four built-in segments

Otis asks two yes-or-no questions about each user. Success asks whether the user is getting value from your product. Affinity asks whether the user is attached to it. The two answers give four segments.

Each axis is a yes-or-no test with a fixed rule. A user's segment changes only when that user's own activity changes.

Power user

High success · high affinity

Getting value and loves the product — the PMF core. These users got a result in the period and show attachment to the product.

What to do: Compare what they do with what other users do. The difference describes what your product is for.

Believer

Low success · high affinity

Loves the product but struggles to get value — high-intent, underserved. These users show attachment, but Otis has no record of a successful result for them in the period.

What to do: Find what blocks them. They are already invested, so removing one specific obstacle can be enough.

Satisfied quiet

High success · low affinity

Getting value but not attached — quietly served, not a champion. These users got a result and show no sign of attachment.

What to do: Look at what would bring them back or deepen their use. Nothing fails for them, so no error points you to this group.

Churning

Low success · low affinity

Neither getting value nor loyal — at risk of leaving. Otis has no record of a successful result and no sign of attachment. Every recently active user who meets neither test is here, including users who have only just arrived.

What to do: Separate new users from long-standing ones before acting. The two need different responses.

Users who get a segment

Every identified user who was active on at least one day in the last 28 days gets exactly one of the four segments. A user is active on a day when your app sent at least one span carrying their user ID.

A user who has been silent for 28 days gets no new segment. Otis keeps showing that user's last segment for up to 30 days.

Success rules

Otis has two rules for success. It uses the first rule when it can, and the second otherwise.

A funnel success. The user has at least one entry in your scoring funnel that ended in success. The scoring funnel is the one your funnel definition marks for scoring, or the first funnel if none is marked. Otis looks at recent entries, and reaches further back when the funnel has a stage that happens only once for each user.
A successful task. The user has at least one task in the last 28 days with a success that your product reported. Examples are positive feedback that your app sent about the work, and an event that your funnel definition lists as a success stage. The outcome that Otis infers from a user's messages doesn't count here.

Affinity rules

Otis also has two rules for affinity.

Investment. The user did at least two different investment events. An investment event is an optional action that shows commitment, such as inviting a teammate, sharing work, or connecting an integration. An event that recurs counts if it happened in the last 28 days. An event that happens only once, such as connecting an integration, counts whenever it happened.
Regular activity. The user was active in at least two of the last four weeks.

Otis takes the investment events from a list declared for your project. Customization describes how to declare one. If there is no list, Otis uses the steps of your funnels other than the scoring funnel, and events that record collaboration, contributions to a knowledge base, or completed setup.

The two rules for affinity measure different things. Investment shows that a user chose to commit to the product. Regular activity shows only that the user keeps coming back, which is also true of a user whose job requires the product. When Otis uses the second rule, power users include users who are merely regular, and believers can include users who return because they must.

Choice of rule

Otis chooses the rule for each question on every run. It moves to the second rule when the first can't separate your users:

For success, when no funnel is defined, when the scoring funnel tracks accounts and not users or artifacts, when the funnel has no recent entries, or when almost all or almost none of your active users have a funnel success.
For affinity, when the project has no usable investment events, or when the rule would select too few or too many of your active users to tell them apart.

When the rule changes for either question, every user is assigned again under the new rule on the same day. A segment's size before that day and after it are measured by different rules and can't be compared.

Projects with no segments

Otis writes no built-in segments for a project on a day when it can't apply either rule for one of the questions. The usual case is a new project with no funnel, no feedback and no success stage yet. A project with no segments hasn't been measured, which is different from a project in which every segment is empty.

Fixed thresholds

The thresholds in these rules are the same for every project. Otis doesn't set them from your users' averages.

A user's segment therefore changes only when that user's own activity changes. It doesn't move because other users became more or less active. A segment can be empty, and its size on one day can be compared with its size on another.

Segments you define

You can define a segment by describing it to Otis in a chat in the Otis app, for example "users who had at least three sessions this week and were frustrated in most of them".

Otis writes the rule and measures it against your current data. It then shows a card with the rule as a sentence, the number of members out of the users it was measured over, some example members, and any warnings. The segment exists once you select Create definition on the card.

Measures available to a rule

A rule compares one or more of these measures with a value, and can combine them with "and", "or" and "not".

Measure	What it counts for each user
Active weeks	How many recent weeks had activity
Sessions	The number of sessions in a period you choose
Funnel stage reached	Whether the user reached a named stage of a named funnel
Funnel outcome	Whether the user has an entry in a named funnel that ended in success, in failure, or with neither
Successful tasks	The number of tasks with a reported success in a period you choose
Signal share	The share of the user's sessions in which a named signal appeared
Investment breadth	The number of different investment events in a period you choose

Signal share counts only sessions that contain a conversation with your AI.

When you ask for a group that these measures can't express, such as one based on a sequence of actions, Otis says so and records the request.

Rules that apply to every segment you define

Only recently active users can be members. Otis evaluates a segment over users who were active in the last 28 days, whatever period the rule itself covers.
Segments can overlap. A user can be in several of your segments and in one built-in segment at the same time.
Otis refuses a segment that can't tell users apart. It won't create a segment that selects almost all or almost none of your users. It also refuses one that matches an existing definition.
A definition can't reuse what decides the built-in segments. Otis refuses a rule that uses the measure currently deciding success or affinity, or any stage of the scoring funnel. Comparing such a group with the built-in segments would compare a rule with itself.
To change a definition, replace it. Remove the definition and create another under a new name.
Removal is permanent. A removed definition's name can't be used again.

Any member of the project can create or remove a definition, in a chat in the Otis app.

Cohorts

A cohort groups users by when Otis first saw them, which is the date of the user's earliest span. Otis decides membership a fixed number of days after that date. That number of days is the cohort's maturity. The result is then frozen, and the user is never evaluated for that cohort again.

Every project has two built-in cohorts, with a maturity of 7 days and of 30 days. Each user joins them on reaching that age. A period, such as the users who first appeared in March, isn't part of a cohort's definition. You choose the period when you ask a question about the cohort.

Cohorts you define

A cohort you define adds a condition that Otis tests at maturity, such as "users who had at least three sessions in their first week". Otis records each user who reaches maturity as a member or as a non-member.

Three rules apply:

The condition can use four of the measures. These are active weeks, sessions, signal share and investment breadth. A cohort can't be defined by a funnel outcome or by successful tasks. A cohort that was selected by an outcome would make any later comparison of outcomes meaningless.
You choose the maturity. It is a number of days, up to a year.
A new cohort starts empty. Otis evaluates only the users who reach maturity after you create it. It doesn't go back over users who matured earlier.

Accounts

Segments and cohorts describe users. For accounts, Otis tracks a stage in the commercial lifecycle.

Reading segments and cohorts

Churning doesn't mean the user has left. The churning segment holds every recently active user who met neither test, and that includes users who arrived yesterday. It differs from the churned stage in the lifecycle, which describes an account that isn't paying and has gone quiet.
A segment restates its own rule. A finding that power users succeed more often than other users is the definition read back. Look at what a segment does differently in other respects.
Check which rule is in use. The meaning of affinity depends on the rule, as described above. Ask Otis which rules your project is on.
A segment shown for a user can be up to 30 days old. Otis shows the most recent segment it assigned in the last 30 days.
A cohort only contains users who have reached maturity. A user who first appeared 10 days ago isn't yet in the 30-day cohort. Their absence doesn't mean they failed a condition.
No segments means not measured. Don't read a missing segment as a count of zero.

Retention of membership

Otis keeps daily segment membership for 365 days. Cohort membership is kept without a time limit, so that a cohort can still be compared after the telemetry behind it has expired.