Segments, signals and cohorts
Saved audiences, per-profile scores and score-based groups, and how they work together.
Three ideas describe groups of customers. A segment says who. A signal says how likely or how valuable. A cohort is a group defined by signals. They can be combined: a segment can include a cohort, and campaigns use segments.
Segments
A segment is a saved, named audience. It stores a question about your profiles, not a list of people. Each time the segment is used, Optima answers the question against current data, so membership is always live.
A segment has:
| Field | Meaning |
|---|---|
| Entity type | The kind of profile it selects, for example customer. |
| Name | Unique for the entity type. |
| Description, Folder, Tags | For organising. Use / in a folder to nest folders. |
| Active | Whether the segment is in use. |
| Filter | The conditions, built in the Filter builder. |
You build and manage segments on the Segments screen in the Engage group. There are two ways to define the audience:
- By hand, in the filter builder. You can filter on identifiers, attributes, computed attributes, signals, cohorts, consent, linked profiles, and aggregates over events, such as the number of purchases in the last 30 days.
- In English. Under Describe in English, type a description such as
German customers who purchased in the last 30 daysand select Suggest audience. Optima drafts a filter, and may propose a name and description, and tells you Audience drafted — review the filter before saving. Nothing is saved until you select Save segment, so always read the filter first. This needs the AI assistants package; without it the console says natural-language translation is unavailable.
Preview sample shows how many profiles match and a sample of them before you commit. The saved segment lists its Members and shows its size.
Where a segment is used shows in its list, as Used by campaigns, offers and audience syncs. Deleting a segment stops it resolving everywhere it is used, which is why the console asks you to confirm.
When you save, the filter must be well formed. Optima also checks it against your schema again each time the segment is used, so a filter that refers to something that no longer exists is reported rather than quietly accepted.
Signals
A signal is a score computed for each profile, for example churn risk, purchase propensity or best hour to send. You create a signal from a built-in template and fill in its parameters, such as which event type or attribute to look at and over what period. The template library is read-only and covers:
| Family | Templates |
|---|---|
| Propensity | churn_risk_score, purchase_propensity |
| RFM (recency, frequency, monetary) | rfm_recency, rfm_frequency, rfm_monetary, rfm_score |
| Engagement | engagement_tier, event_streak |
| Value | lifetime_value_est, next_event_probability |
| Campaign optimisation | best_send_hour, channel_affinity |
The last two power send-time and channel optimisation in campaigns.
A signal definition has:
| Field | Meaning |
|---|---|
| Name and entity type | Unique together. |
| Template and Parameters | Which computation to run and how. |
| Recompute interval (seconds) | How often the value is refreshed. The default is one day. 0 recomputes on every read. |
| Transform expression | Optional. Adjusts the computed value before it is stored, for example to round it or turn it into a flag. See the expressions reference. |
| Output type | Set by the template: a decimal, an integer, a string or a boolean. |
| When a value is stale | Either serve the old value as it is, or exclude it from decisioning so segments do not match it and offers that use it are skipped. |
Every stored value shows when it was Computed, until when it is Valid until, and whether it is Fresh or Stale. A signal that was never computed shows as stale with no value. You can Recompute one profile's signal from its Signals & Cohorts tab and read its History, or run Recompute all for a definition.
Server-side integrations that read signal values through the CDP API can choose between two behaviours with the signal_refresh setting. The default, inline, refreshes any stale value before answering, so the answer is current but can take longer. With cached, the answer uses the last stored value and its Computed and Fresh or Stale information, so it is fast; only a signal that has never been computed is calculated on the spot. Choose cached when a slightly old value is fine.
You manage signals on the Signals screen in the Analyze group, on the Signals tab. The Template library tab lists the available templates.
Cohorts
A cohort is a named group of profiles defined by an expression over signal values, for example:
churn_risk_score > 0.7 && rfm_frequency < 2A cohort belongs to an entity type, and its expression can only use that entity's signals. Profile attributes are not available in it. A name that is not a signal matches nobody rather than failing. An invalid expression is refused when you save.
Cohorts are managed on the Cohorts tab of the Signals screen. A cohort's membership is kept up to date on a schedule you set with Refresh interval (seconds), with a minimum of 60 seconds. The list shows when it was last refreshed, whether it is awaiting first refresh, and whether the last refresh failed. After you change a cohort's expression, its membership is rebuilt shortly afterwards.
Use a cohort in three places:
- In a segment, with the cohort condition in the filter builder.
- On the Profiles screen, to list its members.
- In reports, to analyse the group.
Membership is judged on the stored signal values, without recomputing them, so repeated reads give consistent answers.
Which one should I use?
| You want to… | Use |
|---|---|
| Target everyone who bought in the last 30 days | A segment |
| Rank customers by how likely they are to leave | A signal |
| Target the customers most likely to leave and rarely active | A cohort of signals, then a segment that includes it |
Label customers champion, active or lapsed | A computed attribute |