Target audiences in Feature Experimentation

Audiences give you control over which users can access your flag rules, for example, A/B tests, targeted deliveries, or multi-armed bandit optimizations.

A single audience is defined based on custom user attributes you create. These can be based on location, device type, subscription plan, and so on. When you call user-based methods like Decide and Track, Optimizely Feature Experimentation determines whether a user qualifies for a flag by evaluating the attributes.

After you have created and implemented a few attributes, you can start building audiences. You can reuse audiences that you create across multiple flags and rules.

Optimizely Feature Experimentation supports several match comparison operations that you can use in audience conditions:

While attributes are not explicitly typed in Optimizely Feature Experimentation, you must pass in values of the correct type and format for your audience conditions to evaluate correctly. An attribute value of the incorrect type will result in a false-y audience evaluation.

The Edit Audience window lets you define audience conditions by dragging and dropping attributes from the side panel or defining complex JSON objects using the Code Mode window.

To create audiences from attributes:

After creating an audience, you can set the target audience or audiences for flag rules.

When adding your audience to a flag, you can use and or or operators (any or all) to create a combination of other audiences.

If you want to use more complex nested logical operators with and or not, you can do so in JSON in Code Mode. See Create advanced audience combinations.

In Feature Experimentation, complete the following:

If the audience is in use in a rule, Feature Experimentation displays a callout telling you how many rules are associated with the audience and warning you that updating the audience may change user targeting and influence experiment results.

 Click the number of rules to view which rules use the audience.

You can use audiences to include users but not exclude them.

For example, you cannot exclude "all Desktop users" by dialing back traffic to 0% for that audience. This is because if you toggle off audience evaluation for Desktop users, they end up in the "Everyone" audience.

To actually exclude or blocklist audiences, you can define advanced audience combinations with not using Code Mode. For information, see Create advanced audience combinations.

If you define multiple, overlapping audiences and multiple rules for a single flag, then how a single user successively evaluates against the rules can get complicated. For information, see Interactions between flag rules.

If you want to use nested logical operators (and, or, not) to create audience combinations, you can do so in Code Mode. Each audience is a rule like User likes salads, and an audience combination is a boolean combination of these rules, like User likes pizza NOT (User likes sandwiches AND User likes soup).

For information, see Create advanced audience combinations.