Skip to main content
A campaign combines who should see a change (segments), what should change (rules), where it applies (contexts) and when it runs (schedule). You manage these through the existing console pages; there is no separate Campaigns page in the sidebar.

​
Example: promote a collection for one weekend

  1. Check that collection items have consistent metadata, such as collection: summer-sale, and accurate availability.
  2. In Relevance → Rules Engine, create a boost rule scoped to the intended context. Match the collection field and choose a modest boost.
  3. Set the campaign’s start and end time. Check the time zone shown in the schedule before saving.
  4. If the campaign is for a subset of users, create a segment, then add it to the rule’s targeting.
  5. Enable the rule and verify that the context’s Rules stage is enabled.
  6. Test an eligible user and a non-eligible user with Explainability, using integer item IDs from Data → Items.
A boost affects eligible candidates; it does not guarantee a fixed position. Use pin when a specific position is the requirement, and test competing pins and filters. Some rule types require a higher plan; see Rules Engine.

​
Other patterns

Use metadata that exists on your items. NSL does not infer a custom field such as published_days_ago unless your integration supplies it or the feature explicitly computes it.

​
Measure a rule change

Create a draft experiment with a control and treatment. Configure the intended rules or recipe for each variant and test both before starting. Do not assume an empty Include rules selection disables all rules: an omitted override leaves the normal rule selection in place. Leave mandatory exclusions active in both variants. An experiment’s include list selects from eligible active rules; it is not a way to activate an inactive rule. See Experiments for result definitions and limitations.

​
End the campaign

Verify that the schedule has ended and make a new request to confirm the promotional effect is gone. Complete the experiment and record the result. Preserve its configuration and history for later review rather than deleting it immediately. A scheduled rule can remain enabled in the editor while its time window prevents it from applying. If an item still appears, distinguish normal model relevance from a campaign rule that is still matching.