Skip to main content

Getting started with DMP Audiences

Audiences represent groups of users defined by event data, profile attributes, or combinations of existing audiences. They can be used for activation and can be exported to external destinations or to Decentriq DCR datasets.

All audiences in the DMP are created and managed from Audiences. See Create an audience and Manage audiences for details.

Behind the scenes​

A few audience-building features rely on processing that happens automatically during the daily DMP pipeline run.

Content categorization​

Every page-view URL collected by the Decentriq Tag is classified against the IAB Content Taxonomy 3.1. Categories are extracted at four hierarchy tiers (tier 1 through tier 4) and attached to the event as additional properties.

This enables event-based audience rules such as "users who visited Sports → Boxing pages in the last 7 days" without the publisher having to tag pages manually.

Decentriq runs the URL classification on its own schedule. Newly observed URLs are typically classified within a day or two and become available in the next pipeline run. If a URL has not yet been classified, the underlying page-view event is still usable in audiences via its other properties — only the IAB category fields are missing.

In the audience rule builder, IAB categories appear as event properties named IAB Category (Tier 1) through IAB Category (Tier 4) on event-based blocks. Each tier represents a deeper level of the taxonomy.

Predicted demographics​

Decentriq can predict demographic traits (e.g., age and gender) for users who only generated anonymous behavioral signals.

How it works:

  1. Training (one-off, with hard data from the publisher) — Decentriq trains a per-publisher ML model using a sample of users for whom the publisher has provided ground-truth demographic data (for example, from registration or subscription records). Training is a one-time activity; the resulting model is then versioned and stored.
  2. Inference (every pipeline run) — During each daily pipeline run, the trained model is applied to all users to produce predicted trait values.
  3. Use in audiences — Predicted traits are surfaced as profile properties in the audience rule builder, with a (modeled) suffix on the property name (e.g., Age (modeled), Gender (modeled)). They can be used in profile-based blocks just like observed traits.

Predicted traits don't replace observed ones: if a user has both an observed age and a (modeled) age, both are available as separate properties in the rule builder.

Setting up demographics prediction requires a conversation with your Customer Success representative to scope the available training data and the traits to model.