Abstract: A processing system receives audience area data defining a plurality of geographical entities. The entries are clustered in response to geographical data for the entities by processing the geographical data with audience movement signals representing audience movement between the entities, to define a plurality of entity clusters. Audience similarity is quantified for selected pluralities of the geographical entities by obtaining similarity scores for entities in different entity clusters. Similar entities are allocated to a holdout group and a preferred group in response to their similarity score. Content is served to the preferred group thereby optimally configuring audience devices for subsequent measurement of audience uplift.