WPP Media is testing “synthetic audiences” for programmatic buying in the U.S. through a partnership with Givsly, a tech company founded in 2019. The audience segments are built partly from charity donation data: where donations happen, how often, and to what kinds of causes.
The pitch is simple enough. Instead of relying only on open-web behavior and demographics, Givsly maps donation patterns by zip code as a proxy for disposable income and “social values.” Givsly’s system is built on the Claude LLM and combines U.S. zip-code mapping with third-party data sources including national census and election results. It also uses nonprofit-provided data showing donation volume, frequency, and charity category by zip code.
WPP Media added its own first- and second-party data, then used the resulting synthetic profiles in PMP deals for four unnamed beauty and fashion brands on CTV and online video inventory. Tara Sadlak, WPP Media’s group director of media delivery, said early tests showed a 2% lift in video completion rates and framed the goal as “discoverability and increased reach,” plus the ability to target around values rather than just demographics or open-web behaviors.
This is still a slightly nervous step for an industry that has mostly used synthetic audiences for planning and research — AI “focus groups,” custom personas — rather than as the basis for live targeting. The usual worry is that LLM-enabled systems might hallucinate, or that thin underlying data might send buyers confidently into a ditch.
Forrester analyst Jay Pattisall put the caveat plainly: “Synthetic audiences are only as good as the data that compiles them.” He also pointed to WPP’s InfoSum capabilities as key for matching client first-party data with third-party signals to enrich and model the segments.
Read more at Digiday.