One of the central challenges in modern marketing is no longer simply gaining access to more data, but gaining access to more usable intelligence, in more places, without forcing that data to cross boundaries it should not cross.
As identity signals continue to weaken, the industry needs a model that can preserve, and even enhance, the richness of consumer understanding while moving beyond reliance on limited individual-level identifiers. A shared embedding space for consumer intelligence offers a compelling path forward.
Embedding spaces
An embedding space turns any signal, whether text, image, video, behaviour or context, into a point whose location encodes its meaning, so related things naturally sit close together regardless of the form they arrived in. Each partner builds its own space from the patterns in its data. Aligning these independent spaces into a common frame is a hard technical problem, but it is one we know how to solve, allowing intelligence to flow between partners while the underlying data stays where it is.
A space like this can hold almost any kind of signal, from interests and behaviours to content, creative, outcomes, context, attention, geography and intent. And because meaning lives in position rather than in a shared key, these signals combine and reinforce one another without ever needing to match on an individual identifier. Advertisers, agencies, publishers and SSPs can each contribute what they know and draw on the collective picture, gaining sharper, more granular insight and stronger performance, even as traditional IDs fade away.
A shared space
This model becomes even more powerful when the same embedding space exists both centrally and at the edge, particularly on the publisher and SSP side. Central intelligence can provide strategic understanding, audience structure and planning logic, while edge intelligence applies that understanding at the point of impression, where decisions are actually made.
In this setup, publishers and SSPs do not need to send sensitive user-level data back to a central system, and central systems do not need to expose client or model intelligence in raw form. Instead, both operate within a shared semantic space, allowing impression-level signals to be interpreted with greater relevance and performance. In practice, this leads to better targeting, more effective contextual decisioning and stronger feedback loops, all without introducing the privacy risks associated with moving data across organizational boundaries.
A simple example.
An advertiser has learned what its most valuable customers look like, not as a list of individuals but as a region of the shared space defined by the interests, behaviours and contextual moments those customers have in common. That region is shared with a publisher’s edge model. When an impression becomes available, the publisher places the live context and on-site signals into the same space and simply measures how close they fall to the advertiser’s target region. A close match means the impression is a strong fit, so it can be prioritised and paired with the creative that best suits that region. No audience list changes hands and no user-level data leaves the publisher, yet the decision is as well informed as if both sides had pooled their data.
This is not a theoretical proposition for WPP. We have been developing these capabilities for many years and have products live in market today that share rich semantic consumer intelligence with publishers and SSPs, powering lossless activation with no sharing of consumers’ data.
In an agentic buying world, ensuring that the ground truth for both Buyer and Seller Agents can be trusted to be compatible and comparable increases trust and reduces inefficiency.
As the industry moves further beyond individual identifiers, this approach gives our clients and partners a durable foundation for understanding and reaching audiences, one that grows richer with every signal it encodes.
