Predactiv today introduced the Predactiv People Model, a new class of AI foundation model built to understand people. While large language models learn from text, and world models learn from images and videos, the Predactiv People Model learns from people’s behavior to uncover the interests, habits, and motivations behind their actions. It is the only model of its kind available to marketers, and it is available today through The Predactiv Data Platform, Platform API, and MCP Server.
The Predactiv People Model solves a problem that language models cannot. Since text has an order, a language model can predict the next word. Human behavior, though, has no order. A purchase on Monday, an article on Thursday, a show last night, and a social post from last month say very little when read in sequence. The connector of these events is the person behind them.
“A language model cannot tell you who reads a web page. It learned the page, not the people who visited it. While language models are a transformer architecture, they learn patterns and stable forces resulting from ordered text,” said Huanjin Chen, Chief Technology Officer at Predactiv. “The People Model is different. It connects behaviors such as shopping habits and brand affinities to a psychographic profile, the most stable and predictive intelligence layer.”
The People Model is distinctive both because of the breadth and quality of the data it uses, as well as the differentiated training process of the model. It first learns the behavior patterns from user activities, then goes a layer deeper to learn what connects those behaviors – the psychographic profiles, which are more stable. This results in a highly precise predictive model.
A new category of AI model
All three model types (language, world, and people) are different flavors of transformer models. A transformer finds patterns in what it is given, and what it is given decides what the model becomes. The math is not what separates the models. The data is. Until now, no company had both the differentiated technology and the scale of open-web data that a people foundation model requires. Predactiv changed that.
The Predactiv People Model uses a hierarchical architecture inspired by the organization of human memory. It combines an event model that reads individual actions, a behavior model that finds the patterns within each type of activity, and a psychographic model that connects them into a single picture of the person. Each layer learns from the others, so the model can infer how a person is likely to behave in one setting from what it knows about them in another.
“The industry has spent two years asking language models to do a job they were never built for,” said Dana Hayes Jr., CEO of Predactiv. “Language models read text. The People Model reads audiences and can predict behaviors with unmatched accuracy. We built it because our data was the perfect input to a differentiated model like this. And, because our clients deserve the highest performing audiences, unique to them.”
Availability
The Predactiv People Model powers The Predactiv Data Platform and is available today through the platform, the Predactiv Platform API, and the Predactiv MCP Server, which lets teams build, size, and activate audiences from inside ChatGPT, Claude, and other AI tools.
Predactiv executives will be at Advertising Week New York, October 5 – 8, 2026. Join our panel session October 5 at 4:10 pm on the Tech Stage. Michael Gorman will be joined by leaders at Yahoo DSP, Nexxen and Madhive.
Learn more at https://predactiv.com/people-model/
