Delivering ad relevance without third-party cookies: advanced techniques for modeling audiences

While a distant “maybe” in 2017, we’re past due to shift away from cookies and towards new, more effective solutions to deliver brands’ messages to consumers. It doesn't matter when—in 2025 or even later—virtually the entire web won't be tied to third-party cookies, and preparation for the shift is well underway.

This development has compelled advertisers and ad tech companies to evolve their strategies for engaging with audiences in meaningful ways. These new methods include strategic investments in clean rooms, browser-based solutions, and leveraging first-party data. Additionally, well-established solutions like contextual targeting and modeled solutions have regained traction with advertisers for their ability to deliver relevant messages to customers.

This technical white paper digs into the motivations and methodologies behind modeled audiences. It covers technical aspects inclusive of model architecture, audience clustering, hierarchical thresholding, and challenges, such as training cookie bias and domain adaptation, highlighting the innovative solutions considered by the Amazon Ads team.