This audio is auto-generated. Please let us know if you have feedback.
Pinterest shared an overview of the latest improvements to its ad serving mode, with the updated model now able to factor in both a user’s offsite conversion history and in-the-moment search behavior, in order to serve more personally relevant promotions.
In a post on the Pinterest Engineering blog, the team said it recently built in a user’s external conversion history to predict future interactions with advertisers and specific products.
But that only worked in retrospect, and didn’t factor in what users are searching for right now. In order to improve its ad relevance system, Pinterest’s team built in a secondary feed of real-time behavior.
This audio is auto-generated. Please let us know if you have feedback.
Pinterest shared an overview of the latest improvements to its ad serving mode, with the updated model now able to factor in both a user’s offsite conversion history and in-the-moment search behavior, in order to serve more personally relevant promotions.
In a post on the Pinterest Engineering blog, the team said it recently built in a user’s external conversion history to predict future interactions with advertisers and specific products.
But that only worked in retrospect, and didn’t factor in what users are searching for right now. In order to improve its ad relevance system, Pinterest’s team built in a secondary feed of real-time behavior.
As per Pinterest: “This approach focuses on three major areas: a new model architecture, a novel training approach, and a hybrid serving flow.”
The updated system is now able to assess more input on related pins, alongside past purchase history, in order to improve ad relevance.
“For the Related Pins surface, the context layer’s input features are derived from the subject Pin (the Pin the user is currently viewing), specifically using features like the aggregated embedding representations of the top interest categoriesof the subject Pin, weighted by their confidence scores,” Pinterest said.
In basic terms, Pinterest has integrated more data points in order to predict likely user interest in each Promoted Pin, including past conversion and site visit history, thus driving improved relevance and response.
How much of an improvement will there be?
Pinterest said its updated model has achieved a 3x to 10x improvement in ad relevance in testing.
“The median relevance of the candidates went up by ~275–300%,” Pinterest said. “On the Related Pins surface overall, the ads relevance metric improved by 1.08%. Furthermore, we observed a significant increase in candidate delivery, with 2x more ads candidates retrieved being delivered to impression.”
It’s a fairly technical overview, but essentially, Pinterest is now able to serve more relevant ads to each user, based on a broader set of factors. This will likely lead to improved performance for Pin campaigns.
Last week, Pinterest reported that it’s up to 631 million monthly actives, a year-over-year increase of 60 million people coming to the app each month. Meanwhile, improvements to its artificial intelligence-powered discovery process continue to drive better response and relevance for users.
