Research & Papers

Pinterest's New AI Picks and Resizes the Ads You See

Your Pinterest feed is about to get better at selling you things.

Deep Dive

Pinterest built PinDCO, a production system for ad creative retrieval and selection on its billion-scale visual discovery platform. At its core is a Creative Component Fusion Network, which scores each ad component — image, title, layout — with its own dedicated tower and component-specific hyperparameters, then fuses them into a creative-level score conditioned on the ad-level prediction. A Pixel-aware Adjustment Module adjusts scores based on creative size, since Pinterest's waterfall grid layout means a creative's rendered size affects nearby content and session-level engagement, encouraging efficient screen real-estate use. A lightweight pre-selection model prunes candidates early. Offline analyses and online A/B tests showed a +3.09% lift in ad click-through rate with positive whole-page metrics, and Pinterest launched PinDCO in its Ads platform.

Key Points
  • Pinterest's new AI scores each piece of an ad separately — the image, the headline, the layout — then picks the best combination for you.
  • Live tests showed a 3.09% jump in ad clicks, and the system is already running on Pinterest's real ad platform.
  • It also resizes ads to fit Pinterest's staggered grid, so fewer ads get squished or crowd out the pins you actually want.

Why It Matters

Ads you scroll past will get better at tempting you to buy — great for Pinterest, costly for your wallet.

📬 Get the top 10 AI stories daily