How Pinterest Built an AI Engine Users Actually Trust

Pinterest, recognized for its user-friendly platform focused on home décor and weekly meal ideas, underpins its simple experience with a sophisticated AI system. Monthly, the platform hosts 600 million users and processes over 500 petabytes of data, necessitating a method to recommend the ideal image from billions, all within real-time.

The greater challenge was ensuring this AI didn't merely maintain user engagement but instead pointed them toward genuine inspiration, not a goal of maximizing time spent on the app. Pinterest, since 2010, has relied on Amazon Web Services (AWS) to build trust and safety into its technology from the start. This foundation allowed the company to experiment with generative AI as it matured.

Amazon Elastic Kubernetes Service (EKS) enables Pinterest to deploy new AI features swiftly across the business. Over 10,000 EC2 G5 computing instances manage real-time recommendations, while more than 600 specialized instances train AI models daily on 18 terabytes of data. Consequently, the system delivers over 10 million AI-powered recommendations per second.

Pinterest Canvas, an original AI model, generates and edits high-resolution images, delivering polished, on-brand backgrounds for product photos for advertisers without manual design work. Visual search tools, powered by Amazon SageMaker, now identify over 2.5 billion objects in photos. A new AI assistant allows users to search and plan by voice, transitioning from a search function to a creative partner.

The strategy has yielded positive results. Pinterest reported a 17% year-over-year revenue increase and an 11% rise in monthly active users. Search fulfillment, a metric indicating how often users find what they're looking for, improved by 230 basis points. AI now drives approximately 70% of how people discover content on the platform.

Pinterest's Chief Architect, Kartik Paramasivam, emphasized that trust is not just a feature but the foundation of Pinterest. The company aims to leverage AI to filter out inappropriate content and personalize user experiences, allowing them to easily find what they want to explore. Pinterest plans to continue relying on generative AI to expand tools like Canvas and its voice assistant, while maintaining its positive-not-addictive philosophy.

This balance between business growth and user trust is noteworthy, as many platforms view responsible AI and revenue growth as a trade-off. Pinterest's success suggests that these goals can be achieved concurrently when AI is built with intent from the outset.