Advertising on the open Internet is hard

There are billions of users, tens of billions of ad impressions per day, thousands of advertisers, millions of publishers, and many intermediaries in between. Every ad decision needs to happen in milliseconds.
‍
Each ad impression is a decision and an opportunity: estimate user intent, conversion probability, creative fit, and bid economics for this advertiser and this user, in this moment.
‍
To be able to do this, AI models need rich data to understand and predict the interactions between users, ads, and content. On the open Internet, training pipelines must be engineered from disparate sources including the interaction of dozens of interdependent ecosystem components connecting millions of apps with no standard data schema, UX, or creative formats.

We have been building toward this problem since 2013

Long before the wave of AI excitement, our team was using deep learning and transformer architectures to ingest signals from multiple sources simultaneously and build rich representations of how users, apps, and creative assets interact.

This is how it works

The result is Moloco’s Compound Ad Recommendation Architecture (CARA) AI, which is purpose-built to navigate this complexity, and deliver outcomes for advertisers. CARA is a compound AI system that orchestrates and optimizes more than 30 AI models, agents, and specialized software across six technical domains. Over the last XX time period, CARA AI has driven XX performance improvements.