Atlas ad-data intelligence layer uses JEV to extract creative attributes
It turns ad creative descriptions into structured attributes, links them to CPA and ROAS, and outputs specific calls on which asset to change and which angle to test.
🌶️ Finally showing the thing I’ve been building for months 🔥 Atlas is an intelligence layer on top of our ad data that gives LLMs the context to make useful decisions: what we advertised, how the creative was built, what happened, and whether the results justify doing more of it. It connects angles, personas, hooks, formats, and layouts to purchases, CPA, ROAS, and offer economics. So when an account starts declining, we can investigate which products and tests are dragging it down and which are carrying it. We’re also using JEV to extract structured attributes from creative descriptions, making more of that creative context measurable. The output goes beyond “make more winning ads.” It can name the static to vary, the video to clone, or the angle to test and produce the actual skill calls with parameters. The part I wanted all along: our next creative batch should learn from the money we already spent.