The Model Was the Easy Part. Keeping It Real Was Not.
Ask five people on a data team what "customer" means, and you'll get five different answers. For years, the tools meant to settle that argument just drew pictures instead: a diagram that never checks itself against what the warehouse actually builds.
Our founders had spent thirty years watching that gap widen, project after project. Every handoff between the diagram, the glossary, and the physical model was a place the model quietly drifted from what was real.
So in 2019 we built something different: one canvas where business and data people model the same thing together, instead of passing sketches between disconnected tools. Not because collaboration looks good on a slide. Because that exact handoff was where our own projects had failed before, and we were done watching it happen.
That instinct is still the whole company. Now that AI agents are the ones reading the model, we call Ellie the backbone for a universal semantic layer, not the layer itself. We're not claiming to solve governance or runtime orchestration; we're the part that keeps the model your agents trust the same one your team actually built and agreed on.
Sixty-plus enterprise teams have modeled their business this way since. Every one of them started the way we did: two teams, one disagreement, and no single place to settle it.
