Analytics as code, from metric to MCP

Background When analytics becomes the bottleneck

Growth like Framer's means everyone from product to finance to sales is asking questions about the numbers every day, and a small analytics team can quickly become the limiting factor on decisions.

Approach The model, not the dashboards

A company whose whole product is letting teams ship websites without waiting on developers understood the shift immediately. Instead of building dashboards, Framer's analytics engineers put their time into the model underneath:

• How every metric is calculated
• Which dimensions it can be cut by
• How the tables join
• Where targets and forecasts live
• The context that makes a number mean the same thing to everyone

The rest of the company builds on top of it: exploring, drilling down, creating their own reports and asking questions in natural language.

Setup Business logic in code

That model is Framer's entire business logic, and it lives in code. Every metric, dimension, join path, permission and piece of context is version-controlled and synced to Steep on every push, so it evolves with the business week by week.

The same semantic layer answers Framer's AI agents through Steep's MCP server, so whoever, or whatever, asks gets the same answer.