Why we built gatlas

Every portfolio company keeps its books somewhere, sells somewhere else, and measures itself in a third place. Multiply that by ten companies and the picture a CFO actually needs lives in forty disconnected tools, none of which talk to each other.

The usual answer is a quarter-end scramble: export spreadsheets from each system, paste them into a master file, reconcile by hand, and hope nobody changed a chart of accounts. It works, barely, and it tells you what happened two weeks ago.

We thought the shape of the problem was wrong. Data shouldn't be something you assemble once a quarter. It should be something you can ask — at any moment, in plain language, and trust the answer.

Three things had to be true

Operators connect their own tools. A holding company shouldn't be collecting and storing the passwords of every company it owns. So onboarding is a setup link: the operator connects their Shopify, their accounting system, their CRM — and we never see a credential.

Every entity stays its own entity. No prefixing tables and hoping. Each company gets its own schema, its own pipelines, its own freshness. Consolidation happens on top, deliberately, not by accident.

The data is documented, not just dumped. Raw tables are tribal knowledge. We turn them into views with names, definitions, and the context a good analyst would give you — so the same trusted numbers serve a person and a machine.

The part we didn't expect

When we put a documented semantic layer in front of a portfolio, the most valuable consumer turned out not to be a dashboard. It was an agent.

Give an LLM raw tables and it guesses. Give it governed views with context, scoped to exactly what it's allowed to see, and it stops guessing — and starts citing. That's the gatlas we ended up building: not a prettier dashboard, but a layer your whole portfolio, and its agents, can trust.

This is an early draft of our story. We'll keep writing as we learn.