Metabase already knows your business rules. Your AI should read them, not guess them.

The hidden asset: your saved questions ARE the business logic

Remember the payroll platform whose AI said 100 employees instead of 80? Their own BI dashboard showed 80 all along — because whoever built that dashboard wrote WHERE status NOT IN ("Terminated","Deleted") into it years ago. The rule existed, in production, in SQL. The AI just never read it. Multiply that by every measure in your instance: revenue-after-refunds, churn cohorts, utilization rates. The semantics are already written down — in Metabase’s API, readable by machine.

This is why a 2025 study on real enterprise data found AI text-to-SQL collapsing on anything beyond simple lookups: the model is asked to reconstruct in seconds what your analysts encoded over years. It is not a fair fight, and it is not necessary.

Why regional teams love Metabase — and why that matters now

Metabase became the default BI of MENA startups for good reasons: open source, self-hosted, cheap, friendly. Most growth-stage platforms in Egypt and the Gulf run it. Which means most of them are sitting on a machine-readable library of their own business rules — the single most valuable AI asset they own, and the least used.

A self-hosted BI layer also solves half the sovereignty problem by itself: the definitions live in your building already. The only missing piece is an AI layer that reads them where they live instead of exporting everything to a cloud model.

The pattern: ingest, don’t rebuild

The smart move has three steps. Read the logic: saved questions, models, and metric definitions come out of Metabase’s API as structured data. Certify it: the business owner signs the rules that matter (headcount, revenue, utilization) — the dashboard SQL becomes the certified query, visible filter included. Answer from it: new questions route through the certified layer instead of improvisation — ten seconds, with the filter shown, instead of ten days in the BI queue or a 25%-wrong guess.

The dashboard stays. The queue shrinks. And the BI team stops being a help desk and becomes what they always were: the authors of the company’s metric layer.

DEBO connector status: P1, on the roadmap

Honest badge: the Metabase connector (saved questions and models → certified definitions) is in the first connector batch, and we will say here when it ships. Metabase-first companies are the natural early adopters — if that is you, the conversation is worth having now: early connectors get shaped by real instances.