The semantic layer is your API for analytics

Why the metrics layer is the most underrated piece of your data platform — and how to design one that scales.

AOAisha Oyelaran·Apr 21, 2026· 10 min read

If your BI tool is a website, your semantic layer is the API that powers it. And, like any good API, it deserves to be designed, versioned, and documented — not accreted.

The teams that treat their semantic layer as a product ship faster, argue less, and onboard new analysts in days instead of weeks. The teams that don't end up with three definitions of 'active customer' and a permanent slack thread arguing about which is right.

Design principles we hold to: metrics are singular and canonical; dimensions are shared; time grains are explicit; ownership is named. If you can't answer 'who owns this metric' in one sentence, you don't have a metric — you have a formula.

The tooling matters less than the discipline. dbt's semantic layer, Cube, LookML, or MetricFlow will all work if you treat them as first-class product surfaces. None of them will save you if you don't.

AO
Written by
Aisha Oyelaran
Analytics Engineering Lead

Aisha specializes in semantic layers, dbt architecture, and getting analytics organizations to actually ship.

More from Aisha

Get the next essay in your inbox.

One well-considered read from our practitioners, every other Friday.

Ready to amplify your data?

Book a 30-minute strategy call. We'll map your data landscape and identify the three highest-leverage moves.

See case studies