← Back to work

Analytics engineering · 2026

dbt E-Commerce Analytics Mart

A tested dbt project turning the real Olist e-commerce dataset into a single source of truth for revenue, customers, and product performance.

models, staging → marts
14
revenue definition everywhere
1
real Olist data — no synthetic
100%

Architecture

Select a stage to explore it.

01 · Raw sources

The real Brazilian e-commerce dataset from Olist — orders, items, payments, products, customers — loaded as raw sources with dbt source definitions and freshness context.

singular data tests — revenue consistency, no negativesGitHub Actions CI against real dataDocker — one-command Postgres + dbt

Visuals

dbt E-Commerce Analytics Mart GitHub repository with model documentation
GitHub repository — 14 dbt models with staging, intermediate, and marts layers

01 · Problem

Every team that computes revenue slightly differently eventually has three dashboards with three answers. The point of an analytics mart is one tested definition of the business — this project builds that for a real e-commerce dataset.

02 · Solution

A dbt project with a disciplined staging → intermediate → marts flow: 14 models, a single revenue macro used everywhere, explicit fact grains, and consumer marts for core, customer, and finance questions — tested end-to-end against real data in CI.

03 · Engineering decisions

Revenue lives in one macro, not in every model that needs it — change it once, every mart agrees. Singular tests encode business invariants (customer spend must reconcile across marts; revenue is never negative) rather than just column-level checks. And the whole project runs against the real Olist data in CI, because models that only pass on toy data haven't been tested.

04 · Results

A warehouse where the order-line fact, the customer dimension, and the monthly revenue series all provably agree — with lineage from raw source to Power BI exposure, and CI that would catch a definition drifting.

05 · Lessons learned

Analytics engineering is API design for data: the grain, the definitions, and the tests are the contract. dbt's real value isn't SQL templating — it's making that contract enforceable.