dbt Analytics Engineering case studies
dbt Analytics Engineering · Put your knowledge to work in context.
This page is an independent study guide for the dbt Analytics Engineering certification. It is written for analytics engineers, BI developers, and data practitioners who build, test, document, and deploy transformation projects with dbt, and who want a structured way to prepare for a certification exam in that space. The dbt Analytics Engineering case studies are 5 cases and about 4,959 words.
1Duplicated order lines and a nightly full refresh that never ends
Orbit Retail runs a 210M-row fct order lines model on Snowflake, materialized as incremental with unique key='order line id' and the default merge strategy. The source orders table receives status updates for up to 10 days after creation, and refund rows land up to 30 days late.
1,007 words, about 5 min2Alert fatigue, a silent column rename, and a build blocked by freshness
Meridian Health's claims dbt project declares 22 sources and 340 data tests. Six sources have freshness checks with warn after: {count: 12, period: hour} and error after: {count: 24, period: hour} .
974 words, about 5 min3Cross-team refs, PII lineage, and a docs gate that has to fit in CI
Northwind Logistics consolidated three teams into one 780-model dbt repo six months ago. Finance, Ops, and Growth each own part of the DAG; there is no groups: configuration, so any model can ref any other, and analysts regularly build dashboards on int models that a team later renames or drops.
960 words, about 5 min4Fifty-minute CI, colliding jobs, and an append model counted twice
Cobalt SaaS runs dbt Cloud with four jobs: nightly-prod at 02:00 UTC ( dbt build ), hourly-metrics on the hour, six-hourly , and ci on every pull request. CI runs dbt build across the whole project and takes 50 minutes, so reviewers merge without waiting for it.
970 words, about 5 min5Backdated corrections, hard deletes, and a point-in-time join
Kestrel Telecom maintains snap subscriptions as a dbt snapshot over a daily feed of plan records, using the timestamp strategy with updated at and a 04:00 UTC run. Two problems surfaced this quarter.
1,048 words, about 5 min