Analytics Engineering Overview

• Describe the analyst-to-engineer data flow and where the gaps are
• Explain ELT vs ETL and why warehouses made transformations shift left
• List the practices that make a warehouse dataset trustworthy

• Untested transformations make every analyst an investigator; tested models make them analysts again.
• One well-modeled dataset serves a hundred dashboards instead of a hundred copy-paste queries.
• Documentation and lineage turn a warehouse into an on-ramp for the whole team.

Analytics engineering is the kitchen prep — washing, measuring, labeling, and refrigerating — so the chefs (analysts) can cook instead of forage.

• Raw data lands in a warehouse staging layer as close to source as possible.
• Staging models lightly clean: type coercion, deduplication, renaming.
• Marts and intermediate models reshape data for business questions.
• Every model gets tests, docs, and lineage; the semantic layer exposes final business metrics.

• dbt defines models in SQL, runs them in the warehouse, and tests each one on every run.
• Pipeline reviews catch duplicated logic before it spreads to four dashboards.
• Data docs include lineage so a metric change is traced to its source instead of argued about.

• Version control every model and review them like code.
• Keep staging close to source; do not mix business logic into loading steps.
• Test what changes: freshness, uniqueness, not-null, and relationship integrity.

• Defining the same metric in five transformations, then watching five numbers disagree.
• Transforming directly from source tables so a logic change rewrites everything downstream.
• Documenting only the final report and losing the reasoning that produced it.

• Two dashboards disagreeing is usually a duplicated-model problem — trace both to their staging parents.
• If a model 'randomly' broke, check freshness first, then upstream schema changes.