Course contents

The Analytics Engineer’s Field GuideContextThe big picture

The big picture

Last updated: October 2026

The Analytics Engineer’s Field Guide

dbt, Snowflake and Airflow

Three jobs make an analytics engineer, three tools answer them, and three engineers in this guide each live one of them.

34 lessons, each ending with a quiz. About 4 hours of reading, 2 to 12 minutes a lesson.

Developed by Ismail Mezzour. Published by DataFrost Consulting.

Part 1 · Context

What this guide covers, and who it follows

Three engineers, three jobs that fit together: transform the data, run it on schedule, and make the warehouse fast and cheap.

· · ·

An analytics engineer turns raw data into tables the business can trust, and keeps them running. Three things make that real, and the guide is built in that order:

  • Transform the data with dbt, so a SQL query becomes a tested, documented, dependency-aware table.
  • Run it on schedule with Airflow, so the whole pipeline refreshes in order, retries on failure, and tells you when something breaks.
  • Make it fast and cheap by tuning Snowflake, so all of that work reads less data and burns fewer credits.

Each one follows a person with that problem.

How to use it: read the parts in order, because each one builds on the last. Every lesson ends with a three-question quiz, and your progress is saved automatically. Already know some of it and want a reference? Every lesson is open, and a lesson counts as read once you reach the end of its text or click Next.

What you need, and which versions
A Snowflake account (the free trial works, pick Enterprise edition to try every feature in part 4), Python for dbt, and Docker for the local Airflow in part 3. The examples target dbt Core 1.10+ (where a feature is newer or older, the lesson says so, e.g. "dbt 1.9+"), Airflow 3.x, and Snowflake as of October 2026. Older versions are flagged where the syntax differs.

The three engineers

A
Alice
Analytics Engineer
"I can write SQL. Turning it into reliable, repeatable tables is the messy part."
T
Thomas
Data Engineer
"Alice's tables are built. My job is the other half: every morning at 06:00, in the right order, retried on a blip, loud on a real failure. No one should have to remember to press a button."
I
Ismail
Data Platform Engineer
"The pipeline works, but it's slow and the bill keeps climbing. My job is making Snowflake read less and cost less."
Three jobs, one pipeline: transform the data, run it on schedule, and keep the warehouse fast and cheap. In practice one person often wears two or three of these hats. The personas split the jobs to teach them clearly, not because you need three engineers.

Problem 1: transforming data (Alice)

Alice writes a SQL query that cleans and joins raw tables. It works once, in her editor. But there's no standard way to turn it into a real table, no test to catch bad data, and no record of what depends on what. Next week she copies an old query, swaps the table names, and hopes. Knowledge lives in her head, not in the platform.

The answer: dbt
You write a SQL SELECT. dbt builds the table, tests it, documents it, and tracks what depends on what. One way to transform data, the same for every engineer.

Problem 2: running it on schedule (Thomas)

Thomas has the tables. Now they need to refresh every morning at 6am, in the right order, and someone has to know the moment one breaks. He starts with a cron job, then a second. Then a model fails at 2am and everything downstream quietly serves stale data until a manager notices at 9. Cron has no memory, no retries, no alerts.

The answer: Airflow
Airflow runs your pipelines on schedule, in the right order, retries on failure, and alerts you when something breaks. One scheduler for the whole platform.

Problem 3: a slow, expensive warehouse (Ismail)

The tables are built and they run on time, but a dashboard takes four minutes to load and the Snowflake bill doubled this quarter. Nobody wrote a bad query on purpose. The same query reads a billion rows when it only needed a million, a warehouse stays on all night doing nothing, and the same numbers get recomputed a thousand times a day. Speed and cost have one cause: work you didn't need to do.

The answer: tuning Snowflake
A handful of habits, pruning, right-sizing the warehouse, and letting the caches do their job, make queries read less and cost less. Part 4 covers how Snowflake works, the habits that make it fast and cheap, and how to run it well in production.

How it fits together: sources, Snowflake, dbt, Airflow

It's one pipeline. Raw data leaves your sources and lands in Snowflake, dbt transforms it there, inside Snowflake, into trusted tables, and those feed your BI dashboards. Airflow drives the whole run on a schedule. Follow the data along the arrows, and each engineer owns one piece.

Sourcesapps · APIs · files

Snowflake

stores the data · runs the SQL
dbt transforms the data, right here → trusted tables
BIdashboards
Airflow orchestrates the whole run, in order, on schedule, retries on failure

Ismail tunes SnowflakeAlice builds with dbtThomas runs Airflow

One sentence to hold onto
Data lands in Snowflake, dbt transforms it there into trusted tables, those feed your BI dashboards, and Airflow runs the whole thing on schedule, while Ismail keeps it fast and cheap. Learn it in that order and each piece builds on the last.

Where the analytics engineer sits

Analytics engineering is the seam between two older roles. The data engineer moves raw data into the warehouse, the analyst turns trusted tables into decisions. You own the middle, turning raw into trusted with software discipline. dbt, Airflow and Snowflake are what let one person hold that whole middle.

Data Engineer
moves raw data in
→
Analytics Engineer
raw → trusted (you)
→
Analyst
trusted → decisions

From here, the guide follows Alice as she learns dbt, then Thomas as he puts it on a schedule with Airflow, then Ismail as he makes Snowflake run all of it fast and cheaply.

Lesson quiz

  1. In this guide's pipeline, in which order do the three jobs come?

    1. Tune Snowflake first, then transform with dbt and schedule it
    2. Schedule with Airflow, transform with dbt, then tune Snowflake
    3. Transform with Airflow, schedule with dbt, then tune Snowflake
    4. Transform with dbt, schedule with Airflow, then tune Snowflake
  2. In the Sources → Snowflake → BI flow, where does the transformation SQL actually execute?

    1. In dbt's own engine, before the result is loaded into Snowflake
    2. Inside Snowflake, on a warehouse dbt sends the SQL to
    3. In the BI tool, each time a dashboard queries the raw tables
    4. On the Airflow worker, which pulls the rows and transforms them
  3. Thomas's 06:00 cron job runs the load script and then dbt. One morning the load fails, dbt runs anyway on yesterday's rows, and nobody hears about it until 09:00. What would Airflow have done differently?

    1. Rebuilt the tables itself, in place of dbt, with its own SQL
    2. Made the load faster by running it inside Snowflake instead
    3. Nothing, since run order and alerts are dbt's job, not Airflow's
    4. Held dbt until the load worked, retried it, and alerted