~/maps/dados

Career map: Data

Data is the area that grew most in roles and salary in recent years, and also the one that produces the most wrong study decisions — because the term covers three different professions. Data engineering builds the path the data travels; analytics and BI turn data into decisions; data science models and predicts. They demand different foundations.

The practical recommendation this map assumes: the first three stages are shared and apply to all three paths. Only from the fourth does the trail split — and engineering currently has the most open roles per candidate, besides being the one that best leverages a software development background.

  • 5 stages
  • 20 topics
  • 9 essential

Essential appears in practically every job ad. Recommended is what gets you past screening. Optional is specialisation — pick one instead of attempting them all.

  1. Stage 1

    Choosing the path

    By the end you know which of the three professions you are studying for — and stop spending time on the other two.

    • essential

      Builds and maintains the pipeline; it is the stage most Brazilian companies are still at, and therefore the one with the most openings.

    • essential

      Turns data into decisions; it is the most accessible entry point for people coming from business areas.

    • essential

      Models and predicts; it demands real statistical grounding and hires the fewest people without prior experience.

      in the glossary

  2. Stage 2

    The shared foundation

    By the end you can extract and join data from any relational database — the skill asked for in 100% of data roles.

    • essential

      Joins, aggregations, window functions and CTEs appear in practically every data technical test.

      in the glossary

    • essential

      Facts, dimensions and star schemas explain why a report is slow and why the number differs between two screens.

    • essential

      Pandas and file handling cover the daily work; it is the second most cited requirement after SQL.

    • recommended

      Distribution, sampling and significance prevent the wrong conclusion — the costliest error in the field, because nobody notices.

      practise

    • essential

      A loose notebook does not reach production; mature data teams work like software teams.

      in the glossary

  3. Stage 3

    From raw to trustworthy data

    By the end you can move data from a source to a destination in an automated, auditable way.

    • essential

      Whether you transform before or after loading decides the cost and flexibility of the whole architecture.

      in the glossary

    • recommended

      Scheduling is easy; the value is in backfilling, handling dependencies and knowing which load failed at three in the morning.

    • recommended

      Increasingly common in job ads because it brings testing and code review into SQL.

    • recommended

      A pipeline that delivers wrong data on time is worse than a late one — and the difference is automated testing.

  4. Stage 4

    Cloud and scale

    By the end you work in the stack that appears in most data engineering roles.

    • essential

      BigQuery, Snowflake and Redshift dominate the ads; per-query billing changes how SQL gets written.

      in the glossary

    • recommended

      Databricks and formats like Delta and Iceberg appear at companies with large volumes and platform teams.

      in the glossary

    • recommended

      Necessary when the data does not fit on one machine — and frequently unnecessary when it does.

    • optional

      The border with DevOps is thin: whoever can package and provision delivers without depending on another team.

      in the glossary

  5. Stage 5

    Delivering value

    By the end your work turns into business decisions — what separates people who get promoted from people who just fill requests.

    • recommended

      Power BI, Looker and Metabase show up in ads; what sets people apart is choosing the chart that answers the question.

    • recommended

      Funnels, cohorts and A/B tests are the shared language between data and product — and what makes the team trust the number.

    • optional

      A model that never reaches production creates no value; the hard part is rarely the algorithm.

      in the glossary

    • optional

      RAG, embeddings and inference cost control entered job ads from 2024 and pay above average.

      practise

From the map to the market

Studying without looking at what is being advertised is studying in the dark. The job panoramas are Portuguese-only for now, since the inventory is the Brazilian market — but the salary comparison works in either language.

Other trails

## faq

Frequently asked questions

Which data career has the most openings?

Data engineering. Before a model or dashboard can exist, someone has to move data from the source system to a trustworthy place — and that is the stage most Brazilian companies are still at. It also leverages a software development background best, because the work is essentially engineering.

Do I need a master’s degree to work with data?

For engineering and analytics, no — what counts is SQL, modelling and some cloud experience. For research-oriented data science, heavy statistical modelling or AI R&D, academic background still opens doors and appears explicitly in some ads. Most market roles, however, ask for delivery, not diplomas.

Can I move into data from another field?

It is one of the most common transitions, and prior business knowledge becomes a real advantage: someone who already understands the sales, logistics or finance process asks better questions of the data. The shortest path is usually to start with analytics and BI inside your own company, using data you already know, and move to engineering later if you want more technical depth.

Should I learn Spark right at the start?

No. Spark solves the problem of data that does not fit on one machine, and most companies do not have that volume — using distributed processing on small data adds complexity without gain. Well-written SQL and a cloud warehouse cover most cases; leave Spark for when the problem actually shows up.