Five kinds of analytics careers, and what actually separates them

Titles lie. A data analyst at a 12-person startup and a data analyst at a bank can share almost no daily tasks. One writes SQL for a marketing team; the other maintains dbt models that forty people depend on.

So this page sorts five kinds of analytics careers by work, not by title: the product analyst, the analytics engineer, the data platform engineer, the product data scientist and the BI developer. Each comes with the named tools and employers that hire for it.

What to take away

The five analytics careers, with the employers and tools behind each

Product analyst. You sit inside sales, marketing, product or finance and answer that function's questions. The work is framing: turning "why did signups drop" into something the data can settle. Meta, Spotify and HubSpot hire heavily for this, and their analysts sit in the business unit rather than a central data team.

Amplitude and Mixpanel supply the product event data those teams argue over. The stack is usually SQL plus a BI tool the function already uses. What separates people is getting someone to act, not writing the query.

Analytics engineer. You turn raw source tables into tested, documented models everyone else builds on. This is software engineering applied to data. The named tools are dbt from dbt Labs, LookML inside Looker at Google Cloud, and SQLMesh from Tobiko Data.

dbt Core is free and open source. dbt Cloud adds scheduling and a browser editor on per-seat plans, and dbt Labs sells an Analytics Engineering Certification. Airbnb, GitLab and Monzo built this layer early, and the title now appears on its own ladders.

The satisfaction is structural: you make other people faster. The frustration is that your best work is invisible when it goes well.

Data platform engineer. You own ingestion, storage, orchestration and the infrastructure everything else runs on. Named systems: Snowflake, Databricks and BigQuery for storage and compute. Kafka, Airflow and Dagster for moving and scheduling the work.

Snowflake bills compute in credits, typically around $2 to $4 per credit by edition and region, with storage charged separately. Databricks bills in DBUs on a similar contract shape. Netflix, Airbnb and Shopify publish engineering write-ups on this kind of work.

Expect more code, less meeting time and an on-call rotation. Failures are loud and time-sensitive. This suits people who can hold a distributed pipeline in their head.

Product data scientist. Experiment design, causal inference, forecasting, segmentation. Named tools are R, Python, and platforms like Statsig or Optimizely for running tests. Netflix, Booking.com and Duolingo run this as a real function, mostly on experimentation systems they built themselves.

Less of this exists than people expect. In many companies it is a fraction of a job rather than a whole one, and most of the day is still data preparation.

BI developer. Dashboards, semantic models, self-service enablement, training people to answer their own questions. Named tools: Tableau, Power BI, Looker and Metabase.

Amazon runs a formal version of this as the Business Intelligence Engineer, a ladder separate from its data science track. Accenture and Deloitte staff the same work on client projects.

Licenses cost real money. Power BI Pro has listed around $10 to $14 per user per month, and Tableau list prices have run about $15 per user per month for Viewer, $42 for Explorer and $75 for Creator.

Metabase's open-source edition is free to host yourself, with paid cloud plans above it. Microsoft's PL-300 exam certifies Power BI analysts, and Tableau offers the Desktop Specialist exam.

It looks like tool work from outside and is mostly product work: watching how people actually behave, designing for that, and retiring what nobody opens.

Most real jobs are two or three of these mixed. Ask which mix during interviews, because the posting will not say.

What actually differs between them

Look down the "failure looks like" row before choosing. You will spend your career preventing that specific thing, so pick one whose prevention you find interesting rather than tedious.

Moving between them

Each move has a specific bridge, and the bridge is rarely a course.

Product analyst to analytics engineer. You can already write SQL. The gap is engineering practice: learn version control properly, get your work reviewed by an engineer, and write tests for logic you already produce. Volunteer to move one piece of dashboard logic upstream into a shared dbt model. That project is the portfolio piece.

Analytics engineer to data platform engineer. Take on the orchestration for models you own, then the on-call for them. Learn how your storage layer charges and performs. The transition is mostly about accepting pager duty.

Product analyst to product data scientist. Practice experiment design inside your current role. Design a proper holdout for something your team is about to launch, write the analysis plan before the data exists, and follow through when the result is inconvenient. Doing this once well beats a certificate.

Anything to BI developer. Watch people use what you built, in silence, without helping. Then rebuild it. This work is judged on adoption, and adoption comes from that loop.

Individual contributor to management. This is a change of profession, not a promotion. Your output becomes other people's output. The tell: do you get more satisfaction from a colleague's good week than from your own good analysis?

Into the field from outside. People arriving from finance, operations or support already know a domain and its questions. Start by being the person on your current team who answers data questions, then move.

How to read a posting

Postings are written by committee and describe a wish, not a job. Useful signals:

A posting that names dbt, Snowflake and Tableau in the same breath is describing two or three of these jobs at once.

Questions worth asking in an interview

Ask the same questions of two interviewers. Divergent answers are more informative than either answer alone.

What to build if you are trying to move

Nobody is impressed by an analysis of a famous public dataset that thousands of people have already picked over. What persuades:

One project with real depth beats five tutorials. Interviewers probe for depth immediately, and breadth without it is obvious within two questions.

Signs a role is mis-scoped before you take it

Where the rest of the picture sits

Certifications come up constantly in this conversation, and the honest test is whether the exam maps to work you will actually do. For the daily craft most of these roles share, see data analysis.

The technical grounding all five rest on is in analytics foundations, and the function most of them sit inside is described in business intelligence. Tooling questions that dominate early careers are in buying analytics tooling, and the definitional work senior roles gravitate toward is in data governance.

For duties and outlook described by a consistent method rather than by commentary, the US Bureau of Labor Statistics entry for data scientists is a better starting point than any article. It reports national medians, not your market, so treat it as a shape rather than an answer.

Common questions

Do I need a specific degree?

Teams vary enormously in how much they weight this, and the weight drops as you accumulate demonstrable work. Quantitative research roles are the most likely to hold a formal expectation. Ask directly in a first conversation rather than guessing.

Is SQL still the core skill?

For everything except pure platform work, yes. It is the common language across all five kinds, and depth in it, including window functions and query performance, pays back more than breadth across tools.

Should I specialize or stay general?

Generalists do well in small teams, where variety is the job. Specialists do well in large ones, where depth is the job. Neither is permanent, and moving between company sizes is the most common way people switch.

How do I know when to leave?

When you have stopped learning and the constraint is the organization rather than your effort: no appetite for the questions you can answer, no path to more scope, no interest in changing either. Those conditions rarely improve by waiting.