Data Science Jobs 2026: Seven Roles, One Wrong Move
Data science split into seven roles with pay from $115K to $245K. Here's how to pick the right lane before you apply.
The worst career advice in data science right now is "become a data scientist." In 2026, that title covers everything from dashboard plumbing to training foundation models — and the pay gap between those poles is roughly $130K.
Levels.fyi and Robert Half's 2026 guide put median U.S. base pay for the seven distinct roles between $115K and $245K. Same field. Same job boards. Very different outcomes.
The seven roles, sorted by what they actually do
| Role | Median base | Core work | |---|---|---| | Analytics engineer | ~$115K | dbt models, warehouse hygiene, metrics layers | | Decision scientist | ~$130K | Experiments, causal inference, product bets | | MLOps engineer | ~$160K | Pipelines, serving, monitoring, cost control | | ML engineer | ~$185K | Model training and production systems | | GenAI engineer | ~$190K | RAG, evals, agent orchestration | | Applied scientist | ~$210K | Research translated into shipped products | | Research scientist | ~$245K | New methods, papers, foundation model work |
The pattern: pay tracks proximity to production models and ambiguity. Analytics engineering is essential and stable, but it's a data infrastructure job wearing a data science name. Research science pays most and hires least.
The trap in the middle
Most job seekers aim for "ML engineer" because it sounds like the safe center. It's the most contested lane. You're competing against people with three years of production Python and a portfolio of deployed models — not Kaggle notebooks.
Meanwhile, GenAI engineer roles are growing faster than qualified applicants, and MLOps remains chronically undersupplied. Both pay above $160K and reward skills you can demonstrate in a weekend project.
Pick a lane in one week
Do this before your next application:
- Audit your last three projects. If they were SQL and dashboards, you're an analytics engineer with a data science resume. Own it or change it.
- Check the eval gap. For GenAI roles, build one RAG system with a real evaluation harness — retrieval metrics, hallucination checks, cost per query. That single artifact beats five certificates.
- For MLOps, ship monitoring. Deploy a model, then instrument drift detection and latency alerts. Hiring managers care more about your rollback plan than your accuracy score.
- Read ten job posts per lane and count required skills. Apply only where you match six of eight.
The field didn't get harder. It got specific. The people landing $180K offers in 2026 aren't generalists — they're specialists who picked a lane and built proof. Pick yours this week, ship one artifact, and let the title follow the work.