Nine postings. One shift.
I read nine sports analytics postings from October and November 2026, across baseball, soccer, American football, motorsport and live entertainment. I wrote down what each one asks for, in its own words.
The job has stopped being "data science." It has become data engineering plus applied LLM systems.
Five of the nine are platform or engineering roles, not analysis roles. And the two best-paid roles are both about building the infrastructure that AI runs on.
What they ask for, by how often it appears
LLMs or generative AI in the working process: 6 of 9. Not as a research topic. As part of how the job gets done. A baseball club's research role asks for "demonstrated ability to use LLMs and generative AI in the research process". Another baseball club describes the daily work as "some blend of manual coding and AI-assisted development".
Pipelines and ETL: 5 of 9. Airflow, dbt, Snowflake, API ingestion, scheduled jobs. Two American football clubs, a motorsport series, a baseball club and a live-entertainment platform.
Data quality, validation and audit: 4 of 9. In one baseball analyst role, the entire first phase is auditing data sources and building validation checks, before anyone models anything.
Tools non-technical people can actually use: 4 of 9. The same baseball role puts it best: "make this data self-serve for non-technical stakeholders, not locked in a notebook only you can run."
Governance, lineage and observability: 4 of 9.
Retrieval, RAG and vector search: 2 of 9. Both are senior roles. An American football club wants datasets prepared "for Generative AI, machine learning, semantic search, and Retrieval-Augmented Generation". A soccer league office wants RAG systems grounded in a semantic layer.
Agentic infrastructure: 2 of 9. The live-entertainment platform wants someone with "hands-on experience with Agentic AI technologies, specifically building infrastructure where autonomous agents and AI models can learn, access data, and execute tasks." That's the job title, not a nice-to-have.
The four things nobody is teaching
This is the interesting part. Almost everything above is taught somewhere, by someone. These four aren't.
1. Putting an agent inside an application real people use. There's plenty of material on calling an API. There's almost nothing on what happens to an interface when one part of it is slow, non-deterministic and sometimes wrong. Where does state live? What does the user look at while it thinks? What happens when it fails halfway?
2. Knowing whether the output is any good. The best-paid role in the set asks for "evaluation frameworks for retrieval quality and LLM output correctness" and "guardrails for explainability, traceability and data attribution". There's almost no practical teaching on it anywhere.
3. Auditing a dataset you inherited. Everyone teaches modeling on clean data. Nobody teaches the first six weeks of a real job: working out what you have, whether it's right, and what's missing.
4. Getting a decision maker to act on the result. A senior soccer analytics role spends more words on communicating findings to senior stakeholders than on any modeling technique.
What I take from it
The postings aren't asking for someone who can fit a model. They're asking for someone who can build the system around the model, prove it's right, and get a person to use it.
That's a different job. It's worth training for on purpose.
The nine roles
- Baseball (three clubs): research, international strategy and performance roles
- Soccer (a league office):
- Senior Manager, Soccer Analytics and Data Science
- Principal AI/ML Engineer, Semantic Data
- American football (two clubs):
- Data Engineer
- AI and Data Platform Engineer
- Motorsport: Data Engineer
- Live entertainment (a European platform): Data Engineering Lead, Agentic Infrastructure
Quotes are verbatim from the postings.
Michael Czahor
Founder, AthlyticZ