Data analyst

A vendor walkthrough and reference implementation, not a report of use at any company — no employer is named as running it. Recorded as capability, which on this site never moves an assessment. What each task link rests on: the template's own instructions cover table discovery, read-only SQL and explaining the findings (writing queries); its prerequisites and step 3 require the team to supply metric definitions, trusted filters and previously reviewed queries, and step 9 saves a correction only when a person asks for it (owning the definitions the tools rely on); and the filters it prints as examples — exclude employees and test accounts, exclude the current incomplete day, mobile checkout tracking changed on 18 August — are an analyst's knowledge of where the data lies, supplied by a person rather than found by the agent. Dated from the cookbook commit that added the example, because the showcase page carries no date.

A demo, benchmark or paper shows the task can be done. Updates what the technology can do — not what employers will do.

Written by the authors of the benchmark, who also have a competing system (Rubicon) to promote — so the interested party here is arguing that the easy benchmarks are wrong. The underlying claim is checkable: Beaver is built from real query logs at MIT's 1,400-table Oracle warehouse and three others, and its leaderboard is public. The four reasons given are structural rather than about model quality — public benchmark data is in the training corpus, real schemas rot into six different columns named 'salary', warehouses carry local idiom, and real queries join two or three tables rather than one.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

The UK's statutory statistics regulator reporting on the official statistics system. It says producers who use AI in the production or communication of official statistics should apply the Code of Practice for Statistics as they would for other statistical methods or models, that its updated guidance will cover quality assurance, human oversight and clear accountability for outputs, and that AI use in official statistics remains mainly research, testing and administrative support rather than routine production. It concerns government statisticians rather than business analysts, and the guidance is still being updated.

Failure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.

A paper by LinkedIn staff about their internal text-to-SQL chatbot, available in LinkedIn's querying platform since July 2024 with steady use by over 300 weekly active users. It relies on a knowledge graph in which dataset owners annotate whether a column is a metric, dimension or attribute and popular tables have human-authored descriptions; these components raised answers scoring as correct or nearly correct from 9% to 48%. In an expert review of 124 production answers, 53% were rated correct or with minor issues, and the most common issue was an incorrect filter (24%); the authors note that a request for the latest click-through rate means different things to different teams. It is one company's internal tool, measured by its builders.

An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.

Grab engineers describing their own internal data tools. Only 20% of the most frequently queried tables had documentation; the team pre-populated AI-generated documentation for the most critical tables and notified data producers to review it, showing it with an 'AI-generated' tag until a producer accepted or edited it, and coverage for those tables rose by 70 percentage points to about 90% before the LLM discovery bot was launched. The bot runs on a vendor platform Grab uses; the figures are the company's own.

An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.

One company, limited release; Uber's platform runs about 1.2 million interactive queries a month, so this is a small share. The post itself flags hallucinated tables and columns and prompt quality as open problems.

An employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.

Entering on SQL skill alone is weaker than it was, because that is the part the tools do. The stronger entry is domain plus data: learn one business area deeply enough to know what its numbers mean and how they break, and use the tools for the querying. Volunteer for the unglamorous work of documenting definitions — it is where the new analyst role is forming, and almost nobody wants it yet.

If most of your week is fulfilling report requests, expect that queue to shrink as requesters self-serve, and expect to be measured on what decisions your work changed rather than how many dashboards you shipped. Your accumulated knowledge of where the data breaks is the asset. Turn it into the semantic layer and the definitions, and you become the reason the self-serve tools give correct answers.

Reshape the role

From answering questions to owning definitions

Self-serve analytics moves the bottleneck from query-writing to whether the metrics underneath are right. The vendors say so themselves: the agent template OpenAI publishes for this job lists the metric definitions, the trusted filters and the previously reviewed queries among the things the team has to supply before it can start. Owning that layer is more durable and more senior than owning a dashboard.

Adjacent move

Analytics engineering

Building the models and pipelines that clean data upstream is where much analyst effort is moving, and it pays better than reporting.

Adjacent move

Embedded analyst in a business function

Finance, operations and product teams increasingly want their own analyst who knows the domain rather than a central service that takes tickets. The domain knowledge is the moat.

Yes, but for a different reason than before. You will write less of it, and you will read and check much more of it. Understanding what a query does is what lets you catch the confident wrong answer the tool produced from an ambiguous schema — and that catch is now the job. Learn SQL the way an editor knows grammar: not to produce every sentence, but to know instantly when one is wrong.

The signal is who runs the query. When managers start answering their own questions in a self-service tool instead of sending them to you, the change has arrived — and the first thing you lose is not the job but the daily contact with the data that made you able to spot when it is wrong.

Writing queries and building dashboards — the task marked automating here. Framing the question and knowing when the data is lying are not, but both depend on being close enough to the work to have the context, and self-service moves you further from it.

It is real work and this site marks it as an emerging task. It is also one job per organisation and smaller than answering questions was. Calling headcount 'redirected' into it is a polite way of describing a smaller team doing different work.