■ CRITICAL RISK ■ Technology
The job as commonly practiced — pull the data, make the chart, ship the deck — yes, and faster than most analysts want to believe. The judgment layer survives: knowing which question matters, and noticing when a clean-looking answer is wrong.
“Pandas isn't just an animal anymore. It's your replacement.”
Our AI replacement risk score — how we score jobs
Strip away the title and most data analyst work is a pipeline: gather requirements from a stakeholder who half-knows what they want, write SQL against a warehouse that's documented nowhere, clean the inevitable mess, build the dashboard in Tableau or Power BI, and present findings to people who will ask why the number differs from the one in last quarter's deck. (The answer is always a definition change nobody wrote down.) Sprinkled through: ad hoc requests, metric maintenance, and A/B test readouts.
Nearly every stage of that pipeline now has an AI directly aimed at it. Language models write competent SQL from plain-English questions, and text-to-query features are built into every major BI platform. Code assistants generate the pandas cleaning script from a description of the mess. Dashboard tools auto-summarize trends and anomalies in prose. The result isn't that analysis disappears — it's that a product manager can self-serve the answer that used to be a Jira ticket and a three-day wait. The ticket-servicing analyst, the largest single population in the field, is the one whose queue is being drained. Our risk score of 94 is aimed squarely at that version of the job.
What resists automation is everything upstream and downstream of the query. Knowing that the stakeholder's question is the wrong question. Knowing the revenue table double-counts refunds and every AI-generated query against it will be confidently wrong. Defining metrics, arbitrating between conflicting sources, translating a finding into a decision the VP will actually make. That work — sometimes called analytics engineering or decision science — is judgment about a specific business, encoded nowhere, and it's what separates the analysts who survive from those who were, functionally, a natural-language interface to SQL. The interface part now ships as a feature.
Automatability: our editorial assessment of current and near-term AI capability
Pressure is already on and compounds through this decade. Text-to-SQL and auto-generated dashboards are shipping in mainstream BI tools now, and companies are visibly slowing junior analyst hiring as self-serve improves. By around 2030 expect the report-and-dashboard tier of the job to be mostly absorbed, with remaining demand concentrated in analytics engineering, metric governance, and decision-facing senior roles. The title survives; the entry-level version of it is what's evaporating.
Riskier than the bootcamp ads suggest. The tasks that entry-level roles are built on — SQL pulls, cleaning, dashboarding — are exactly what AI tools now do well, and junior openings are tightening. The field still needs people, but it increasingly hires for judgment: metric design, experimentation, domain fluency. If you enter, plan to sprint past the report-building stage fast, because that rung of the ladder is dissolving.
Both, depending on which analyst. The version of the role that services a queue of query and dashboard requests is being replaced by self-serve AI features in BI tools. The version that defines what should be measured, catches bad data before it drives a bad decision, and argues findings in the room gets more valuable, because AI floods companies with plausible-looking numbers that someone must sanity-check.
Three moves pay off: analytics engineering (dbt, warehouse modeling, semantic layers), because AI query tools are only as good as the models underneath; experimentation and causal inference, which remain genuinely hard; and stakeholder skills — turning analysis into decisions. Also become an expert user of the AI tools themselves. The analyst who reviews and directs machine output beats the one competing with it keystroke for keystroke.
Because demand for answers isn't demand for analysts. Companies want insights, and AI collapses the labor between question and answer — a PM typing a question into a BI chatbot replaces a ticket an analyst would have serviced. Data volume grows while headcount-per-insight shrinks. Our 94 rates the traditional pull-clean-chart job description; the judgment-heavy senior work is a different, safer occupation wearing the same title.