Ask a modern text-to-SQL model for "monthly active users by plan tier, broken out by acquisition channel, for the last six months" and you'll get a clean query back in under ten seconds. Five years ago, that same request was a half-day ticket for a junior analyst. If the job was just writing SQL, data analysts would be in serious trouble right now. They're not — but the job is changing shape fast.
What AI has already automated
The mechanical layer of analytics is essentially solved. Tools like Hex Magic, Mode's AI assistant, Snowflake Cortex, and ThoughtSpot Sage all handle:
- Translating natural language questions into SQL against a schema they've been given
- Generating starter dashboards from a table
- Writing narrative summaries of a metric change
- Flagging anomalies in a time series
- Cleaning column names, casting types, basic transformations
Any analyst whose week was dominated by "can you pull me a report on X" is watching their job description evaporate. Stack Overflow's 2025 developer survey found 68% of data professionals now use AI assistants daily for query writing, up from 24% in 2023.
But Coursera's 2026 workforce report still shows data analyst roles projected to grow 23% through 2030 — well above average. The BLS projects 36,000 net new analyst jobs by 2032. Something doesn't add up unless you look at what the work actually is.
What AI can't do (and what that leaves you)
Three parts of analytics remain deeply human, and they're the parts that produce real value.
Defining the right question. A stakeholder asks "why is churn up?" The correct response is rarely to pull a churn dashboard. It's usually: which cohort, over what window, compared to what baseline, and what decision will the answer drive? An analyst who skips this step produces the query the stakeholder asked for instead of the answer they needed. AI does not yet push back on bad questions; it happily generates bad answers to them.
Judging what's real. A number moved. Is it signal or noise? Is the data source reliable right now? Did a tracking change two weeks ago corrupt the baseline? Did a sales campaign inflate a metric for reasons unrelated to product? These are judgment calls based on institutional knowledge an LLM doesn't have access to and can't build from a schema alone.
Communicating results. The highest-paid analysts are the ones whose insights actually change decisions. That means framing a finding, anticipating objections, connecting it to business context, and telling a story that lands with an exec who has 15 minutes before their next meeting. AI can draft the slide. It cannot read the room.
The salary split is widening
Levels.fyi data shows a stark bifurcation. "SQL monkey" analyst roles — dashboard builders at non-technical companies — are seeing flat comp and shrinking headcount. Analytics engineers and senior analysts who own data models, drive experimentation programs, and advise on strategy are seeing 12–20% comp growth year-over-year.
The skills premium in 2026 belongs to analysts who can: build and maintain semantic layers (dbt, Malloy), design and run experiments rigorously, work fluently with AI tools while being critical of their outputs, and translate between business and data teams.
Five things to do this quarter
-
Move up the stack. If your week is 80% ad-hoc queries, push actively toward building reusable data models, owning a metric definition, or running an experiment end-to-end. The tickets-per-day treadmill is the most automatable version of the role.
-
Learn experimentation properly. Not A/B test basics — actual power analysis, sequential testing, CUPED variance reduction, and how to avoid peeking. Experimentation is the single highest-leverage thing an analyst can own, and AI tools still produce statistically naive results by default.
-
Get fluent with AI, then skeptical. Use the tools daily for boilerplate queries and first-pass summaries. But always verify the joins, check the filters, and sanity-check the numbers. The analysts who blindly trust AI output are the ones producing the wrong answer fast.
-
Own a business outcome. Pick one metric that matters to the company. Make yourself the person who knows more about it than anyone else. Track its movements, understand its drivers, and drive changes that move it.
-
Write like you care. Analysts who communicate clearly get promoted. Analysts who dump a dashboard and walk away do not. Practice writing executive summaries that land in three sentences.
What this means for your applications
Resumes in 2026 should lead with outcomes: experiments run, metrics moved, decisions informed. The data scientist resume example works for senior analyst roles too — the framing is the same. Pair it with a data scientist cover letter example that connects your work to real business outcomes.
For interview prep, the data scientist interview questions guide covers both the SQL fluency screens and the case-style stakeholder questions that actually decide hires. Comp ranges are in the data scientist salary guide.
The real answer
AI is not replacing data analysts. It is replacing the least valuable 40% of what many analysts were doing. The analysts who spent their careers being the SQL person are in trouble. The analysts who invested in statistics, experimentation, domain expertise, and communication are having their best years on record. The field is splitting, and it's splitting in a direction that rewards the analysts who were underpriced all along.
