Will AI Replace Data Analysts?

What the employment figures say

The nearest official series is not a clean match for this job. The US Bureau of Labor Statistics publishes Data Scientists (SOC 15-2051), reporting 245,900 jobs in 2024 and projecting 34% employment growth from 2024 to 2034 with about 23,400 annual openings, and it does not attribute any of those figures to AI ^{(US Bureau of Labor Statistics 2025)}. That growth sits beside a high exposure score without contradicting it: BLS has no separate series for Data Analysts or Business Intelligence Analysts (O*NET 15-2051.01), and the Data Scientists category is substantially broader than the analyst role ^{(US Bureau of Labor Statistics 2025)}. The UK picture at analyst level is less comfortable — a DSIT and LinkedIn measure recorded Data Analyst hiring down 15% in April 2026, though LinkedIn is a commercial recruitment platform and the report itself states its members are not representative of the UK workforce ^{(Department for Science 2026)}.

What the research says about data analysts

Now

The applicability research gives the data scientist analog 0.357, and it is routine querying and reporting that sit at the exposed end of that figure ^{(Microsoft Research 2025)}. O*NET's description of Business Intelligence Analysts (15-2051.01) reads almost as a list of what current tools target: querying repositories, producing periodic and executive reports, maintaining dashboards and BI tools, collecting data, and communicating trends to stakeholders ^{(O*NET OnLine 2026)}. Adoption inside the profession is already broad — dbt Labs, which sells analytics-engineering software and did not survey a representative sample, found 80% of 459 respondents using AI day to day against 30% previously, and 70% using it for analytics development ^{(dbt Labs 2025)}. The hiring side has begun to move in the UK, where a DSIT and LinkedIn measure placed Data Analyst among the steepest declines at 15% in April 2026, with 30 of 38 occupations declining and 8 growing ^{(Department for Science 2026)}.

The next few years

The relevant shift over the next few years is from tools that speed up a step to systems that plan, use tools and run multi-step work unattended, which changes the question from whether AI can help with a task to whether it can complete it without you ^{(Anthropic 2026)}. In the coding-adjacent group this role sits in, the migration from chat to agentic APIs is documented, but augmentation upside and demand are the highest of any category, so exposure is high while the net employment effect is cushioned ^{(Anthropic Economic Index 2026)}. The official projection is consistent with that cushioning: 34% growth for the broader Data Scientists category from 2024 to 2034, with no AI attribution attached ^{(US Bureau of Labor Statistics 2025)}. Meanwhile nearly 45% of US Data and Analytics job postings mentioned AI-related terms in January 2026, which is a change in what employers ask for rather than evidence of jobs appearing or disappearing ^{(Indeed Hiring Lab 2026)}.

Longer term

Longer term, the binding constraint on fully automated analysis is who is accountable for the number. SEC material on Sarbanes-Oxley Section 302 places certification of financial and other information in quarterly and annual reports, together with responsibility for internal controls, with a company's principal executive and financial officers, so automated reporting still runs inside a human-controlled certification framework ^{(US Securities and Exchange Commission 2002)}. The EU AI Act pushes the same way, requiring effective human oversight of high-risk systems under Article 14 and high-quality, relevant, representative and complete datasets with documented governance under Article 10 ^{(EUR-Lex 2024)}. That makes data quality, documentation, monitoring and escalation part of analyst work around regulated systems rather than something the systems absorb ^{(EUR-Lex 2024)}.

What is exposed, and what protects it

● Where the exposure is

The exposed core is the routine end of the workflow: querying, scheduled reporting and dashboard upkeep, where the data scientist analog scores 0.357 on applicability ^{(Microsoft Research 2025)}. Preparation is the other soft target — Alteryx, which sells data-preparation software to the people it surveyed, found 76% of 1,400 respondents using spreadsheets as their primary cleaning tool and 45% spending more than six hours a week on cleansing and preparation ^{(Alteryx 2025)}. Grupo Casas Bahia reports tasks that previously took 5-6 hours now taking 2 minutes once business users could ask questions in natural language, in a case published by the vendor selling the automation ^{(Databricks 2026)}. Agent benchmarks point the same way: on InfiAgent-DABench's 311 questions over 55 CSV files, GPT-4 reaches 74.60% accuracy using a file-interacting agent and code executor ^{(arXiv 2024)}.

○ What holds

Reliability is the moat that is actually holding, alongside accountability. A Frontiers in Artificial Intelligence study of ChatGPT 3.5 on statistical-analysis tasks found repeated failures: 5 of 5 answers supplied a basis for specifications but no solution, 4 of 4 were incorrect in syntax or procedure, and one task produced 0 of 3 usable syntax ^{(Frontiers in Artificial Intelligence 2025)}. The DS-1000 benchmark of 1,000 data-science code-generation problems across 7 Python libraries reports 43.3% accuracy for Codex-002, which is partial automation rather than autonomy ^{(Proceedings of Machine Learning Research 2023)}. Demand for the underlying skill has been persistent too — a UK government survey found 48% of businesses recruiting for data skills, Data Analyst the most commonly recruited data role at 12%, 9% reporting difficulty filling it, and an estimated 215,000-234,000 UK roles incorporating hard data skills ^{(UK Department for Digital 2021)}. And dbt Labs, a vendor that benefits from data-team spending, reports 40% of respondents seeing headcount increases against 14% previously, with 57% still spending most of the workday maintaining and organising datasets ^{(dbt Labs 2025)}.

Which tasks go first

What AI already covers is the front half of the job: pulling data from repositories, writing routine transformation code, cleaning and reshaping files, and generating the recurring reports and dashboards that O*NET lists as core Business Intelligence Analyst work ^{(O*NET OnLine 2026)}. The agentic shift moves the next layer, with systems that plan, spawn subagents and run multi-step work on a schedule rather than waiting for a prompt at each step ^{(Nous Research 2026)}. Self-service is where that lands first in practice: Snowflake reports a Fireblocks internal agent fielding more than 2,000 analysis queries a month, 40-50% of all queries, framed as capacity equal to two full-time analysts, with no dismissals stated ^{(Snowflake 2026)}. Databricks describes the Webmotors data team shifting toward complex queries rather than shrinking, alongside a 72% year-over-year reduction in manual tickets and 200 analyst hours saved per month ^{(Databricks 2025)}. What stays human is the part that carries consequences: choosing which question is worth asking, judging whether an output is wrong, and standing behind numbers that feed certified reporting ^{(US Securities and Exchange Commission 2002)}. The Skills England definition of the occupation — collecting, organising and studying data to provide business insight, mapped to SOC 2020 code 3544 with a median salary of GBP 33,977 — describes a role whose value now concentrates in the last word of that sentence ^{(Skills England 2026)}.

What to do about it

Own the question, not the query

The automatable surface is retrieval and reporting; the part that resists it is deciding what should be measured and whether an answer is credible. Alteryx, a vendor selling analytics automation to the professionals it surveyed, reports 87% of 1,400 respondents saying their influence increased and 94% reporting an impact on strategic decision-making ^{(Alteryx 2025)}. Read that as a direction of travel rather than a measurement, and build the stakeholder-facing half of the role on purpose.

Take the governance and assurance work

Regulated AI systems generate analyst-shaped work that the tools cannot sign off on themselves: dataset quality, documentation, monitoring and escalation under Article 10, with effective human oversight required under Article 14 ^{(EUR-Lex 2024)}. Financial reporting has the same structure, where certification and internal-control responsibility rest with named officers rather than with a pipeline ^{(US Securities and Exchange Commission 2002)}. An analyst who can evidence where a number came from is harder to route around.

Supervise the code as well as write it

dbt Labs reports 80% of respondents using AI day to day, up from 30%, and 70% using it for analytics development — from a vendor with a commercial interest in that adoption ^{(dbt Labs 2025)}. The benchmark picture explains why review still matters: DS-1000 records 43.3% accuracy for Codex-002 across 1,000 data-science problems ^{(Proceedings of Machine Learning Research 2023)}. The transferable skill is catching a plausible but wrong output before it reaches a decision.

Tools and courses that fit this situation

Below are paid tools and courses that fit this occupation's position: skills training on the analytics and AI side, and a resume service for anyone moving roles. None of them removes the exposure described above — they change what you can offer around it.

Some links above are affiliate links. If you buy through them, Jobs After AI may earn a commission at no extra cost to you. It does not change which tools appear here.

Common questions

How is the 38 score for data analysts calculated?

It rests on the data scientist analog in Microsoft's applicability research, which scores 0.357, with routine querying and reporting at the exposed end of that measure ^{(Microsoft Research 2025)}. The task surface it maps onto is the one O*NET describes for Business Intelligence Analysts, 15-2051.01: querying repositories, producing reports, maintaining dashboards and communicating trends ^{(O*NET OnLine 2026)}. Confidence in the placement is high, and the method treats the analyst role as more exposed than data scientist because more of its content is routine.

Is this about data analysts losing jobs, or about the tasks changing?

Mostly the tasks. Exposure measures what AI can do, not what employers have already done — BLS projects 34% growth for the broader Data Scientists category from 2024 to 2034 and attributes none of it to AI ^{(US Bureau of Labor Statistics 2025)}, and the vendor cases describing large query volumes moving to agents disclose no dismissals ^{(Snowflake 2026)}. The clearest employment-side signal is a hiring one: Data Analyst hiring down 15% in April 2026 in a measure that cannot determine whether AI caused or merely compounded the decline ^{(Department for Science 2026)}.

Which tasks of a data analyst can AI already do?

Data preparation and routine reporting are furthest along. Alteryx, which sells preparation software, found 45% of respondents spending more than six hours a week on cleansing and preparation and 76% doing it primarily in spreadsheets — a large manual burden its own products target ^{(Alteryx 2025)}. On the reporting side, Grupo Casas Bahia describes natural-language self-service cutting tasks that took 5-6 hours down to 2 minutes, in a case published by the automation vendor ^{(Databricks 2026)}.

Which parts of the job survive?

The parts where being wrong is expensive. A peer-reviewed test of ChatGPT 3.5 on statistical analysis found 4 of 4 answers incorrect in syntax or procedure and one task returning 0 of 3 usable syntax ^{(Frontiers in Artificial Intelligence 2025)}, and DS-1000 puts Codex-002 at 43.3% accuracy across 1,000 data-science problems ^{(Proceedings of Machine Learning Research 2023)}. Certification of reported figures also stays with named executives under Sarbanes-Oxley Section 302, which keeps a human inside the reporting chain ^{(US Securities and Exchange Commission 2002)}.

What do data analysts retrain toward?

Two directions have evidence behind them. One is analytics engineering and AI-assisted development, where dbt Labs reports 70% of respondents already using AI for analytics development ^{(dbt Labs 2025)}. The other is AI governance and data quality work, which the EU AI Act makes mandatory around high-risk systems through Article 10 data-governance duties and Article 14 human oversight ^{(EUR-Lex 2024)}.

Related occupations

The evidence behind this page

How this score was built

The AI-Proof Score is an estimate of how exposed an occupation's core tasks are to AI over roughly the next three to five years, on a 0–100 scale where higher is safer. It measures the exposure of tasks, which sits upstream of, and is softer than, job loss — it is not a probability that this job disappears, and not a claim about any individual doing it. The full derivation, the four research families behind it and its limitations are set out in the methodology, and every occupation is ranked in the AI Job Impact Report 2026.