The future of data analysts
An answer can be technically correct and still use the wrong definition. For data analysts exploring AI, that is a practical place to start: test a small analysis, inspect its assumptions and decide which checks must stay in your workflow.
Technology. Updated
These examples concern business reporting and analysis. “Data analyst” covers many roles; business intelligence is one useful reference, not a universal job description. The task labels below are our editorial estimates, not measured replacement probabilities.
AI-assisted editorial guide by Future of My Job.
How AI is changing the work of data analysts
Ways to examine your own workflow, with its context and limitations in view.
Break an analysis into its question, data, transformation and interpretation. A drafting tool may be useful for one step without being reliable for the others. Before trying a generated query or summary, write the metric definition and the result you expect from a small test case.
The exercise below deliberately includes a duplicate and a refunded order. It shows why a plausible total is not enough: the analyst has to decide what counts, document that choice and verify that the output follows it. Those checks matter whether the draft comes from an AI tool, a colleague or an existing dashboard.
The useful question is not whether a tool can produce SQL or a chart, but whether the whole checked workflow meets your standard. Measure correction time and missed assumptions as well as speed. Do not infer a hiring forecast or a personal job-loss timeline from one task experiment.
Catch the wrong revenue total
Budget 30 minutes. This invented export contains a duplicate row and a refunded order. Use it to test whether an AI-assisted answer follows a clear metric definition.
| Order | Amount ($) | Refunded |
|---|---|---|
| O101 | 100 | No |
| O102 | 200 | Yes |
| O102 | 200 | Yes |
| O103 | 150 | No |
Define and calculate (5 min)
Define net revenue here as the sum of unique, non-refunded orders. Write the expected answer by hand before using a tool.
Ask for a checked explanation (10 min)
Copy the fictional table and prompt below into an approved tool. Ask for the calculation and assumptions, not a query against a real database.
Test the reasoning (10 min)
Check deduplication, refund handling and the denominator against the answer checks. Ask how the result would change if the question were gross sales instead.
Keep the useful evidence (5 min)
Record the result, missed assumptions and total review time. Decide which instruction or check you would retain for a future approved task.
A prompt to try with the fictional table
Analyze only this fictional table. Define net revenue as the sum of unique, non-refunded orders. The repeated O102 row is a duplicate export, not a second purchase. Show net revenue, the number of included orders and average revenue per included order. Explain exclusions. Also show gross sales across unique orders, including refunded orders. Do not infer trends or causes from these four rows.
Show the answer checks
- Net revenue is 100 + 150 = 250 dollars. Two orders are included, so average revenue per included order is 125 dollars.
- Gross sales across the three unique orders are 450 dollars. Summing all four exported rows gives 650 dollars and double-counts O102.
- O102 is counted once for gross sales and excluded for net revenue. Deduplication and refund handling are separate decisions.
- The example defines a full refund. Partial refunds, taxes, currencies and reporting periods would need explicit rules in a real metric.
- Four invented rows cannot establish a trend, customer behavior or a reason for refunds. Reject explanations that claim otherwise.
Keep a metric definition and a tiny known-answer test alongside any reusable analysis. The useful output is a result you can explain and verify.
Get my free work snapshotData analyst tasks most exposed to AI
Our relative estimates of where drafting or other assistance may be useful. They are editorial judgments, not measured automation rates. Actual usefulness depends on the tool, inputs and checks.
Drafting a query from an explicit metric definition
HighWhy AI affects it
A candidate for assistance when tables, joins and the expected result are clear. Review access, query scope and edge cases; never run unfamiliar output against production data unchecked.
Preparing a first-pass summary of a small table
HighWhy AI affects it
A bounded output is easy to compare with known answers. Check units, denominators, missing values and whether the text adds unsupported causes.
Documenting a transformation or recurring report
HighWhy AI affects it
A draft can help explain a supplied process. Verify that the explanation matches the actual transformation and does not invent business rules.
Resolving duplicate and missing records
ModerateWhy AI affects it
Detection can follow explicit rules. Choosing the correct entity, deduplication key and treatment of missingness depends on what the data represents.
Defining a metric with stakeholders
LowWhy AI affects it
A definition involves the decision someone needs to make, trade-offs and agreement. A tool can support the discussion without settling those choices.
Deciding whether evidence supports a recommendation
LowWhy AI affects it
This requires understanding limits, alternative explanations and consequences. Low estimated exposure is not a guarantee that the role or its tasks will stay unchanged.
What stays valuable in data analyst work
Capabilities to practice and examine in your own work. None guarantees job security.
Defining the question
Specify the decision, population, time period and metric before producing an answer. A correct calculation cannot repair an undefined question.
Understanding the data
Check what a row represents, how records enter the system and where duplicates or missing values come from.
Designing a verification test
Use a small known-answer example, reconcile totals and test a boundary case before trusting a generated transformation.
Explaining limits and trade-offs
Separate a pattern in the data from a causal claim. Communicate what the analysis can and cannot support.
AI tool categories for data analysts
Starting points for an approved experiment. Check the specific tool’s capabilities and your workplace’s data policy.
Query and code assistants
What it affects
An option for drafting a query or transformation from a clear specification. Use a safe local example and review generated code before execution.
Data preparation tools
What it affects
An option for repeatable cleaning steps. Supply the entity definitions and check how the tool handles duplicates, missing values and types.
Visualization assistants
What it affects
An option for an initial chart. Verify aggregation, axes, units and whether the visual answers the intended question.
Text analysis assistants
What it affects
An option for a first-pass summary of permitted material. Compare it with the source and check whether important exceptions were omitted.
Where data analyst skills transfer
Options to investigate against your interests, experience and local requirements. These are not personal recommendations or promises of demand.
Analytics engineering
Why it fits
Worth exploring if you enjoy tested transformations, data models and reliable reporting. Compare actual vacancies with your SQL, software and platform experience.
Product or decision analytics
Why it fits
An option if you enjoy defining questions and working with decision makers. Investigate experiment design, domain knowledge and communication requirements.
Data quality or governance
Why it fits
An option if definitions, lineage and controls interest you. Responsibilities and required experience vary by organization; this is not a promise of an easier or safer role.
Evidence and its limits
These sources inform the guide. Our task labels and exercises are editorial interpretations, not findings validated by these researchers.
Business Intelligence Analysts: tasks and work activities
O*NET OnLine, US Department of Labor sponsored resource. Accessed .
Describes reporting, analysis and communication tasks in a US business-intelligence occupation. It is one reference for a broad job title, not a definition of every data analyst role or a measure of AI performance.
Generative AI and Jobs: A Refined Global Index of Occupational Exposure (2025)
International Labour Organization. Accessed .
This task-based study finds that job transformation is the most likely overall effect of generative AI. Occupational exposure is not a forecast of an individual losing a job, and the study does not validate this site’s scores.
Navigating the Jagged Technological Frontier (2026 publication)
Harvard Business School / Organization Science. Accessed .
In an experiment with consultants using GPT-4, AI assistance helped on some tasks and harmed performance on others. It supports checking quality task by task; it does not establish a productivity gain for analysts or for newer tools.
Common questions about AI and data analysts
Will AI replace data analysts?
There is no defensible yes-or-no answer for every analyst. A role combines tasks with different requirements. Occupational exposure research can inform questions about those tasks, but it cannot tell you what your employer will do. Test a bounded workflow and record what still needs your judgment.
Should I still learn SQL if I can ask AI to write it?
Understanding the query helps you verify it. Practice joins, filtering, aggregation and the meaning of a row so you can spot a plausible but incorrect result. A generated query is a draft to check, not evidence that the metric is right.
What is a safe first AI experiment for an analyst?
Use a tiny fictional dataset with an answer you can calculate yourself. Specify the metric, ask for the calculation, then check duplicates, denominators and assumptions. The example on this page is designed for that purpose.
Can I paste a production dataset into an AI assistant?
Follow your organization’s data and tool policies. Use public or invented data for this exercise. Avoid customer information, internal identifiers, credentials and confidential material; an approved product may still have restrictions on specific accounts or features.
Illustrative AI exposure profile for data analysts
Illustrative: a fictional work profile run through our scoring model. This is not a measured average for the occupation.
These are self-assessment indices, not job-loss probabilities or measured percentages of tasks. How the estimate works
- Low0–34
- Moderate35–54
- High55–74
- Very high75–100
Your own task mix may produce a different result. No score can tell you when your job will change.
Five dimensions on the same 0 to 100 scale. These model indices summarize the fictional answers: relative task exposure, potential for AI assistance, demands for human input and readiness to adapt. They do not measure AI capability.
- AI exposure
- 69
- Automation exposure
- 66
- AI augmentation potential
- 82
- Human resilience
- 40
- Transition readiness
- 78
See where your own work stands
Fifteen questions about the work you actually do, scored with the same rules as the profile above. An estimate, not a prediction.
Free · ~3 minutes · No account required