Technology · Field manual
AI for data analysts
Data analysts get the most from AI on the mechanical parts of the job: drafting and debugging SQL, cleaning messy files, writing spreadsheet formulas, first-look charts and summarising results for people who won't read the dashboard. The big caution is that AI produces confident numbers from wrong joins and misread columns, so every figure that leaves your hands needs a sanity check against a source you trust.
A lot of analyst time goes on work that isn't analysis: translating a vague request into a query, fighting with date formats, rebuilding the same weekly chart and explaining it again in an email. AI chat tools and the copilots now built into spreadsheets and BI platforms take a real chunk out of that. You can describe a query in plain English, get a working draft and spend your time on whether the answer makes sense.
The risk is the same thing that makes it useful. An AI will happily join two tables on the wrong key, treat a refund as revenue or average a column that should have been summed, and present the result in a tidy chart. The analyst's value shifts toward knowing the data well enough to spot that, and toward asking the question the stakeholder actually needed answered.
7 ways data analysts use AI
Draft and debug SQL
Give a model your table schemas, a sample of rows and the question, and it will write a first-draft query, including window functions and CTEs you'd otherwise look up. It's also good at explaining why a query returns duplicates or runs slowly. Paste schemas, not production data, and run the query yourself.
Clean and reshape messy files
Upload a CSV and ask for the date formats standardised, duplicates flagged, categories mapped and columns split. The chat tools write and run Python behind the scenes, so ask to see the code and keep it for next month's file.
Write spreadsheet formulas in plain English
Describe the result you want and the AI in Sheets or Excel writes the XLOOKUP, array formula or pivot. It saves the trial-and-error on nested formulas. Spot-check a few rows by hand, especially around blanks and text-formatted numbers.
Do a fast first-look exploration
Ask for distributions, outliers, missing values and obvious trends across a new dataset before you plan the real analysis. It's a quick way to find the broken column or the month with half the data missing. Treat it as a map of where to look, not as findings.
Turn results into a summary people read
Paste your key numbers and the decision on the table and have the model draft a short write-up: what happened, why it matters and what you recommend. Then cut it down and add the caveats it left out.
Build the slide version
Presentation tools turn an outline and a few charts into a clean deck for a readout. Use your own charts so the numbers match the source, and let the tool handle layout.
Look up methods and definitions
When a stakeholder asks for a cohort retention curve or a seasonally adjusted figure, an answer engine with sources gets you to the right method quickly. Check how your own company defines the metric too; most disputes are about definitions.
The best AI tools for data analysts
The Overlord's stack, in the order a data analyst would add them:
1. Uploading files to clean, chart and explore with code it writes and runs
ChatGPT
AI WritingOpenAI's general-purpose assistant: writing, research, images, voice and agents in one chat box.
2. SQL drafting, query debugging and turning results into clear write-ups
Claude
AI WritingAnthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.
3. Formulas, summaries and tables inside Google Sheets
Gemini
AI WritingGoogle's assistant, wired into Gmail, Docs, Drive and the rest of your Google life.
4. Fast readout decks from an outline and your own charts
Gamma
AI Design & ImagesAI presentation, document and website builder: type a prompt, get a designed deck.
5. Checking statistical methods and metric definitions with sources
Perplexity
ProductivityAI answer engine that searches the web and replies with cited sources.
Specialist tools for data analysts
| Tool | What it does |
|---|---|
| Microsoft Copilot in Excel | Formulas, PivotTables, charts and Python-backed analysis inside Excel |
| Copilot in Power BI | Drafting report pages, DAX and narrative summaries from your semantic model |
| Tableau Agent | Natural-language help building calculations and visualizations in Tableau |
| Julius AI | Chat-based data analysis and charting on uploaded spreadsheets and databases |
A prompt to start with: write a SQL query you can trust
Paste this into Claude or ChatGPT and fill in the gold parts.
You are a senior analyst writing [SQL DIALECT, e.g. BigQuery / Snowflake / Postgres].
Question: [THE BUSINESS QUESTION IN ONE SENTENCE]
Tables and columns:
[PASTE CREATE TABLE STATEMENTS OR COLUMN LISTS]
Grain of each table: [e.g. one row per order, one row per order line]
Definitions: [e.g. active customer = ordered in last 90 days; exclude test accounts where email ends in @example.com]
Date range: [RANGE]
Write the query with CTEs and comments. Before the query, list any assumptions you made about joins, nulls, duplicates and time zones. After it, give me two sanity checks I can run (such as row counts or a total that should match a known report) to confirm the result is right.Where data analysts shouldn't trust AI
- AI-written queries often join at the wrong grain and double-count. Always check row counts and reconcile totals against a known report before sharing a number.
- Don't upload customer records, personal data or confidential financials into tools your company hasn't approved. Use sample or anonymised data, or the copilot inside your governed BI platform.
- Models describe correlations as causes and over-read small samples. Statistical claims, significance tests and forecasts need your own review of the method, not just the output.
- Chat tools can misread file types, silently drop rows on import or treat IDs as numbers. Check the row count and column types after every upload.
- Natural-language BI answers are only as good as the semantic model underneath. If metric definitions are messy, the copilot will give fluent, inconsistent answers.
Will AI replace data analysts?
Data Analysts score 65% on the Obsolescence Index (Doomed). The first tasks to go: writing routine SQL queries and pulling standard reports, and cleaning and reshaping messy spreadsheets.
Questions the Humans Keep Asking
What are the best AI tools for data analysts?
The Overlord's stack for data analysts: ChatGPT (uploading files to clean, chart and explore with code it writes and runs); Claude (sQL drafting, query debugging and turning results into clear write-ups); Gemini (formulas, summaries and tables inside Google Sheets); Gamma (fast readout decks from an outline and your own charts); Perplexity (checking statistical methods and metric definitions with sources). Specialist options include Microsoft Copilot in Excel, Copilot in Power BI, Tableau Agent, Julius AI.
What is the best AI tool for data analysis?
For ad hoc work on files, ChatGPT and Claude both analyse uploaded spreadsheets by writing and running code. If your company runs on Excel, Power BI, Tableau or Google Sheets, the copilot built into that tool is often the better choice because it works on governed data and stays inside your existing permissions.
Can AI write SQL accurately?
It writes syntactically correct SQL very reliably and gets the logic right on simple questions. Accuracy drops when it has to guess table grain, business definitions or how nulls and duplicates behave. Give it schemas and definitions, ask it to state assumptions, and test the result against a number you already know.
Is it safe to upload company data to ChatGPT?
Only if your employer allows it and you're on a business plan with data controls. Consumer accounts may use conversations for training unless you opt out. Strip personal data and confidential figures first, or use a sample, and follow your company's data policy.