14 AI skills for
data analysts

For: Data / business analyst · Problem it solves: Simple business questions wait days in the data team's queue because nobody else can write SQL.

Outcome preview

• You paste: BigQuery · monthly active users for 2026 · events(user_id, event_name, event_ts)
• You get: MAU = distinct users with any event in the calendar month (UTC).
• SELECT DATE_TRUNC(DATE(event_ts), MONTH) AS month, COUNT(DISTINCT user_id) AS mau FROM events WHERE event_ts >= '2026-01-01' GROUP BY month ORDER BY month;
• Sanity check: no month should exceed total sign-ups to date.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send your question in plain English, the table and column names, and your database (PostgreSQL, MySQL, BigQuery, Snowflake).

# Skill: SQL from plain English

You are running the "SQL from plain English" skill for me. I work as: Data / business analyst. What I need from you: a correct, commented SQL query for my database, with the assumptions it makes spelled out.

## When I send you my question in plain English, the table and column names, and my database (PostgreSQL, MySQL, BigQuery, Snowflake)
1. Restate the question as a precise metric definition.
2. Write the query in the right dialect with comments.
3. List assumptions (time zone, duplicates, nulls, what counts as active).
4. Give a quick sanity check to run on the result.

## Reply in this format
- Metric definition
- SQL (code)
- Assumptions
- Sanity check

## Rules
- Never invent columns: ask if a needed column isn't listed.
- Use explicit JOINs and date ranges.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send my question in plain English, the table and column names, and my database (PostgreSQL, MySQL, BigQuery, Snowflake).

Immediate result: A correct, commented SQL query for your database, with the assumptions it makes spelled out. Copy it or download SKILL.md

For: Analyst or Excel / Sheets user · Problem it solves: You know what you want the spreadsheet to do but not the formula, and nested IFs break.

Outcome preview

• You paste: Google Sheets: total of column C where column A is 'Mumbai' and column B is in October 2026
• You get: =SUMIFS(C:C, A:A, "Mumbai", B:B, ">="&DATE(2026,10,1), B:B, "<"&DATE(2026,11,1))
• Each pair is a condition: city = Mumbai, date ≥ 1 Oct and < 1 Nov.
• Error #VALUE! usually means column B holds text, not dates.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send what you want to calculate, your column layout (e.g. A = date, B = amount) and Excel or Google Sheets.

# Skill: Spreadsheet formula helper

You are running the "Spreadsheet formula helper" skill for me. I work as: Analyst or Excel / Sheets user. What I need from you: the exact formula for Excel or Google Sheets, explained, with an example of the output.

## When I send you what I want to calculate, my column layout (e.g. A = date, B = amount) and Excel or Google Sheets
1. Pick the simplest function that works (XLOOKUP, SUMIFS, FILTER, LET, QUERY).
2. Write the formula for the cell references given.
3. Explain each part.
4. Show the result on a small example table.

## Reply in this format
- Formula (code)
- How it works
- Example
- Common errors and fixes

## Rules
- Say if the function isn't available in older Excel versions and give an alternative.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send what I want to calculate, my column layout (e.g. A = date, B = amount) and Excel or Google Sheets.

Immediate result: The exact formula for Excel or Google Sheets, explained, with an example of the output. Copy it or download SKILL.md

For: Growth / product analyst · Problem it solves: Teams call A/B tests on gut feel, ship false winners and miss real ones.

Outcome preview

• You paste: Control 10,000 visitors / 400 sign-ups · Variant 10,000 / 460 · 14 days
• You get: CR: 4.0% vs 4.6% · lift +15%
• z ≈ 2.09, p ≈ 0.037 · 95% CI for difference: +0.04 to +1.16 points
• Verdict: likely winner — ship, and watch week-two retention.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send visitors and conversions for control and each variant, and the test duration.

# Skill: A/B test reader

You are running the "A/B test reader" skill for me. I work as: Growth / product analyst. What I need from you: a plain-English verdict with lift, confidence interval, p-value and whether the sample is big enough.

## When I send you visitors and conversions for control and each variant, and the test duration
1. Compute conversion rates and relative lift.
2. Run a two-proportion z-test and give the 95% confidence interval, showing the formulas.
3. Check sample size against the minimum detectable effect.
4. Give a ship / keep running / stop verdict with caveats (peeking, novelty, seasonality).

## Reply in this format
- Results table
- Statistics (with formulas)
- Verdict
- Caveats
- Next test idea

## Rules
- Show the arithmetic so it can be checked.
- Never call significance on fewer than 100 conversions per arm without a warning.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send visitors and conversions for control and each variant, and the test duration.

Immediate result: A plain-English verdict with lift, confidence interval, p-value and whether the sample is big enough. Copy it or download SKILL.md

For: Data analyst · Problem it solves: A messy export has duplicates, mixed formats and blanks, and cleaning it by hand is error-prone.

Outcome preview

• You paste: pandas · columns: name, phone, city, signup_date (formats: 03/10/26, 2026-10-03, '3 Oct')
• You get: signup_date: 3 formats → pd.to_datetime(df.signup_date, dayfirst=True, format='mixed')
• city: 'Bombay'/'Mumbai '/'mumbai' → strip, title-case, map Bombay → Mumbai
• Check: rows before = rows after + duplicates removed (logged).

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send column names and 10–20 sample rows (anonymised), plus the tool you use.

# Skill: Data cleaning plan

You are running the "Data cleaning plan" skill for me. I work as: Data analyst. What I need from you: a step-by-step cleaning plan with the code (pandas, SQL or spreadsheet steps) to do it.

## When I send you column names and 10–20 sample rows (anonymised), plus the tool I use
1. Profile each column: type, format issues, blanks, outliers, duplicates.
2. Propose a cleaning rule per issue.
3. Write the code or steps in the chosen tool.
4. Add checks that prove the cleaned data is right (row counts, totals).

## Reply in this format
- Column profile table
- Cleaning rules
- Code
- Validation checks

## Rules
- Never drop rows silently: log what's removed and why.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send column names and 10–20 sample rows (anonymised), plus the tool I use.

Immediate result: A step-by-step cleaning plan with the code (pandas, SQL or spreadsheet steps) to do it. Copy it or download SKILL.md

For: Analytics lead / PM · Problem it solves: Dashboards get built with every metric and answer no question.

Outcome preview

• You paste: Weekly dashboard for the head of a food-delivery city team
• You get: Q1: Are we delivering on time? → On-time rate = orders delivered ≤ promised ETA ÷ delivered orders
• Chart: line, 8 weeks, with target line at 92%
• Filter: zone, hour of day.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send who the dashboard is for, the decisions they make and the data sources available.

# Skill: Dashboard spec writer

You are running the "Dashboard spec writer" skill for me. I work as: Analytics lead / PM. What I need from you: a dashboard spec: the questions it answers, metrics with definitions, charts, filters and layout.

## When I send you who the dashboard is for, the decisions they make and the data sources available
1. List the three to five questions the audience asks weekly.
2. Define one metric per question, with formula and source.
3. Choose the right chart for each and the default time range.
4. Sketch the layout top to bottom and the filters.

## Reply in this format
- Audience and decisions
- Metric table: name, definition, formula, source
- Charts
- Filters
- Layout (wireframe in text)
- Alerts

## Rules
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send who the dashboard is for, the decisions they make and the data sources available.

Immediate result: A dashboard spec: the questions it answers, metrics with definitions, charts, filters and layout. Copy it or download SKILL.md

For: Analytics engineer / data analyst · Problem it solves: Two teams report different numbers for 'active users' because nobody wrote the definition down.

Outcome preview

• You paste: 'Activation rate' for a B2B SaaS — people mean different things
• You get: Activation rate = new workspaces that invite ≥ 1 teammate and create ≥ 1 project within 7 days ÷ new workspaces
• Excludes: internal and test workspaces (domain list).
• Edge: workspaces created on day 6 of the window are counted next week.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send the metric name and how your company currently talks about it.

# Skill: Metric definition doc

You are running the "Metric definition doc" skill for me. I work as: Analytics engineer / data analyst. What I need from you: a precise metric definition: formula, inclusions, exclusions, grain, owner and SQL.

## When I send you the metric name and how my company currently talks about it
1. Write a one-sentence business definition.
2. Write the exact formula, grain and time window.
3. List inclusions, exclusions and edge cases.
4. Give example SQL and a list of related metrics.

## Reply in this format
- Definition
- Formula
- Grain and window
- Includes / excludes
- Edge cases
- SQL
- Owner and review date

## Rules
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send the metric name and how my company currently talks about it.

Immediate result: A precise metric definition: formula, inclusions, exclusions, grain, owner and SQL. Copy it or download SKILL.md

For: Research / insights analyst · Problem it solves: Hundreds of open-text survey answers go unread, and only the scores get reported.

Outcome preview

• You paste: NPS survey, 340 answers to 'What's the main reason for your score?'
• You get: Slow delivery — 92 (27%) · 'Third late order this month'
• Detractors mention delivery 3× more than promoters.
• Action: show a realistic ETA at checkout, not the best case.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send the survey question and the answers (anonymised), plus any score column.

# Skill: Survey analysis

You are running the "Survey analysis" skill for me. I work as: Research / insights analyst. What I need from you: coded themes with counts, representative quotes and what to do about each.

## When I send you the survey question and the answers (anonymised), plus any score column
1. Code each answer into themes (an answer can have several).
2. Count and rank themes; split by score group if provided.
3. Pick two verbatim quotes per theme.
4. Recommend actions for the top three themes.

## Reply in this format
- Theme table: theme, count, % of answers, example quotes
- Differences by score group
- Recommendations
- Method note

## Rules
- Counts must add up from the input.
- Quotes verbatim.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send the survey question and the answers (anonymised), plus any score column.

Immediate result: Coded themes with counts, representative quotes and what to do about each. Copy it or download SKILL.md

For: Analyst presenting data · Problem it solves: The wrong chart hides the point — pie charts with twelve slices, dual axes that mislead.

Outcome preview

• You paste: Revenue by 9 product lines for 2025 vs 2026 · message: 3 lines drove all the growth
• You get: Chart: slope chart (or sorted dot plot), 9 lines, 2025 → 2026.
• Why not grouped bars: 18 bars make the 3 winners hard to spot.
• Title: 'Three product lines drove all of 2026's growth' · colour only those three.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send a sample of your data (columns and a few rows) and the one message you want the audience to get.

# Skill: Chart picker

You are running the "Chart picker" skill for me. I work as: Analyst presenting data. What I need from you: the best chart for my data and message, with labelling advice and a mock-up in text.

## When I send you a sample of my data (columns and a few rows) and the one message I want the audience to get
1. Identify the comparison type: over time, ranking, part-to-whole, distribution, relationship.
2. Recommend one chart and say why the obvious alternative is worse.
3. Specify axes, sorting, colour emphasis and title (the title states the message).
4. Give the code for the chart in Python or the steps in Excel/Sheets.

## Reply in this format
- Recommended chart
- Why not the alternative
- Spec
- Code or steps

## Rules
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send a sample of my data (columns and a few rows) and the one message I want the audience to get.

Immediate result: The best chart for your data and message, with labelling advice and a mock-up in text. Copy it or download SKILL.md

For: Growth analyst · Problem it solves: A retention table full of percentages doesn't tell you what's improving or why.

Outcome preview

• You paste: Monthly cohorts Jan–Jun 2026, months 0–5
• You get: Biggest drop: month 0 → 1 (100% → 38%).
• Improving: May and June cohorts keep 46% at month 1 vs 34% for Jan.
• Flattens at ~22% from month 3 — a real core, worth growing.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send a cohort retention table (sign-up cohort by week or month, % retained per period).

# Skill: Cohort retention read

You are running the "Cohort retention read" skill for me. I work as: Growth analyst. What I need from you: a plain-English read of my cohort table: where users drop, which cohorts improved and what to test.

## When I send you a cohort retention table (sign-up cohort by week or month, % retained per period)
1. Find the biggest drop between periods.
2. Compare recent cohorts to older ones at the same age.
3. Check whether the curve flattens (a sign of product–market fit) and where.
4. Suggest hypotheses and experiments for the biggest drop.

## Reply in this format
- Headline finding
- Biggest drop
- Cohort trend
- Does it flatten?
- Hypotheses and tests

## Rules
- Don't compare cohorts at different ages.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send a cohort retention table (sign-up cohort by week or month, % retained per period).

Immediate result: A plain-English read of your cohort table: where users drop, which cohorts improved and what to test. Copy it or download SKILL.md

For: Data analyst · Problem it solves: Repeating the same analysis in a spreadsheet every week is slow and error-prone.

Outcome preview

• You paste: sales.csv (date, region, product, units, price) · weekly revenue by region + chart
• You get: df['revenue'] = df.units * df.price
• weekly = df.groupby([pd.Grouper(key='date', freq='W-MON'), 'region']).revenue.sum().unstack()
• weekly.plot(); plt.savefig('weekly_revenue.png', dpi=200)

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send the file type and columns, the analysis you want and the output (table, chart, Excel).

# Skill: Python analysis script

You are running the "Python analysis script" skill for me. I work as: Data analyst. What I need from you: a commented pandas script that loads my file, does the analysis and saves a chart and summary.

## When I send you the file type and columns, the analysis I want and the output (table, chart, Excel)
1. Write a script that loads the data and validates columns.
2. Perform the analysis step by step with comments.
3. Save outputs (CSV/Excel and a PNG chart).
4. Explain how to run it and change the parameters.

## Reply in this format
- Script (code)
- How to run
- Parameters to change
- Assumptions

## Rules
- Use pandas and matplotlib only unless asked.
- Fail with a clear message if a column is missing.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send the file type and columns, the analysis I want and the output (table, chart, Excel).

Immediate result: A commented pandas script that loads your file, does the analysis and saves a chart and summary. Copy it or download SKILL.md

For: Analyst / business owner · Problem it solves: A key number dropped overnight and everyone has a theory.

Outcome preview

• You paste: Checkout conversion fell from 3.1% to 2.2% on Tuesday
• You get: Check 1: did the analytics event or app version change Tuesday?
• Decompose: sessions → add-to-cart → checkout start → payment success
• If only Android 5.12 dropped: the release is the prime suspect.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send the metric, how much it changed, when, and any data you have by segment.

# Skill: KPI drop investigator

You are running the "KPI drop investigator" skill for me. I work as: Analyst / business owner. What I need from you: a structured investigation plan that rules out data issues first, then segments to find the cause.

## When I send you the metric, how much it changed, when, and any data I have by segment
1. Check data problems first: tracking changes, pipeline delays, definition changes.
2. Break the metric into components (e.g. traffic × conversion × order value).
3. Segment by platform, country, channel, version and new vs returning.
4. Rank the likely causes and the query to confirm each.

## Reply in this format
- Is it real? (data checks)
- Metric decomposition
- Segments to check
- Likely causes ranked
- Queries to run

## Rules
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send the metric, how much it changed, when, and any data I have by segment.

Immediate result: A structured investigation plan that rules out data issues first, then segments to find the cause. Copy it or download SKILL.md

For: Analyst presenting to leadership · Problem it solves: Analysis gets presented as charts with no story, and leaders can't act on it.

Outcome preview

• You paste: Findings: repeat customers 22% of buyers but 51% of revenue; repeat rate fell 4 pts
• You get: Headline: Repeat customers are a fifth of buyers but half our revenue — and we're losing them.
• Recommendation: fund a 6-week retention test before more acquisition spend.
• Caveat: one quarter of data.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send your findings, numbers and charts described in text, and who the audience is.

# Skill: Data story writer

You are running the "Data story writer" skill for me. I work as: Analyst presenting to leadership. What I need from you: a one-page summary: the headline, three supporting points, the 'so what' and a recommendation.

## When I send you my findings, numbers and charts described in text, and who the audience is
1. Write the headline as a full sentence with the key number.
2. Order three supporting points by importance (pyramid principle).
3. State the implication and the recommendation.
4. List caveats and what you'd need to be more certain.

## Reply in this format
- Headline
- Supporting points
- So what
- Recommendation
- Caveats

## Rules
- Use only numbers provided.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send my findings, numbers and charts described in text, and who the audience is.

Immediate result: A one-page summary: the headline, three supporting points, the 'so what' and a recommendation. Copy it or download SKILL.md

For: Analyst / planner · Problem it solves: Teams need a forecast by Friday and either guess or overbuild a model.

Outcome preview

• You paste: Monthly orders Oct 2025–Sep 2026, forecast Oct–Dec 2026
• You get: History: +3% a month, with a Diwali spike (Oct–Nov ≈ 1.4× normal).
• Method: seasonal naive × trend — simple and explainable.
• Base Nov 2026: 18,900 · low 16,800 · high 21,000

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send historical monthly numbers (at least 12 months ideally) and what you need to forecast.

# Skill: Simple forecast builder

You are running the "Simple forecast builder" skill for me. I work as: Analyst / planner. What I need from you: a transparent forecast with method, formulas, base, high and low cases.

## When I send you historical monthly numbers (at least 12 months ideally) and what I need to forecast
1. Describe the trend and seasonality in the history.
2. Pick a simple method (moving average, growth rate, seasonal naive) and say why.
3. Produce base, high and low forecasts with formulas you can paste into a spreadsheet.
4. List the assumptions that would change the forecast most.

## Reply in this format
- What the history shows
- Method
- Forecast table (base/high/low)
- Spreadsheet formulas
- Key assumptions

## Rules
- Never present a forecast as certain; always give a range.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send historical monthly numbers (at least 12 months ideally) and what I need to forecast.

Immediate result: A transparent forecast with method, formulas, base, high and low cases. Copy it or download SKILL.md

For: Product analyst · Problem it solves: Analytics is full of badly named events nobody trusts, and key actions aren't tracked at all.

Outcome preview

• You paste: Food delivery app: search, add to cart, checkout, payment, rating
• You get: search_performed — on results load — {query_length, results_count, filters_used}
• payment_failed — gateway error — {method: upi|card|cod, error_code}
• Exclude: delivery address, phone, typed search text.

Install: add it to your assistant in one tap, or copy the text below as your first message in any AI assistant — then send your product's key flows and the questions you want analytics to answer.

# Skill: Event tracking plan

You are running the "Event tracking plan" skill for me. I work as: Product analyst. What I need from you: a tracking plan: events, properties, naming convention and the questions each answers.

## When I send you my product's key flows and the questions I want analytics to answer
1. Set an object_action naming convention.
2. List the events needed for each question.
3. Define properties with types and allowed values.
4. Flag personal data that must not be sent.

## Reply in this format
- Naming convention
- Event table: event, trigger, properties, question it answers
- Personal data to exclude
- QA steps

## Rules
- Never track passwords, payment card data or free text that may contain personal data.
- Use only the facts I give you; mark anything you had to assume with [ASSUMPTION].
- If something essential is missing, ask me one short question, then continue.
- No preamble or filler: start with the result.

Confirm you're ready in one line, then wait for me to send my product's key flows and the questions I want analytics to answer.

Immediate result: A tracking plan: events, properties, naming convention and the questions each answers. Copy it or download SKILL.md

Want proof you can do this work? These free tests check the skills on this page, and each has a career guide for the role: how hiring works, what it pays and how to prepare.

• 20 AI skills for software engineers
• 14 AI skills for product managers
• 12 AI skills for finance and accounting teams
• All 200 AI skills, by job — how skills work and the full series