Module 4 of 8 · Intermediate · 10 min read

Data analysis and spreadsheets

Clean, analyse and chart data with AI that writes and runs code or formulas, and check every number.

Key takeaways

1Let the AI compute with code or formulas, not in its head.
2Describe your columns and your question; ask it to check data quality first.
3Ask for the method and the code so results can be reproduced.
4Sanity-check totals and samples against the raw data before sharing.

01Where AI helps

The key shift: modern assistants can run code on your file. That makes their numbers far more reliable than mental arithmetic.

Writing spreadsheet formulas, pivot tables and scripts from plain-language requests.
Cleaning messy data: duplicates, inconsistent names, dates and formats.
Exploratory analysis: summaries, segments, trends and outliers.
Charts and dashboards, with explanations of what they show.
Explaining statistics and checking whether an analysis is sound.

02The workflow

Most bad analyses come from skipping step 2.

State the business question and the decision it informs.
Share the data (or a sample plus column descriptions) and ask for a data-quality check.
Agree the cleaning steps, then the analysis method.
Have the AI run the analysis with code and show its working.
Verify key figures against the raw data yourself.
Turn results into a short summary with charts and caveats.

03Tools to reach for

For sensitive data, check your organisation's policy and the tool's data handling before uploading.

Chat assistants with code execution, which can load files and produce charts.
AI features in spreadsheet apps for formulas, cleaning and summaries.
Notebooks with AI coding assistants for repeatable analysis.
Business intelligence tools with natural-language querying.

04Prompt: data-quality check

Before any analysis, ask what is wrong with the data.

Example

I've uploaded {{file}}. Columns: {{column descriptions}}. Before any analysis, use code to check data quality: row count, missing values per column, duplicates, inconsistent categories (e.g. spelling variants), impossible values (negative quantities, future dates), and outliers. Show a short table of issues and propose a cleaning step for each. Don't change the data until I approve.

05Prompt: the analysis

Ask for the method, the code and the caveats, not just a conclusion.

Example

Question: {{question, e.g. which customer segments drove the drop in repeat purchases last quarter?}} Using the cleaned data, run the analysis with code. Show: the method in plain English, the code, a summary table, and one clear chart. State caveats such as small sample sizes, missing data, and correlation versus causation. Finish with three plain-language findings, each with the number that supports it.

06Prompt: spreadsheet formulas

Describe the layout precisely so the formula fits your sheet.

Example

In {{spreadsheet app}}, sheet 'Orders' has: A = order date, B = customer ID, C = region, D = amount. Write a formula for sheet 'Summary' cell B2 that sums amounts for the region in A2 for the month in B1 (a date). Explain how it works, and give a version that ignores blank or text values in column D.

07Pitfalls

Treat AI output like a colleague's analysis: check the method and spot-check the numbers.

Mental maths: a model that answers without running code may get numbers wrong. Require code or formulas.
Hidden assumptions: ask what it did with missing values and outliers.
Confident causation: correlation in your data rarely proves cause. Ask for caveats.
Privacy: remove personal data you don't need before uploading.

Tip

Pick two or three key figures and recompute them yourself with a filter or pivot table. If they match, trust the rest more; if not, find out why.

Finished reading?

Mark it complete to track your progress, then test yourself with the AI at Work: Business & Professional quiz.

Next: Research and summarising →