What you will learn
- Clean and reshape messy tables before analysing them
- Summarise a dataset in plain language with the key numbers
- Spot trends, outliers, and patterns worth investigating
- Turn results into charts and simple reports
- Write spreadsheet formulas from a plain description
- Check every number before you quote it
Start by understanding the data you have
Before asking for analysis, look at the data yourself. How many rows and columns are there? What does each column mean? Are there missing values, duplicate rows, or dates in different formats? Claude can help with all of these, but it helps more when you can describe the shape of the data and the question you actually need answered.
State the question in one sentence before you start. "Which products grew fastest last quarter?" leads to a better analysis than "analyse this spreadsheet." A clear question decides which columns matter, which time period counts, and what a useful answer looks like.
Clean the data first
Most real datasets are messy: extra spaces, inconsistent spellings, blank cells, and numbers stored as text. Ask Claude to help you standardise these before you draw conclusions. For example, it can write a formula that trims spaces, a rule that maps variant spellings to one standard value, or a short script that converts a date column into one consistent format.
Keep a copy of the original data untouched. Work on a copy, and note each cleaning step you apply. That way, if a later result looks odd, you can trace it back to a cleaning decision instead of wondering whether the raw data was wrong.
Ask for summaries, then dig into the interesting parts
A good first analysis is a plain summary: totals, averages, counts by category, and the highest and lowest values. Read that summary and look for anything surprising. A category that is far larger than the rest, a month that breaks the pattern, or a value of zero where you expected sales are all worth a second question.
Follow up on one finding at a time. Ask what drives it, whether it holds across different time periods, and whether a different way of grouping the data changes the picture. This step-by-step digging is what separates a real finding from a number that merely looks interesting.
Here is a sales table with columns: date, product, region, units, and revenue. First, list total revenue and units by product for the last three months. Then tell me which product grew fastest compared with the previous three months, and flag any region where revenue fell while units rose. Show your working so I can check each number.
Charts and reports with artifacts
Claude's artifacts can turn a set of results into a chart, a table, or a short report you can read and share. Ask for one chart per question rather than a dashboard of everything at once. A single bar chart of revenue by product is easier to check, and easier for a reader to understand, than a page of mixed visuals.
Label every chart with its source and its time period. A chart without a date range or a definition of what counts as revenue will be misread sooner or later, usually by someone quoting it in a meeting. If a number in the chart surprises you, ask Claude to show the rows behind it.
Formulas and repeatable steps
If you work in a spreadsheet, describe the calculation you want in plain words and ask Claude to write the formula. Give the sheet layout: which column holds what, and where the result should go. Test the formula on a few rows you can check by hand before copying it down a whole column.
For work you repeat every week or month, ask Claude to write the steps out as a short checklist or a small script. A repeatable process beats a one-off answer, because next month's data will need the same cleaning and the same checks.
Tip: Never quote a number from an analysis you have not checked. Recompute the headline totals yourself, or ask Claude to show the rows and the arithmetic behind them, before the number goes into a report or a decision.
What to watch out for
Watch for small samples dressed up as trends, averages hiding a split between two very different groups, and comparisons across periods that are not really comparable, such as a holiday month against an ordinary one. When a result would change a decision, check it a second way: a different grouping, a different time window, or a manual spot check of the underlying rows.
Claude is strong at structuring analysis and explaining results, but it cannot see data you have not given it, and it cannot know your business context unless you state it. The final judgement about what a number means is still yours.
Frequently asked questions
Can Claude analyse my spreadsheet directly?
Yes, you can share the data or paste the relevant table and ask for summaries, groupings, and comparisons. For large or sensitive files, share only the columns the analysis needs.
How do I know the numbers are right?
Ask Claude to show its working and the rows behind each total, then spot-check the headline figures yourself. Recompute anything that will go into a decision or a published report.
Can Claude make charts from my data?
Yes, through artifacts it can build charts and simple reports from the results. Ask for one chart per question and label each one with its source and time period.
What if my data is messy?
Clean it first: standardise spellings, fix date formats, and handle blank cells. Work on a copy and keep the original untouched so every cleaning step can be traced.
Is Claude a replacement for a data analyst?
For everyday summaries and one-off questions it goes a long way. For decisions with real money or risk behind them, a human analyst checking the work is still the safer choice.