Working with Data & Analysis
Analyze, Transform, and Extract Insights
Data is everywhere. Most teams have more of it than they know what to do with. Sales spreadsheets that grow every month. Customer feedback sitting in a CSV. Survey results that nobody has had time to analyse. Financial reports that tell a story — if only someone could read it. The bottleneck is not data. It is analysis.
Claude compresses the time from raw data to meaningful insight through six capabilities: exploratory data analysis (understanding structure and quality), statistical analysis (descriptive stats, correlations, time series), Excel and Sheets automation (formulas and Apps Scripts), data visualization (chart code generation), and turning numbers into plain-language narratives. Always start with a data overview before any specific analysis, and always end with the business question — what decision does this data inform?
This chapter covers what Claude can and cannot do with data, how to use the Analysis Tool for deeper work, and structured workflows for every major data task — from exploratory analysis through to stakeholder-ready narratives.
What Can Claude Do with Data — and What Are the Limits?
- Exploratory analysis — Structure, patterns, outliers, data quality
- Descriptive statistics — Mean, median, SD, quartiles, distribution
- Correlation analysis — Relationships between variables
- Time series — Trend, seasonality, anomaly detection
- Cohort analysis — Retention and segmentation
- Cleaning and transformation — Standardise, reshape, derive columns
- Visualisation code — Python, JavaScript, multiple libraries
- Data narratives — Numbers into plain-language stories
- Very large datasets — Works best up to ~50MB; larger needs specialist tools
- Real-time data — Uploaded data is static; no live database connections
- Complex ML modelling — Basic stats yes; training custom models, no
- Guaranteed accuracy — Verify critical calculations; occasional errors on complex multi-step arithmetic
What Is the Analysis Tool and When Should It Be Used?
The Analysis Tool (also called Code Interpreter) is a significant upgrade for data work. When enabled, Claude can execute code, produce real charts and tables from actual files, handle larger datasets, and iterate on results within the same session.
- Quick calculations on small datasets you can paste
- You want a framework or approach, not executed results
- You're asking conceptual questions about data
- You need code to run yourself later
- You have an actual data file to upload and analyse
- You want real rendered charts and visualisations
- The dataset is too large to paste into conversation
- You want iterative analysis: explore → find something → go deeper
How Should Every Data Analysis with Claude Begin?
Before any specific analysis, always start with a data overview. This surfaces quality issues before they corrupt results, ensures Claude's understanding of the data matches reality, and identifies the most valuable analyses to run.
Addressing Data Quality
After the overview, fix quality issues before analysis:
How Do You Run Statistical Analysis with Claude?
How Do You Use Claude Effectively with Excel and Google Sheets?
The Excel Analysis Workflow
When working with an Excel file, follow three steps:
Writing Complex Excel Formulas
Claude can write formulas that would take an hour to research manually:
Real example:
Google Sheets Automation
How Do You Generate Data Visualisations with Claude?
Choosing the Right Chart Type
| Goal | Chart Type |
|---|---|
| Trends over time | Line chart |
| Comparing categories | Bar chart (vertical) or column chart (horizontal for many categories) |
| Part-to-whole relationships | Pie chart (max 5 segments) or stacked bar |
| Relationship between two variables | Scatter plot |
| Distribution of values | Histogram or box plot |
| Matrix patterns (cohorts, correlations) | Heatmap |
Generating Visualisation Code
Multi-Chart Dashboard
How Do You Turn Data Findings into a Clear Narrative for Any Audience?
One of Claude's most underused data capabilities is turning numbers into stories. Data only becomes valuable when stakeholders understand it — and most stakeholders do not think in statistics.
Turning a Dashboard into a Briefing
What Are the Most Effective Data Analysis Prompting Patterns?
The Business Question First
Instead of "analyse this data," start with the decision that needs to be made:
The Anomaly Investigation
The Comparison Request
What Are the Most Common Data Analysis Mistakes to Avoid?
Mistake 1: Skipping Exploratory Data Analysis
Mistake 2: Confusing Correlation and Causation
Mistake 3: Not Specifying Business Context
Mistake 4: Ignoring Data Quality
Mistake 5: Not Asking "So What?"
Analysis should answer something specific. The most powerful prompt in data work is not "analyse this" — it is "what does this mean for [the decision]?" Every table of numbers should end in an action or a recommendation.
- Start with EDA always — Understand the data before analysing it; quality issues caught early prevent corrupted results
- Business question first — Analysis should answer something specific and connect to a decision
- The Analysis Tool unlocks more — Execute code, produce real charts, handle larger files iteratively
- Excel and Sheets work well — Complex formulas, Google Apps Scripts, and narrative summaries all within reach
- Visualisation matches the message — Choose the chart type that fits what is being shown, not what looks impressive
- Numbers into narratives — Data only becomes valuable when stakeholders understand it in plain language
- Always ask "so what?" — Connect findings to decisions and actions; analysis without application is just interesting reading
Beginner: Take any spreadsheet used regularly. Ask Claude for a data overview and quality check. See what it surfaces that was previously unnoticed.
Intermediate: Pick a dataset with at least 3 months of data. Ask Claude for time series analysis and a visualisation. Use the narrative template to write an executive summary from the findings.
Advanced: Frame a real business question. Design the full workflow — EDA → cleaning → statistical analysis → visualisation → narrative → business recommendations. Execute it with Claude end to end.
Reflection questions: What did EDA reveal that was not previously known about the data? Did correlation analysis surface any relationships that were unexpected? How did turning the numbers into a narrative change the understanding of them?
Claude can perform exploratory data analysis (understanding structure, patterns, and quality issues), descriptive statistics (mean, median, standard deviation, percentiles), correlation analysis between variables, time series trend analysis including seasonality and anomaly detection, cohort analysis and segmentation, data cleaning and transformation, and generating Python or JavaScript visualisation code. It can also turn quantitative findings into plain-language narratives for any audience.
The Analysis Tool (also called Code Interpreter) allows Claude to execute code, produce real charts and tables from actual data files, handle larger datasets that would exceed the plain chat context limit, and iterate on results within a single session. Use plain chat when you need quick calculations on small pasted datasets, conceptual guidance, or code to run yourself later. Use the Analysis Tool when you have an actual data file to upload, want real rendered visualisations, or need iterative exploratory analysis.
Always start with a data overview request before any specific analysis. Ask Claude to describe the file contents, column names and data types, row count, time period covered, any obvious data quality issues such as missing values or inconsistent formats, and which analyses would be most valuable. This prevents wasted analysis on dirty data, surfaces issues that could corrupt results, and ensures Claude's understanding of the data structure matches reality before deeper work begins.
Claude can write complex Excel and Google Sheets formulas that would take an hour to research manually — including nested IF logic, VLOOKUP replacements, dynamic array formulas, and conditional calculations. It can also write Google Apps Scripts for automated recurring workflows such as weekly email summaries, data aggregation across sheets, and conditional formatting triggers. When uploading an Excel file directly, Claude can read values across all sheets, analyse the data, and produce narrative summaries.
Provide Claude with the key metrics or analysis findings and specify the document type (executive summary, board update, team report), the intended audience (CFO, all-hands, investors, technical team), and the desired length. Ask Claude to lead with the most important finding, use specific numbers to support claims, explain trends in plain language, and end with clear implications or recommended actions. Avoid passive voice and hedging language — specify that active voice and confident conclusions are required.