Extract insights from a small dataset

The problem

You have a spreadsheet with 50-500 rows of data (donations, beneficiaries, events) and want to understand what it tells you, but most analytics tools assume thousands of records. Traditional statistical analysis won't work with such small samples, but you still need insights to inform decisions.

The solution

Use a structured LLM prompting approach to analyse small datasets directly. Export your data to CSV, anonymise any personal information, then upload to Claude or ChatGPT with specific analytical prompts. Ask the AI to identify patterns, outliers, trends over time, and segments in your data.

What you get

A written analysis identifying (a) patterns in timing or behaviour, (b) unusual values that warrant investigation, (c) natural segments or groups you can see, (d) trends over time, and (e) actionable recommendations. The AI can also generate simple visualizations or summary statistics.

Before you start

• Your dataset in a spreadsheet format (50-500 rows works best)
• Basic understanding of what the columns represent
• A Claude or ChatGPT account (free tier is fine)
• Data anonymised if it contains personal information

When to use this

• Dataset is too small for statistical significance in traditional ML (50-500 rows)
• You need quick insights without building analytics infrastructure
• One-off analysis rather than ongoing monitoring
• You want to explore the data before deciding whether deeper analysis is warranted
• Traditional tools are overkill for the size of your dataset

When not to use this

• Data contains sensitive PII that cannot be anonymised
• You need reproducible, auditable analysis for formal evaluation
• Dataset exceeds AI context limits (~100KB of text, roughly 1000 rows)
• You need statistical significance testing or p-values
• This will be an ongoing monthly analysis (set up proper analytics instead)

Steps

1. 1

   Clean and anonymise your data

   Open your dataset and remove or hash any personal identifiers: names, email addresses, phone numbers, addresses. If you have beneficiary IDs, replace them with generic labels (Person A, Person B). Remember that anonymisation means more than removing names - consider indirect identifiers like unique postcodes, rare job titles, or unusual combinations of traits that could identify someone in a small community. Remove completely empty rows or columns. Save a clean copy specifically for this analysis.

2. 2

   Export to CSV format

   Save your spreadsheet as CSV (comma-separated values). This format works best with AI tools. Keep the header row with column names, as these help the AI understand what each field represents. Check the file size - it should be under 100KB for free tier AI tools.

3. 3

   Craft your initial prompt

   Paste the CSV data into Claude or ChatGPT and use a structured prompt: "Analyse this [type of data, e.g., donation data]. Identify: (a) any patterns in timing or frequency, (b) unusual values or outliers, (c) natural segments or groups, (d) trends over time if applicable, (e) actionable insights or recommendations." Be specific about what you want to learn.

4. 4

   Review the initial analysis

   Read through what the AI identifies. Check if the patterns it finds make sense given your domain knowledge. Are the outliers genuinely unusual or just data entry errors? Do the segments it identifies feel meaningful? Note down anything that surprises you or needs investigation.

5. 5

   Ask targeted follow-up questions

   Drill into specific findings. Try: "Can you show me the 5 most unusual values in the [column name] field and why they're unusual?" or "What's the distribution of [field] across the segments you identified?" or "Are there any seasonal patterns in this data?" The AI can pivot quickly to new questions.

6. 6

   Request visualisation suggestions(optional)

   Ask: "What would be the most effective chart to visualise this pattern?" or "Can you give me the data I'd need to create a simple chart in Google Sheets?" The AI can suggest appropriate visualizations and prepare data in the right format for charting.

7. 7

   Document your insights and limitations

   Write up the key findings in a short report. Be explicit about the small sample size and what this means: "Based on analysis of 127 donations from Q3 2024. This sample is too small for statistical significance but reveals patterns worth investigating further." Always note the limitations.

Tools

Resources