Automated Descriptive Statistics

Understand Your Data in an Instant

Descriptive statistics are the foundation of any analysis. Mean, median, standard deviation, distribution: these indicators tell you what your data contains before you even try to explain anything. Skipping this step is like building a recommendation on an average skewed by three extreme values, then defending it with a conviction the numbers don’t deserve. With ChatGPT, this diagnosis takes a single sentence, and the interpretation follows immediately.

The Express Statistical Summary

The simplest command to get started:

Give me a complete statistical summary of this dataset.

ChatGPT generally relies on the pandas describe() function and comments on the result. For each numerical column, you get the following indicators:

• count — number of non-empty values
• mean — average
• std — standard deviation (measure of dispersion)
• min / max — extreme values
• 25% / 50% / 75% — quartiles (median = 50%)

Cross-referencing these rows is already insightful. If count is 8,400 in a file of 10,000 rows, 16% of the data is missing, and any average calculated applies only to part of the portfolio. If the mean differs significantly from the median, the distribution is skewed. Instead of doing this mental math yourself, just ask:

Interpret these statistics: which columns have high dispersion?
Are there outliers? Which distributions appear skewed?

Statistics by Category

Global averages almost always mask disparities. An average basket size of €180 could hide a corporate segment at €900 and an individual segment at €60. In that case, the average describes no real customer, and any decision based on it misses both populations entirely.

Calculate the mean, median, and standard deviation of order amounts
for each product category.

You still need to choose the indicator that answers your question, because each measures something different.

Distributions and Frequencies

For a numerical variable, the shape of the distribution tells you more than the average. A histogram shows whether your amounts cluster around a single value or spread across two distinct peaks, a sign of two overlapping purchasing behaviors in the same file.

Show the distribution of the "order_amount" column
with a histogram and indicate whether it follows a normal distribution.

ChatGPT generates the histogram and can apply a normality test, such as the Shapiro-Wilk test, which influences the choice of subsequent tests. On large samples, this test flags even the slightest deviation from normality, so check the histogram’s shape before drawing conclusions. For categorical variables, the logic shifts to frequencies:

What is the distribution of clients by industry sector?
Display absolute and relative frequencies.

You get the count and percentage for each category—enough to spot that one sector makes up 60% of the portfolio, meaning 60% of your risk if that sector slows down.

Correlations Between Variables

Correlations reveal relationships between your numerical columns and open up new avenues for analysis.

Calculate the correlation matrix and identify the pairs
of variables with the highest correlations (positive and negative).

ChatGPT generates the matrix, highlights significant correlations, and explains what each means in your business context. Three precautions are necessary when reading the results. First, correlation does not imply causation: a joint rise in advertising budget and sales might just be the common effect of seasonality. Second, extreme values distort the coefficient—ask for the calculation with and without outliers; the difference is often instructive. Finally, categorical variables are not included by default. To include customer segments or sales channels, request prior encoding. To go further—comparing groups with a t-test or ANOVA, or modeling a relationship with regression—describe what you want to understand, ask ChatGPT for the appropriate technique, and have it explain the result.

Detecting Outliers

An outlier is not necessarily an error. An order for €95,000 could be a data entry mistake or the year’s first major enterprise contract. The handling differs completely: in one case you correct it, in the other you highlight it.

Identify outliers in "order_amount"
using the IQR method. List the affected rows
and indicate whether they appear to be errors or legitimate cases.

ChatGPT calculates the IQR bounds (Q1 - 1.5 * IQR and Q3 + 1.5 * IQR) and lists the out-of-bounds values with their context. You then decide row by row rather than discarding the whole batch—and accidentally deleting the very clients who made your year.

Automating the Descriptive Report

Once you have mastered the approach, condense it into a single prompt:

Generate a complete descriptive statistical report for this file.
Include: overview, statistics per numerical column,
distribution of categorical variables, main correlations,
detected outliers, and recommendations for the next analysis step.

This prompt produces a structured document that you should review before sharing. Keep it handy: applied to the same export every month, it becomes your standard starting point and makes periods comparable.

Your Turn

Your first statistical diagnosis.

1. Request the statistical summary of your file, then its interpretation
2. Compare the mean and median on your primary amount column
3. Calculate the same indicators by category
4. Display the distribution of this column and identify outliers
5. Classify each outlier: error to correct or legitimate case

Official sources: Extracting Insights with ChatGPT Data Analysis · Data analysis with ChatGPT (OpenAI Help Center)

Key Takeaways

• The statistical summary is your first step after cleaning
• Always segment by category—global averages hide disparities
• The median is often more reliable than the mean for real-world data
• Correlations reveal avenues for analysis, not certainties
• Ask ChatGPT to interpret the numbers, not just calculate them