Exploratory data analysis is how an analyst gets to know a dataset before trusting it with a question.
It exists because data rarely matches its description, and surprises are cheaper to find early.
It also produces the questions that later analysis tests properly.
A typical pass starts with the shape of the data: rows, columns, types and blanks.
Each numeric column gets summary statistics and a histogram, and each category gets value counts.
Key pairs of columns get scatter plots or grouped summaries, such as revenue by region.
Data recorded over time gets a line chart, to spot trends, seasons and breaks.
Outliers, impossible values and odd spikes are noted and investigated, not deleted on sight.
Each finding is written down as a question, such as why churn doubles in March.
Exploration is open-ended, so it is easy to find patterns that are only noise.
Treating an exploratory pattern as proven is the multiple testing problem in disguise.
Findings should be confirmed on fresh data or with a planned test before anyone acts on them.
Keeping exploration in a notebook, with notes beside each chart, makes it easy to revisit.
In practice, a short summary of what was explored and what was surprising helps the rest of the team.
The usual mistake is skipping exploration under time pressure, then finding a data problem after the results are published.