How to choose the right chart for a business question
Choose a chart from the question and grain: bars for category comparisons, lines for ordered time trends, histograms or box plots for distributions, scatter plots for relationships, and labeled stacked bars for composition. Use a table when exact values or many dimensions matter. Avoid crowded pies, truncated bar axes, and dual axes that imply unsupported relationships.
The decisions that matter
- Write the question before choosing a chart.
- Aggregate to the grain the chart claims to show.
- Prefer position and length over area or decorative effects.
- Expose filters, units, denominators, and time coverage.
Question-to-chart guide
| Question | Start with | Avoid |
|---|---|---|
| Compare categories | Sorted bar or dot | 3D columns and crowded pies |
| Change over time | Line or small multiples | Unordered category lines |
| Distribution | Histogram or box plot | Average alone |
| Relationship | Scatter with context | Dual axes implying cause |
| Exact values | Table | Forcing a visual |
Where the difficulty begins
A chart can be valid yet answer the wrong question. Teams use bars for time, lines for unordered groups, pies with too many slices, or averages that hide distributions and sample size.
Chart selection is reasoning
The same columns support several questions. Revenue by month is a trend; revenue by market is a comparison; order-value variation is a distribution. Starting from types alone misses the decision.
Aggregation can mislead more than design
A beautiful chart of the wrong aggregate is wrong. Lifetime orders should not be summed across snapshots. Rates should be recomputed from totals rather than averaged across unequal groups.
From the field · Operations manager
Choose five charts for five different questions from one operations dataset
The same table contains monthly revenue, market, service type, delivery time, rating, and cancellation status. A single default bar chart cannot honestly answer ranking, trend, distribution, relationship, and composition questions.
Columns used in the example
month · market · service_type · revenue · delivery_minutes · rating · cancelled
Follow it with your own file
- Use sorted bars for revenue by market, with a zero baseline.
- Use a line for monthly revenue because time has a meaningful order.
- Use a histogram or box plot for delivery-time spread rather than only the average.
- Use a scatter plot for delivery time versus rating and a 100% stacked bar only when composition is the question.
Check: State the analytical question beside each chart, confirm the what one row represents and aggregation, and check that axes, sorting, labels, and missing categories do not change the conclusion.
Build the result step by step
- State the decision question: Identify what the reader must compare or decide, including measure, dimension, period, filter, and unit.
- Confirm analytical grain: Ensure each mark represents the intended unit. Do not sum snapshots or average already-aggregated percentages.
- Select the simplest valid encoding: Match question to chart family, sort deliberately, use honest axes, and label units and denominators.
- Test the interpretation: Ask a colleague to state the finding and caveats. Revise if the chart suggests more than evidence supports.
What may be distorting the answer
- The title names a dataset rather than a question.
- Readers ask for denominator or period.
- A dual axis makes unrelated series align.
- An average hides important variation.
Check the source before you publish
- Question stated
- Grain verified
- Aggregation appropriate
- Units labeled
- Axis honest
- Denominator visible
- Table considered
Doing these steps in Prepyr
Upload a CSV working copy in Prepyr. Describe the result you need: Choose the clearest chart for a comparison, trend, distribution, relationship, or composition question. Check the columns selected, preview any proposed changes, and compare the output with the source before applying or exporting it.
Prepyr steps used here: Chart recommendations · AI chart builder · Evidence notes · Report export
A chart that lets readers recover the main comparison quickly and understand the evidence limits.
Questions readers usually ask next
When should I use a pie chart?
For a small number of mutually exclusive parts of one whole when approximate composition is enough. Bars are usually easier.
Should a bar chart start at zero?
Usually yes because bar length encodes magnitude. If not, make the break unmistakable.
When is a table better?
When readers need exact values, lookup, many dimensions, or audit detail rather than rapid pattern recognition.