Different Types of Charts and When Each One Works

Most chart advice starts with the chart and works backwards, which is why so many reports open with a pie chart carrying twelve slices. The different types of charts are not a menu of decorations. They are answers to a small set of questions: how do these compare, what is this made of, how is it spread out, do these two things move together, and what changed. Pick the question first and the chart type stops being a matter of taste. It becomes something closer to a lookup.

This guide organizes chart types by the question each one answers, explains the failure modes of the popular ones, and is blunt about the cases where a chart is the wrong deliverable entirely.

Start with the question, not the chart

Before you open a charting tool, write the sentence you want a reader to say out loud after three seconds of looking. Not "here is revenue by region" but "the West is carrying the quarter and the Northeast fell off a cliff in May." That sentence tells you the comparison you need to make visible, and the comparison tells you the encoding.

Encoding is the technical word for how a number becomes a visual property: position along a scale, length, angle, area, color intensity, or shape. These are not interchangeable. In their 1984 work on graphical perception, William Cleveland and Robert McGill tested how accurately people read values from different encodings and found position along a common scale to be the most accurate, with angle and area considerably worse. That single finding explains most chart advice you have ever read, including the general preference for bars over pies.

So the practical rule: encode the thing you most want read precisely as position or length. Push secondary information into color, shape, or size, and accept that readers will judge those loosely.

The different types of charts, grouped by the question they answer

Here is the lookup table. Find your question in the left column, and use the failure column to check yourself before you ship.

Read the table as a starting point, not a verdict. A histogram and a box plot both answer "how is this spread out," and the choice between them depends on your audience and how many groups you are comparing at once. For a broader visual walkthrough of individual chart types with concrete use cases, our companion piece on examples of charts works through them one at a time.

Comparison charts: bars, and the things that beat bars

Comparison is the most common question in business reporting, and the bar chart is the correct default answer more often than any other chart type. Length starting from a shared baseline is easy to read, robust to color blindness, and works at any scale.

Two rules improve almost every bar chart. First, sort by value unless the category has a natural order such as months or survey scale points. An alphabetical bar chart makes the reader do the ranking that the chart was supposed to do for them. Second, start the axis at zero. Bar length encodes the value, so truncating the axis makes a 3 percent difference look like a 60 percent difference. Line charts are a different case, since they encode change rather than magnitude, and a non-zero baseline there is often defensible.

Use horizontal bars when the labels are long, which is most of the time with product names, campaign names, or survey responses. Vertical columns are better when the categories are short and ordered, such as months.

Grouped and stacked bars

Grouped bars compare several series across the same categories, for example this year against last year for each region. They stay readable to roughly three series. Past that, the eye has to hop across clusters and the comparison collapses.

Stacked bars answer a different question. They show composition within each category while still allowing rough comparison of totals. The catch: only the bottom segment sits on a shared baseline, so only that segment can be compared precisely across bars. Put the series you most care about comparing at the bottom.

Bullet charts and dot plots

When each item has an actual value and a target, a bullet chart beats a bar plus a legend. It shows the measure as a bar, the target as a perpendicular line, and qualitative ranges as background bands. It fits in a table row.

Dot plots, sometimes called Cleveland dot plots, are underused. When you are comparing two points per category, such as before and after, a dot plot with a connecting line reads faster than paired bars and uses a fraction of the ink. It also handles thirty categories in a space where bars would need scrolling.

Composition charts: parts of a whole

Composition questions sound simple and cause the most bad charts.

Pie charts are not evil. They are fine for two or three slices where one slice is obviously dominant and the exact values do not matter. Beyond that, readers are judging angles, which they do poorly, and comparing slices across two pies, which is worse. If your pie needs a legend, percentage labels, and a callout box, the pie is doing no work and a sorted bar chart will do it better.

Donut charts have the same limitations as pies with a hole in the middle. The hole is useful for a total, which is a real advantage in a dashboard tile, but it does not fix the angle-reading problem.

Stacked bars over time

The 100 percent stacked bar is the workhorse for "how has the mix changed." Each bar is normalized to full height, so you see share rather than volume. Pair it with a plain line or column chart of the total, because share and volume tell different stories and a normalized chart deliberately hides one of them. A category can grow every quarter in absolute terms while its share falls.

Treemaps and waterfalls

Treemaps encode value as rectangle area inside a hierarchy. They handle many categories in a compact space and show nesting, for example spend by department and then by vendor. Area is read imprecisely, so use treemaps for structure and outlier spotting, not for close comparisons.

Waterfall charts explain how a starting number became an ending number through a sequence of additions and subtractions. They are the right chart for bridging last quarter's revenue to this quarter's: new business up, expansion up, churn down, and there is your gap. Keep the steps to a handful and label each one with its signed value.

Distribution charts: the shape averages hide

An average is a summary that can be true and useless at the same time. Two support teams can both average four hours to first response, where one is consistently at four hours and the other splits between twenty minutes and eleven hours. Only a distribution chart shows the difference.

A histogram buckets a single numeric variable into bins and plots counts. The bin width is a real decision, not a default: too wide and you flatten the shape, too narrow and you show noise. Try a few widths before you conclude anything, especially about whether a distribution has two peaks.

A box plot, developed by John Tukey, compresses a distribution to median, quartiles, and outliers. Its strength is comparing many groups side by side, for example cycle time across fifteen teams in one view. Its weakness is that it cannot show bimodality at all: two very different distributions can produce identical boxes.

A violin plot or a strip plot with jittered points fixes that by showing the actual density or the actual observations. With fewer than a few hundred points, plot the points. Readers trust dots they can count, and outliers become individual cases rather than statistical artifacts.

Relationship charts: scatter plots and their cousins

The scatter plot is the only chart type that shows the raw joint behavior of two numeric variables without summarizing it away. Every point is a record. Clusters, gaps, curves, and outliers all become visible at once.

Two cautions. Overplotting is the first: with tens of thousands of points, the middle turns into a solid blob and you lose exactly the density information you came for. Fix it with transparency, hex binning, or sampling. The second caution is interpretive. A trend line through a scatter plot describes association, and nothing about the chart establishes which variable causes which, or whether a third variable drives both.

Bubble charts add a third variable as point area. They work when the third variable has a wide range and only rough comparison is needed, for example deal count as size on a plot of win rate against average deal value. Size the bubbles by area, not diameter, or you will exaggerate large values dramatically.

Heatmaps show a numeric value across two categorical dimensions using color intensity. They are excellent for pattern spotting at a glance: hour of day against day of week for support volume, or cohort against month for retention. Use a sequential color scale for values that run from low to high, and a diverging scale only when there is a meaningful midpoint such as zero or a target.

Change over time: line charts and their relatives

Time series data has a default and the default is the line chart. Position on both axes, continuity implying sequence, and a shape that reads instantly. Keep it to roughly five series before the lines start to tangle.

Some specifics that matter. Do not connect points across a gap in the data, since a straight line across a missing month is a claim you did not verify. Do not use a line for unordered categories, because the connecting segments imply a progression that does not exist. And when you plot a cumulative total, say so in the title, because cumulative lines only ever go up and readers habitually read rising lines as good news.

Area charts fill under the line to emphasize volume. Stacked area charts show composition over time, with the same baseline caveat as stacked bars: only the bottom band is easy to read.

Slope charts and small multiples

A slope chart plots two time points and connects each category with a line. When your real question is "who moved and in which direction between these two dates," it is far clearer than a full time series, because it removes everything except the change.

Small multiples, the grid of identical miniature charts with a shared scale that Edward Tufte popularized, are the correct answer to "I have twenty series and they all matter." Twenty lines in one panel is unreadable. Twenty small panels in a grid is scannable, and the shared axis keeps the comparison honest.

Sparklines

A sparkline is a tiny line chart with no axes, embedded next to a number in a table. It answers "is this number's trajectory normal" without spending a whole chart on it. Sparklines belong in tables, not on their own.

Different types of charts that usually fail, and what to use instead

Some chart types show up constantly and rarely earn their space.

• 3D bar and pie charts. Perspective distorts length and angle, which are the encodings doing the work. Nothing is gained. Use the flat version.
• Radar or spider charts. The shape depends on the arbitrary order of the axes, and the enclosed area has no real meaning. For comparing entities across several metrics, a small multiples grid of bar charts or a heatmap is clearer.
• Dual-axis charts. Two different scales on left and right let you manufacture any crossover point you like by adjusting the scales. Use two stacked panels sharing an x axis instead.
• Gauges and speedometers. Enormous canvas, one number, no context. A bullet chart or a big number with a sparkline delivers more in less space.
• Word clouds. Size encodes frequency, position encodes nothing, and long words look bigger than short ones. A sorted bar chart of the top terms is strictly more informative.
• Pie charts with more than five slices. Sort the categories, group the tail into "other," and use bars.

The pattern across all six: each one trades an accurate encoding for a decorative one. When you catch yourself defending a chart because it looks interesting, that is the trade you are making.

Your data shape decides more than the chart picker does

Charting tools assume tidy data: one row per observation, one column per variable, consistent types down each column. Most real exports are not tidy. They arrive wide, with a column per month, or with mixed formats in one column, or with subtotal rows sitting in the same table as detail rows. If you have ever produced a bar chart where "Total" was the tallest bar, you have met this problem.

Getting the shape right before charting is most of the work. Our guide to rows and columns in data covers the tidy layout that charting libraries expect and how to pivot between wide and long. When your grouping logic depends on conditions, such as bucketing deal sizes or labeling accounts by status, SQL CASE WHEN is usually the cleanest place to define those buckets so that every chart drawn from that query agrees. And if the same metric keeps coming out differently depending on who built the query, the fix is upstream in your model rather than in the chart, which is the territory covered in our overview of data modelling tools.

A useful discipline: name the metric definition in the chart subtitle. "Active accounts, defined as at least one login in the trailing 30 days" prevents an entire category of meeting argument.

Where charts stop and answers start

Not every question deserves a chart. Plenty of the reports people build are not really visualizations, they are recurring questions someone got tired of asking. "Did anything unusual happen in support volume this week." "Which deals slipped past their close date." "What changed in the pipeline since Monday." A dashboard makes you go look. A written answer comes to you.

Be clear about the split. If you need drag and drop dashboards, interactive filtering, and a library of visualization types, use a dedicated business intelligence tool. Metabase, Looker Studio, Tableau, Power BI, and Grafana all occupy that space, and their capabilities and pricing change often, so check current terms as of 2026 before committing. Skopx does not build dashboards or visualizations and will not replace those tools.

What Skopx does is the other half: it connects to nearly 1,000 business tools plus PostgreSQL, MySQL, and MongoDB, and answers questions in chat with citations back to the source. Skopx catches what falls between your tools, which in practice is the recurring question that never justified a dashboard of its own.

A concrete example. In Skopx chat you could type:

Query the Postgres sales database for revenue by region for last quarter and the same quarter last year, list every region that moved more than 10 percent in either direction, and tell me which single region drove the largest share of the total change.

What comes back is a written answer with the regional figures, the percentage moves, and the ranked contribution to the total, with the source it queried cited so you can check it. That is a paragraph you can paste into an update, not a chart you have to interpret. If you want the same answer every Monday without asking, you can describe that in chat as a workflow: a schedule trigger, the database query, an AI step summarizing the result on your own API key, and a step posting it to Slack. Workflows are described in plain English rather than assembled in a builder, and every run is inspectable step by step. They have real limits worth knowing: a maximum of 20 steps, no loops, no custom code steps, and no human approval steps. More detail lives on the workflows page.

Skopx is paid from day one, with no free tier: Solo is $5 per month and Team is $16 per seat per month with no seat cap. Full details are on the pricing page, and the integrations directory lists what it connects to.

Frequently asked questions

What are the main different types of charts?

They fall into five families by the question they answer. Comparison charts, mainly bar and column charts, compare categories. Composition charts, such as stacked bars, treemaps, and waterfalls, break a total into parts. Distribution charts, such as histograms and box plots, show how values spread. Relationship charts, such as scatter plots and heatmaps, show how two variables move together. Time series charts, such as line, area, and slope charts, show change. Maps and flow diagrams like Sankeys handle location and stage transitions.

How do I choose between a bar chart and a pie chart?

Use a bar chart in almost every case. Bars encode value as length from a shared baseline, which people read accurately, while pies encode value as angle, which people read poorly. Reserve pie charts for two or three slices where one clearly dominates and precise values are not the point. If you need a legend and printed percentages for the pie to be readable, the pie is not doing the work.

When should I use a line chart instead of a bar chart?

Use a line chart when the x axis is continuous and ordered, typically time, and you want to show the shape of change. Use bars when the categories are discrete and you want readers to compare magnitudes. If the categories have no natural sequence, do not use a line at all, because the connecting segments imply a progression that does not exist in the data.

Do I need a BI tool to make charts, or can a spreadsheet handle it?

A spreadsheet is genuinely fine for a chart that a few people look at occasionally, and it is faster to produce. Move to a dedicated BI tool when charts need to refresh automatically from a live source, when several people need consistent metric definitions, or when readers need to filter and drill down. Move to a code library such as Vega-Lite, Observable Plot, or matplotlib when you need precise control or reproducibility.

Can Skopx build charts and dashboards?

No. Skopx does not build dashboards or visualizations, and a dedicated BI tool is the right choice for that job. Skopx connects to your tools and databases and answers questions in chat with cited sources, delivers a daily morning brief on what changed and what is slipping, and runs workflows you describe in plain English. It fits the recurring questions and alerts around your dashboards rather than replacing them.

How many series can one chart hold before it stops working?

As a working guide: about three series for grouped bars, five for a line chart, and five slices for a pie. Past those limits, split into small multiples with a shared scale, or move to a heatmap or a sorted table. Adding a fourteenth line to a chart does not add information, it removes it.