Examples of Charts: Choosing the Right One for Your Data

Most chart advice starts with the picture and works backwards. That is the wrong direction. The useful examples of charts are the ones tied to a question: you have a number that changed, or a set of things you want ranked, or two variables you suspect are related, and the chart is just the fastest honest way to show it. Pick the question first and the chart mostly picks itself.

This is a working catalog. For each chart type you get what question it answers, the shape of data it needs, a worked example, and the specific way it lies when you use it wrong. There is a decision table near the end you can bookmark. And there is one section on what to do when the answer you need does not actually require a chart at all, which is more often than people admit.

The selection heuristic: five questions, not fifty chart types

Before you open a chart picker, answer one question about your data. Almost every legitimate visualization falls out of the answer.

1. How did this change over time? You need a time axis. Line, area, or a column chart with time on the x axis.
2. How do these compare to each other? You need a categorical axis and a shared scale. Bar chart, almost always horizontal, sorted by value.
3. What is the composition of this whole? Stacked bar, treemap, or a very small pie. This is the category where people go most wrong.
4. How is this distributed? Histogram, box plot, or a strip plot. Averages hide here.
5. Are these two things related? Scatter plot, or a scatter with a third encoded dimension.

There is a sixth, less common one: where is this happening? which means a map, and maps carry their own trap that we will get to.

If your question does not map cleanly to one of these, that is a signal. Usually it means you have two questions bolted together and you need two charts. A single chart that answers "how did revenue change over time, and how does it split by region, and which regions are growing fastest" is a chart nobody can read. Three simple charts beat one clever one every time.

The other half of the heuristic is about the data shape underneath. If you cannot describe your dataset as a clean table where every row is one observation and every column is one attribute, you are not ready to chart it. That reshaping step is where most bad visualizations are born, and it is worth reading Rows and Columns: The Foundation Everyone Skips before you spend an afternoon fighting a chart library that is only doing what your malformed table told it to.

Examples of charts for change over time

Line chart

Question it answers: how did a continuous quantity move across a time range?

Data shape: one date column, one or more numeric columns. Regular intervals.

Worked example: weekly active users for the last 18 months, one line per plan tier. The eye reads slope, so you see the moment Team-tier growth flattened without needing anyone to point it out.

When it misleads: three ways. First, a truncated y axis. Starting at 94 instead of 0 turns a 3 percent dip into a cliff. This is sometimes legitimate (for a metric that lives between 94 and 99, zero is meaningless whitespace) but you must label the truncation. Second, irregular intervals plotted as if regular, which distorts every slope on the chart. Third, connecting points across a gap in the data, which invents a trend that was never measured. If February is missing, break the line.

Line count limit: past five or six lines, a line chart becomes a plate of spaghetti. If you need twelve series, use small multiples: twelve tiny identical line charts in a grid, same axes, arranged so the reader scans them. It is boring and it works.

Area chart and stacked area

Question it answers: how did a total change, and how did its parts contribute?

Worked example: monthly support ticket volume split into billing, technical, and onboarding. The top edge shows total load. The bands show what is driving it.

When it misleads: only the bottom band sits on a flat baseline. Every band above it rides on the ones below, so its shape is distorted by their movement. Readers consistently misjudge the middle bands. If the comparison between categories is the point, use separate lines. Use stacked area only when the total is the headline and the split is secondary.

The 100 percent stacked area, where every column sums to full height, is worse in a specific way: it hides the total entirely. Your "technical tickets share" can fall from 60 percent to 40 percent while the absolute number of technical tickets doubles. Show the absolute chart alongside it or expect someone to draw the wrong conclusion in a meeting.

Examples of charts for comparison

Horizontal bar chart

Question it answers: how do discrete items rank against each other?

Data shape: one categorical column, one numeric column.

Worked example: deals closed per account executive last quarter, sorted descending. Horizontal because names are horizontal text, and sorted because unsorted bar charts force the reader to do the ranking manually.

When it misleads: truncating the y axis on a bar chart is not a judgment call, it is a lie. Bars encode value by length. If the axis starts at 40, a bar of 50 looks half the size of a bar of 60, when the real ratio is 5 to 6. Line charts survive truncation because they encode by position. Bars do not. Start bars at zero, always.

Second failure: bars for things that are not comparable. Revenue per region in local currency, unconverted, produces a chart that is arithmetically valid and completely meaningless.

Grouped bar chart

Question it answers: how do items compare across a second dimension?

Worked example: ticket resolution time by priority level, grouped by quarter. Four priorities, four quarters, sixteen bars.

When it misleads: it degrades fast. Beyond roughly four groups of four, no reader can hold the comparison. And a grouped bar chart quietly asks the reader to make two comparisons at once, within group and across group, and the layout privileges one. Decide which comparison matters and group by the other.

Bullet chart

Question it answers: how does an actual value compare to a target and to qualitative bands?

Worked example: each sales rep's quarter-to-date number as a bar, a vertical tick for quota, and shaded background bands for below, at, and above pace. Dense, honest, and it replaces a wall of gauges.

When it misleads: rarely, which is why it is underused. The main risk is unlabeled bands: if nobody knows what "green" means, the chart is decoration.

Examples of charts for composition

This is where the worst examples of charts live, so it is worth being blunt.

Pie chart

Question it answers: what fraction of a whole is this one slice?

Worked example: a legitimate one. Three segments: 62 percent of signups came from organic, 27 percent from referral, 11 percent from paid. Three slices, one obviously dominant, the point is "organic is most of it."

When it misleads: the human eye compares angles and areas badly and compares lengths well. Two slices at 23 and 26 percent are indistinguishable. A pie with nine slices is a lookup table wearing a costume. And a pie whose values do not sum to a meaningful whole (percentages of respondents who could pick multiple answers, say) is simply invalid.

Practical rule: three slices, or use a sorted bar chart. Donut charts inherit every one of these problems and add a hole. The only defensible reason to use a donut is putting the total in the middle.

Treemap

Question it answers: how does a whole break down hierarchically when there are many parts?

Worked example: cloud spend by service, nested inside team. One glance shows that one team's data transfer line dominates the bill.

When it misleads: area comparison is imprecise, and long thin rectangles read as smaller than square ones of identical area. Treemaps are for spotting the dominant blocks, not for reading values. Label them with numbers.

Waterfall chart

Question it answers: how did we get from a starting number to an ending number?

Worked example: starting ARR, plus new business, plus expansion, minus contraction, minus churn, equals ending ARR. The bridge is the entire point and no other chart tells this story as clearly.

When it misleads: when the categories overlap or double count. A waterfall implies the steps are mutually exclusive and jointly exhaustive. If they are not, the chart asserts something false about your business. The categorization logic behind a waterfall is usually a pile of conditional bucketing, and getting that logic right is the real work: see SQL CASE WHEN: Patterns, Pitfalls, and Better Alternatives for how those buckets go wrong and how to keep them mutually exclusive.

Examples of charts for distribution and relationship

Histogram

Question it answers: how are values spread across a range?

Worked example: time to first response on support tickets. The mean says 4.2 hours. The histogram shows two humps: a large cluster under 30 minutes and a second cluster at 20-plus hours, which is tickets that arrived overnight. The average describes almost nobody's actual experience.

When it misleads: bin width is a free parameter and it changes the story. Too wide, the bimodality vanishes. Too narrow, you get noise that looks like structure. Try at least three bin widths before you believe any shape, and state the bin width on the chart.

Box plot

Question it answers: how do distributions compare across groups?

Worked example: page load time by region, one box per region. Medians, spread, and outliers side by side.

When it misleads: a box plot cannot show bimodality. Two very different distributions can produce identical boxes. If your audience is not fluent in quartiles, overlay the actual points (a strip or beeswarm) or use small multiple histograms instead.

Scatter plot

Question it answers: are two continuous variables related, and how?

Worked example: number of integrations a workspace has connected on the x axis, weekly queries run on the y axis, one dot per workspace. Clusters, outliers, and the general shape all appear at once.

When it misleads: the correlation-causation trap is famous and still catches people, particularly once a trend line is drawn, because a fitted line looks like a claim. Overplotting is the quieter problem: with 50,000 points, the dense middle is a solid blob and you have lost all information about density. Use transparency, sampling, or a hexbin. And be careful with a third variable encoded as bubble size, since readers judge bubble area poorly. Encode the third variable as color if you can.

The decision table

Two entries deserve a note. Dual axis line charts put two different units on left and right axes, and because the axis scales are arbitrary, you can manufacture or erase an apparent correlation by rescaling. If you need to compare two differently scaled series, index both to 100 at a common starting point and use one axis. Choropleth maps of raw counts almost always end up showing where people live. Normalize per capita or per household, or use a proportional symbol map instead.

When a chart is not the answer

A surprising share of chart requests are actually questions with a one-sentence answer. "Which customers dropped usage this month" does not need a visualization, it needs a list of seven names and their numbers. Building a dashboard for it means somebody has to remember to open the dashboard.

There are three real jobs hiding under "make me a chart," and they want different tools:

• Exploration. You do not know what is in the data. You genuinely need charts, and you need to make dozens of them quickly and throw most away. A notebook or a BI tool is right for this.
• Monitoring. A known metric, watched over time. A dashboard is right, but only if someone actually looks at it. Otherwise you want an alert.
• Answering a specific question. Someone asked something and needs a correct answer with its source. This does not need a chart at all, and it is where dashboards are worst, because the answer sits three filters deep in a view built for a different question.

Skopx sits squarely in that third category, and it is worth being precise about that: Skopx is not a BI tool and does not build dashboards or visualizations. If your job is exploratory visual analysis or a wall of charts for a leadership review, use a dedicated BI tool. What Skopx does is answer the question in chat across your connected tools and databases, including PostgreSQL, MySQL, and MongoDB, and cite where the answer came from.

A concrete example. In Skopx chat you would type:

Query the production Postgres for support tickets created in the last 90 days, bucket first response time into under 30 minutes, 30 minutes to 4 hours, 4 to 24 hours, and over 24 hours, and tell me how the distribution differs between weekday and weekend arrivals.

What comes back is the bucketed counts for each group with the SQL it ran shown alongside, so you can check the logic. That is the histogram question from earlier, answered as numbers, which is what you needed to know whether the overnight cluster is real before anyone builds a chart of it. If it turns out to be worth watching every week, you can describe a scheduled workflow in chat that reruns the query and posts the buckets to Slack every Monday, no builder to learn: see skopx.com/workflows. Workflows have real limits worth knowing up front: they are acyclic, capped at 20 steps, have no human approval step and no custom code step, and any AI step runs on your own provider key.

Skopx is a paid product with no free tier: Solo is $5 per month, Team is $16 per seat per month with no seat cap, and billing starts on day one. Full details are at skopx.com/pricing. AI runs on your own provider key with no markup on the AI cost.

Getting the underlying data right first

Every failure mode in this article has a data twin. Truncated axes are a presentation problem. Wrong denominators, double counted categories, and mismatched grain are data problems, and they produce charts that are beautifully rendered and completely wrong.

The most common one is grain mismatch: joining an order-level table to a customer-level table and then summing revenue, which multiplies revenue by the number of matching rows. The chart looks fine. The number is inflated. Nobody catches it until finance does.

Sorting this out before you visualize is unglamorous and it is the entire ballgame. If you are choosing tooling for that layer, Data Modelling Tools: Picking One for How Your Team Works covers the tradeoffs. And if you want a deeper reference on the chart families themselves, including the less common ones this catalog skipped, Different Types of Charts and When Each One Works is the companion piece.

Frequently asked questions

What are the most useful examples of charts for a business report?

Four cover the large majority of real reporting: a line chart for the trend, a sorted horizontal bar chart for the ranking, a waterfall for how a headline number moved between periods, and a plain large number with a comparison for the single metric everyone actually asks about. If you need more than these, you are usually answering more than one question and should split the report.

When is a pie chart acceptable?

When there are three or fewer slices, they sum to a genuine whole, and one is clearly dominant. The point of the chart should be "this part is most of it," not a precise comparison between slices. For anything with four or more categories, a sorted bar chart is easier to read and takes less space.

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

Ask whether the x axis is continuous. Time is continuous, so connecting points with a line asserts something meaningful about the space between them. Categories are not continuous, and a line drawn between "Marketing" and "Sales" implies a relationship that does not exist. Time gets a line, categories get bars.

What is the single most common way charts mislead?

The non zero baseline on a bar chart. Bars encode value by length, so cutting the axis breaks the proportional relationship the reader is trusting. Close behind it is the dual axis chart, where two arbitrary scales can be tuned until unrelated series appear to move together.

Can Skopx build charts or dashboards for me?

No. Skopx is not a BI or dashboard tool and does not build visualizations. It answers questions across your connected tools and databases in chat with cited sources, generates documents and reports, sends a daily brief on what changed and what is slipping, and runs workflows you describe in plain English. For visual analysis, pair it with a dedicated BI tool. You can see the connected tool coverage at skopx.com/integrations, which spans nearly 1,000 integrations.

How many charts belong on one dashboard?

Fewer than you think. A useful working limit is five to seven, each answering a distinct question, arranged so the most important one is top left. If a dashboard is genuinely being read every day it will stabilize at that size on its own. If it keeps growing, that is usually a sign that people want answers to ad hoc questions and a dashboard is the wrong shape for that need.