How to present data in a presentation: one question, one chart, one finding

Disclosure: NOXI publishes this guide and is one of the tools it mentions; the method works in any presentation software.

Most data slides fail for the same reason: they show everything the analyst found instead of the one thing the audience needs to see. This guide gives you the method, a chart chooser, the rules for numbers, scales and labels, a checklist and a worked example: a fictional quarterly KPI review rebuilt so that every title is a finding. The company and its figures are invented; the arithmetic is real and shown in full.

The method: question, comparison, chart, title

Presenting data well is four decisions made before any design work: the question the slide answers, the comparison that answers it, the chart that shows that comparison most directly, and the title that states the answer. Formatting comes last and matters least.

Make the decisions in order, because each one constrains the next. The last column follows this guide's worked example.

The UK Government Analysis Function's chart guidance puts the same test from the other end: if you cannot write down the message of your chart in a few sentences, reconsider the chart. The charts you made while analysing help you find a pattern; the charts you present show one pattern to people who have seconds to read it.

Count slides by questions, not by charts: four real questions need four data slides plus a summary and a decision, however many charts the dashboard holds. For the arc of a whole deck, our guide to consulting-style slides covers the answer-first structure and the ghost-deck method.

Anatomy of a data slide: six parts

A finished data slide has six parts: a title that states the finding, a subtitle with the measure, unit and period, a visible baseline or comparison, one accent colour on the point that matters, an annotation that explains it, and a source line. Anything else must earn its place.

The Analysis Function recommends the same split for published charts: a headline title that communicates the main message, and a statistical subtitle that says what the data is, the geography it covers and the time period shown. On a slide, the subtitle is one short line under the title.

Chart chooser: match the chart to the question

Choose a chart by the question it must answer: a line for change over time, sorted bars to compare categories, a waterfall to explain how one total became another, a stacked bar for parts of a whole, a scatter plot for a relationship, and a table when people need exact values. A chart chosen for variety usually needs a speaker to explain it.

The example titles below are illustrative. Each one states a finding the chart beside it can prove.

For a larger catalogue, the Financial Times' Visual Vocabulary groups chart types by nine relationships in the data: deviation, correlation, ranking, distribution, change over time, magnitude, part-to-whole, spatial and flow. Use it when none of the rows above fits.

Bar chart or pie chart?

Use a bar chart by default. People compare positions and lengths more accurately than angles and areas: in a 1986 experiment that followed their 1984 paper on graphical perception, William Cleveland and Robert McGill found position judgments the most accurate, length second, angle and slope third and area last. Jeffrey Heer and Michael Bostock repeated the comparison with online participants in 2010 and again found angle (pie) and circular-area (bubble) judgments worse than position judgments.

A pie still works for a simple part-to-whole message. The Analysis Function allows one only for five categories or fewer that add up to a meaningful whole, with one dominant category; if the parts are similar in size, it advises a bar chart. "Two-thirds of support tickets now arrive by chat" works as a two-slice pie; eight similar slices do not.

Write the title as the finding

A data slide's title should state what the data shows, as a full sentence with the number and the comparison, for example "Q3 revenue missed plan by 8.6%", rather than label the topic ("Q3 revenue"). Someone who reads only the titles should be able to follow the whole argument.

There is evidence that it helps. In a 2006 study in a large geoscience course at Virginia Tech, Michael Alley and colleagues taught some sections from slides with mostly phrase headlines and others from slides whose headlines stated the main assertion as a sentence. On 15 exam questions about those assertions, the sentence-headline groups averaged 79% correct against 69%, a statistically significant difference. The redesigned slides also replaced bullet lists with visual evidence, so the result supports the pattern as a whole.

A reliable formula is [measure] [moved how much] [against what] [because, after or so what]. Five rewrites:

Keep titles under about 15 words and two lines, with the rounded number the audience should remember and a verb: grew, fell, missed, held. Match the wording to the evidence: write "after" when you have timing, and "because" only when the slide or the appendix shows why. "No clear difference between the two pilots after six weeks" is a finding too.

Give every number a comparison

A number on its own is not a finding; it needs a comparison that tells the audience whether it is good, bad or normal. Pick the comparison that matches the decision: plan or target for performance reviews, the prior period for momentum, the same period last year for seasonal businesses, and a benchmark for competitive position.

Show one comparison per chart. Actual, plan, last year and last quarter in four colours leave the audience guessing which one the title means; put the comparison the meeting is about in the chart and the others in the subtitle or a scorecard. When a figure is a forecast or an estimate, label it as one; our guide to presenting forecasts and uncertainty covers ranges, scenarios and revisions.

If you show actuals, plans and forecasts together, use one visual convention everywhere. The International Business Communication Standards (IBCS) draw actuals as solid dark shapes, plans as outlined shapes and forecasts as hatched shapes, and show a previous period in a lighter tone when it is the comparison, under rule UN 3.2, "Unify scenarios" (IBCS working group note). The anatomy figure above uses the same idea: an outlined plan bar and a solid actual bar.

Get the numbers right: units, rounding and percentages

Round numbers on slides to the precision a decision needs, usually two or three significant digits; put the unit and period in the subtitle; use the same number of decimal places for numbers that are compared; and write "percentage points" when a rate changes by subtraction. In practice, many credibility problems in data decks are number problems, not chart problems.

Rounded parts may not add up to the rounded total; add "Figures may not sum due to rounding" rather than adjusting a component to force it. Before any design work, recalculate every percentage from its inputs, confirm that segments add up to the total and check that all figures share one data cutoff. Our guide to turning spreadsheet data into a briefing deck covers preparing the workbook so these checks are quick.

Scales and axes: when to start at zero

Bar and column charts should start at zero, because readers compare bar lengths and a cut axis exaggerates differences. Line charts may use a non-zero axis when the movement is the message, but mark the break, label the scale and keep the same scale on slides that will be compared.

The Analysis Function calls breaking the axis on a bar chart "problematic and highly controversial" and suggests a Cleveland dot plot when a zero baseline hides the story; it accepts a broken y-axis on a line chart when necessary. Research supports caution even with warnings: in crowdsourced experiments, Michael Correll, Enrico Bertini and Steven Franconeri found that truncated y-axes made differences look larger, and that the effect persisted when the truncation was flagged visually.

• Bars from zero. If the differences that matter are small relative to the values, use a dot plot, or plot the differences themselves as variance bars around zero.
• One scale for comparable charts. Small multiples and slides that will be compared side by side share the same axis range.
• No dual axes. Two y-axes suggest a relationship created by the choice of scales; the Analysis Function advises separate charts, each with its own y-axis.
• Time spaced as time. A missing month is a gap in the line, not a join.
• Log scales only for audiences who read them, labelled as logarithmic.
• No 3D. Perspective changes the apparent size of bars and slices.

Honest scales sometimes make a difference look small. That is information: in the worked example, the traffic and basket steps are slivers beside the conversion step.

Highlight and annotate the point that matters

Draw context in grey and the element the title talks about in one accent colour, label lines and bars directly instead of using a legend, and write a short annotation on the chart where the important thing happened. The audience should see the finding before you say it.

• One accent colour. If everything is coloured, nothing is highlighted.
• Direct labels. Series names at the ends of lines, values on or beside bars; the Analysis Function prefers this to legends.
• Annotations that explain. Mark launches, price changes, outages and definition changes where they apply, in a few words.
• Four lines at most. The Analysis Function warns that more than four lines often clutters a chart; use small multiples instead.
• Less decoration. No shadows, 3D or boxed legends; a few light gridlines at most.

Never let colour carry the meaning alone. The Web Content Accessibility Guidelines require that colour is not the only visual means of conveying information, and that chart elements needed to understand the content have a contrast ratio of at least 3:1 against adjacent colours. It is practical too: the US National Eye Institute says about 1 in 12 men have a colour vision deficiency, and the most common type makes red and green hard to tell apart. A red-and-green variance table should also use signs, arrows or labels. For the full review of an exported deck, see our presentation accessibility checklist.

Tables or charts: when to show the numbers

Use a table when the audience needs exact values or will look up individual numbers, and a chart when they need to see a pattern, a trend or a difference. In a live presentation, a table works best when you highlight the row or cell the title is about.

The Analysis Function's table guidance draws the same line: tables to compare values or to show values of very different sizes, charts for patterns, trends and relationships.

Format slide tables the way statisticians do: right-align numbers and their headings, keep one precision per column, put units in the headers (or the title, if all columns share one), keep gridlines light and put totals at the bottom or right. Then add one emphasis: bold the row the title is about. A KPI scorecard is such a table with one row per KPI, ordered by importance, and status shown with a word or symbol as well as colour. Our QBR presentation guide covers scorecard design for quarterly business reviews in more depth.

Sources, definitions and the appendix

Every data slide needs a source line that names where the numbers come from and for which period, and every data deck needs an appendix with the definitions, full tables and cuts that people will ask about. The main deck carries the argument; the appendix carries the proof.

Write the source line in the same place and size on every slide. The Analysis Function's format for published charts is the publication or dataset, then the organisation, with a direct link rather than a homepage. Inside a company, "Source: order system and web analytics, Q3 2026; team analysis" is enough.

The appendix holds, each slide titled by the question it answers: metric definitions ("Conversion rate = orders ÷ sessions"), method notes such as the order of steps in a bridge, the full scorecard and detailed tables, the segment cuts people are likely to request, and data-quality notes. Figures from outside your organisation need their publisher and year on the slide, and someone should open each source before the deck circulates, especially when an AI tool supplied the number. Our guide to AI presentations with research and citations covers checking them.

How to talk through a data slide

When presenting a chart live, say the finding first, then orient the audience (what is plotted, the units, what the colours mean), point to the evidence, and finish with what it means for the decision. As a rule of thumb, a data slide needs 60 to 90 seconds; if it needs five minutes, it probably holds more than one finding.

1. Claim. "Mobile conversion fell sharply in August."
2. Orient. "This shows conversion by device, by month. Mobile is in red, desktop in grey."
3. Point. "Mobile was at 2.20% in July, the same as in Q2. From August it is 1.45%. Desktop stayed at 3.00% throughout."
4. Explain. "The drop starts when the new mobile checkout launched on 4 August; the desktop checkout did not change."
5. So what. "At current traffic, that costs about 2,760 orders a month. The decision on slide 7 is how we test the fix."

Do not read every number: the audience will remember one or two. Reveal complex charts in stages, such as the plan bar first, then each step of the bridge. A pre-read has no presenter, so put the explanation on the slide as an annotation. To plan the length of the talk, see our slides-per-minute table.

Worked example: a quarterly KPI review, before and after

Quillmoor Home is a fictional online homeware retailer reviewing Q3 2026 (July to September) with its leadership team. The first draft opened with a slide titled "Q3 KPIs" that put five metrics on one clustered chart, followed by topic-titled charts such as "Conversion rate". The rebuilt section answers the review's real question, and every number below can be recalculated.

The numbers

Two true headlines compete: revenue grew 17.9% on last year, and it missed plan by 8.6%. A review against plan exists to answer the second, so the deck leads with the miss.

Step 1: split the miss into drivers

Because revenue is the product of three drivers, the gap to plan can be split into a traffic effect, a conversion effect and a basket-size effect, changing one driver at a time:

The steps reconcile: $2,880,000 + $64,000 − $353,280 + $40,480 = $2,631,200. Traffic and basket size added $104,480; lower conversion removed $353,280. The order of steps matters a little: here the 5,520 lost orders (1,840,000 × 0.30%) are priced at the planned $64.00, while putting conversion last would price them at $65.00 and show −$358,800 for conversion and +$46,000 for basket size. Neither is wrong; state the order in a footnote and keep it every quarter.

Step 2: find where the conversion fell

Splitting conversion by device and month shows where it fell:

Desktop had 230,000 sessions a month and converted at 3.00% throughout, as it did in Q2. Mobile converted at 2.20% in July, its Q2 rate, then at 1.45% in August and September, after a redesigned mobile checkout launched on 4 August. Had mobile held 2.20%, August and September would have produced 2 × 368,000 × 0.75% = 5,520 more orders: exactly the conversion shortfall in the bridge.

At August–September traffic, each further month at 1.45% instead of 2.20% costs about 2,760 orders (368,000 × 0.75%), or about $179,400 at the Q3 average order value of $65.00. That number turns the diagnosis into a decision.

Step 3: rebuild the slide

The title says "after", not "because". The timing, the device split and the unchanged desktop flow support the checkout explanation without proving it, so the deck proposes a test rather than declaring a cause.

The rebuilt section, slide by slide

Read the titles in sequence and the review makes sense without a single chart: the test for any data deck.

We also generated the after version as a deck in NOXI from the slide titles and numbers above and exported it to PDF without edits. Two of its slides:

Data-slide checklist

Run this checklist on every data slide before the deck circulates. If a slide fails the first two checks, fix them before touching the design.

3 AI prompts for data slides

Use them in NOXI after attaching your spreadsheet, or in any AI tool that has your data. Replace the brackets.

How to build data slides in NOXI from your spreadsheet

NOXI turns an Excel or CSV file into a structured deck with charts, in the style of your own template. It is free, with no card and no watermark; uploading files needs a free account, and a full deck uses about 25 of the 1,000 monthly credits.

Frequently asked questions

What is the best way to present data in a presentation?

Give each slide one question and one comparison, choose the chart that answers that question, and write the answer as the slide title with its number, such as "Revenue missed plan by 8.6%". Show the baseline the number is judged against, highlight the point that matters in one colour, annotate it, state the unit and period, and add a source line. Put definitions and detailed tables in an appendix.

Which chart is best for comparing values?

A bar chart sorted by value is usually best for comparing categories, because people judge lengths and positions on a common scale more accurately than angles or areas. Use horizontal bars when labels are long, start the axis at zero and highlight the bar the title is about. To compare against a target, add a target line; to compare change over time, use a line chart.

Should I use a bar chart or a pie chart?

Use a bar chart by default. A pie chart suits only a simple part-to-whole message with a few parts that add up to a meaningful whole and one clearly dominant part; the UK Analysis Function sets the limit at five categories. When slices are similar in size, readers cannot rank them reliably by angle, so a sorted bar chart shows the same data more clearly.

How many numbers or charts should one slide have?

One chart and one finding per slide is the reliable rule for a live presentation, with the one or two numbers that prove the title labelled on the chart. A scorecard table can hold five to seven KPIs if one row is highlighted. Pre-read decks can carry more detail, but each slide should still lead with a single finding in its title.

Should the y-axis always start at zero?

For bar and column charts, yes, because readers compare bar lengths and a cut axis exaggerates differences. Line charts may use a non-zero axis when the movement is the message, provided the axis is labelled and any break is clearly marked. If zero hides a meaningful difference in a bar chart, switch to a dot plot or plot the differences as variance bars.

When should I use a table instead of a chart?

Use a table when the audience needs exact values, will look up individual numbers, or must compare several measures with different units, as in a KPI scorecard. Use a chart when the message is a pattern, a trend or a difference. On a slide, right-align numbers, keep consistent decimals in each column, put units in the headers and highlight the row the title is about.

How do you present survey results in a presentation?

Show the question wording, the number of respondents and the survey dates in the subtitle, and give percentages with their base, such as "64% of 412 respondents". For agreement scales, use 100% stacked bars ordered from most negative to most positive. Compare groups only when the samples are large enough to support the difference, and title each slide with the finding, not the question.

Sources & methodology

How this guide was made: it applies published data-visualisation guidance and graphical-perception research to presentation practice. The method, chart chooser, title formula, timings and checklist are editorial recommendations, not a standard. Quillmoor Home, its figures and its checkout change are fictional; the arithmetic is real. All sources were checked on 9 October 2026.

• [1] Headline title and statistical subtitle; stating the chart's message; zero baselines for bars and the dot-plot alternative; axis breaks on line charts; pie-chart conditions; dual axes; line limits; direct labelling; source format — UK Government Analysis Function, Data visualisation: charts (published 19 May 2022).
• [2] When to use tables; right-aligned figures, consistent precision, units in headers, totals and gridlines — UK Government Analysis Function, Data visualisation: tables (published 19 May 2022).
• [3] Percentage points versus percentage change; consistent decimal places in a sequence of percentages — Office for National Statistics, Service manual: Percentages (no date displayed).
• [4] Accuracy of basic graphical judgments: position, then length, then angle and slope, then area — William S. Cleveland and Robert McGill, An experiment in graphical perception, International Journal of Man-Machine Studies 25, 491–500 (1986).
• [5] Crowdsourced replication of the position, length, angle and area comparisons — Jeffrey Heer and Michael Bostock, Crowdsourcing Graphical Perception: Using Mechanical Turk to Assess Visualization Design, CHI 2010.
• [6] Truncated y-axes increase perceived effect sizes, even when the truncation is signalled — Michael Correll, Enrico Bertini and Steven Franconeri, Truncating the Y-Axis: Threat or Menace? (CHI 2020; arXiv, revised January 2020).
• [7] Sentence headlines and audience recall in a geoscience course (79% against 69% on 15 questions) — Michael Alley, Madeline Schreiber, Katrina Ramsdell and John Muffo, How the Design of Headlines in Presentation Slides Affects Audience Retention, Technical Communication 53(2), 225–234 (May 2006).
• [8] Colour must not be the only visual means of conveying information — W3C, Understanding WCAG 2.2: Success Criterion 1.4.1 Use of Color.
• [9] A 3:1 contrast ratio for graphical objects needed to understand content — W3C, Understanding WCAG 2.2: Success Criterion 1.4.11 Non-text Contrast.
• [10] About 1 in 12 men have a colour vision deficiency; the most common type affects red and green — US National Eye Institute, Color Blindness (last updated 5 November 2025).
• [11] Chart types grouped by nine data relationships — Financial Times Visual Journalism Team, Visual Vocabulary (no date displayed).
• [12] Rule UN 3.2, "Unify scenarios": actuals solid, plans outlined, forecasts hatched, previous periods lighter in comparisons — IBCS Association, Work group 3 for IBCS Version 1.2: Application of semantic notation (12 December 2018).

Build your next data deck in NOXI, free: attach your spreadsheet, match your company template, review finding titles in the outline, check every chart in the data editor, and export an editable PowerPoint or PDF.