Business Visualizations
A business visualization is the graphic representation of an analysis result, built for a business stakeholder so that they see the conclusion and decide on that basis. Two questions govern the choice. Does the visual assert a conclusion or look for one? Does it carry measured data or an idea? Crossing the two answers gives four cases. Each calls for a different family of visuals: a process model in one, a scatter plot in another. The deliverable is the finished visual, its chart family settled, its scale set, its message carried by a title written as a sentence and its accompanying text drafted, so that it reads without its author. The technique applies when the results of an analytics initiative are interpreted and reported.
Goal
A business visualization is built after the analysis, for a business stakeholder who has to decide from what they see in it. The charts an analyst produces for their own understanding of the data during the analysis are technical visualizations. The distinction is one of audience and moment: the same scatter plot is a diagnostic tool on Monday and a reporting exhibit on Thursday.
The problem addressed is the step from result to decision. An analysis produces a gap, a trend or a correlation; the committee that has to rule has a few minutes and rarely any statistical training. A table of numbers leaves the committee the comparison work a chart would do for it. The same chart, badly scaled, shows a conclusion the data does not carry: that is the limitation the IIBA guide lists first.
The finished visual then travels in a presentation, a report or a dashboard.
Usage
When to use it
- An analysis result to take to a committee: a gap read in three seconds is worth a page of commentary.
- A counter-intuitive conclusion to get accepted: showing the data moves the discussion from opinion to measurement.
- Recurring operational monitoring: a dashboard gives the current state without going through an analysis request.
- A trade-off between costed options: the comparison of three variants is easier to see than to read.
- Stakeholders unevenly comfortable with numbers: the visual gives them all the same basis for discussion.
When not to use it
- Analysis still running: the chart serves the analyst and belongs to technical visualizations.
- Values to be read off one by one: a reader looking for an amount wants a sorted table.
- Disagreement on how the number is defined: no chart form settles that, so go through metrics and key performance indicators.
Description
Two questions before choosing a form
Two closed questions set the purpose of the visual. Is it declarative or exploratory? The first asserts an established conclusion, the second looks for or confirms a fact in front of its audience. Is its content conceptual or data-driven? On one side an idea or a structure, on the other measured values. Crossing the two answers gives four cells, each calling for a different family of visuals. The model is Scott Berinato's, published in the Harvard Business Review and then in Good Charts; the IIBA guide takes it up as the technique's first element.
Idea illustration
conceptual · declarative
Process model, infographic, framework diagram, maturity curve.
Everyday dataviz
data-driven · declarative
Bar chart, line chart, pie chart.
Idea generation
conceptual · exploratory
Mind map, mental model, whiteboard sketch.
Visual discovery
data-driven · exploratory
Interactive visualization, small multiples.
Idea illustration occupies the conceptual and declarative cell. The visual breaks a complex idea down into a simple image, by metaphor, by analogy or by structure. Nothing in it is measured; it convinces only if the reader accepts the reasoning.
Idea generation occupies the conceptual and exploratory cell. The visual, a mind map for instance, is a way of thinking as a group during a workshop; its finished form matters little.
Everyday dataviz occupies the declarative data cell. The forms here are the ones everybody can read, one idea per visual. This is the cell most of an analyst's work falls into and the one where the discipline of scale and message is decided.
Visual discovery occupies the exploratory data cell. Here a new fact is sought or a hypothesis tested in front of the stakeholders, which means changing the breakdown, the filter or the form during the session. The same work done by the analyst alone belongs to technical visualizations.
The message dictates the chart family
With the purpose settled, the choice of form turns on the type of message. The IIBA guide distinguishes four intentions, the same ones Andrew Abela uses in his Chart Chooser, a widely reproduced selection table.
| Intention | What the reader has to see | Chart families |
|---|---|---|
| Comparison | which category is larger, by how much, at what date | bars, grouped bars, line charts, slope chart, butterfly chart |
| Composition | what a total is made of and what made it move | stacked bars, waterfall, stacked areas |
| Distribution | how the values spread around their centre | histogram, density curve, scatter plot |
| Relationship | whether two quantities vary together and in which direction | scatter plot, heat map, chord diagram, Venn diagram |
These four intentions frame the decision without closing it. The Visual Vocabulary of the Financial Times cuts the same ground into nine intentions, among them deviation from a reference, ranking and part of a whole: when the obvious form fails, it names the alternatives. A change in ranking is read on a slope chart, whereas two sets of bars leave the reader to do the matching.
The level of detail comes from the stakeholders
The IIBA guide makes the level of detail depend on what is known about the stakeholders: their areas of interest, their motivation, their technical acumen and the time they have. The same result is therefore reported at two levels depending on the room. A cantonal government IT programme has spent CHF 2'400'000 of the CHF 2'600'000 allocated. The steering committee, which decides whether to continue or stop, reads a single number, 92%, with the amount spelled out. The project office, which has to explain the rate of spending, reads the cumulative monthly curve with the allocation line on it.
Recipients are asked what decision they have to take; their answer gives the level of detail.
The key message and the accompanying text
A chart delivered bare leaves its reader to take from it what they like. The IIBA guide asks that the context, the notable statistics and the legends be explained and that the key message be written into the visual without cluttering it.
| Element | What it carries | Example |
|---|---|---|
| Title | the message, written as a complete sentence | "The portfolio renews 2.2 points below the industry average" |
| Subtitle | the scope, the period, the unit | "Household contents policies, French-speaking Switzerland, 2025, as % of policies falling due" |
| Reference line | the benchmark that gives the value its meaning | industry average line, target line, contractual threshold |
| Annotation | the fact the eye would miss | a mark on the April drop, with its cause in four words |
| Legend | the encoding, placed against the series rather than in a separate block | series label set at the end of the line |
| Source note | where the data comes from and the extraction date | "Policy administration system, extracted 3 March" |
A descriptive heading such as "2025 renewal rate" makes every reader work out a conclusion of their own. A title written as a sentence fixes the reading and makes the author answerable for the claim. A visual chained to others to build an argument belongs to data storytelling.
In what order?
The order of the sequence matters more than its duration. It begins with writing the message in one sentence, before any software: a message that will not fit into one indicates an unfinished analysis or two visuals to produce. Placement in one of the four quadrants follows, which rules out the unsuitable families, then the choice of intention, which points to two or three candidate forms.
The setting-up accounts for the rest of the work: axis baseline, bounds, category order, unit and rounding, colours that carry meaning. The text layer is written last. The final check is run on a reader who took no part in the analysis: they are shown the visual without commentary and asked what they read in it. The gap between their answer and the written message says what is left to adjust.
What makes the technique fail
The scale that decides for the reader
The IIBA guide illustrates the bias with two charts identical in data and different in vertical scale. This is the most widespread defect and the easiest to commit by accident, because spreadsheets frame their axes on the range of the values. The rule comes from the encoding, as Stephen Few establishes: a bar encodes its value by length, so its axis starts at zero, or the chart misstates the ratio between two bars; a line encodes by position and slope, so its axis may be truncated provided the truncation is signalled. Two vertical axes on one chart produce the same effect and worse, since where the two series cross depends on the settings. The same care applies to a set of small multiples, the same form repeated for each segment: scales that differ from one panel to the next make any comparison impossible.
The chart chosen before the message
The analyst starts from the dataset to hand and looks for the form that shows it off. The IIBA guide takes the example of a study of sales by region: the single-year series gives the most flattering visual, whereas the result concerns a region falling behind its peers after a year of growth, which calls for a comparison year on year and between regions. The remedy is the order: the message sentence first, the form after.
Complexity taken for rigour
The IIBA guide puts function before form: a third dimension rendered in perspective makes the coordinates unreadable, whereas the same variable carried by the size of the points reads without effort. The principle covers simulated depth, multi-ring pie charts, decorative gauges and eight-colour palettes. A visual that needs an explanation to be read has shifted its cost onto its reader.
The dashboard that grows
A dashboard fits on one screen and is read at a glance: that is Stephen Few's working definition. Requests nevertheless pile up one after another, each of them reasonable, until twenty tiles fill a screen nobody looks at any more. Every addition is therefore handled as a trade-off: what comes in either displaces an existing tile or goes into a detailed report reachable from the dashboard.
The visual left on its own
The IIBA guide notes that business visuals are for the most part summaries, so that a minor but significant trend disappears in the aggregation and the volume of data handled makes error likely. Two safeguards answer this: the source note, which allows a return to the dataset, and a look back at that dataset before circulation to check that the aggregation has not absorbed the trend the visual is meant to show.
AI considerations
A language model lowers the cost of iterating on form. From a natural-language description and a data extract, it produces the chart specification (Vega-Lite, ggplot2, matplotlib) and then the next variant in seconds, which makes trying four forms practical where the analyst stopped at one. It also drafts the title variants and the annotations, which the analyst then keeps or rewrites.
A multimodal model is a reviewer. Given a screenshot of a visual, with no context from the analysis, it says what it reads first: it stands in for an uninformed reader at little cost, and a gap with the written message signals a setting to revisit. It does not replace review by someone who knows the business, who alone spots that an April spike comes from the system migration.
Three decisions stay with the analyst. The scale setting first: a model reproduces prevailing practice, where the truncated axis is common, and an instruction asking for a more striking visual produces the very bias the technique fights. The level of detail next, which depends on what is known about the room. Confidentiality last: a management dashboard reveals more about the organisation than the file that feeds it, and sending it to an online service is settled before the first screenshot.
Examples
An insurer in French-speaking Switzerland compares its annual renewal rate in household contents insurance, 88.4%, with the industry average, 90.6%. The result is a gap of 2.2 points. The visual is addressed to the executive committee, which has to decide whether to open a customer-retention budget for the coming financial year. The message is written before any form: the portfolio renews 2.2 points below the industry average.
The purpose is declarative and data-driven, so everyday dataviz; the intention is a comparison between two values at the same date. Two bars and a reference line answer it. The analyst had the monthly series for the last five years and set the line chart aside: it answers the question of the trend, which the committee is not asking that day. The scatter plot crossing average premium and renewal by segment belongs to visual discovery: it calls for a working session, whereas the committee has three minutes. The same data is then carried on two bar charts that differ only in the baseline of the vertical axis.
Truncated axis
Zero baseline
The portfolio renews 2.2 points below the industry average
Household contents policies, French-speaking Switzerland, 2025, as % of policies falling due
Policy administration system, extracted 3 March
The left panel shows a collapse, the right panel a measurable gap that the reference line places. Both are accurate as far as the numbers go. The second also carries the sentence the analyst stands behind, the reference line that gives the scale of the judgement and the annotation that names the gap, which is what makes it the deliverable of the technique.
Visualizations
The grid of purposes is drawn as four cells of equal area, the declarative versus exploratory axis vertically and the conceptual versus data-driven axis horizontally. Each axis carries its two poles as labels, with no graduation. Each cell carries its name and the reading of the two answers that lead to it.
A before-and-after pair is drawn as two twin panels in which one variable alone changes, here the baseline of the axis. The graduations stay readable on both panels; otherwise the comparison cannot be verified. The compliant panel carries its full text layer: title written as a sentence, reference line, annotation and source note.
Cost
| Phase | Level | Rationale |
|---|---|---|
| Preparation | Medium | The analysis is finished and the data available. The effort goes into formulating the message and into what has to be known about the recipients to set the level of detail. |
| Execution | Low | A visual is produced in minutes once the family is chosen. The cost sits in iterating on form and setting the scales. |
| Documentation | Low | The visual is itself the deliverable; what remains is the source note and the definition of the indicators shown. A recurring dashboard shifts the cost to its maintenance. |
Tooling
The spreadsheet (Excel, Google Sheets) produces most of the business visuals in circulation. Its defaults work against the technique: automatic palette, perspective options. The rule of use is to set the baseline of the axis, the order of the categories and the labels by hand.
BI platforms (Power BI, Tableau, Qlik) serve the recurring visual and the dashboard connected to its source. They bring refresh and access control; in exchange, their templates impose a layout that is not always the message's.
Code libraries (Vega-Lite, ggplot2, matplotlib, D3) make the visual reproducible and versioned. They become necessary as soon as the same chart is regenerated every month; the specification is then the reference document.
The pencil sketch remains the first tool. Drawing the form before opening it in software costs two minutes and kills bad ideas while they are still easy to throw away. The Visual Vocabulary and the Chart Chooser do the same job on the selection side.
Sources
- IIBA, Guide to Business Data Analytics, §3.2 Business Visualizations: the purpose of the technique, the distinction from technical visualizations, the design guidelines, the model of purposes, the four intentions and their chart families, the place of the key message, together with the strengths and limitations stated there.
- Scott Berinato, "Visualizations That Really Work", Harvard Business Review, June 2016, developed in Good Charts: The HBR Guide to Making Smarter, More Persuasive Data Visualizations, Harvard Business Review Press, 2016: the original publication of the four-cell model the guide takes up, with its two axes and the names of its quadrants.
- Andrew Abela, Chart Chooser, Extreme Presentation Method, 2006: the selection table that ties a family of charts to each of the four intentions. Abela credits Say It With Charts by Gene Zelazny, p. 27 of the 4th edition, for the typology of messages.
- Stephen Few, Information Dashboard Design: Displaying Data for At-a-Glance Monitoring, 2nd edition, Analytics Press, 2013: the axis baseline rule derived from the mode of encoding, the critique of ornament and the definition of a dashboard as a display that fits on one screen.
- Financial Times Visual Journalism Team, Visual Vocabulary: the catalogue of chart forms organised by intention, kept up to date and freely accessible.