Show data so people understand it, not to impress them.
One day to learn how to choose the right visualisation, strip it down and make it readable for everyone, on a desktop screen or a phone, whether the chart is built by hand or suggested by an AI.
The agenda
The 2012 workshop ran to 106 slides centred on QlikView. Its foundations still hold, because human perception has not changed. Everything else has been revised: the tools, the screens, accessibility, AI, and what research has learned since.
Each participant brings a report or dashboard they actually use. We diagnose it in the morning and rebuild it in the afternoon during the makeover workshop.
Analysts, BI developers (Qlik Sense, Power BI, Tableau) and business users who design or commission reports. No technical prerequisites.
Why design matters
A badly designed report is not just ugly: it wastes time, hides the signal, and eventually nobody opens it. Design decides whether an application gets adopted.
“Effective data presentations are rooted in a constant awareness of the goal, which is to communicate, not to entertain or impress.”
Stephen Few, the opening quote of the 2012 workshop, still true
2012 → 2026: what changed
- QlikView on a desktop, 1024 × 768 screen, no scrolling
- The BI developer designs every object alone
- One dominant reference: Stephen Few
- Colours picked “by eye”
- Dashboard = one fixed screen, a “How to” page for help
- Qlik Sense / Qlik Cloud, Power BI, Tableau: responsive sheets, phones, dark mode
- AI generates charts from a question (Qlik Answers, Copilot, Tableau Pulse)
- Several complementary frameworks: Few, Knaflic (storytelling), IBCS, FT Visual Vocabulary
- Palettes validated for colour blindness, WCAG contrast, European Accessibility Act in force since June 2025
- Overview first, then filter, then details on demand, with alerts and pushed summaries
Our eyes haven't changed
Preattentive attributes, Gestalt, limited working memory: the whole perception part of 2012 still applies.
The screen has changed
We no longer design for a fixed resolution. A dashboard must reflow and stay readable at phone width.
The author has changed
An AI can produce a first draft in seconds. The skill shifts towards reviewing and correcting.
The evidence has matured
Pie charts, uncertainty, encodings: recent research nuances some of the absolute rules of 2012.
Take out the report you brought and answer three questions in writing:
- What decision does this report help make, and who makes it?
- What should you see in 5 seconds?
- When did you last open it, and what for?
Visual perception
We don't see with our eyes, we see with our brain. Understanding how it processes an image lets you decide what will jump out and what will stay in the background.
Iconic memory
Under half a second, automatic and unconscious. This is where preattentive processing happens: we spot things without searching.
Working memory
Temporary and very limited: today it is estimated at about 4 items at once (Cowan), rather than the often-quoted “7 ± 2”. A legend with 9 colours overloads it.
Long-term memory
Learned conventions: red for negative, time from left to right, green for “good”. Build on them rather than contradict them.
Click the odd one out as soon as you see it. Your reaction time is shown. Then try the “conjunction”: colour and shape combined.
Which attribute for which data?
The 2012 table sorted attributes into “quantitative: yes / no”. The experiments of Cleveland & McGill (1984), replicated at scale by Heer & Bostock (2010), allow a finer ranking, which Tamara Munzner summarised: from most to least accurate for reading a quantity.
What percentage of the larger marked value (●) is the smaller one?
Gestalt principles
“Gestalt” means “form” or “pattern”. The brain spontaneously groups what it sees, and we use these groupings to organise a dashboard without adding frames or lines.
Key takeaways
- One preattentive attribute at a time to make information stand out. A conjunction (colour + shape) no longer pops out.
- Quantities go on positions and lengths; categories on hues, shapes or areas.
- Group by proximity and alignment before adding a frame. Enclosure is the strongest grouping: save it for what really matters.
Colour & accessibility
Colour is the most powerful tool and the most misused. The 2012 rules remain right: soft colours, a neutral background, one colour = one meaning. We now add a verifiable method so that everyone can read the chart.
Every colour has a job
Readable by everyone
~8% of men
and about 0.5% of women (of Northern European descent) have a colour vision deficiency, most often red-green. In a room of 25 people, statistically at least one participant.
WCAG 2.2
1.4.1: colour is never the only way to convey information. 1.4.11: graphical marks at 3:1 against their surroundings. 1.4.3: text at 4.5:1.
Validated palettes
Okabe-Ito (categorical), viridis or cividis (sequential), and palettes built in a perceptually uniform space (OKLCH), then tested with a simulator. Never a rainbow (“jet”).
Key takeaways
- Gray by default, one colour for what matters. The same entity keeps the same colour on every page, even after filtering.
- Sequential = one hue from light to dark. Diverging = two opposite hues and a neutral gray in the middle.
- Status colours (good, warning, critical) are reserved and always paired with an icon or a label.
- Dark mode is designed: you pick other shades for the dark background, you don't just invert.
Choosing the right chart
Start by writing the message in one sentence. The chart type follows from what the reader needs to do with that message: compare, track, locate, relate.
A table when…
- people look up individual values
- precise values are required
- several units of measure are mixed
- detail and totals are shown together
A chart when…
- the message is in the shape of the data: trend, gap, exception
- you want to reveal relationships between sets of values
- you compare magnitudes at a glance
The visual vocabulary
The 2012 “Graph Selection Matrix” (Few, Abela) was taken up and expanded by the Financial Times graphics team in its Visual Vocabulary: nine families based on the relationship you want to show. Click a thumbnail to open its card in the decision tree.
The 2012 charts, revisited
Integrity: don't mislead
A chart can be accurate and still give a false impression. Distances on screen must stay proportional to the differences in the data.
Other traps to check
- Bars and areas start at zero. A line chart may zoom in, provided this is clearly shown.
- Bubbles and pictograms: the value scales the area, not the radius. Doubling the radius quadruples the surface.
- Maps: normalise (per capita, in %), otherwise the map simply shows where people live.
- Time window: showing “since the trough” or “since the peak” changes the story. Justify the period chosen.
- Misleading averages: look at the distribution before summarising with a single figure.
Data-ink & tables
Edward Tufte called “data-ink” the ink that actually represents data. Remove the rest, then strengthen what matters.
Tables: the 2012 rules still apply
Delineate
White space and very light rules rather than a full grid. Alternating row shading only for very wide tables.
Align and format
Numbers right, text left, tabular figures. Same number of decimals, unit in the header (k€, M€), thousands separator.
Organise
A useful sort, columns in reading order, subtotals per group, a well-placed total. An icon or a colour flags exceptions, sparingly.
Dashboards
“A dashboard is a visual display of the most important information needed to achieve one or more objectives, consolidated and arranged on a single screen so the information can be monitored at a glance.” Few's definition still holds; what has changed is the size of the screen.
- Dashboard: the summary sheet
- Analysis: the thematic sheets
- Reporting: the detailed tables
- Buttons, conditional show and document chaining to navigate
- Overview first: 3 to 6 indicators, with their comparison (vs target, vs last year)
- Zoom and filter: a single filter bar that drives the whole page
- Details on demand: tooltips, side panels, contextual navigation, a question asked to the AI
- Alerts and summaries pushed (e-mail, Teams, mobile) so you no longer have to open the tool
Layout rules
The 5-second test
Show the dashboard for 5 seconds, hide it, ask what people remember. If it isn't the main message, the hierarchy needs work.
Z-pattern reading
The essentials top left. We read left to right and top to bottom: the order of the objects is the order of the story.
Mobile first
On a phone: two views and a few figures per screen, stacked. No hover with a finger: labels rather than tooltips, targets of at least 24 px.
Consistency
Same place for filters, refresh date and current selections on every sheet. A shared template, like the 2012 one.
Context
A number alone says nothing. Always a reference: target, previous period, average or trend.
Performance
Still relevant: avoid heavy expressions, use calculation conditions, prepare aggregates upstream. A slow dashboard doesn't get used.
The IBCS framework (SUCCESS)
The International Business Communication Standards (version 1.2) propose a unified notation for management reports: actual solid, budget outlined, forecast hatched, variances in green and red. Useful for standardising financial reporting.
Storytelling & AI
A chart answers a question. The title should give the answer. And when an AI proposes the chart, it is up to you to check that it answered the right question.
The six lessons of Storytelling with Data
Working with an AI
What it does well
- A first draft of a chart from a natural-language question
- Summarising a report page, spotting anomalies
- Writing an expression (DAX, Qlik) or a data-preparation script
- Suggesting several forms for the same message
What you must check
- Invented joins, fabricated columns, silently applied filters, stale data
- The “default” chart: dual axis, 9-slice pie, rainbow palette
- A confident tone that doesn't guarantee the figure is right
Makeover
Each pair rebuilds one screen of the report brought in the morning. You present the before and after in 2 minutes, and the group scores it with the grid below.
Scoring grid: the common mistakes
Stephen Few's list presented in 2012, completed with the typical mistakes of 2026. Tick what the screen avoids.
End-of-day quiz
Ten questions to check what stuck, then a round table: what will you change in your report on Monday?
References
The 2012 sources remain the foundation. Additions in this edition are marked 2026.
- Stephen Few, Information Dashboard Design, O'Reilly 2006 (2nd ed. Analytics Press 2013)
- Stephen Few, Show Me the Numbers, Analytics Press, 2nd ed. 2012
- Stephen Few, Common Mistakes in Data Presentation, Perceptual Edge, 2004
- Stephen Few, Practical Rules for Using Color in Charts, 2008
- Colin Ware, Information Visualization: Perception for Design
- Edward Tufte, The Visual Display of Quantitative Information
- A. Abela, Chart Suggestions, A Thought-Starter, 2006
- Cleveland & McGill, Graphical Perception, JASA, 1984
- Heer & Bostock, Crowdsourcing Graphical Perception, CHI 2010 2026
- Tamara Munzner, Visualization Analysis and Design, CRC Press, 2014 2026
- Cole Nussbaumer Knaflic, Storytelling with Data, Wiley, 2015 2026
- Wexler, Shaffer & Cotgreave, The Big Book of Dashboards, Wiley, 2017 2026
- Financial Times, Visual Vocabulary 2026
- Skau & Kosara, Arcs, Angles, or Areas, EuroVis 2016 2026
- Kosara, study of 3D pie charts 2026
- Hullman, Resnick & Adar, Hypothetical Outcome Plots, PLOS ONE 2015 2026
- Padilla, Kay & Hullman, Uncertainty Visualization, 2020 2026
- IBCS Association, IBCS Standards 1.2 2026
- W3C, WCAG 2.2, success criteria 1.4.1, 1.4.3 and 1.4.11 2026
- Okabe & Ito, Color Universal Design; viridis, cividis 2026
- Publications Office of the EU, Data visualisation guide 2026
- Qlik, Structuring an application using sheets 2026
- Tableau, Mobile dashboard design: less is more on the small screen 2026