Why this session exists
Most executives read dashboards the way most people read contracts: skim, nod, sign. And just like contracts, the expensive surprises live in the parts you skimmed. The good news is that dashboards are far more standardized than contracts - once you know the anatomy, every well-built dashboard reads the same way, and every badly built one exposes itself within five questions. No code tonight, just trained eyes.
The anatomy of a dashboard 8 min live
Stephen Few's core insight: a dashboard is a single screen designed for a glance, and good ones share a grammar. KPI cards across the top answer "how are we doing right now" - a level plus its change. Trends sit in the middle answering "which way are we heading". Breakdowns live below answering "driven by what". And filters hug the edge, quietly reshaping everything else. Learn the grammar once and you can sight-read any dashboard your teams produce.
LiveKPI cards: level, delta, target4 min▶
The big numbers across the top are where every dashboard conversation starts, and a complete KPI card carries three ingredients:
- Level - the number itself. $412K. On its own, almost meaningless.
- Delta - the change against a comparison period. +4% vs last month tells you direction and speed.
- Target - the number leadership committed to. Against target is the only comparison that carries consequences.
A card showing only the level is Few's "inadequate context" mistake wearing a suit: is $412K good? Against what? When you review a dashboard proposal, count the cards that show all three ingredients. Missing deltas and targets are not a style gap - they are the difference between reporting and managing.
Self-studyPreattentive attributes - why your eye jumps4 min read▶
Few's chapter 5 borrows a result from vision science: some visual differences register before conscious attention - in well under a second, effortlessly. The big three for dashboards:
- Position - where something sits on a common scale. The most accurate encoding we have, which is why bar and line charts beat everything else for comparison.
- Length - how long a bar is. Nearly as accurate, and the reason truncated axes (bars starting at 80 instead of 0) are a lie your eye cannot help believing.
- Color - one distinct hue in a field of sameness. Your eye jumps to it whether you want to or not.
This is why one gold bar among blue ones works: the designer spent the single loudest visual signal on the single thing that matters. And it is why a dashboard where everything is colorful directs your eye nowhere - spending the signal everywhere is the same as spending it nowhere. When your eye jumps somewhere on a dashboard, someone chose that. Ask yourself whether they chose well.
The interrogation protocol 9 min live
Knaflic's first rule of data storytelling is that context decides meaning - the same number is great news or a crisis depending on what surrounds it. Dashboards strip context to stay glanceable, which means the reader has to restore it. That is the interrogation protocol: five questions, in order, before you trust any number enough to act on it. It works on dashboards, board packs, vendor pitches, and news charts alike.
LiveSnapshot vs trend - the most common misread4 min▶
The single most common executive misread: judging a snapshot when the decision needs a trajectory. A KPI card can glow green - best month ever - while the trend line right below it shows growth decelerating for five straight months. The snapshot says celebrate; the trajectory says worry. Both are on the same screen, and the eye anchors on the big number.
- Snapshots answer: where are we right now? Good for status, targets, accountability.
- Trends answer: where are we heading? Good for decisions, which are always about the future.
- The reflex: every time a snapshot impresses you, immediately look for its trend. A good month can hide a bad trajectory - and a bad month can hide a recovery already underway.
Self-studyAverages hide, distributions tell3 min read▶
"Average order value is $128" sounds like your typical customer spends $128. But the same average comes from everyone spending around $128 - or from a small crowd of $400 whales and a long tail of $40 one-timers. Those two businesses need opposite strategies, and the average cannot tell them apart.
- Ask for the spread: a median next to the mean is the cheapest fix - a big gap means whales are dragging the average.
- Ask for segments: "average by plan" or "average by cohort" often dissolves the mystery in one chart.
- The tell: any strategy pitched off a single average deserves one question - "what does the distribution look like?" Watch how quickly the room gets quieter.
The filtered dashboard that approved a budget. A retail COO approved a seven-figure expansion for a "star" region off a dashboard showing it far outgrowing all others. Weeks later an analyst noticed the view had been left filtered to online orders only - a filter someone applied in a meeting months earlier that quietly persisted for every viewer after. Include stores, and the star region was mid-pack. Nothing was miscalculated; the data was flawless. Question 4 - "what filters are on?" - would have caught it in five seconds, before the money moved. That is why filters sit on the interrogation ladder at all: they are the only part of a dashboard that changes what every other part says without leaving a visible trace in the numbers themselves.
Interrogate the Daybreak exec view ★ 12 min · everyone interrogates
Here is Daybreak's revenue trend - the chart their leadership sees every Monday. Run the full protocol on it, live. This is the same warehouse your builder colleagues use, so what you find is what they find - you are just finding it without writing a line of SQL.
Read the level first. Switch the chart to KPI for a moment: one big governed number. That is the snapshot. Impressive or not, you cannot judge it yet - snapshot without trajectory is half a story.
Now the trend - spot the dip. Back to the line chart. Growth, growth, and then March 2026 sags. A snapshot in February would have looked wonderful. This is the snapshot-vs-trend trap live on real data.
Climb the ladder. What exactly is revenue here - press Show SQL and see the definition. As of when - this playground computes live on click. Out of what - it is a raw total, no denominator. Filters - none applied. Compared to what - no target line, so bring your own baseline. Ninety seconds, and you now know this chart better than most people know their own dashboards.
The leader move: switch the measure to Orders. Does March look as bad? Look closely - orders dip, but revenue dips harder. Same customers, still buying, spending less per order. That points at basket size, not demand. You just narrowed an investigation without writing a query - the builder track chases this exact thread to its answer in b9.
Your turn: is Pro really better? ★ 10 min · you decide
New chart: average order value by subscription plan. At first glance, one plan looks like the clear winner - the kind of chart that launches a "migrate everyone to Pro" initiative in the wrong meeting. Before you would let that initiative launch, interrogate.
LiveQ1 · Is Pro really "better"?4 min▶
Work the ladder before you answer. The measure is an average - so ask what the distribution hides (a few whale orders can crown a plan). It is also a rate, not a size - a plan can win on average order value and still be tiny. Test that: switch the measure to Revenue, then Customers. Does the "winner" keep winning? A plan that wins AOV but loses on customer count is a premium niche, not a growth engine - and those need very different investments.
LiveQ2 · What would you ask before acting?4 min▶
Suppose someone proposes pushing every Basic customer toward Pro on the strength of this chart. Draft your questions, then compare with the ladder's answers:
- Definition: does AOV here include one-off orders, or subscription renewals too?
- Denominator: average per order or per customer? A Pro member ordering rarely but big differs from one ordering weekly.
- Baseline and causality: do Pro customers spend more because of the plan - or did big spenders simply choose Pro? A dashboard cannot answer that (session 1's rule); an experiment or analysis can. Approving the migration without asking is the filtered-dashboard story waiting to happen again.
Try it yourself - this week ◐ 20-30 min total
- Run the five-question protocol on two real dashboards you receive - one you trust, one you barely look at. Write the answers down; note which questions the dashboards themselves cannot answer.
- Check the filter rails: are any filters applied by default? Ask who set them and whether every viewer knows.
- Find one KPI card in your world showing only a level - no delta, no target - and ask its owner for both. You have just improved a dashboard without touching a tool.
- Bring to a3: one number you now distrust and which of the five questions broke it. Session a3 is about why the same metric returns different numbers from different teams - and how "one number, one truth" actually gets built.
Official sources covered
This session carries the perception and layout science of Few and the context discipline of Knaflic - the reading half. The writing half (commissioning and design) arrives in a4. This page covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · A region looks like a runaway winner on the dashboard, but something feels off. Which interrogation question is most likely to expose the problem?
Filters change what every chart says while leaving no visible trace in the numbers - the budget-approval story in Part 2 turned on exactly this. Always check the rail first.
2 · The monthly KPI card shows the best revenue month in company history. Before celebrating, a trained reader immediately...
Snapshots answer "where are we"; decisions need "where are we heading". Every impressive snapshot earns an instant look at its trend - a good month can hide a bad trajectory.
3 · A KPI card shows "$412K" and nothing else. What is missing before it can support a decision?
Level, delta, target - the three ingredients of a complete KPI card. A lone level is Few's "inadequate context" mistake: is $412K good? Against what? Says who?