learn-business-intelligence-with-phoebe / Builder session 5 of 10
Learn Business Intelligence with Phoebe · Builder track · Session 5 of 10

Chart choice + dashboard grammar

You have a modeled star (b3) and governed measures (b4). Now: which picture? Most bad dashboards are not bad data - they are the right number in the wrong chart, buried under clutter. Tonight you get a decision rule that picks the chart from the question, Stephen Few's compact library of dashboard-native displays, and a declutter pass you can run on any chart in two minutes.

🟡 Builder track Practitioners: analysts · DE · DS · PMs Runs in your browser · no install 45 min live + self-study
0-3 · Recap 3-20 · Chart grammar 20-42 · Build-along: chart doctor 42-45 · Q&A
Part 0

Stage 4: visualize - but grammar first

Stage 4 of the five-stage workflow: Visualize. Every BI tool offers forty chart types; a working dashboard uses about six of them, chosen by rule, not taste. This session is the grammar - which chart answers which question, which displays earn their pixels on a dashboard, and what to delete. b6 then assembles these pieces into the actual Daybreak dashboard.

Live - presented in session Self-study - read after class ▶ Mini-BI - interactive playground Official sources covered
★ What you walk out with today A question-to-chart decision rule you can apply in five seconds, Few's compact dashboard library (KPI card, bullet graph, sparkline), and a declutter checklist that turns any default chart into a professional one. Plus a formally logged mystery: the March dip, spotted on a proper trend line at last.
Part 1 · covers Knaflic ch2 "Choosing an Effective Visual", Few ch7-8

The question picks the chart 7 min live

You never choose a chart - the question already chose it. Trend over time → line. Compare categories → bar, horizontal when labels are long. Part-of-whole → stacked bar, or honestly just a table; pie only when there are 2-3 slices. Distribution → histogram. Relationship between two measures → scatter. Exact lookup → a table beats every chart. Memorize the six mappings and you out-chart most of the industry.

Trend over time? → line chart Compare categories? → bar, horizontal if long Part-of-whole? → stacked bar (pie ≤ 3) Distribution? → histogram Two measures related? → scatter plot Exact values needed? → a table wins Say the question out loud first. If it starts with "how has", it is a line. If "which", it is a bar.
🔍 Click to zoom - six questions, six charts: the whole decision rule on one screen
LiveThe pie intervention4 min

Daybreak sells in eight cities. Picture those eight as pie slices: to rank them you must compare angles, and human eyes are terrible at angles. Is the 11% slice bigger than the 13% one? You genuinely cannot tell without the labels doing all the work - at which point the pie is decoration around a list of numbers.

Now the same data as a sorted bar chart. Ranking is instant, differences are lengths (which eyes judge brilliantly), and the top city announces itself in gold:

Real world

The polite rule of thumb. When a stakeholder asks for a pie, do not lecture - ask how many slices. Two or three big ones ("Pro vs Basic") and a pie is fine, even friendly. More than that, show them the sorted bar next to it and let their own eyes make the argument. The bar wins the meeting every time.

Self-studyWhen a table beats every chart3 min read

Charts show shape: trends, gaps, outliers. But some questions have no shape - they are lookups. "What was Berlin's exact revenue in April?" A chart makes the reader squint at an axis; a table hands over the number.

  • Choose a table when: the reader needs precise values, will look up individual rows, or must compare across multiple units (revenue AND orders AND AOV per city - three scales, one grid).
  • Make it a good table: right-align numbers, consistent decimals, sort by the column that matters, and no zebra-striping heavier than the data.
  • Hybrid trick: a table with one embedded sparkline column (Part 2) gives lookup precision AND trend shape - a Few favorite for dashboards.
Part 2 · covers Few ch8 "An Ideal Library of Graphs", ch9 bullet graphs, ch10 sparklines

Few's dashboard library 6 min live

A dashboard is not a slideshow - it is a dense, single-screen display where every square centimeter is contested. Stephen Few's "Information Dashboard Design" gives the dashboard-native displays that earn their space: the KPI card (a level, its delta, its target), the bullet graph (his invention - a measure against target and qualitative bands, replacing the space-hogging gauge), and the sparkline (Tufte's word-sized trend, perfect inside a table cell).

Bullet graph Revenue vs target Bands = poor / ok / good ranges Dark bar = the measure Gold tick = the target All of a gauge, in one flat row Sparkline row Revenue 48.2k Orders 1,204 Word-sized trend in a table cell: shape only - no axes, no labels, current value printed beside it KPI card MONTHLY REVENUE $412k ▲ 6% vs last month Level + delta + target context. Never a lonely number - a number with its comparison attached These three carry most executive dashboards. b6 builds Daybreak's top row out of exactly these.
🔍 Click to zoom - the three dashboard-native displays: bullet graph, sparkline, KPI card
LiveBullet graphs - the gauge that respects space3 min

Dashboards inherited round gauges from car dashboards, and Few spent a chapter politely demolishing them: a gauge burns a big square of screen to show one number, and its needle angle is as hard to read as a pie slice. His replacement, the bullet graph, shows strictly more in a fraction of the space:

  • The bar is the measure - length, the encoding eyes read best.
  • The gold tick is the target - you see at a glance whether the bar crossed it.
  • The background bands are qualitative ranges (poor / ok / good) - context without a legend.
  • It stacks: five bullet graphs fit in one gauge's footprint, so a whole KPI family scans in a single eye sweep.
In Power BI / In Tableau Tableau ships a native bullet graph (Show Me, or a bar with a reference line and distribution bands). Power BI gets there with a bar plus target line, or a bullet visual from AppSource. Either way the recipe is the same: measure bar, target line, quiet bands.
Self-studySparklines: trend, not values3 min read

A sparkline is a line chart shrunk until only the shape survives - no axes, no gridlines, no labels, roughly the size of a word. That is not a compromise; it is the point. In a table of ten metrics, a sparkline column answers "which of these is trending badly?" in one glance, and the printed value beside it handles precision.

  • Use for: direction and volatility at a glance, especially one-per-row in tables.
  • Do not use for: reading values, comparing magnitudes across rows (each sparkline scales to its own range), or as a main chart.
  • Mark sparingly: one dot for the latest point or the minimum is plenty - our playground's gold minimum marker follows the same instinct.
Part 3 · covers Knaflic ch3 "Clutter Is Your Enemy", IBCS Simplify

Declutter: every pixel fights for attention 5 min live

Knaflic's rule: anything on the chart that does not carry information is clutter, and clutter taxes the reader. Default tool output arrives pre-cluttered - heavy gridlines, borders, legends, a rainbow of series colors. The professional move is subtraction: delete until only data and the minimum scaffolding remain, then spend one accent color on the single point you want remembered, gray for everything else. Our playground's gold top bar and gold minimum marker are exactly that convention, automated.

LiveThe declutter pass3 min

Run this checklist top to bottom on any chart. Two minutes, every time, before anything ships:

ElementBefore (tool default)After (decluttered)
GridlinesDark, both axes, every tickFaint horizontal only - or none
Chart border + backgroundBoxed and shadedGone; whitespace separates
3D effectsBeveled 3D barsNever. No exceptions, ever
LegendFloating box, eyes ping-pongDirect labels on the data
Series colorsEvery bar its own colorOne accent for the point, gray context
Axis labelsRotated 45°, unreadableHorizontal bar chart instead
Data labelsOn every pointOnly where the reader needs the value
Title"Chart 1" or "Revenue"The takeaway: "Revenue dipped in March"
Real world

The squint test. Lean back and squint at your chart until the labels blur. Whatever still stands out is what the reader's eye will visit first. If it is a legend box or a bright gridline rather than your data point, the pixels are fighting the message - delete and re-squint.

Self-studyIBCS in one paragraph2 min read

IBCS (the International Business Communication Standards) turns these instincts into a shared notation for business reporting, organized as the SUCCESS families. Two of them are tonight's session in standards form: Express - choose the proper visualization for the message - and Simplify - avoid clutter, remove everything not needed to convey the message. The deeper IBCS idea is consistent notation: the same thing always looks the same, so a reader who learned one report can read them all - actuals, plan, and forecast each get one fixed visual treatment across every page. If your company reports monthly to a board, IBCS is worth an afternoon of reading.

Demo 1 of 2

Five questions, five right charts ★ 12 min · everyone builds

One playground, five reconfigurations. For each question below, set the dimension, measure, and chart type - then say out loud why that chart wins before moving on. The saying-out-loud is the training.

"How is revenue trending?" Month · Revenue · Line. Time on the x-axis, and there it is: the playground marks the minimum gold, and the minimum is March. Log it formally - "2026-03 revenue dip, cause unknown." b9 is the investigation; tonight we only chart it honestly.

"Which cities lead?" City · Revenue · Bar. A ranking question, so lengths, sorted. A line here would invent a trend between cities that does not exist.

"What is the category mix?" Category · Revenue · Bar. Part-of-whole with three categories - a pie would survive, but the bar still ranks faster and handles the small slice better.

"How big are we overall?" KPI · Revenue. No dimension at all - one governed number. Resist the urge to decorate it.

"Do Pro customers order bigger baskets?" Plan · AOV · Bar. Note the measure choice from b4: average order value, not total revenue - a comparison of behavior, not of size.

Real world

Chart choice is an interview question. "Select an appropriate visualization" is literally a tested skill on the PL-300, and some version of "here is a question, which chart?" appears in most BI interviews. The decision flow from Part 1 is the answer key - question type first, chart second, taste never.

Demo 2 of 2

Your turn: chart-doctor ★ 10 min · build your own

Three patients. Two need a chart chosen; one arrives with the wrong chart already fitted, and your job is to name the disease before you cure it.

LiveQ1 · Revenue by channel - would a pie work here?3 min

Daybreak sells through a handful of channels. Build the bar, count the categories, and make the call: is this one of the rare legitimate pie cases, or does the bar still win? Defend your answer with the slice-count rule.

LiveQ2 · Buying customers per month3 min

Trend question → line. And remember the b4 warning: this measure is a COUNT(DISTINCT ...), semi-additive - you cannot sum these months to get "customers this half-year". Does the customer trend echo the March revenue dip, or does it tell a different story? Note what you see; it is evidence for b9.

Self-studyQ3 · The wrong chart, on purpose3 min

Below is a deliberate malpractice case: cities on a line chart. Before touching anything, describe the disease in one sentence. (Diagnosis: a line implies order and continuity between points - but Amsterdam to Berlin is not a journey, cities have no time order, so the line invents a trend that does not exist.) Then fix the patient: switch the chart type to bar and watch the fake trend become an honest ranking.

Homework

Try it yourself - this week ◐ 20-30 min total

Source material

Official sources covered

Tonight leaned on the two canonical visualization books plus the reporting standard, all tool-agnostic. This page covers:

Knaflic · Storytelling with Data, ch2-3Part 1 (choosing an effective visual) + Part 3 (clutter is your enemy); ch7 narrative returns in b7
Few · Information Dashboard Design 2e, ch7-10Part 1 (advantages of graphics, ideal library) + Part 2 (bullet graphs, sparklines)
PL-300 · Select an appropriate visualDemo 1 · the decision flow is the exam answer key, tool-agnostic
IBCS · SUCCESS families: Express + SimplifyPart 3 · the one-paragraph version; full notation system stays with IBCS
Check yourself

Three questions before you go 🎯 ◐ 90 seconds

1 · "How has monthly revenue developed since January?" The chart is...

"How has ... developed" is a trend question, and trend over time → line. Bars per month work too, but the line shows the shape of change - which is what was asked.

2 · Why does a pie chart fail with 8 city slices?

The failure is perceptual: angle comparison is hard, length comparison is easy. That is why the sorted bar chart wins the makeover - same data, better encoding.

3 · The professional color rule for a chart is...

Color is attention. Spend it on the one thing you want remembered and mute everything else - the playground's gold top bar and gold minimum marker are this rule, automated.

Builder session 5 cheat sheet · pin this

The ruleThe question picks the chart: trend → line, compare → bar, distribution → histogram, relationship → scatter, lookup → table.
Bar chartsRanking and comparison. Sort them, go horizontal when labels are long, never 3D.
Pie rule2-3 slices maximum. Beyond that, angles become unreadable - use a sorted bar or a table.
KPI cardLevel + delta + target. Never a lonely number - always attach its comparison.
Bullet graphFew's gauge replacement: measure bar, gold target tick, quiet bands. Five fit where one gauge sat.
SparklineWord-sized trend for table cells. Shape only - no axes, value printed beside it.
DeclutterDelete gridlines, borders, 3D, legends-when-labels-work. One accent color, gray context. Squint test.
Running projectThe March dip is now formally logged from the trend line. b6 builds the dashboard; b9 solves the mystery.