§3.1
Exploratory Visualization and Dashboards
Part II — visual evidence makes patterns visible before any method is named.
The CEO opens the new Bean & Basket executive dashboard. Twelve tiles, six charts, three filter dropdowns, two KPI rings, and a heatmap. Each one looks polished. After ninety seconds of scanning, he closes the tab. He could not have answered, from the dashboard, why revenue was flat last quarter — not because the answer wasn't in there, but because nothing on the page told him where to look first. The dashboard had charts. It did not have a question. A good dashboard is the opposite: every panel answers one specific question, and the panels are ordered so that the answer to the first leads naturally to the second, which leads to the third. The CEO's eye should never have to wander. The page should walk him through the analysis.
The executive question: what should a manager notice first?
A dashboard is a sequence of business questions. Not a collection of charts, not a status report, not a data buffet — a sequence. The first panel asks an executive question, the second answers it, the third asks the natural follow-up, and so on until the page ends at a recommended action. When a manager looks at a well-built dashboard, she does not scan; she reads, in the order the page is laid out, and the order makes sense.
The instinct to "show everything" is the most common mistake in dashboard design. Twelve tiles, each showing a different KPI, each presented as if equally important, force the viewer to do the prioritization work that the dashboard was supposed to do for her. The defense is to design the page like an analyst writes a memo: start with the question, state the headline number, show the trend, break it down where the trend gets interesting, and end with what to do about it. The same five-step arc, every page.
Before getting to the arc, the smaller question: which chart for which question? Figure 1 is a working table. It is not exhaustive — every analytics team has its own variants — but it is the minimum vocabulary a manager should expect a dashboard designer to know. The right chart is determined by the question, not by what the data team happens to have a template for.
| Business question | Chart | Use when |
|---|---|---|
| How is one numeric variable distributed? | Histogram | Spread, skew, multiple modes |
| Which category is largest? | Bar chart (sorted) | Comparing values across a small set of categories |
| How does a metric change over time? | Line chart | Continuous time series; >5 points |
| How do two numeric variables relate? | Scatterplot | Looking for correlation, clusters, outliers |
| Where is performance highest? | Choropleth / map | Geographic differences are part of the story |
| Where do users drop off? | Funnel | Conversion across a fixed sequence of stages |
| What changed between two periods? | Slope chart / variance bar | Two snapshots, want to see direction and magnitude of change |
| Which products drive most revenue? | Pareto (sorted bar + cumulative line) | Long-tail distributions; '80/20' |
| Two breakdowns of the same metric? | Stacked or 100% bar | Composition matters more than absolute values |
| Movement over time, broken by group? | Multi-line or small multiples | 3+ series; small multiples if the comparison is the story |
Figure 1 names ten question-chart pairs. The pattern across all ten is that the question shape determines the chart shape: continuous time becomes a line; a fixed sequence becomes a funnel; categorical comparison becomes a sorted bar. There is no chart in Figure 1 whose primary job is decoration. Every entry exists because some specific kind of business question is hard to answer without that visual.
A dashboard is one or more of these charts arranged in an order. The order is the part most teams get wrong. Below is the order that has worked for fifty years and will work for fifty more: the executive question on top, the KPI tile that lands the headline, the trend that puts the headline in context, the breakdowns that explain the trend, and the recommended action at the bottom.
The dashboard arc, in six steps
Every dashboard starts with a question, written in plain English at the top of the page. Not "Sales overview." Not "Q1 performance." Something specific: "Why did revenue flatten in Q1, after eight quarters of growth?" The question names what the rest of the page is trying to answer, and gives the reader a frame for the numbers that follow.
One number, large, with a comparison. Last quarter's revenue against the quarter before, or against the same quarter last year. The KPI tile is the dashboard's headline. Used well, it gives a viewer the answer to the executive question in about two seconds.
The headline number
One number, large, with a comparison — the answer in two seconds.
Bean & Basket chain-wide revenue, Q1 2024. The badge is the comparison that turns a label into a KPI.
A line chart of the same metric over the last 8–12 periods. Without the trend, "+0.7%" reads as steady; with the trend, it reads as the first non-growth quarter after a year of 4–6% gains. The chart in this position is almost always a line chart, almost always with the most recent point flagged, and almost always bare — no clutter, no second metric, no overlaid bar chart. The trend's whole job is to give context for the KPI above.
Is +0.7% steady, or a break in the pattern?
Eight quarters of 4–6% growth, then the line goes flat. The last point is flagged.
Once the trend establishes that something interesting happened, the next panel decomposes it. Revenue by region. Revenue by product category. Revenue by store cohort. The right breakdown is the one whose variance explains the trend — usually two or three breakdowns are tried, and the one where the differences are largest stays on the dashboard. The chart is almost always a sorted bar, sometimes a small-multiples grid if there are too many groups.
Where is the flat total coming from?
Revenue by region. Suburban is the lone grower; every other region shrank — the trend is a mix shift.
The breakdown identifies the where; the drilldown finds the what. If the regional breakdown surfaces Suburban as the breakout (Chapter 2's lesson made concrete), the drilldown digs into Suburban: customer segments, hour-of-day patterns, product mix. This is where exploratory visualization lives — scatterplots, small multiples, anything that admits new information into the analysis. Most viewers do not need this panel every day. Analysts do.
What inside Suburban is growing?
Small multiples by region × daypart. Suburban's weekday-morning commuter block towers over everything — Campus has comparable traffic but a weak morning.
The last panel is the only one that is not a chart. It is a paragraph — sometimes two — that translates the four chart panels into a recommendation. "Suburban grew on weekday-morning commuter traffic. Recommend doubling weekday-morning staffing and piloting a 7am promo at the Campus store, where commuter foot traffic is comparable but conversion lags by 11 points." This is the panel most dashboards omit. It is also the only panel that turns the dashboard from a status report into a decision document.
So what do we do about it?
The only panel that is not a chart — it turns the four panels above into a decision.
Here is the whole arc on one page — the executive question on top, then the five panels reading straight down to the recommended action. Hover any chart for the underlying numbers. Notice that each panel answers the question the panel above it raises.
Step 1 — the executive question
Why did revenue flatten in Q1, after eight quarters of growth?
One sentence at the top. Everything below is the page answering it, in order.
The headline number
One number, large, with a comparison — the answer in two seconds.
Bean & Basket chain-wide revenue, Q1 2024. The badge is the comparison that turns a label into a KPI.
Is +0.7% steady, or a break in the pattern?
Eight quarters of 4–6% growth, then the line goes flat. The last point is flagged.
Where is the flat total coming from?
Revenue by region. Suburban is the lone grower; every other region shrank — the trend is a mix shift.
What inside Suburban is growing?
Small multiples by region × daypart. Suburban's weekday-morning commuter block towers over everything — Campus has comparable traffic but a weak morning.
So what do we do about it?
The only panel that is not a chart — it turns the four panels above into a decision.
The six steps above are not specific to a particular tool. They work in Looker, in Tableau, in a printed PDF, in a one-pager taped to a wall. What makes the dashboard work is not the platform; it is the sequence. Without a sequence, the same six panels are a buffet; with a sequence, they are a memo.
The difference is entirely in the ordering. Toggle below between the buffet (the same four panels as equal tiles, no obvious place to start) and the memo (the arc). The data never changes — only whether the page does the prioritizing for you.
Buffet: same panels, no order
Four equal tiles. Each is fine alone; together they make you do the prioritizing.
Breakdown · revenue by region
KPI · chain revenue
Drilldown · Suburban by daypart
Trend · revenue by quarter
Nothing about the data changed between the two views — only the order. The buffet asks you to find the story; the memo tells it to you, top to bottom.
The deeper pathology behind all three is the same: building dashboards from the data the team has instead of the questions the manager has. The fix is to start every dashboard project with the question, written down, before any chart template is opened. A team that does this has dashboards that survive their first review. A team that doesn't builds beautiful tools that get closed after ninety seconds.
Concept check
These three questions span choosing a chart from the business question, looking at a full distribution before trusting one summary number, and the difference between exploring data and presenting an executive answer.
- 1.A regional VP asks: "which of our twelve metro markets grew fastest in home values since 2020?" Your housing dataset has monthly price levels per market, but the markets start at very different price points, so a raw multi-line chart is dominated by the expensive coastal markets sitting at the top. Which chart best answers the VP's actual question?
- 2.Across all Progresso store-months, average monthly unit volume is about 900 units, and the finance team proposes staffing and inventory plans assuming a typical store-month sells around 900. Before signing off, you plot the full distribution of store-month volumes. Which finding would most change how you act on that 900 figure?
- 3.You are deep in the drilldown panel of a dashboard, where the article says exploratory visualization lives. You build a scatterplot of store foot traffic against conversion and notice three odd stores far from the cloud of points. A colleague says, "great, drop those three outliers straight onto the executive KPI row so leadership sees them." What is the soundest response?