Dashboard Design for Decisions

BUS220 - Business Intelligence and Analytics | Week 6

Oleh Omelchenko

2026-04-20

Dashboards are built around workflows

“You’ve spent two weeks building calculations that answer complex analytical questions — LOD expressions, table calculations, running totals. You can compute the right numbers. The question now is: how do you arrange them into something a busy person can actually use? That’s a different design problem than making a single chart look good.”

Monday morning, 15 minutes before standup

You’re a regional manager. You need to know:

1. Are we hitting target this month?
2. Which stores are underperforming?
3. Is the new product launch getting traction?
4. Did returns spike last week?

No single chart answers all four. You need a coordinated set of views that lets you assess the situation and decide what to do.

“Four questions, each needing different data, different visualizations, different levels of detail. This is what dashboards exist for — not showing data, but supporting a decision workflow. The Monday morning scenario should feel familiar from your metrics discussions in week 4.”

What makes dashboards different from charts

Single chart	Dashboard
One focal point	4-8 views competing for attention
Answers one question	Supports a sequence of questions
Static once built	Interactive elements create new failure modes

The design challenge shifts from “make this chart clear” to “make this collection of charts tell a coherent story, with the right view getting attention first.”

“You covered chart design in your dataviz course — encoding, chart selection, color, labels. Dashboards introduce problems that don’t exist for individual charts: multiple views competing for attention, users who follow a workflow rather than reading a single chart, and interactivity that can confuse as easily as it clarifies.”

Explanatory vs. exploratory

“CwD Chapter 7, which you’ll read before practice, draws a distinction that shapes every dashboard design decision. Let me introduce the core idea now so the reading has a frame.”

Two communication modes

Explanatory

You’ve found something. You guide the audience to a conclusion.

• You control the order and emphasis
• The narrative is curated
• The main question: “So what?” — what action should the audience take?

Quarterly reviews, board reports, stakeholder updates.

Exploratory

You give users tools to answer their own questions.

• The user drives, not the designer
• Flexible: filters, parameters, drill-downs
• The main challenge: signposts — how does the user know what they can do?

Marketing analysts investigating campaigns, finance teams exploring cost variance.

“Neither mode is better — they serve different situations. Explanatory work requires clarity: you’ve done the analysis, now communicate the finding. Exploratory work requires flexibility: the user has questions you can’t anticipate, so you give them tools to investigate.”

“The mistake students make most often: building in exploratory mode (lots of filters, multiple views, flexibility) but without signposts — so the user doesn’t know what’s interactive. Or building in explanatory mode but without a clear ‘so what’ — the data is there, but the insight isn’t stated.”

The “So what?” test

Questions to ask yourself when building explanatory work:

• Does my message have more than one part? If so, which charts tell each part?
• What do I want the audience to take in first? What order should I tell the story?
• What is the most important takeaway, and how can I make sure they remember it?
• Why is my message important? So what?

If you can’t articulate what action your audience should take, the dashboard is incomplete.

“The ‘So what?’ question is the most important and the most often skipped. You’ve built a beautiful dashboard showing revenue by region over time. Great. So what? Should the audience increase investment in the top region? Investigate the declining one? Change strategy? If the dashboard doesn’t point toward an action, it’s data display, not communication.”

Signposts for exploratory work

Notice the instruction panel on the left. When a dashboard is interactive, users need cues about what’s clickable — signposts. Written instructions are clear, but take space.

“This is a Prep Air dashboard with written instructions in the left panel: ‘Use Top N control to pick the number of routes to analyse and select any route to filter the contextual numbers above.’ Clear for first-time users, but the text block takes up real estate that could go to data.”

Signposts: symbols instead of text

Interactive key with symbols — compact, scales across dashboards:

Notice the small key at the bottom left: icons for click-to-filter and hover-to-reveal. Less text, same information — but users need to learn the system.

If you produce work regularly for the same audience, use the same symbols consistently so interactions become intuitive over time.

“Same dashboard, different signpost approach. The interactive key at the bottom left uses two small icons to indicate what’s clickable and what responds to hover. It takes a fraction of the space that written instructions did.”

“The trade-off: symbols are compact but require learning. Written instructions are clear but add cognitive load and take up space. The choice depends on your audience — first-time users need explicit guidance, frequent users need less. If you produce dashboards regularly for the same team, establish a consistent icon vocabulary.”

What a dashboard actually does

Think about a car dashboard: it monitors conditions and facilitates understanding. Business dashboards do the same two things.

Monitoring — is everything OK right now?

The most frequently checked information should be largest and most prominent, like the speedometer on a car.

Facilitating understanding — why is something happening?

Charts that provide context, breakdown, comparison. This is where interactivity earns its place.

A KPI row with contextual numbers and trend indicators:

The user checks this in seconds. Trend arrows answer “up or down?” before you read the numbers.

“Most organizational dashboards serve both purposes. The KPI row at the top monitors: are we on track? The charts below facilitate understanding: why are we off track, and what changed?”

“This is the Prep Air KPI row from CwD Ch 7: Total Ticket Revenue, Total Tickets Sold, Average Ticket Revenue. Each has a trend indicator — the triangle arrows showing whether the metric is up or down versus target. The user’s eye goes straight to the indicators. This is the monitoring layer.”

“The car dashboard analogy makes the design priority concrete: the speedometer is large and central because the driver checks it constantly. The fuel gauge is smaller and peripheral because it’s checked less often. Same principle for your dashboard layout.”

Layout as information architecture

“Now that you know what kind of dashboard you’re building, the next question is: where does everything go? Layout isn’t decoration — it’s the information architecture that determines how your audience processes the content.”

The newspaper front page

A well-designed dashboard works like a newspaper:

• Headline — the most important thing, big and prominent → KPI summary row
• Key stories — enough detail to understand the situation → main charts
• Supporting context — for readers who want to go deeper → detail tables, secondary views

The information design principle behind this: overview first, zoom and filter, then details on demand (Ben Shneiderman).

“Shneiderman coined this mantra in the 1990s for information visualization, and it’s still the most useful rule for dashboard layout. You want the big picture first — am I in trouble or not? Then you want to know where the trouble is. Then you want specific data points to act on.”

“The executive dashboard from BBoD Ch 5 follows this pattern: KPI cards at top showing total sales, profit, quantity. Trend charts and regional breakdowns in the middle. Individual transaction details available on drill-through.”

Layout skeleton

KPI 1

KPI 2

KPI 3

KPI 4

Main trend chart
“Where and why?”

Breakdown chart
“Which segment?”

Detail table / secondary views
“Show me specifics”

“This is a skeleton, not a template. The proportions and number of charts depend on your content. But the vertical flow — headline KPIs, then main analysis, then supporting detail — matches how people process complex information. You want the big picture first.”

A real example: Prep Air ticket analysis

Trace the layout: KPI row at the top (headline), daily revenue sidebar on the left, weekday bar charts and target heatmap in the center (key stories), revenue by destination on the right (breakdown). Overview → detail, top to bottom.

“This is a dashboard from CwD Chapter 7 for the fictional airline Prep Air. Notice the structure: contextual KPI numbers at the very top — total ticket revenue, tickets sold, average revenue, each with trend arrows. Below that, the main analysis is split into three sections: weekday revenue vs. target as bar charts, a monthly heatmap showing above/below target by day, and ticket revenue by destination as line charts.”

“This follows the overview → detail pattern. The user checks the KPIs first (are we on track?), then looks at the weekday pattern (when does it happen?), then checks the destination breakdown (where’s the issue?).”

“We’ll come back to this dashboard in the interactivity section — it has filter actions that let the user click a destination to drill down.”

The Z-pattern

Without explicit visual cues, eyes move in a Z-pattern: top-left → top-right → diagonally down → bottom-left → bottom-right.

KPIs at the top where the eye lands first. Filters in the left sidebar, found before the data. Main charts in the center and right, following the natural reading flow.

“This isn’t a rigid rule, but it’s a useful default. If you’re not sure where to put something, ask: how often will the user look at this? Frequently checked = top-left, prominent. Rarely checked = bottom-right, smaller. The Z-pattern also explains why filters at the bottom of a dashboard are a common design mistake — the user never scrolls down to find them.”

Dashboard grouping with spacing

Charts about revenue — grouped together by proximity:

Revenue trend

Revenue by region

Charts about customers — separated by white space:

Customer segments

Retention rate

No borders needed. Spacing alone creates the grouping.

The white space between the revenue and customer sections tells the user: “These are different topics.”

The tight spacing within each group tells the user: “These are related views.”

“You know this principle from your dataviz course — proximity groups related elements. But in dashboards you’re grouping entire charts, not just data points. A revenue trend next to a revenue breakdown says ‘these are both about revenue.’ A customer chart separated by white space signals ‘this is a different topic.’ You don’t need boxes or borders — spacing does the work.”

Common layout mistakes

Every available view crammed onto one page. Tiny text, no hierarchy, no entry point.

Mistake	Fix
The data dump	If you need 12 views, use 2-3 pages
Filters at the bottom	Filters go at the top or in a sidebar
No grid	Use a consistent column grid
No visual hierarchy	The most important view should be the largest

“The data dump is the most common mistake. Students build a workbook with 10 worksheets and put all of them on one dashboard. The result is a wall of small, unreadable charts that nobody can make sense of. If you need that many views, split them across pages — that’s what navigation is for.”

“The ‘no visual hierarchy’ mistake is subtler. If every chart is the same size, you’re telling the user ‘everything here is equally important.’ That’s almost never true. Size reflects priority — make the most important view the largest.”

Exercise: sketch a dashboard layout

A parcel delivery company’s regional manager checks a dashboard every morning. They need to answer:

1. How many parcels were delivered yesterday vs. the daily target?
2. Which distribution hubs are falling behind on delivery times?
3. What’s the trend in customer complaints this week?
4. Are any vehicle routes consistently over capacity?

Sketch a rough layout — just boxes and labels.

“Students don’t need data for this — the exercise is about spatial thinking. They should decide where the overview goes (KPI row: parcel count vs. target, complaint trend), where the detail goes (hub-level breakdown, route capacity), and what gets the most visual weight. When volunteers share, ask: ‘Where would the user look first? Is that where you want them to look first?’ If the answer is no, the layout needs adjustment.”

When interactivity helps vs. when it hurts

“You’ve read about parameters and sets in PT Ch 14-15. You know the mechanics. Now the harder question: when should you use them, and when should you not?”

The interactivity spectrum

	Static report	Light interactivity	Heavy interactivity
Controls	None	1-2 (metric switcher, date filter)	Multiple filters, drill-downs
Audience	Specific finding for a specific audience	Same analysis from different angles	Analytical users investigating freely
Cognitive load	Low	Low	High

More interactivity means more flexibility but also more cognitive load. The mistake is defaulting to heavy interactivity because the tools make it easy.

Tableau lets you add 20 filters in 5 minutes. That doesn’t mean you should.

“A static report is appropriate when you have a specific finding for a specific audience. A quarterly board deck doesn’t need 15 filters — the story is curated.”

“Light interactivity is the sweet spot for most dashboards: a metric switcher, a date filter, maybe a category selector. One or two controls with clear purpose.”

“Heavy interactivity is appropriate when the audience is analytical and needs to investigate freely. But it demands the most design discipline — every additional control increases cognitive load and creates potential confusion.”

Good interactivity reduces cognitive load

1. The metric switcher (parameter)

Instead of three separate charts for Revenue, Profit, and Orders — one chart with a parameter control. The user sees one clean view at a time. A parameter used well is a decluttering tool.

2. Cross-filtering (dashboard action)

Click a region on the map, and the bar chart updates to show only that region. Two views working together — overview provides context, detail provides specifics.

3. Highlight action

Hover over a category in one chart, and the same category lights up across all views. The user traces a dimension across perspectives without losing the overall context.

“The metric switcher is the most important reframe for students. They often think of parameters as adding complexity. But a well-designed parameter that replaces three charts with one actually reduces the number of things competing for attention. It simplifies.”

“Cross-filtering is probably the most useful interactive pattern in Tableau. It connects views — click on the overview to update the detail. This directly supports the overview-then-detail workflow.”

Good interactivity in action

The Prep Air dashboard you saw earlier — now the user clicks on Paris:

All destinations

Filtered to Paris

Same dashboard, one click. The KPIs update, the charts narrow to Paris data. The layout stays stable — nothing jumps around.

“Same dashboard, one click. The KPIs update to Paris-specific numbers: revenue drops from £57.9M to £7.7M because we’re now looking at one route. The line chart highlights Paris. The weekday heatmap shows Paris-specific performance.”

“The chain of questions this enables: What’s overall revenue? — on track. What about Paris, our new route? — click to filter. Are there weak days? — Mondays below target. Is this new or ongoing? — check the monthly heatmap. Four clicks from high-level check to a specific operational finding.”

“Notice what’s NOT interactive: the overall layout stays stable. The KPI row is still at the top, the charts are still in the same positions. Good interactivity feels natural — you click something and the dashboard responds in the way you expected. The interactivity saved Claire three rounds of ‘can you build me another chart?’”

The question chain

One dashboard, four questions answered through interactivity:

1. What’s overall revenue? → check the KPI row → on track
2. What about Paris, our new route? → click to filter → below target
3. Are there weak days of the week? → read the heatmap → Mondays
4. Is this a new pattern or ongoing? → compare months → started in March

Each interaction answers a natural follow-up question. The user follows a thread, not clicking randomly.

“This is the test for good interactivity: does each click answer a question the user actually has? If the user is clicking to explore rather than to answer, the interaction is doing its job. If they’re clicking because they’re confused about what the controls do, you have a signpost problem.”

Bad interactivity: the filter wall

Long checkbox lists for currency and country. The sidebar is bigger than the chart.

Dropdowns for rarely used filters, checkboxes only for the ones that matter. Same data, less friction.

“The filter wall is the most common anti-pattern in student dashboards. It happens because the student wants to show they can build filters — so they add one for every dimension. But a dashboard with long checkbox lists is intimidating. The fix: pick the 2-3 most important filters, collapse the rest to dropdowns, and set good defaults so the user sees a useful view before they touch anything.”

Bad interactivity: the invisible interaction

A filter action fires when you click a bar, but nothing indicates the bar is clickable. The user either never discovers the interactivity or is surprised when the dashboard suddenly changes.

This is the signposts problem: if an interaction isn’t obvious, it doesn’t exist for most users.

“You know where the color legend is because you put it on the dashboard. You know what to click because you coded those interactions.” — But the user doesn’t know any of that.

“The invisible interaction is subtler than the filter wall. BBoD Chapter 32 makes the point: the designer knows everything is interactive because they built it. The user doesn’t. Sit behind someone and watch them use your dashboard — you’ll quickly see what’s actually discoverable and what’s invisible.”

The zombie dashboard

This Fitbit dashboard was motivating at first — it answered questions, changed behavior, helped form new habits.

After a year, the questions had evolved: What were my best days ever? How does activity vary by weather or work schedule?

But the dashboard hadn’t changed. Same step count, same layout, day after day.

“My goals and related questions were changing, but Fitbit’s dashboard didn’t.”

Dashboards that don’t evolve become zombies — updating daily, answering questions nobody asks anymore.

“This happens in every organization. A dashboard gets built, people use it for a while, their questions change, but the dashboard keeps refreshing. It shuffles along like the living dead — technically alive, practically useless.”

“BBoD Ch 32 recommends four steps to prevent zombie dashboards: review your KPIs regularly — if the reliability metric never misses target, why monitor it? Track usage with data — which dashboards are popular, which were popular but aren’t anymore? Talk to actual users — sit behind them and watch them use it. And consider building starter dashboards that users can customize, rather than shipping a finished product they can’t change.”

The three-question test

Every interactive element you add should pass:

1. Is it discoverable?

Will the user notice this control without being told? Use signposts: labels, icons, or a small interactive key.

2. Does it answer a “So what?”

What business question does this filter serve? If you can’t name the question, the filter probably shouldn’t be there.

3. Will it still matter in 3 months?

Build for questions users will keep asking. One-off investigations belong in a separate analysis, not on the permanent dashboard.

“This is the portable checklist from today’s lecture. Apply it to every parameter, filter, and action you build in practice this week.”

“Question 2 is the one students most often skip. It’s easy to add a Ship Mode filter because Tableau makes it easy. But if nobody’s business decision depends on ship mode, it’s occupying space and attention that could go to something useful.”

“Question 3 comes from the BBoD Ch 32 insight. Dashboards are living products, not one-time artifacts. If you’re building something for a one-time investigation, that’s an analysis — don’t put it on the permanent dashboard.”

Exercise: audit a familiar dashboard

Think about the dashboard you built in week 4-5 practice, or the executive dashboard you read about in week 4.

• What interactive elements does it have?
• Do they pass the three-question test?
• Which elements would survive the 3-month test?

“The point of this exercise is to apply the framework to something concrete. Students should be critical of their own work — that’s where the learning happens. When pairs share, ask: ‘Was there a filter or control that you added just because you could? What would happen if you removed it?’”

Visual design at the dashboard level

“You know about preattentive attributes and decluttering from your dataviz course. SWD Chapters 3 and 4 deepen that foundation. I won’t re-teach those concepts — instead, let’s focus on how they play out differently when you’re managing multiple coordinated views.”

Color consistency across views

Inconsistent

Technology is blue in one chart, green in the other. Two palettes = two things to memorize.

Consistent

Blue = Technology, red = Furniture. Everywhere.

Practical tip: assign colors at the data source level in Tableau (Edit Colors → Assign Palette). Don’t rely on automatic assignment — it shifts when you add or remove dimensions.

“Short-term memory holds about 4 chunks. If your user is spending one of those chunks remembering that blue means Technology in the bar chart but orange means Technology in the line chart, that’s cognitive budget wasted on decoding instead of understanding.”

“The Tableau tip is immediately practical: if you right-click a color legend and assign specific colors to specific values, those assignments stick across all worksheets in the workbook. If you rely on automatic assignment, the colors can shift when you change dimensions — which breaks consistency.”

Visual hierarchy across views

Equal sizing = no hierarchy

Chart A

Chart B

Chart C

Chart D

“Everything is equally important” — which means nothing stands out.

Sized by importance

Main analysis

Support A

Support B

The eye goes to the largest view first. Size reflects priority.

“If every chart is the same size, you’re telling the user everything is equally important. That’s almost never true. The main analysis chart should get more space than supporting views. The KPI row should be visually distinct from the detail section. Size is a preattentive attribute — your brain processes it before conscious thought. Use it at the dashboard level to create reading order.”

Conditional formatting in KPI rows

Without conditional formatting

$1.2M
Revenue

8.2%
Margin

342
Returns

Which one needs attention? You have to read all three.

With conditional formatting

$1.2M
▲ 3.5%

8.2%
▲ 0.4pp

342
▼ 28%

Returns jumps out immediately — preattentive processing at work.

“This is the ‘count the 3s’ principle from SWD Chapter 4, applied at the dashboard level. Without conditional formatting, finding the off-track metric is like counting grey 3s — you have to scan everything. With conditional formatting, the alert color makes the problem announce itself before conscious processing begins.”

“The Prep Air example in CwD Ch 7 does this too: contextual numbers at the top with triangle indicators showing whether metrics are above or below target. The user’s eye goes straight to the indicator arrows.”

Decluttering at the dashboard level

SWD Ch 3’s decluttering process applies to each chart. At the dashboard level there’s an additional layer:

What to check	Why
Redundant axes	Two charts with the same time axis side by side may only need labels on one
Multiple legends	If three charts use the same color encoding, one legend covers all of them
Repeated titles	If the dashboard title says “Sales Performance by Region,” the chart title “Sales by Region” adds nothing
Over-labeled KPI row	If a KPI card shows “$1.2M Revenue,” the bar chart below doesn’t also need an axis label saying “Revenue”

Each removal reduces the total number of elements competing for attention.

“These are decisions you won’t encounter when designing a single chart — they only arise when multiple views share a dashboard. The principle is the same as SWD Ch 3: every element should earn its place. But at the dashboard level, you’re looking for redundancy across views, not just within a single chart.”

“Removing a redundant axis might feel risky — won’t the user lose context? Usually not. Visual alignment (the continuity principle) creates a shared reference between adjacent charts. If two bar charts are vertically aligned and share a category axis, the user reads the axis labels once and applies them to both.”

Today’s takeaways

Dashboard design starts with the audience’s workflow.

Know whether you’re building explanatory or exploratory — it shapes every design choice.

Layout is information architecture: overview first, then detail.

Every interactive element should pass three tests: discoverable, answers a “so what,” survives 3 months.

Visual design at the dashboard level means consistency across views, hierarchy through sizing, and removing redundancy.

“Five ideas to take into practice this week. The most immediately useful is the three-question test — apply it to every parameter, filter, and action you build.”