Syllabus topics: Introduction to dashboards; when to use dashboards; dashboard components; principles of effective visualization and dashboarding; advanced visualizations — parameters, slicers, filters, drilldowns, graphs and maps; dashboard design — layout, alignment, accessibility; publishing dashboards — Power BI Service, Tableau Public; storytelling and insight communication; build a complete BI dashboard using either tool. Case study — a business decision-making scenario such as sales forecasting or budgeting.
THE BIG IDEA
A dashboard is a single screen of the few numbers a specific person needs to do their job, updated automatically.
Every word is a constraint that gets violated in practice:
FORMULA
| Dashboard | Report | Scorecard | |
|---|---|---|---|
| Purpose | Monitor | Analyse | Track against targets |
| Size | One screen | Many pages | One screen |
| Time | Current status | A period, in depth | Progress toward a goal |
| Detail | Summary, with drill-down | Full detail | KPIs vs targets only |
| Question | "Is anything wrong?" | "What exactly happened?" | "Are we on track?" |
| Frequency | Glanced at daily | Read occasionally | Reviewed monthly |
A scorecard is a dashboard where every metric has a target. Balanced Scorecard — Kaplan and Norton's four perspectives (financial, customer, internal process, learning and growth) — is worth naming if the question asks about strategic dashboards.
A genuinely good answer names these, because it shows judgement:
| Situation | Use instead |
|---|---|
| The question is asked once | An analysis, a slide, an email |
| The answer needs a paragraph of explanation | A written report |
| Nobody will act on any value it can show | Nothing. Do not build it |
| The data is unreliable | Fix the data first — a dashboard makes bad data authoritative |
| The user needs row-level detail | A paginated report or an export |
| The decision is made once a year | A one-off analysis |
The test from Unit 1 §1.3 applies here too: when this number moves, who does what? If nobody can answer, the dashboard should not be built.
| Component | Purpose | Guidance |
|---|---|---|
| KPI cards | The headline numbers | 3–5, top row, with comparison to target or prior period |
| Trend chart | Direction over time | Almost always earns its space |
| Breakdown | Composition by category | Bar, ranked. Rarely a pie |
| Detail table | The rows behind the summary | Bottom, or on a drill-through page |
| Slicers / filters | Let the user narrow | Left or top edge, consistently placed |
| Title and timestamp | What this is, and how fresh | The timestamp is not optional |
| Legend | Decode the colours | Better still, label directly and delete it |
"Revenue ₹12,880" tells nobody anything. ₹12,880, up 8% on last quarter, against a target of ₹14,000 supports a decision.
Every KPI card should carry at least one of:
This is the highest-value single rule in the unit and it is cheap to apply.
FORMULA
| Principle | Meaning | Violation |
|---|---|---|
| Purpose first | Every visual answers a stated question | "Because we had the data" |
| Data-ink ratio (Tufte) | Maximise ink that carries data; delete the rest | 3-D bars, heavy gridlines, drop shadows |
| Choose the right encoding | Position > length > angle > area > colour, in accuracy | A pie where a bar belongs |
| Zero baseline on bars | Bar length is the message | Truncated axis exaggerating a difference |
| Consistent colour meaning | One colour, one thing, across the whole dashboard | Red meaning "loss" here and "region A" there |
| Order deliberately | Sort by value unless there is a natural order | Alphabetical by accident |
| Label directly | Put labels on the marks | Forcing a trip to the legend |
| Show uncertainty | A small denominator is not a fact | 25% attrition from a team of four |
| Progressive disclosure | Summary first, detail on demand | Everything at once |
KEY INSIGHT
Open any default chart and delete: the border, the background fill, heavy gridlines, the redundant legend, and every decimal place nobody reads. The chart gets easier to read every time. Nothing you delete in that list has ever carried information.
Not banned, but narrow: parts of one whole, five slices or fewer, and only when "roughly half" is the message rather than a ranking. Humans compare angles badly and lengths well. Two pies side by side are worse still — nobody can compare across them.
A ranked bar chart is the right answer to most questions a pie is used for.
| Requirement | What to do |
|---|---|
| Colour blindness (~8% of men) | Never encode by colour alone; add shape, label or position. Avoid red/green pairs |
| Contrast | At least 4.5:1 for text against its background |
| Text size | Nothing below 10–12 pt; dashboards get shown on projectors |
| Alt text | Set it on every visual — screen readers use it |
| Tab order | Set it, so keyboard users move sensibly |
| Not by colour alone | A red cell must also carry a symbol or a number |
Red/green for good/bad is the single commonest accessibility failure, and it is exactly the pair most affected by the most common colour blindness. Use blue/orange, or add symbols.
FORMULA
| Feature | What it does | Changes |
|---|---|---|
| Filter | Restricts data for a visual, page or report | The data shown |
| Slicer | An on-canvas filter the user can see and click | The data shown |
| Parameter | A user-chosen value feeding a calculation | What is calculated |
| Drilldown | Moves down a hierarchy in place | The level of detail |
| Drill-through | Jumps to another page filtered to the selection | The page |
A slicer is a filter the user can see. Both restrict data; a filter lives in the Filters pane (and may be hidden), a slicer occupies canvas space and invites interaction. Put on canvas the two or three the user changes often; leave the rest in the pane.
THE BIG IDEA
A parameter changes an input to a calculation, not the rows shown. That is the distinction, and it is what makes parameters the what-if feature — and therefore the model-management component of a DSS (Unit 1 §1.6) living inside a BI tool.
Filter: "Show me the South region" -> fewer rows
Parameter: "What if we raise prices by 5%?" -> different numbers, same rows
In Power BI: Modeling → New parameter → Numeric range, which creates a table and a measure you use in DAX. In Tableau: a parameter plus a calculated field that references it.
Projected Revenue = [Total Revenue] * (1 + 'Price Change'[Price Change Value])Say the DSS connection in the viva. A what-if parameter is a small DSS, and it ties Unit 5 back to Unit 1.
The syllabus lists "Graphs and Maps" alongside the interactive features, and both mean something more specific here than the basic charts of Unit 2.
| Visual | What it adds | Use it when |
|---|---|---|
| Combo chart (column + line) | Two units on one canvas — revenue as columns, margin % as a line | A total and a rate must be read together |
| Waterfall | Shows how a total got from A to B, step by step | Explaining a variance: budget → actual |
| Scatter / bubble | Two measures, plus size and colour | Looking for a relationship, not a ranking |
| Decomposition tree | Interactive, user-chosen drill path | "Why is this number what it is?" — the user picks the order |
| Key influencers | Ranks what drives a metric | An automated first pass at a diagnostic question |
| Small multiples | The same chart repeated per category | Comparing shapes across many categories |
| Gauge / KPI | Value against a target | A target genuinely exists. Otherwise it is decoration |
Maps, specifically:
| Map type | Encodes | Watch for |
|---|---|---|
| Filled (choropleth) | A value as area colour | Area is not population. Large empty districts dominate the eye |
| Bubble / symbol | A value as circle size at a point | Overlapping bubbles in dense cities |
| Density / heat | Concentration of points | Good for "where", useless for "how much" |
| Shape map | Custom regions from a shapefile | Needed for sales territories, which are not administrative areas |
A map is the most seductive visual on the list and the most often misused. If the question is "which region sold most?", a ranked bar chart answers it better — you can read the order instantly, which no map allows.
Use a map when location itself is the variable: distance to a store, clustering, coverage gaps, routing. "Which of our districts have no outlet within 20 km?" is a map question. "Rank the districts by sales" is not.
The choropleth trap is worth stating: colouring districts by total sales makes large rural districts look important because they are big on screen. Normalise — sales per capita, or per outlet — or use bubbles, whose size you control.
Drilldown needs a hierarchy defined in the model (Unit 4 §4.6). Given
Region → City → Store, the user expands from region to city to store in place.
| Control | Effect |
|---|---|
| Drill down (single item) | Expand the selected item one level |
| Expand all | Add the next level for every item |
| Drill up | Back a level |
| Drill through | Jump to a detail page filtered to the selection |
Drill-through is the right answer to "users want the underlying rows". Keep the summary clean, and put the detail table on a drill-through page rather than on the dashboard.
FORMULA
Readers of a left-to-right script scan in an F: across the top, across again lower, then down the left edge. Design for it.
+--------------------------------------------------+
| Title As at 27-08-2026 |
+--------------------------------------------------+
| [KPI] [KPI] [KPI] [KPI] | <- most important, top-left
+--------------------------------------------------+
| | |
| Trend over time | Breakdown | <- supporting
| | by category |
+------------------------------+-------------------+
| Detail table / exceptions | <- detail, on demand
+--------------------------------------------------+
| [slicers] |
+--------------------------------------------------+
Top-left is the most valuable real estate on the screen. Put the number the user came for there. The commonest layout mistake is putting the company logo in it.
Align to a grid. Misalignment by three pixels reads as carelessness even when nobody consciously notices it.
Use a consistent gutter between visuals — one spacing, everywhere.
Limit the palette. One accent colour, a neutral, and semantic colours reserved for meaning.
One font, two or three sizes. Never more than two fonts.
Round sensibly. ₹12.9K on a card; ₹12,880 in the detail table. Never ₹12,880.0000.
Whitespace is not wasted space. It is what makes the rest readable.
Scroll test: if the dashboard scrolls, it is not a dashboard. Split it, or cut it.
Five-second test: show it to someone for five seconds, take it away, and ask what the main message was. If they cannot say, the hierarchy is wrong — not their attention.
| Power BI Service | Tableau Public | |
|---|---|---|
| Publish from | Desktop → Publish → workspace | Desktop → Server → Tableau Public → Save |
| Who can see it | Whoever you grant access to | Everyone on the internet |
| Cost to share | Pro licence both sides | Free |
| Refresh | Scheduled (8/day Pro, 48 Premium); gateway for on-prem | Manual re-publish, or a linked Google Sheet |
| Row-level security | Yes | No |
| Right for | Real organisational data | Portfolios, coursework, public data |
Tableau Public publishes to the open web and allows download of the workbook. For lab experiments 8, 9 and 12 that is intended and fine. For anything containing real student, employee or customer data it is a data breach. Check what is in the extract before you press Save.
KEY INSIGHT
| After publishing | Why |
|---|---|
| Set scheduled refresh and alert on failure | A silently stale dashboard is worse than none |
| Add the data-as-at timestamp to the canvas | Users must see freshness without asking |
| Check usage metrics after a month | Nobody opening it is the finding |
| Write one paragraph of what it is for | Six months on, nobody remembers |
THE BIG IDEA
A chart shows what happened. A story says what it means and what to do.
The gap between them is where BI either earns its budget or does not.
FORMULA
Context -> Complication -> Cause -> Consequence -> Call to action
| Step | Says | Example |
|---|---|---|
| Context | The normal state | "Revenue runs ₹12–13 lakh a quarter" |
| Complication | What changed | "Q2 fell 11% in South" |
| Cause | Why | "Two large accounts churned in April" |
| Consequence | Why it matters | "That is 8% of annual revenue if unrecovered" |
| Call to action | What to do | "Assign a retention owner to the top 10 accounts this quarter" |
The call to action is what distinguishes BI from reporting, and it is what the ten-mark storytelling question wants to see.
Lead with the finding, not the method. "South fell 11%" first; how you calculated it only if asked.
One message per visual. If a chart needs two sentences to explain, it is two charts.
Annotate on the chart. An arrow saying "price change here" beats a paragraph below it.
Quantify the consequence in the units the audience cares about — rupees, customers, days. Not percentages alone.
Say what you do not know. "This is two months of data; the trend may not hold" builds more trust than false confidence.
Recommend something. An analysis with no recommendation puts the work back on the audience.
Statistical Foundations for Data Science taught it and BI is where it gets violated. A dashboard showing two lines moving together will be read as cause and effect by whoever sees it. If you cannot support the causal claim, do not put the two lines on one chart — or annotate it explicitly. This is a legitimate exam point about ethical insight communication.
The syllabus sets a decision-making scenario here. Sales forecasting exercises every part of the unit.
The decision. How much stock to buy and what quota to set per region for next quarter. Decided by the sales director, quarterly.
The dashboard:
| Zone | Contents |
|---|---|
| KPI row | Revenue QTD vs target · Forecast for quarter-end · Variance % · Pipeline coverage |
| Trend | Actual by month, forecast continuing it as a dashed line with a confidence band |
| Breakdown | Revenue by region, ranked, with target markers |
| What-if | Parameters: growth rate, price change, win rate |
| Detail | Drill-through to accounts, for the account owner |
Show the forecast's uncertainty. A single forecast line will be treated as a promise. A band — or three scenarios, low/expected/high — communicates what a point estimate cannot. This is Statistical Foundations for Data Science's confidence interval doing its actual job.
Do not draw the forecast in the same style as the actuals. Dashed, lighter, and clearly labelled, with a vertical rule at "today". Otherwise people will quote a forecast as an actual within the week.
Budget variance needs both absolute and percentage. A region 50% under budget on ₹2 lakh matters less than one 5% under on ₹2 crore. Show both, and sort by the absolute figure, because that is the one that decides where attention goes.
The what-if parameter is what makes it a decision tool rather than a report. Let the director move the growth-rate slider and watch the quarter-end forecast move; that closes the loop back to Unit 1's Decision Support System, and saying so is a good way to end a ten-mark answer.
PROBLEM 1
What makes a dashboard effective? List and explain the principles, and describe a good layout. (10 marks)
Solution.
Open with the definition and its constraints: one screen, few visuals, a specific person, something they act on, refreshed automatically.
Then the principles from §5.3 — purpose first; data-ink ratio; right encoding (position beats length beats angle beats area beats colour); zero baseline on bars; consistent colour meaning; deliberate ordering; direct labelling; show uncertainty; progressive disclosure. Explain each in a line.
Then the layout, drawn: title and timestamp at the top; 3–5 KPI cards on the top row, most important top-left; trend and breakdown in the middle; detail at the bottom or on a drill-through page; slicers in a consistent position. Mention the F-pattern as the justification.
Add accessibility, because most answers omit it: never encode by colour alone, 4.5:1 contrast, alt text, and avoid red/green, which is both the commonest choice and the worst for the commonest colour blindness.
Close with the two tests — the scroll test (if it scrolls it is a report) and the five-second test (if the viewer cannot state the message, the hierarchy is wrong).
PROBLEM 2
Distinguish filters, slicers, parameters and drilldowns. (10 marks)
Solution.
Give the table from §5.4, then make the two distinctions that carry the marks:
Slicer vs filter: both restrict which rows are shown; a slicer is a filter the user can see and click, occupying canvas space. Filters live in the pane and may be hidden. Put the two or three most-used on canvas.
Filter vs parameter — the important one: a filter changes which rows are shown; a parameter changes what is calculated. Give the contrast:
Filter: "Show me the South region" -> fewer rows, same measures
Parameter: "What if we raise prices by 5%?" -> same rows, different numbers
Drilldown vs drill-through: drilldown moves down a hierarchy in place (Region → City → Store) and needs a hierarchy defined in the model; drill-through jumps to a different page filtered to the selection, and is the right way to give users row-level detail without cluttering the dashboard.
Finish with the connection worth stating: a what-if parameter is the model component of a Decision Support System (Unit 1 §1.6) inside a BI tool.
PROBLEM 3
Design a dashboard for sales forecasting. Describe its components and the traps. (10 marks)
Solution.
Start with the decision, not the charts: how much stock to buy and what quota to set per region, decided quarterly by the sales director. Everything follows from that.
Components: a KPI row (revenue QTD vs target, forecast to quarter-end, variance %, pipeline coverage); a trend chart with actuals solid and forecast dashed; a ranked regional breakdown with target markers; what-if parameters for growth rate and price change; drill-through to account detail.
The three traps, which is where the marks are:
Show uncertainty. A single forecast line is read as a promise. Use a confidence band or low/expected/high scenarios — Statistical Foundations for Data Science's confidence interval doing its job.
Style the forecast differently. Dashed, lighter, labelled, with a vertical rule at today. Otherwise a forecast gets quoted as an actual.
Show variance in both rupees and percent, sorted by rupees. 50% under on ₹2 lakh matters less than 5% under on ₹2 crore.
Close on the parameter: it is what makes this a decision tool rather than a report, and it is Unit 1's DSS model component living inside a BI dashboard.
Two marks
Five marks
Ten marks
COMMON ERRORS
Defining a dashboard as "several charts on a page". One screen, few visuals, one audience, something they act on, refreshed automatically.
A KPI with no comparison. "₹12,880" is not information. Against target, prior period or peer — always.
Saying a parameter filters data. It changes an input to a calculation. That distinction is the question.
Truncating a bar chart axis. Length is the message; a truncated axis lies with it.
Red/green for good/bad. The commonest choice and the worst for the commonest colour blindness.
Forgetting the timestamp. Users cannot judge freshness, so they assume it is current — and eventually they are wrong.
Putting a detail table on the dashboard. Use drill-through.
Ending an analysis with a chart. Finish with a recommendation, or the work lands back on the audience.
Omitting accessibility. It is in the syllabus — "layout, alignment, accessibility" — and most answers skip the third word.