Two dashboards display the same dataset and calculate the same totals. Yet they lead two managers toward opposite conclusions. How can both dashboards be mathematically correct?

Presentation is not merely the last decorative step in BI. How information is selected, aggregated, visualized, and contextualized can determine whether a technically correct analysis becomes a good or bad decision.
A supermarket dashboard can show an accurate sales total, refresh on schedule, and reconcile perfectly with the data warehouse. Every calculation might be ✨right✨
The next step feels obvious: give management the chart and let the numbers speak for themselves.
Except the number can tell us how much without showing where, when, or why. A company-wide total can conceal differences between products, branches, and periods. A trend can be made hard to notice by a poor visual encoding. And an impressive KPI can be irrelevant to the decision someone actually has to make.
The arithmetic can be correct while the interpretation is wrong.
That is the uncomfortable responsibility of the final Business Intelligence capability: presentation. Not decorating the analysis, but making its meaning accessible without quietly changing it.
A Cube Lets Us Ask Different Questions of the Same Facts
The natural starting point is Online Analytical Processing (OLAP). Operational systems use Online Transaction Processing (OLTP) to record business activity. OLAP supports analytical queries, often over consolidated data in a warehouse, and lets us inspect facts across multiple dimensions.
Think of a sales measure such as dollars_sold, organized by time , product , and location. We could display total annual sales, but the underlying multidimensional view supports more than that one answer. A cube is a conceptual way to organize those dimensions; it is not limited to three physical axes.
These operations are useful because aggregation is a choice. Roll up from city to country and a local variation may disappear into the total. Drill down again and it becomes visible. Both views can be mathematically correct; they answer different questions.
The first constraint is therefore the level of detail at which a claim remains meaningful. OLAP gives us ways to move between levels. It does not decide which level management ought to trust.
Four Datasets Can Agree on the Numbers and Disagree on the Shape
Now we reach a more unsettling case. Suppose we compute a mean, a variance, a correlation, and a regression line. It would be reasonable to expect those summaries to describe the relationship well enough for a quick decision.
Anscombe’s Quartet is a direct challenge to that intuition. Francis Anscombe constructed four small datasets whose key summary statistics and fitted straight lines are nearly identical. Their plotted shapes, however, are very different.

https://plotnine.org/gallery/anscombes-quartet_files/figure-html/cell-10-output-2.png
In one panel the relationship is approximately linear. Another is curved. A third includes an influential outlier. In the fourth, most observations share nearly the same horizontal position, while a single distant point strongly influences the fitted line.
The summaries did not lie. They compressed the observations into quantities that could not preserve every important feature of their arrangement.
A regression line can be correct as a calculation without being a sufficient description of the data. Visual analytics matters because a human can inspect structure that a few aggregates conceal. It does not replace statistical analysis; it helps reveal what the statistics deserve to be questioned about.
The Chart Has Its Own Rules of Meaning
Once we accept that looking at the data matters, another tempting assumption appears: any readable chart should do. Put the numbers in a pie chart, bar chart, or scatterplot, add some colors, and move on.
But visual marks are an encoding. They translate attributes into position, length, size, color value, color hue, texture, orientation, or shape. Different encodings support different judgments, and different data types permit different comparisons.
- Nominal : Categories without an inherent order, such as product types. The display must distinguish identities without implying a ranking.
- Ordinal : Ordered categories, such as satisfaction levels. The display must preserve order without assuming equal gaps.
- Interval : Equal differences, but no meaningful absolute zero, such as Celsius temperature. The display must represent differences accurately.
- Ratio : Quantities with a meaningful zero, such as units sold or salary. The display must support meaningful magnitude and proportion comparisons.
This is why position on a common scale is generally more precise for quantitative comparisons than asking readers to estimate areas or shades of color. Jacques Bertin’s visual variables provide a language for these choices; Cleveland and McGill’s graphical-perception work provides experimental evidence about how accurately people decode them. Later research on visual communication explains why important numeric differences usually deserve strong positional encodings.
For example, if the question is how salary varies across locations, placing locations along a categorical axis and salaries on a common numerical axis is usually more useful than mapping salary solely to color. Color can identify categories or draw attention; it is not a substitute for a precise scale when the magnitude is the point.
There are two further responsibilities: expressiveness , which means showing what the data supports without adding misleading implications, and consistency , which means preserving comparable scales and meanings across the display. Changing baselines, using inconsistent units, or omitting the context of a measure can make legitimate values feel more dramatic than they are.
The chart is not merely showing data. It is asking the reader’s visual system to make a particular comparison. The design should make the right comparison easy.
One Screen Cannot Serve Every Decision
A performance dashboard is often compared with a car dashboard: it makes important conditions visible without requiring the driver to inspect the engine. In a business, it acts as an organizational magnifying glass.
It is tempting to make that magnifying glass as comprehensive as possible: more gauges, more charts, more real-time numbers. Unfortunately, a dashboard that only monitors can leave the user staring at a problem without a path to understand or manage it.
The Three Threes framework makes this limitation explicit. A capable performance dashboard brings together three different axes:
- Three applications: monitoring reveals performance; analysis investigates exceptions; management helps coordinate a response.
- Three layers: a graphical monitoring view, a multidimensional analysis view, and detailed operational information beneath it.
- Three types: operational, tactical, and strategic dashboards, emphasizing different decisions and audiences.
These are not nine unrelated screens. They describe how the same information-delivery system can support different kinds of work. Wayne Eckerson’s treatment of performance dashboards explains why the monitoring, analytical, and management responsibilities belong together.
An operational user might need a prompt about today’s inventory. A departmental manager may need a trend and its causes. An executive may need to know whether the organization’s strategy is on course. The Origami Software Engineer may need to know the information needed to investigate exceptions. These views can share underlying data without presenting identical detail.
The failure is subtle: giving everyone the same correct dashboard is not the same as giving everyone the information they need.
A Number Becomes a Performance Measure Only in Context
Now suppose the dashboard displays a sales figure. Even if its chart is clear and the user can drill into the data, one question remains: is the result good?
That requires a goal, a comparison, or a target. A Key Performance Indicator (KPI) expresses progress toward something the organization has decided matters. A balanced scorecard places measures against objectives and targets, so the gap between actual and desired performance becomes visible.
The familiar balanced scorecard, developed by Kaplan and Norton , looks beyond financial outcomes through four perspectives: financial , customer , internal business process , and learning and growth. It addresses a real limitation of measuring success solely through what has already appeared in financial results. The original balanced-scorecard approach connects these perspectives to strategy rather than treating them as a decorative collection of metrics.
For a supermarket, revenue may matter, but so might customer satisfaction, availability of products, and the organization’s ability to improve its processes. Those are examples of possible concerns, not a claim that any particular company tracks these exact measures.
The progression is vision -> strategic objectives -> measurable indicators and targets -> observed results -> feedback and learning. Translating the vision, communicating and linking objectives, planning, and learning from results are part of making the scorecard useful.
Without a target, a dashboard can tell us what happened. Without a relevant KPI, it may not tell us whether that outcome advances the strategy. Choosing a metric is already a decision about what deserves attention.
Real-Time Numbers Are Not the Same as Real-Time Decisions
It is easy to mistake freshness for usefulness. A dashboard updates continuously, so surely the organization can act continuously too.
Not necessarily. A real-time feed with no context may simply deliver confusion faster. Attractive animations cannot compensate for weak KPI definitions. A lightweight dashboard that cannot support deeper analysis may become a dead end. And an alert is not particularly helpful if no one is responsible for the next step.
The case of Western Digital makes the alternative concrete. Its Vigilant Information System (VIS) combined BI across multiple sources with management dashboards that supported automated alerts and responses. Instead of depending only on a user to request a report, the system could initiate attention when conditions warranted it. This was a shift from passive information delivery toward sense and respond.
Its decision cycle used OODA: Observe, Orient, Decide, Act.
The published Western Digital case emphasizes a surprisingly important constraint: the time taken by the OODA loop should match the organization’s ability to respond. Chasing zero latency everywhere is not automatically helpful. If a decision and the process it changes require more time, a faster refresh may add cost and noise without improving the response.
It also makes clear that the system was not simply a technology installation. The dashboards had to support coordination, learning, and management action. Its reported operational improvements were associated with both the VIS and changes to business processes, not a prettier display alone.
The Last Step in BI Is Not Merely the Last Screen
Return to the accurate sales dashboard. Its total might be correct. Its chart might even be attractive. But a responsible decision still needs the appropriate level of detail, an encoding the viewer can interpret, an audience-specific way to investigate exceptions, and a measure connected to an actual objective.
That is why presentation belongs beside organizational memory , information integration , and insight creation as a genuine Business Intelligence capability. It is the boundary where carefully processed evidence meets human judgment.
The surprising lesson is that this boundary can introduce errors without corrupting a single row. We can roll up the wrong dimension, emphasize the wrong visual property, highlight a metric without its target, or deliver an alert faster than anyone can use it.
Correct information is necessary. Making the right meaning available to the right decision-maker is another engineering problem entirely.
That’s not failure.
That’s evolution.
The “I liked this” Starter Pack:
Don’t let your fingers get lazy now.
- Like : It tells me this was worth writing.
- A Comment: Tell me your thoughts, your favorite snack, or a better title for this blog.
- Boost it: Especially with that one developer who definitely needs this.
Thanks for being here. It genuinely helps more than you know!
— Aaroophan Varatharajan
Find me elsewhere:
- Professional stuff: linkedin.com/in/Aaroophan
- Code stuff: github.com/Aaroophan
- UI stuff: aaroophan.dev/Aaroophan
- Life stuff: instagram.com/Aaroophan



Top comments (0)