The Most Dangerous Number on a Dashboard Is the One Nobody Questions
A dashboard is meant to help people make better decisions. That sounds harmless enough. Put the important figures on one screen, add a few traffic-light colours, perhaps a reassuring arrow pointing upwards, and let everyone get on with their day.
The trouble starts when a number becomes familiar.
Once a figure appears every morning in the same corner of the same dashboard, people stop asking what it means, where it came from, whether it is complete, and whether it is still answering the question they actually have. It acquires the quiet authority of a laminated sign in an office kitchen. Nobody knows who put it there, but nobody fancies being the first to remove it.
The most dangerous number on a dashboard is not necessarily the lowest one, the highest one, or the red flashing one. It is the one nobody questions.
A number is a model, not the world
Every metric is a compressed version of reality. That is useful. Nobody wants to inspect ten million database rows before deciding whether a service is healthy. But compression always leaves something out.
Take a temperature-monitoring dashboard showing 99.8% device availability. At first glance, that is splendid. Tea and biscuits all round. Yet availability might mean the device checked in at least once during the previous day. It might exclude units deliberately placed into maintenance mode. It might count a gateway as healthy even when its sensor inputs have stopped updating. It might be calculated in UTC while the operations team is reviewing local business days.
None of those choices is automatically wrong. They are definitions. The danger comes when the definition is invisible.
A device can be connected but reporting stale data. A sensor can provide a value that is electrically plausible but physically absurd. A dashboard can show green while a fridge is quietly warming up because the last valid reading arrived six hours ago and the system has not distinguished last received from currently trustworthy.
That distinction matters. Especially when the dashboard is helping someone decide whether stock is safe, an alarm was handled, or a site needs attention.
The comfort of a clean line
Humans rather like a tidy chart. A line trending gently upwards makes us feel we have grasped the situation. A single percentage feels manageable. A red-amber-green status is practically a cuddle in spreadsheet form.
But a clean chart can conceal a messy process.
Consider an average. Average temperature, average response time, average ticket resolution, average energy consumption. Averages are useful summaries, but they are also very capable of hiding the thing that should worry you. If nine readings are perfectly normal and one exceeds a critical limit, the average may look almost saintly. The product or patient or process affected by that one reading may take a less charitable view.
The same goes for averages in disability and accessibility reporting. A website might report that most pages meet a given accessibility check, while the one crucial task - applying for support, paying a bill, booking an appointment - remains impossible with a keyboard or screen reader. An overall score offers very little comfort when the door you need is still locked.
Good dashboards pair summary measures with evidence of spread and exception. Show the maximum as well as the mean. Show late data separately from missing data. Show the count behind a percentage. Let people see whether a target was comfortably met or technically scraped by while several systems were on fire behind a tasteful green icon.
Ask the mildly irritating questions
The best thing a team can do is normalise a few questions that may feel slightly awkward at first:
- What exactly does this number count?
- What does it exclude?
- When was it last calculated?
- What happens if the source data arrives late or twice?
- Has the definition changed since last month?
- Which decision would be different if this figure were wrong?
That last question is particularly valuable. If no decision changes, the metric may be decorative. Decorative metrics are not evil. A little operational décor harms nobody. But they should not be allowed to dress up as decision support merely because they have a gradient fill and a confident-looking font.
For important measures, make the calculation inspectable. A user should be able to move from a headline figure to the underlying population, filters, time window and data-quality rules. This does not mean every dashboard needs to become a database administration course. It means the route to an explanation must exist.
A tooltip saying “Availability: 99.8%” is not an explanation. It is a number wearing a tiny hat.
Design for challenge, not compliance
Dashboard designers often optimise for fast reading, which is sensible. The screen must help a busy person see what requires attention. Yet fast reading should not mean blind obedience.
Use labels that reveal the rule: “devices reporting in the last 15 minutes” is better than “online”. Flag partial data clearly. Show a timestamp for the freshest source record, not simply the time the page refreshed. A dashboard that refreshes every 30 seconds while its underlying feed is delayed by four hours is very efficient at displaying old news.
Thresholds deserve the same care. A red line at 8°C, 80%, or 30 minutes can look as though it descended from a mountain engraved on stone tablets. Usually it came from a policy, a risk assessment, a commercial target, or someone's best judgement on a Tuesday afternoon. Record why it exists, who owns it and when it should be reviewed.
There is warmth in this discipline. It respects the person looking at the screen by giving them enough context to think, rather than asking them to accept a verdict from a colourful rectangle.
A useful dashboard does not demand faith. It invites curiosity. The number may still be alarming, reassuring or inconveniently complicated. But when someone asks, “Are we sure?”, the system should have an answer better than, “Well, it has always been there.”