Dashboard signal selection works by choosing the few measures that deserve immediate attention.
This sounds like a visual-design problem.
It is really a decision-design problem.
A dashboard has limited space, but the deeper scarcity is human attention. Every metric placed on the primary screen competes with every other metric for interpretation time.
The goal is therefore not to represent the whole system.
The goal is to represent the smallest set of signals that can tell the viewer whether the system is working, drifting or demanding investigation.
This article is the first pillar of How Dashboards Fail.
The Core Idea: A Signal Earns Space by Changing a Decision
A useful dashboard signal should answer at least one of these questions.
- Is the system achieving its purpose?
- Is something important changing?
- Is a threshold approaching?
- Is one group or component behaving differently?
- Does somebody need to investigate?
- Does somebody need to act?
If a metric cannot affect interpretation, priority or action, it probably belongs in a report, drill-down page or archive rather than the primary dashboard.
Start From Purpose, Not Data Availability
The GOV.UK Service Manual recommends defining the purpose of the service first, then choosing a small number of metrics that answer whether the service is working.
This principle generalises well.
Available data are not automatically important data.
A dashboard built from whatever the system happens to record will usually overrepresent operational convenience and underrepresent the real objective.
Separate Outcome Signals From Activity Signals
Activity tells us what the system did.
Outcome tells us what changed because of it.
Both can matter.
The problem is when activity becomes a substitute for outcome.
A tutor may conduct twelve lessons. A publishing team may produce fifty articles. A service desk may close four hundred tickets.
None of those numbers alone proves the intended capability, reach or reliability improved.
Use Activity as a Mechanism Signal
Activity metrics are strongest when they help explain outcome.
If student independence fell while prompting increased, the activity signal is diagnostic.
If site traffic rose while conversion stayed flat, acquisition volume may explain the changed workload.
The dashboard should distinguish ‘what happened’ from ‘what we did’.
Choose Leading and Lagging Signals
Lagging signal
Shows the result after the system has already produced it.
Examples include final exam marks, monthly revenue, completed transactions or published error rates.
Leading signal
Shows a condition that tends to precede the outcome.
Examples include prerequisite instability, queue growth, unresolved blockers, missed practice or declining completion quality.
A good dashboard often needs both.
Lagging signals tell us whether the outcome happened. Leading signals create time to intervene.
Do Not Pretend a Leading Signal Is a Guarantee
A leading indicator is useful because it moves before the outcome.
It is not proof that the outcome will follow.
The relationship should be monitored and recalibrated.
Otherwise the organisation begins managing a proxy that may have stopped predicting what matters.
Protect Against Proxy Drift
A proxy is a measure used because the true objective is difficult to observe directly.
Examples include practice completion as a proxy for learning effort or page views as a proxy for audience interest.
Proxies are practical.
They become dangerous when the proxy improves while the underlying objective does not.
Always keep at least one outcome check capable of disagreeing with the proxy.
Prefer Measures With a Clear Interpretation
A metric should have a definition that the intended viewer can understand.
If the number requires a ten-minute explanation every time, it may be too specialised for the primary layer.
Complex measures can still exist deeper in the system.
The dashboard surface should favour signals whose meaning is stable and communicable.
Prefer Measures With a Stable Definition
If the metric calculation changes frequently, the trend becomes difficult to interpret.
The signal may still be necessary.
When the definition changes materially, mark the break rather than pretending the series is continuous.
Prefer Signals With Reasonable Data Quality
A theoretically perfect KPI is weak if half the data are missing.
Signal selection should consider whether the measurement process is reliable enough to support the decision.
Where data quality is weak but the metric is still important, show the uncertainty rather than hiding it.
Prefer Signals That Arrive in Time
A metric that becomes available after the decision window closes cannot serve as an operational signal.
It may still be useful for retrospective review.
Dashboard placement should match the latency of the measure.
Use a Small Number of Primary KPIs
The GOV.UK Service Manual suggests choosing a small number of KPIs that answer whether the service is working, while keeping other measures available for deeper analysis.
That is a useful dashboard architecture.
Primary screen: few decision-critical signals.
Secondary layers: richer diagnostic measures.
Avoid the KPI Inflation Problem
A KPI is supposed to be key.
If a dashboard contains thirty ‘key’ metrics, the word has lost meaning.
Not every useful measurement is a KPI.
Reserve the primary category for signals that materially represent system purpose or risk.
Use Signal Families
A strong dashboard can group signals by function rather than by data source.
- Outcome — is the goal being achieved?
- Demand — what load is arriving?
- Capacity — can the system absorb it?
- Quality — is the work reliable?
- Risk — what may fail next?
- Flow — where is work blocked?
- Recovery — is the system returning to normal?
This prevents one convenient data source from dominating the screen.
Use One Signal to Challenge Another
Pairs can reveal contradictions.
For example: output rises while error rate rises too.
Completion improves while satisfaction falls.
Marks improve while prompting remains high.
Traffic grows while useful enquiries decline.
Contradictory signals often reveal gaming, quality trade-offs or hidden dependencies.
Do Not Select Metrics by Visual Appeal
A metric that produces a beautiful chart does not earn importance.
Some of the strongest signals are simple counts, ratios or yes/no states.
The dashboard should optimise understanding, not chart diversity.
Select Signals by Consequence
Rare conditions may deserve primary visibility when their consequence is high.
A severe safety issue, critical publication error or exam-readiness collapse can matter more than a frequently changing low-impact metric.
Frequency and consequence should both influence placement.
Select Signals by Reversibility
When a decision becomes expensive to reverse, earlier warning deserves more visibility.
A learner heading toward a major examination with a broken prerequisite needs an earlier signal than a low-stakes worksheet error.
A publication with a canonical collision needs a pre-release signal because repair becomes more expensive after indexing and linking.
Select Signals by Propagation
An upstream problem deserves attention when it can damage many downstream outcomes.
This links directly to the dependency-criticality architecture.
A small failure in a shared prerequisite may be a stronger signal than several downstream symptoms.
Select Signals by Actionability
A signal is stronger when the viewer can do something useful with it.
This does not mean every metric needs an immediate fix.
It means the signal should route somewhere: investigate, escalate, verify, reallocate, wait or decide.
A metric that only creates anxiety is a poor operational signal.
Distinguish Monitoring Signals From Diagnostic Measures
Monitoring asks whether attention is needed.
Diagnosis asks why.
The primary dashboard should favour monitoring signals.
Diagnostic measures belong behind drill-down unless they themselves are critical.
Signal Selection in Education
A learning dashboard can easily overvalue visible administrative data: attendance, homework completion and test marks.
Those matter.
A more capable view may also track whether errors repeat, whether prerequisite gaps are shrinking and whether the learner can perform with less prompting.
The right signal depends on the instructional decision.
Signal Selection in Mathematics
An A-Math dashboard should not automatically treat chapter scores as the only signal.
If a calculus score falls, the system may need a prerequisite-stability signal showing algebraic manipulation, equation solving or graph interpretation.
Upstream signals can make the dashboard diagnostically useful.
Signal Selection in English
An overall writing mark is a lagging outcome.
Useful supporting signals might include task fulfilment, evidence selection, sentence-control errors, editing recovery and time completion.
These should not all become primary KPIs.
Choose the measures that change the next teaching move.
Signal Selection in Publishing
Article count is an activity metric.
Traffic is an outcome signal of reach, but not necessarily quality.
Collision state, unresolved factual blockers, indexability and internal-link completeness may be stronger operational signals during production.
The dashboard changes because the decision changes.
Signal Selection in Family Learning
Avoid turning family life into a sensor network.
A few high-leverage signals are enough.
Sleep stability, start resistance, homework completion pattern and upcoming assessment load can be more useful than minute-by-minute tracking.
The dashboard should support calm decisions.
The Signal Selection Matrix
- Purpose — what system objective does the metric represent?
- Decision — what decision could the metric change?
- Type — outcome, activity, leading, lagging or risk?
- Data quality — is the measure trustworthy enough?
- Latency — does it arrive before the decision window closes?
- Consequence — what happens if the signal is missed?
- Propagation — how much downstream work can it affect?
- Actionability — where does the signal route the viewer?
- Redundancy — is another metric already telling the same story?
- Retirement — when would this measure stop being useful?
The Removal Test
Take one metric off the dashboard.
What decision becomes harder?
If no important decision changes, the metric may belong elsewhere.
Repeat until the primary screen becomes sparse enough that important movement is obvious.
The Contradiction Test
For every primary signal, ask which other signal could prove that its apparent improvement is misleading.
This protects against proxy gaming.
A good dashboard does not merely congratulate itself.
It contains enough independent evidence to detect when one metric is being optimised at the expense of the purpose.
The Deeper Principle: Key Means Decision-Relevant
Dashboard signal selection is successful when each primary metric has a reason to exist.
The reason is not that data were available.
The reason is that the metric helps the viewer detect something important early enough to think or act differently.
Across the eduKate Ecosystem
eduKateSG’s How Data-Informed Instruction Works keeps classroom evidence tied to the next teaching move. How Warnings Fail owns salience and actionability once a signal becomes exceptional. How Dependency Criticality Works helps identify upstream measures with broad downstream leverage. Why Tuition Tracking Systems Fail Without Learning State Diagnostics applies the same principle directly to tuition tracking.
Sources and Further Reading
GOV.UK Service Manual — How to Set Performance Metrics for Your Service
GOV.UK Service Standard — Define What Success Looks Like and Publish Performance Data
Continue the Series
How Dashboards Fail | Why More Metrics Can Produce Less Control
How Dashboard Context Works | Give Every Number a Baseline, Trend, Segment and Time Window