Leading Indicators: 4 Distortions That Turn Early Warnings Into Dashboard Theater

A safety dashboard can be full of green numbers while the same exposure keeps returning to the field. The problem is not always poor data collection. Often, the organization has selected measures that reward visible activity while leaving the quality of the decision untested.

That distinction matters because leaders use leading indicators to see trouble before an injury, process loss, or serious incident. If the indicator only confirms that a meeting happened or a form was completed, it gives management a reassuring record instead of an early warning. In projects supported by Andreza Araujo, the more useful question has been whether the information changed a decision close to the work.

Why dashboard activity is not the same as prevention

Leading indicators are often described as measures that precede harm. That definition is incomplete. A measure can appear before an event and still have no preventive value if it is weakly connected to exposure, insensitive to control quality, or ignored when its result is uncomfortable.

James Reason’s work on organizational accidents helps explain the gap. Harm becomes more likely when latent conditions, local decisions, and failed defenses align. A dashboard that counts only the visible action, without testing the defense, can miss the conditions that make the event possible.

The four distortions below are common because each one is easy to report. They are also dangerous because they let an organization claim prevention before it has demonstrated control.

Distortion 1: Counting activity as evidence of control

The first distortion appears when completed activity is treated as proof that the risk has been reduced. A site may report hundreds of safety conversations, inspections, or toolbox meetings, even though the recurring hazard that prompted them remains unchanged.

Activity matters, but it answers a narrow question. It tells leaders whether a process occurred. It does not tell them whether the process found a meaningful exposure, whether someone owned the response, or whether the control became more reliable after the response.

Consider a weekly observation target. Supervisors who are measured on volume may select easy tasks, avoid difficult conversations, or record the same low-risk behaviors because those entries are faster to close. The dashboard improves while the selection process becomes less honest.

The correction is not to eliminate observations. It is to separate activity from effect. Report the number completed, then show the percentage that identified a material condition, received a decision within the agreed time, and passed a later verification. The final measure is harder to produce, which is precisely why it says more.

Distortion 2: Treating closure as improvement

A corrective action can be closed in a system without changing the work. The owner may upload a photograph, rewrite a procedure, or deliver a short briefing, while the physical arrangement, staffing pressure, or maintenance condition that created the exposure remains in place.

Closure is an administrative state. Improvement is a field condition. Confusing the two creates a dashboard that rewards speed of disposal, even when the action has only moved the problem from an open list into a completed archive.

This distortion is especially visible when every action receives the same deadline and the same evidence rule. A damaged guard, an unclear handover, and a missing sign may all be closed with a document, even though the risk owner needs a different proof for each one.

Leaders should ask what changed at the point of exposure and how the organization knows. For a physical control, the proof may be a field verification after production restarts. For a management control, the proof may be a decision record showing that the owner changed the plan, staffing, or authorization condition. The indicator should reward verified change, not an attractive closure rate.

Distortion 3: Averaging away the dangerous work

Averages make dashboards easy to read, but they can hide the task, shift, contractor group, or location where risk is concentrated. A site may show a healthy monthly average for pre-job reviews while one night crew repeatedly starts work with incomplete isolation information.

The average is not false. It is incomplete. When a measure combines routine and high-consequence work, strong performance in low-risk tasks can dilute the signal from a smaller set of exposures that deserve executive attention.

Managers also lose the story of change when they collapse a result too early. A monthly score may look stable even though performance deteriorated after a supervisor change, a contractor mobilization, a shutdown, or a production target revision. Those transitions are not statistical noise. They are context that explains why a control is holding or failing.

Keep the aggregate view for orientation, but segment the indicator by the conditions that can change the decision. Review high-energy tasks, abnormal operations, shift transitions, contractor interfaces, and repeat findings separately when those dimensions are relevant. A useful dashboard makes concentration visible before the average becomes an excuse to wait.

Distortion 4: Measuring belief instead of operating behavior

Survey scores, confidence ratings, and self-reported commitment can reveal perception, but perception is not the same as control. A workforce may agree that safety matters while still lacking time, equipment, authority, or supervisor support to perform the safer option.

The distortion begins when leaders use a favorable belief score as evidence that the system is healthy. The result can be a campaign to improve attitudes when the real barrier sits in planning, maintenance, workload, or conflicting production instructions.

Edgar Schein’s work on organizational culture is useful here because culture is expressed through repeated decisions and shared assumptions. The strongest test is not what people say during a survey. It is what the operation does when a deadline, an abnormal condition, or a technical concern interrupts the plan.

Pair perception measures with observable decisions. Check whether workers can stop and restart work without penalty, whether supervisors escalate a control gap, whether technical dissent changes the plan, and whether recurring concerns receive a visible response. The belief measure then has a job. It helps explain behavior rather than pretending to replace it.

What to do now when the dashboard looks too healthy

Start with one indicator that leaders already trust. Trace it from collection to decision and ask where the chain becomes uncertain. Does the number identify a real exposure? Does it distinguish routine activity from high-value work? Does someone own the response? Does the response have a threshold? Can the field prove that the condition changed?

Do not add three new metrics to compensate for one weak metric. That response creates more reporting work and can make the original distortion harder to detect. Instead, add one quality test that exposes the missing link, such as repeat-condition rate, verification after closure, response time for critical findings, or the share of actions that changed a physical or operational barrier.

Andreza Araujo’s experience across more than 25 years of executive EHS work supports a practical standard. A safety measure earns its place when it helps a leader decide where to intervene, who must act, and what evidence will show that the intervention worked. A number that cannot answer those questions may still describe effort, but it should not be presented as prevention.

The purpose of a leading indicator is not to make the dashboard look forward-looking. It is to make weak control visible while there is still time to change the work. When leaders test for distortion before celebrating the score, the dashboard becomes less comfortable and more useful.