Safety Indicator Explained: 4 Tests That Separate Activity From Control

A dashboard can be full of green cells while a serious exposure remains poorly controlled. The problem is not that the organization measures too little. It is that the measures describe activity without showing whether the activity changed the conditions in which people work.

A supervisor may report that every planned observation was completed, yet the same lifting aid may still be unavailable, the exclusion zone may still be crossed, and the escalation route may still be unclear. The report looks active because the process produced records. It does not necessarily show that the risk became smaller.

A safety indicator is a defined measure that makes a relevant risk condition, control state, or decision-quality problem visible. It earns its place on a dashboard when a credible change in the measure leads a specific audience to make a different decision.

Why activity counts are mistaken for safety evidence

Activity counts are attractive because they are easy to collect. A completed training session has an attendance list, an inspection has a form, and a safety meeting has minutes. Those records matter for governance, but they answer a narrow question. They show that an administrative event occurred.

The harder question is whether the event improved the work. When a training session is counted, the metric may say nothing about whether workers can recognize the changed hazard, whether the procedure fits the task, or whether supervision corrects a conflict between production and control. The activity has been completed, but the control claim remains untested.

James Reason’s analysis of organizational accidents helps explain the gap. Visible events sit above less visible conditions, including weak design, unclear responsibility, normalization of deviation, and delayed escalation. A dashboard that only counts visible activity can therefore report movement while leaving the conditions underneath untouched.

Test 1: Does the indicator stay close to the exposure?

The first test asks whether the measure is connected to the hazard that can produce harm. Distance is not automatically a defect, because some system measures are necessary, but a measure becomes weak when its relationship to exposure cannot be explained without several assumptions.

“Number of toolbox talks delivered” is distant from the exposure created by a suspended load. “Percentage of critical lifts with a verified exclusion zone before movement” is closer. The second measure still needs a clear verification method, yet it asks about a condition that exists at the point where the hazard is active.

Use the exposure test by naming the event the organization is trying to prevent, the control that should interrupt it, and the observable condition that proves the control is in place. If the indicator cannot complete that chain, it may be an administrative activity rather than a risk measure.

Test 2: What evidence supports the number?

A number can be precise and still be poorly grounded. The evidence test checks how the value was produced, who observed the condition, when the observation occurred, and whether another competent reviewer could reach a similar conclusion.

For example, a record that says “100 percent compliant” may represent a careful field verification, a supervisor’s unchecked declaration, or a form completed after the job. Those are different evidence states even though they create the same percentage. The metric needs a method that makes the distinction visible.

ISO 45001:2018 places emphasis on monitoring, measurement, analysis, and evaluation, but the standard does not turn every collected number into proof of effectiveness. The organization still has to define what is measured, how it is checked, and how the result is interpreted in relation to operational control.

Test 3: Which decision changes when the result moves?

A safety indicator should have an owner, a review cadence, and a decision rule. Without those three elements, the measure becomes a status update that consumes attention without directing action.

Suppose a dashboard reports that verification of machine guarding fell from 96 percent to 82 percent. The leadership team should already know what happens next. The responsible manager may need to pause selected work, assign a control-recovery plan, increase field verification, or review whether the metric reflects a change in the work rather than a change in reporting.

The decision does not need to be automatic, and it should not be reduced to a rigid threshold when context matters. It does need to be explicit enough that a green result cannot close the conversation before someone has tested what the number means.

Test 4: Can the result improve without safer work?

The fourth test looks for cosmetic improvement. Any measure that becomes a target can attract behaviors that raise the score without strengthening the control. The question is not whether people are dishonest. It is whether the system makes the easier path look successful.

A team may increase inspection completion by shortening inspections, avoid reporting difficult observations, reclassify a control gap as an opportunity, or schedule reviews where the exposure is easiest to find. Each action can improve the dashboard while weakening the connection between the number and the work.

To test resistance to cosmetic improvement, ask how the measure could be made better in a week without changing equipment, task design, supervision, or worker exposure. If the answer is obvious, pair the activity measure with a field-proof measure and review the two together.

How should leaders pair activity with control evidence?

Activity measures should not be discarded. They become more useful when they are placed beside evidence that tests the claim behind the activity. Training completion can sit beside a task demonstration. Inspection completion can sit beside the percentage of critical defects corrected before exposure. A meeting count can sit beside the number of unresolved high-consequence decisions that reached an owner.

The pairing principle is simple, although its application requires discipline. Every activity measure should have a companion question about effectiveness, and every effectiveness measure should have a route to action. This prevents the dashboard from becoming a museum of completed tasks.

What Andreza Araujo’s safety-culture lens adds

Andreza Araujo’s book Safety Culture: From Theory to Practice treats culture as something people experience through decisions, routines, and consequences. That perspective matters for measurement because a metric is also a message. It tells the workforce what leaders notice, reward, question, and tolerate.

If leaders praise a full calendar of observations while ignoring repeated control failures, the organization learns that visible activity is safer for the reporter than uncomfortable evidence. If leaders ask what the measure reveals about work design and then fund the correction, measurement becomes part of the control system.

The strongest dashboard is therefore not the one with the most indicators. It is the one whose measures make important conditions discussable and whose review process gives someone the authority and resources to respond.

A practical review for the next dashboard meeting

Before the next monthly review, choose one indicator that the team considers successful and run the four tests against it. Write down the exposure it represents, the evidence behind the value, the decision that changes when it moves, and the ways the result could improve without safer work.

• Keep the measure when all four answers are specific and reviewable.
• Redesign the measure when it depends on activity alone or when evidence quality is unclear.
• Pair it with a field-proof measure when the activity remains useful for planning or accountability.
• Assign a named decision owner so the dashboard leads somewhere beyond discussion.

That review can be completed with one operational team and one real exposure. It does not require a new software platform. It requires leaders to stop treating a number as self-explanatory.

Why a green dashboard can still be a warning

A green result is reassuring only when the measure is close to the exposure, supported by credible evidence, connected to a decision, and difficult to improve cosmetically. Without those conditions, green may describe reporting discipline rather than control strength.

Safety measurement becomes more useful when it makes the uncomfortable question easier to ask. Are people completing the process, or is the process changing the conditions that create risk? The answer should be visible in the metric design before the next incident makes the distinction unavoidable.

Headline Podcast publishes practical conversations and analysis about safety leadership, safety culture, and risk decisions. Explore more safety articles on Headline Podcast.