Behavioral Observation: How 250+ Projects Reframed Unsafe Acts

Behavioral observation is a structured field conversation that tests whether work conditions, leadership follow-up and worker choices are protecting people in real time. In mature programs, it is not a form count or a hunt for unsafe acts. It is evidence about how risk is being managed where the work actually happens.

In more than 250 cultural transformation projects supported by Andreza Araujo, one pattern keeps returning: behavioral observation fails when the organization treats behavior as the whole story. The worker is visible, the missing barrier is less visible, and the pressure behind the shortcut is often invisible unless the observer knows how to ask.

This case-style article uses that project base to show how companies can reframe observation from a compliance ritual into a practical control conversation. The category is safe behavior, but the thesis is wider. Unsafe behavior may be the first thing the supervisor sees, although it is rarely the first thing the system created.

Initial scenario

Many observation programs begin with good intent and weak design. The EHS team asks supervisors to complete a target number of observations per month, leaders celebrate rising totals, and the dashboard starts to look active. On paper, the organization has more eyes in the field. In practice, it may only have more forms.

The failure appears when the observation card records the same items repeatedly: no handrail use, exposed hands, missing eye contact at intersections, tools left in the line of fire, or a worker bypassing a pre-task check. The form names the visible behavior, but it does not explain why that behavior made sense to the person at that moment.

Andreza Araujo's position in *Muito Alem do Zero* (Far Beyond Zero) is useful here because she treats behavior as a reflection of context and system, not only individual intent. People are not the weak link by default. Very often, they are holding together a system whose planning, staffing, equipment or supervision has already become fragile.

That is why an observation program can produce volume without learning. If a supervisor sees a shortcut and writes only "unsafe act," the company loses the chance to see production pressure, unclear standards, missing tools, fatigue, fear of delay, poor layout, or a rule that no longer matches the task.

Decision

The turning point in these projects was the decision to stop treating observation count as the main indicator. The count remained useful as a participation signal, but it stopped being the headline. The stronger question became whether each observation improved the quality of field control.

That decision changes the observer's job. The observer no longer arrives as an inspector with a checklist. The observer arrives as a leader of a short conversation whose purpose is to understand work, detect weak signals and restore control before the next event escalates.

As Andreza Araujo argues in *Cultura de Seguranca: Da Teoria a Pratica* (Safety Culture: From Theory to Practice), behavioral observation should operate as a structured conversation of active care, not as a punitive form. The difference is visible in the first question. "Why did you do that?" often sounds like accusation. "What made this the easiest way to do the task today?" opens the system.

This shift also protects the worker's voice. A worker who expects blame will hide adaptation. A worker who recognizes care may explain that the tool crib is too far away, the job plan missed a change, the deadline is unrealistic, or the supervisor praised the last person who finished early by skipping the same control.

Execution

The execution model used three practical changes. First, observers were calibrated around the quality of questions, not just the number of cards closed. Second, every observation had to identify at least one condition that shaped the behavior. Third, supervisors had to close the loop with the crew when a barrier, tool, instruction or routine changed because of what was observed.

This is where many companies struggle. They train observers to see body position, PPE and procedure compliance, but they do not train them to notice context. A better program asks whether the person had the right equipment, enough time, clear authority to stop, a realistic work method, and a supervisor who rewards the safe pace rather than the fastest finish.

The link with safety observer calibration is direct. Calibration means two observers can watch the same task and reach a similar conclusion about risk quality, not only about whether a rule was broken. Without that shared standard, observation data becomes personal opinion dressed as evidence.

The second link is with shortcut normalization. When a shortcut happens once, it may be a poor decision. When it happens every week and production still applauds the result, it has become a management signal. The observation program has to expose that signal before it hardens into culture.

Measured result

The measured result in this case is not a single invented reduction rate, and it should not be presented that way. The verified base is the 250+ transformation projects supported by Andreza Araujo and the recurring change observed inside them: stronger programs move from counting observed behaviors to classifying control quality, follow-up quality and learning quality.

That distinction matters for YMYL safety decisions. A site can double its observations and still leave critical risk untouched. Another site can run fewer observations, but use them to identify a recurring barrier gap in line-of-fire work, correct the work method, brief supervisors and verify the change in the field. The second site has learned more.

The most important metric becomes the percentage of observations that generated a useful action, a clarified rule, a removed obstacle, or a verified conversation with the crew. That is a harder number to inflate because it depends on visible management work after the card is closed.

Generalizable lesson one

The first lesson is that behavior should start the conversation, not end it. James Reason's Swiss cheese model helps leaders understand why the visible act is often the final opening in a longer chain of latent conditions. Andreza's work reaches the same practical conclusion without excusing the act itself.

If a worker bypasses a control, the supervisor still has to intervene. The trap is stopping at moral judgment. A mature intervention asks what made the bypass attractive, normal or necessary, because the same condition will keep producing the same choice in another person tomorrow.

Generalizable lesson two

The second lesson is that field dialogue needs a rhythm. A good observation is short enough to fit the workday and serious enough to change something. It respects the worker's time while refusing to reduce safety to a polite chat.

That rhythm connects with the pre-job change brief. When conditions shift, the supervisor needs a fast routine to re-open risk perception before the task continues. Observation should feed that brief, because the best observations often reveal where the next change conversation is needed.

Generalizable lesson three

The third lesson is that cognitive load is part of behavior. People miss cues when they are tired, overloaded, rushed, distracted or forced to hold too many exceptions in memory. Calling that carelessness may satisfy frustration, but it does not repair the exposure.

For supervisors, the better link is with cognitive load in safety. An observation program that ignores mental workload will over-diagnose attitude and under-diagnose work design. In field operations, that error can leave serious risk untouched.

What to apply in your operation

Start by auditing the last fifty observation cards. Separate them into three groups: cards that only name behavior, cards that identify a condition behind the behavior, and cards that led to verified follow-up. The distribution will tell you whether the program is learning or only recording.

Then change the supervisor script. Require one question about context, one question about the easiest safe way to do the job, and one agreement about what will happen next. The agreement matters because workers stop trusting observation when every conversation disappears into a database.

Finally, review the dashboard with operations, not only EHS. If production pressure, tool availability, staffing or layout appears repeatedly, the owner of the fix may not be the safety department. Safe behavior is built in the field, but many of its causes are designed in planning meetings.

FAQ

Is behavioral observation the same as BBS?

Behavioral observation can be part of a BBS program, but it should not be reduced to scoring unsafe acts. In Andreza Araujo's approach, the strongest observation is a conversation that connects behavior, context and control quality.

What is the main mistake in behavioral observation programs?

The main mistake is treating the number of completed observations as proof of prevention. Volume matters only when the observation produces learning, barrier restoration, better supervision or a clearer work method.

Should supervisors correct unsafe behavior immediately?

Yes, supervisors should intervene immediately when exposure is present. The intervention should stop the risk and then ask what made the behavior likely, because correction without learning leaves the same exposure available for the next person.

How should EHS measure observation quality?

EHS should track whether the card identified a condition behind the behavior, whether the right owner acted on it, whether the crew received feedback, and whether the field change was verified after the conversation.

Where should a company start?

Start with observer calibration and supervisor follow-up. If observers do not ask consistent questions and supervisors do not close the loop, the program becomes paperwork with a human face.