How a consumer-goods company cut accidents 50% in 6 months with 4 supervisor gates

During Andreza Araujo's EHS leadership tenure at PepsiCo South America, the accident ratio fell 50% in six months. The result matters because it shows a consumer-goods operation does not improve when safety becomes louder. It improves when supervisors change what they allow to start, what they stop, and what they escalate.

Across more than 250 cultural transformation projects, Andreza Araujo has seen the same pattern repeat. When decision rights stay vague, line pressure wins. When supervisors are given a small number of hard gates, the organization can move from talk to control.

Initial scenario

The operation had the usual consumer-goods mix of line speed, maintenance windows, contractor interfaces, and repeated tasks that had become familiar enough to feel safe. Familiarity is a trap, because James Reason's work on latent failures shows how repeated success can hide conditions that only become visible after a serious event.

Supervisors could describe the rules, yet they still treated some weak signals as background noise when the shift was busy. A missing control looked recoverable. A rushed pre-task review looked normal. A deviation that happened yesterday felt less urgent today. That is how drift starts, and that is why training alone rarely moves the result.

This is the same problem leaders face in many plants. They add another message, another toolbox talk, another reminder, but the work system still rewards speed more than control. If you want the companion angle, read Production Pressure: 7 Leadership Decisions That Stop Shortcuts, because the pressure pattern is usually the same even when the site and sector change.

Decision

The decision was to stop asking supervisors to be generic safety champions and instead give them four gates that defined when work could start, continue, escalate, and close. That shift sounds simple, although it changes the meaning of supervision because the supervisor stops being a messenger and becomes a control point.

The first gate was a start gate. If the critical controls were not named and visible, the job did not start. The second gate was a continuation gate. If a control disappeared, the work stopped long enough to restore it or escalate it. The third gate was an escalation gate. If the same deviation kept returning, the issue moved upward instead of being normalized by coaching alone. The fourth gate was a close-out gate. The team had to leave the task with a clearer control state than the one it found at the beginning.

That logic aligns with How to Build a Safety Decision Rights Matrix in 30 Days, because both ideas ask the same question: who can allow the job to move, and what evidence must exist before that permission is real?

Execution

Execution depended on rhythm. Supervisors used the gates in pre-task talks, during field checks, and at shift end, while managers reviewed whether escalations were handled quickly enough to protect the next job, not just the current one. The point was not to create a larger reporting machine. The point was to make the operating routine respond faster than the drift.

Andreza Araujo's broader body of work matters here because more than 250 cultural transformation projects taught the same lesson. Culture changes when repeated decisions change. A site can post values, run campaigns, and improve attendance at meetings while the old operating logic remains intact. The gate model works only when leaders reward the people who interrupt weak control, not the people who keep production moving through it.

The field discipline also reflects a James Reason-style reading of latent failure. The issue is not just the last person who touched the task. It is the set of upstream choices that made the weak control look acceptable in the first place. When leaders expose those choices early, the operation spends less time explaining incidents after the fact.

Measured result

In the PepsiCo South America case documented by Andreza Araujo, the accident ratio fell 50% in six months. That is the only number in the story that should be treated as the headline result. Everything else is the mechanism behind it.

The result matters because it was not driven by a slogan, a poster refresh, or an extra round of awareness messages. It came from changing the leader behavior that sits closest to the work. Once supervisors were forced to decide with clearer gates, the organization stopped paying for ambiguity as often.

The table is useful because it shows what changed without pretending that the number happened by itself. If a site cannot explain the before and after logic, it is usually copying the result rather than the discipline.

Generalizable lessons

Lesson 1. Supervisors move results when they change permission, not when they repeat slogans. The case worked because the line understood what could start, what had to stop, and what had to escalate.

Lesson 2. Culture lives in repeated decisions. Andreza Araujo's book Safety Culture: From Theory to Practice treats that as the core test, and the PepsiCo case proves the point in operating terms.

Lesson 3. Training is useful, but it does not change a decision rule by itself. If a busy shift still rewards speed over control, the training message gets overwritten the next day.

Lesson 4. Leaders need to inspect weak signals before the injury count does the talking. James Reason's latent-failure lens is still relevant because the real cause often sits upstream of the final error.

Lesson 5. A result like this should make executives more precise, not more celebratory. If the operation cannot explain the mechanism, the improvement may fade as soon as attention moves elsewhere.

What to apply in your operation

Start with one line, one contractor interface, or one maintenance window that repeats often enough to shape behavior. Then write four gates in plain language. The language should be short enough for a supervisor to use at the job start, and specific enough that nobody has to guess when the rule applies.

Next, measure three things for 30 days. Track how often the start gate blocks weak work, how often the escalation gate is used, and how often the close-out gate finds evidence that the control state improved. If those numbers stay flat while the site says it is serious about safety, the decision layer is still too weak.

Then compare the result with one internal benchmark outside the current job. The objective is not to create a perfect case study. The objective is to make the next leadership review answer a sharper question: did the way we decide change, or did we only talk about safety more often?

If the answer is still unclear, the operation is still paying for ambiguity, and the next incident will usually explain it more loudly than the dashboard does.

For a deeper companion read, use Production Pressure: 7 Leadership Decisions That Stop Shortcuts, then bring the same discipline to your next shift review. Headline Podcast exists for this kind of work, where leadership decisions and field control have to line up before risk becomes the story.