Twenty-five years ago, a researcher named Amy Edmondson published a study that became the most-cited construct in modern organizational science. The finding, in plain language: teams that openly discuss problems perform better than teams that don't.
That sentence sounds obvious. It is not.
The reason Edmondson's work mattered is that, before the study, most leaders assumed the better teams were the quieter ones — fewer reported issues meant fewer problems, fewer problems meant a healthier operation. Edmondson's work, conducted in hospital settings, found something the prevailing assumption couldn't explain: the best teams reported more errors than the worst teams. Not because they made more errors. Because they were willing to name them.
The construct she named for this — psychological safety — has been validated across hundreds of subsequent studies, in healthcare, manufacturing, technology, professional services, and education. The research is well past the contested stage. A 2017 meta-analysis covering 136 independent samples confirmed the original finding: teams with higher psychological safety report more problems, learn from them faster, and outperform teams that don't.
That is the research. Here is what it looks like in the field.
What healthy looks like in measurable terms
The instrument most often used to measure psychological safety is a seven-item questionnaire on a seven-point scale. Workplace teams across the published studies tend to score around 5.0 on the seven-point scale — that's the empirical mean, with a standard deviation of about 0.9. Healthy teams cluster between 4.6 and 5.4.
The number that matters more for an operations leader is the cutoff. Across most published studies, when a team scores below 4.0, voice behavior measurably declines and problem reporting drops. The team stops bringing things up — not because the things stopped happening, but because the system stopped expecting that bringing them up would matter.
A few vertical notes from the research, because the baseline is not the same everywhere:
- Healthcare teams tend to run slightly lower — 4.6 to 4.9 — due to the hierarchy steepness in clinical environments. This is well-documented and built into the validated healthcare-specific versions of the instrument.
- Manufacturing and frontline service teams track close to the general workplace mean.
- Professional services tend to vary widely depending on partnership structure and seniority dynamics.
The published numbers are yours to compare against. You don't need to run the survey to use them. You can ask, honestly: where would I estimate my team would land? If your honest answer is "well below 5," the research has a name for what's happening — and it is not a people problem.
What it looks like in a real moment
A regional manager runs a Friday standup at a multi-site service operation. Twelve sites report up to her. Eleven of the twelve report green for the week. The twelfth — call it Site 7 — also reports green, but the manager who runs Site 7 lingers afterward and says, in a low voice: "We're fine for now, but I'm worried about the staffing model. It's not sustainable."
The regional manager nods. She has heard a version of this sentence from Site 7's manager twice before — once in March, once in May. Each time it was raised quietly, after the meeting. Each time it was followed by a few weeks of stability. Each time, no formal report was filed. No agenda item appeared at the next standup.
Then, in late July, two associates at Site 7 quit in the same week. The exit interviews cite the staffing model. Within thirty days, a third associate quits. Within sixty days, Site 7 is in active recovery — escalations up, customer complaints up, the regional manager spending two days a week on-site to stabilize.
What happened? Nothing dramatic. No one hid anything. The Site 7 manager raised the issue — quietly, three times, in the way the system rewarded raising it. The regional manager registered the concern — informally, as a thing to keep watching. The standup, the dashboard, and the weekly green report all reflected reality as the system measured it. The system worked exactly as designed. The cost of what the system was not designed to catch arrived in the turnover numbers ninety days later.
If you've ever sat in a meeting where the room agreed with something you suspected the room didn't actually believe — where the green report was technically accurate but the room knew, collectively and quietly, that the green was thinner than the report suggested — you already know this pattern.
The Site 7 manager was not afraid to speak up. He spoke up three times. He just spoke up in the channel the system had quietly taught him produced no result. The dashboard caught the turnover. It did not catch the three quiet conversations that predicted it.
That gap is what Edmondson's research is describing. And it is what most operational systems are not built to see.
Why this costs money
The financial story of psychological safety is rarely told in operator language. It usually shows up dressed in HR vocabulary — engagement, retention, culture — that sits a comfortable distance from the P&L. Translated back into operator language, the math is direct. When the reporting threshold rises by even a small amount — when even 1 in 5 of the issues that operating conditions should produce stops getting named — three things happen on a predictable lag.
The same problem solves twice. A real condition gets fixed locally without the structural cause surfacing. The condition reappears in 60 to 120 days, in the same site or a similar one. The cost of solving it the second time is paid by the same team, plus the customers affected during the second occurrence, plus the operational leader's attention spent triaging it again.
Tenured staff start to leave. Not because they are angry. Because they have observed, over time, that the reporting channel they used to trust no longer produces a result. Tenured staff exits are the most expensive turnover in any service organization — they take institutional knowledge with them, and the replacement cycle takes 90 to 180 days to recover even partial competence. Industry data puts the fully-loaded replacement cost of a frontline service employee somewhere between 30 percent and 150 percent of annual compensation, depending on the role.
The dashboard goes greener. Counterintuitively, when a reporting system erodes, the surface metrics improve. Fewer issues raised means fewer issues in the queue. The number that should worry an operations leader is not the rising-complaint number — it is the stable or declining complaint number that is no longer tracking against an operational reality that everyone on the floor knows is getting worse.
The third point is the hardest one for most operations leaders to internalize, because the dashboard was built to reward what it just showed them.
What this is not
This is not a recommendation to run an annual engagement survey. The research community itself has been clear for years that annual surveys, while useful as benchmarks, capture psychological safety at a tempo that misses most of what an operations leader needs to act on. The construct is stable enough across months that an annual survey is not wrong; it is just slow.
This is not a recommendation to fix individuals. Edmondson's research is explicit on this point: psychological safety is a team-level and structural construct. It is produced by the conditions a leader creates, the consistency with which raised issues are addressed, and the visible track record of what happens when someone speaks up. It is not a personality trait. The Site 7 manager did not need coaching; the channel needed repairing.
And this is not a recommendation to do nothing. The pattern is detectable. There are signals — behavioral, observable, structural — that show up before the lagging indicators do. Reading those signals is what the rest of this series is about.
What to do this week
One question, for an operations leader reading this:
In the last quarter, what is the most important thing someone on my team told me — informally, quietly, after the meeting — that did not make it onto the agenda the following week?
If you can name it, the system is healthier than the dashboard knows. If you can't, that absence is itself the signal.
Next week, the series moves into what happens after training — and why the most carefully designed training programs lose 60 percent of their effect inside ninety days for reasons that have very little to do with the training itself.
Sigma G Signal publishes weekly. If you're a COO, VP of Operations, or operations leader in a service organization, this series is written with you in mind.
— Greg
References
1. Edmondson, A. C. (1999). Psychological Safety and Learning Behavior in Work Teams. Administrative Science Quarterly, 44(2), 350–383.
2. Frazier, M. L., Fainshmidt, S., Klinger, R. L., Pezeshkan, A., & Vracheva, V. (2017). Psychological Safety: A Meta-Analytic Review and Extension. Personnel Psychology, 70(1), 113–165. (136 independent samples; 22,000+ individuals; ~5,000 groups.)