Amazon is legendary for moving fast. What gets talked about less is what happens to the people inside that speed when the infrastructure hasn’t caught up.
I experienced this firsthand as part of a data analysis team at Alexa Shopping. Our standard was clear: maintain 95% personal accuracy so the team could hit 90% overall. On established projects, that was achievable. We knew the data, we knew the process, and we had SOPs built from months of real-world application. The standard made sense because the conditions supported it.
Then a new project would come in.
Sometimes we had full context: here’s the data, here’s the question we’re asking, here’s what good looks like. But often, we had half the picture. We had the question without the data context or the data without a clear question. Sometimes it was an SOP that covered the straightforward cases and went silent on everything else. And in the gaps, we were expected to use our judgment while still being held to the same 95% accuracy standard we met on projects we could do in our sleep.
This is where change management and psychological safety intersect in a way that rarely gets named clearly.
Launching a change before it’s fully defined is not the same as launching a change that’s ready.
What we were experiencing had a name, even if no one used it at the time: we were being measured against a mature process standard while operating inside an embryonic one. The SOP was incomplete not because someone was negligent, but because the project was new. The context was missing not because it didn’t matter, but because it hadn’t been fully worked out yet. We were, in effect, building the plane while flying it and being graded on our landing.
In change management terms, this is a failure at the Knowledge and Ability stages of the ADKAR model before the change has even had a chance to stabilize. You cannot build ability on incomplete knowledge. And you cannot hold people to a performance standard that assumes capability they haven’t yet had the conditions to develop.
We pushed back. As a team, we asked questions amongst ourselves first, then escalated to leadership when the concern was shared. We asked for context. We asked for assurance that accuracy expectations would be adjusted during the learning phase. To leadership’s credit, those assurances were generally given and, in my experience, honored. My accuracy dropped to 88% on some of those projects. I was not penalized.
But here is the critical point: the psychological safety issue existed before we knew that outcome. We operated under genuine uncertainty about whether we would be protected. The burden of proof was on us to advocate for reasonable conditions, not on the system to provide them by default. That is not psychological safety. That is compliance under pressure with a favorable result.
Psychological safety isn’t about whether things turn out fine in the end. It’s about whether people feel safe enough to act, speak up, and absorb uncertainty before the outcome is known. When people are afraid to flag a problem because they don’t know if it will be held against them, the organization loses the most valuable change intelligence it has: early warning signals from the people closest to the work.
The teams and organizations that manage change well understand that readiness is a precondition for performance, not a byproduct of it. Before you set a standard, ask whether the conditions exist to meet it. Before you launch a process, ask whether the people executing it have what they need. This includes not just the SOP, but the context, the clarity, and the explicit protection to learn without penalty during the period when the process itself is still being defined.
Speed is an asset. But speed without readiness doesn’t accelerate success. It accelerates failure. Worse than that, it does so quietly, because people learn very quickly not to say so.
So I’ll ask you this: when your organization launches something new, how do you distinguish between a performance problem and a readiness problem? And are the people closest to the work safe enough to tell you which one it actually is?