A recommendation lands on someone’s desk, generated by a dashboard, a colleague, or increasingly an AI model, and something about it feels off. The data looks thin, the context feels wrong, or a piece of institutional knowledge that never made it into the model suggests a different path entirely. What happens in that moment, whether the person raises the concern or quietly goes along with it, says more about an organization’s culture than almost any stated value on a wall poster ever could.
As recommendations of every kind multiply across dashboards, reports, and AI tools, the ability to override a bad one is becoming one of the more consequential skills an organization can build into its culture, and one of the most fragile, because it depends entirely on whether people believe speaking up is actually safe.
Where Good Recommendations Go Wrong
Research from Michael Luca and Amy Edmondson, published in the Harvard Business Review, found that leaders faced with a data-backed recommendation tend to fall into one of two unhelpful extremes: treating the evidence as gospel or dismissing it outright, rather than doing the harder work of interrogating whether the finding actually applies to the situation in front of them. Both extremes remove judgment from the equation entirely, and both make it far less likely that anyone in the room raises the obvious question of whether this particular recommendation fits this particular context.
That interrogation only happens if someone is willing to ask it out loud, in front of colleagues, possibly contradicting a tool or a person with more seniority. Whether anyone does that depends less on how smart or careful the individual is and more on whether the room around them has made that kind of pushback feel survivable.
What Psychological Safety Actually Means
Amy Edmondson and Michaela Kerrissey’s more recent research, published in the Harvard Business Review, pushes back on several popular misreadings of psychological safety that quietly undermine it in practice. Psychological safety does not mean being nice, and it is not the same as everyone getting their way. It does not require lowering the bar on performance, and it is not something a policy document can install on its own. It is, instead, a shared belief that a team is safe for interpersonal risk-taking, which includes the specific risk of telling a senior colleague, or an algorithm’s confident output, that something looks wrong.
That distinction matters because organizations often believe they have built psychological safety simply by being generally pleasant places to work, while the actual test only shows up in the specific, uncomfortable moment when someone needs to contradict a recommendation that everyone else in the room seems ready to accept.
FutureThink’s Approach: Naming What Gets Fired vs. What Gets Promoted
FutureThink’s Lead with Impact programming is built specifically around fostering trust, collaboration, and psychological safety rather than treating those as side effects of good management. One exercise drawn from that work, described in a FutureThink piece on the ideas that get people fired or promoted, asks participants to name the boldest possible idea they are holding back, the one they suspect might raise eyebrows, before judgment is allowed into the room. The exercise works precisely because it surfaces the gap between what people privately believe and what they feel safe saying, and that same gap is exactly what determines whether someone will flag a bad recommendation or quietly let it pass.
Applied to the override question directly, the same principle holds. A team that has practiced naming its boldest, most uncomfortable ideas in a low-stakes setting is far better positioned to name a genuine concern about a high-stakes recommendation when it actually matters, because the muscle for saying the uncomfortable thing has already been exercised.
Why This Cannot Be the First Thing Leaders Cut
Recent research covered in the Harvard Business Review found that programs supporting psychological safety are often among the first things cut when budgets tighten, precisely the moment when the ability to catch a bad decision matters most. Constrained resources tend to increase reliance on automated recommendations and shortcuts, which raises rather than lowers the value of a team that still feels able to question them.
Building that culture is less about a single policy and more about repetition: normalizing dissent in low-stakes settings, rewarding the person who raised the uncomfortable question even when the recommendation turned out to be right, and treating an override, when it is well reasoned, as evidence the system is working rather than as friction to be managed away. Organizations that get this right will not necessarily generate fewer bad recommendations. They will simply be far more likely to catch them before they become expensive mistakes.