He Wasn’t Ignoring It.
He Couldn’t Feel It.

The gap shows up in the data. 76% of executives say their employees are enthusiastic about AI adoption. 31% of employees agree. 80% of executives say their teams are well-informed about AI strategy. 29% of employees agree. Same organizations. Same rollouts. Two completely different versions of how it’s going (Lovich & Meier, HBR, 2025 / BCG AI at Work, 2025, 10,600 employees, 11 countries).

The pattern has a name in the research. Two names, actually. The mum effect is the upward filter. People who escalate bad news up the chain expect to pay for it, so the bad news stops climbing (Smith & Keil, 2003; Park, Im & Keil, 2008). The deaf effect is the downward complement. When the bad news does reach the top, the people up there discount it (Cuellar, Keil & Johnson, 2006). MIT Sloan compiled the operational version from 14 empirical studies on IT project status reporting. Same finding. The signal gets edited on the way up and discounted on arrival (Keil et al., MIT Sloan, 2014).

The pattern goes back further. The original 1973 paper on upward communication found the same thing in plainer terms. People with the most to gain from being seen well filter the most. People with the least to lose tell the most truth (Athanassiades, 1973). The hierarchy produces the filter. The filter operates regardless of who sits inside it.

Culture shapes how steep the filter gets. Late-1960s IBM survey work named the variable, power distance, the degree to which less-powerful members accept that power gets distributed unevenly (Hofstede, 1980). Modern numbers from 17,300 middle managers across 62 societies confirmed the pattern (GLOBE Project, 2004). The steeper the hierarchy, the more layers the signal has to climb. The more layers, the more editing.

The view from the executive seat is documented too. As leaders rise, peers stop challenging them, direct reports filter feedback, boards focus on outcomes instead of behavior. The insulation builds quietly (Ashkenas, HBR, 2017). CEO peer groups have a phrase for it, the CEO bubble. The room around the CEO can’t be the honest mirror it looks like, because everyone in it has their own interests to protect (Vistage, 2017, practitioner). A CEIBS strategy professor put the mechanism more directly. Each level processes information “to align with superiors’ preferences more safely rather than to convey facts more accurately,” which he called “survival wisdom” (Tsai, CEIBS, 2025). Not moral failure. A rational response to incentive.

The brain can be tricked about where the body is. Stroke someone’s hidden hand and a rubber hand in front of them at the same time. Within a couple of minutes they start to feel the brush on the rubber hand instead of their real one. They call it proprioceptive drift (Botvinick & Cohen, Nature, 1998). Same effect at the leadership level, scaled to an org chart. The 45-point AI enthusiasm gap is what proprioceptive drift looks like in a boardroom.

The pattern is about the people. The middle manager edits the message before passing it up. The director reframes “the team can’t deliver this” as “on schedule.” The frontline worker marks the problem closed because flagging it open reopens a conversation that costs political capital. None of them is wrong. Each is reading the environment correctly. The aggregate is an executive who can’t feel the friction. The architecture is the only variable that can change without firing every honest person in the chain.

When the perception gap shows up, the standard response is to add a listening mechanism. Skip-level meetings. CEO town halls. Anonymous feedback systems. Executive shadow programs. Leadership offsites with frontline workers. Each of these moves the executive closer to the floor for a measured period. None of them changes the architecture that filters the signal on the days the executive isn’t visiting.

Skip-levels surface what the worker is willing to say to someone two levels up. That’s a different conversation than what the worker would say to a peer. The power gradient is still in the room. The filter is still operating.

Town halls scale the visibility problem. The worker who speaks at a town hall is the one already comfortable speaking in front of the most senior person in the building. The silent majority stays silent. The dashboard reads “engaged session” because the people willing to engage engaged.

Anonymous feedback systems trade attribution for usefulness. The feedback that arrives is unattributed by definition, which means it can’t be followed up, contextualized or acted on without re-identifying the source. The architecture either over-investigates and breaks the anonymity promise or under-investigates and the feedback dies in a queue.

Executive shadow programs put a senior leader on the floor for a day. The day is observed. The week before and the week after aren’t. The team performs the version of the work that’s safe to be seen. The executive collects the impression of a system functioning well.

None of these mechanisms touch the structural incentive. The reason the signal gets filtered isn’t a shortage of listening channels. It’s that the people upstream have correctly read what gets rewarded and what gets penalized in their environment. Adding more channels for the executive to listen doesn’t change the math for the worker deciding what to say.

The thing that fixes the signal isn’t another communications surface. It’s an architecture that makes the truth survivable for the person who has to send it.

The Dignity-Required System names three concepts for this friction front. Each one shifts the diagnostic from “the leader needs to listen more” to “the architecture is producing sensory deprivation at the top.” Each one expands inline below.

The 45-point AI enthusiasm gap isn’t a communications problem. It’s a structural one. The fix isn’t another channel for the executive to listen on. It’s an architecture that changes what the worker can afford to say.

That looks like specific structural changes. A reporting line that lets a frontline worker flag a problem without their manager being the routing hop. Promotion criteria that explicitly reward managers for surfacing bad news up the chain instead of catching it on the way down. A documented response when a project status reports red that doesn’t punish the team that reported it. The pattern isn’t about training executives to ask better questions. It’s about changing what gets rewarded between the floor and the boardroom.

Culture matters here. The 17,300-respondent GLOBE study found power distance varies sharply across societies (GLOBE, 2004). A US-headquartered organization rolling out in a high-power-distance market is dealing with a steeper local hierarchy than the headquarters culture was built to detect. If the rollout assumes the same signal velocity at every site, the gap at the perimeter is wider than the gap at the center.

Governance matters too. The pattern operates across three layers. The strategic governor sets direction. The contextual governor (which is what middle management actually is) translates direction into operational meaning. The moral-execution governor decides what happens when the direction collides with reality on the floor. The IT governance research documents the same three-layer structure (Parent & Reich, 2009). The management research documents middle managers occupying exactly the contextual layer, translating strategic intent into operational action and synthesizing operational signal into strategic input (Wooldridge, Schmid & Floyd, 2008). Each layer is a filter. Each filter is a person reading the environment correctly. The architecture has to fund the translation work at each layer, or the signal gets compressed at every hop.

Diversity matters here too, for a reason that isn’t the usual one. Companies in the top quartile for ethnic and cultural diversity on executive teams were 36% more likely to outperform on profitability than those in the fourth quartile (McKinsey Diversity Wins, 2020). Heterogeneous executive teams filter information differently. A board where every member shares the same background, the same career path and the same incentives produces a uniform filter. Heterogeneity introduces variance, and variance catches signal a uniform frame would have dismissed.

Six of the eight documented causes of UK government IT project failure are governance issues. Board-level decisions made without operational reality represented in the room (Parent & Reich, 2009). The fix isn’t another board update. The fix is restructuring the governance forum so the operational layer has a seat at the table, with the authority to slow a decision when the strategic and operational views disagree.

The thing worth measuring at the executive layer isn’t listening-program participation. It’s trust signals. Voluntary candid disclosure from middle managers who used to filter. Board-level disagreement that surfaces operational reality before a decision lands. The frequency with which executive sponsors report being surprised by what frontline workers tell them. The third one is the diagnostic.