Return-to-office edicts and AI strategies hog the airtime, yet the quiet failure sits elsewhere: decisions still queue for permission. That queue is getting longer while managers’ energy drains. In Gallup’s latest global survey, only 20% of employees report being engaged, and the firm estimates the productivity cost at about 9% of global GDP, roughly 10 trillion dollars. Manager engagement fell from 27% to 22% in a single year and is down nine points since 2022, an uncomfortable signal for any model that concentrates decisions in managerial bottlenecks (Gallup, 2025). These are surveys, not causal proof, but they frame a hard design question. Where should decisions live when volatility is the norm?
What turbulence does to decision rights
The best field evidence we have on this question comes from firms that had to make hard calls fast. Aghion, Bloom, Lucking, Sadun and Van Reenen studied thousands of plants across ten OECD countries and the United States. They compared firms that had delegated more decision authority from headquarters to local plant managers before the Great Recession with those that had held decisions at the center. In the sectors hit hardest by the crisis, the decentralized firms outperformed their centralized peers (Aghion, Bloom, Lucking, Sadun and Van Reenen, 2021).
This was not a one-off anecdote. The authors triangulated across two large datasets, the World Management Survey and the US Census Bureau’s Management and Organizational Practices Survey, and used a difference-in-differences design. They tested alternative explanations, such as a greater fear of bankruptcy among centralized firms or shifting coordination costs, and the decentralization effect held. The mechanism was consistent with a simple idea. Under turbulence, local information becomes more valuable, so pushing authority to where that information resides pays off. The paper corroborated this “localist” view using independent measures of turbulence, including product churn and stock market volatility (Aghion, Bloom, Lucking, Sadun and Van Reenen, 2021). The authors are careful about causality, yet the pattern is robust and actionable. When the environment becomes less predictable, central clearance slows you down at exactly the wrong time.
The promise and the boundary of self-management
It is tempting to take this result and declare hierarchy obsolete. That would ignore what we know about where hierarchy still works. Lee and Edmondson define self-managing organizations as those that decentralize authority radically, formally and systematically, and that remove the manager–subordinate relationship as a structural feature. In their synthesis, traditional managerial hierarchy continues to perform well for reliable execution of known tasks in stable contexts. The same structure tends to become rigid and to mute cross-functional problem solving when work is complex and non-routine. It also reinforces status differences that can stunt people’s development. These are theoretically grounded claims, not experimental findings, and the authors are explicit about the empirical limits, including a shortage of large-sample causal studies of self-managing designs (Lee and Edmondson, 2017).
Read together with the crisis evidence, the implication is practical rather than doctrinal. You do not need to abolish managers to gain the benefits of distributed decision-making. You do need to be explicit about which decisions live where, and to shift rights toward the edge as uncertainty rises. In stable domains that prize repeatable quality, central standards and supervisory clearance can still be the right design. In messy domains that cut across functions, pushing decisions down and across is the bet that matches the task.
Safety is the operating system for distributed judgment
Authority on paper is not the same as authority in practice. People only use discretion if they believe they can speak up, surface errors and take interpersonal risks without being punished. Edmondson’s field study of 51 work teams inside a manufacturing company introduced the construct of team psychological safety as precisely that shared belief. In the data, psychological safety was associated with more learning behavior, and learning behavior mediated the relationship between safety and team performance. Team efficacy did not explain learning once safety was accounted for (Edmondson, 1999). The study is cross-sectional and set in a single firm, so we should be careful about causal claims. Even so, the mechanism it highlights is directly relevant to distributed decision-making.
When you distribute decisions, you distribute the need to admit uncertainty and correct course quickly. That is learning. Without psychological safety, local autonomy can devolve into silence, blame and private workarounds. With it, teams are more likely to spot anomalies, ask for help and integrate perspectives across functions. If the Aghion et al. result tells you where to place decisions under turbulence, Edmondson’s research tells you what social conditions let that design work.
The practical read for 2026
The short run looks choppy. That makes the design work urgent. Here is a pragmatic way to start, informed by the evidence but framed as guidance rather than law.
- Decide where uncertainty is high, then push rights there. Use the turbulence logic from Aghion and colleagues as your litmus test. High product change, volatile demand and information that sits with the front line point to decentralization. Stable, tightly specified work tilts you back toward centralized control (Aghion, Bloom, Lucking, Sadun and Van Reenen, 2021; Lee and Edmondson, 2017).
- Redefine the managerial role. Move from approver to system steward. Managers curate guardrails, standards and escalation paths, and they help teams coordinate across boundaries. This addresses the engagement problem where it matters. In Gallup’s data, best-practice organizations report 79% of managers engaged, almost four times the global average of 22%, which signals that design choices, not just mood or macro cycles, drive manager energy (Gallup, 2025).
- Build psychological safety into routines. Borrow directly from Edmondson’s insight. Make it normal to surface errors, run quick after-action reviews and ask naive questions. Reward the detection of problems early rather than punishing them late. This is how distributed judgment learns fast enough to be worth it (Edmondson, 1999).
- Make information visible. Local authority without shared context creates local optima. Publish simple decision logs and metric dashboards so adjacent teams can see what changed and why. This supports the cross-functional problem solving that hierarchical structures often suppress in complex work (Lee and Edmondson, 2017).
You can read Gallup’s figures as indictment or invitation. The survey is not a randomized trial, but it quantifies a gap between common practice and what is possible when organizations are designed for energy and clarity. The cost of indecision now is higher than the cost of a fast call that you correct with learning. The firms that will like their odds in the next shock are already doing the quiet work of moving decisions to where the information lives and making it safe to use that discretion in public.
Sources
- Aghion, Bloom, Lucking, Sadun and Van Reenen (2021). Turbulence, Firm Decentralization, and Growth in Bad Times. In sectors hit hardest by the Great Recession, firms that had delegated more to local managers before the shock outperformed, consistent with the value of local information under turbulence. https://www.aeaweb.org/articles?id=10.1257%2Fapp.20180752
- Lee and Edmondson (2017). Self-managing organizations: Exploring the limits of less-hierarchical organizing. Defines self-managing organizations and synthesizes where hierarchy helps or hinders, highlighting contextual limits to radical decentralization. https://static1.squarespace.com/static/5b3cd0cb1137a6d4c0c877c9/t/5c919b0afa0d60694eebc96b/1553046287400/Final+ROB.pdf
- Edmondson (1999). Psychological Safety and Learning Behavior in Work Teams. Introduces team psychological safety and shows its association with learning behaviors that mediate performance in a field study of 51 teams. https://journals.sagepub.com/doi/10.2307/2666999
- Gallup (2025). State of the Global Workplace 2025. Reports global engagement at 20%, an estimated productivity cost near 9% of world GDP, and a sharp drop in manager engagement since 2022, underscoring design pressure on decision-making models. https://www.gallup.com/workplace/659610/state-global-workplace.aspx