Pushing decisions down improves speed and quality in theory. The conditions that make it work in practice are more demanding than they appear.
The Promise and the Gap
Distributed decision-making — the structural approach in which decisions are made as close as possible to the operational reality they affect, by the people with the most relevant knowledge — has genuine advantages over centralised decision-making that are well-established in both theory and practice. Faster decisions, better-calibrated decisions, more engaged decision-makers who own the outcomes they produce, reduced bottlenecks at the centre of large organisations. These advantages are real. The gap between the promise of distributed decision-making and its typical implementation reality is also real, and it is produced by underestimating how demanding the conditions are under which distributed decision-making actually delivers its advantages.
The Conditions
The first condition is that the distributed decision-makers must actually have the relevant knowledge. This is true for operational decisions in stable, well-understood domains. It is not true for decisions that require strategic context, institutional history, or cross-functional knowledge that is held at the centre rather than at the edges. Pushing these decisions down to people who lack the required knowledge does not produce better-calibrated decisions — it produces faster but worse-calibrated ones, made by people who are doing their best with incomplete information.
The second condition is that the incentive structure of the distributed decision-makers must be aligned with the institution's overall objectives. When this alignment exists, distributed decision-makers who optimise for their own performance produce outcomes that serve the institution. When it does not, distributed decision-makers who optimise for their own performance produce locally rational decisions that are collectively suboptimal — the unit that maximises its own metrics at the expense of institutional coherence, the team that prioritises its own objectives over cross-functional collaboration, the region that builds practices optimised for local conditions in ways that prevent the institutional learning that would come from standardisation.
The third condition is that the institution has the feedback infrastructure to detect and correct distributed decisions that are going wrong before the errors compound to the point of material damage. Without this infrastructure, the distributed decision-making advantage — faster response to local conditions — is accompanied by the distributed decision-making risk — errors that propagate uncorrected across many decisions before centralised oversight can identify them.
Distributed decision-making works when the conditions that make it work are present. Installing it without those conditions produces decentralised decision-making — which is a different thing, with worse outcomes than either centralised decision-making or genuine distribution.
Discussion