Decisions Not Made: Owner Attention, the Intrinsic Value of Authority, and the Limits to Organizational Scale

Owner-managed firms frequently stop growing well before technology, demand, or finance would require them to. This paper develops a span-of-control model in which the binding constraint is the owner's attention and the central choice is which decisions to stop making. Delegating a decision frees attention at the cost of an oversight requirement and a decision-quality loss. The optimal policy is a threshold rule: the owner delegates every decision whose delegation loss falls below the attention it frees, valued at the average return to the owner's time. Four further results follow. First, behavioral sources of under-delegation (an intrinsic value of authority and a salience-driven overweighting of delegation errors) are equivalent to a reduction in the perceived value of the business, and the authority rent distorts behavior only to the extent that oversight is costly. Second, under-delegation is first order for scale but second order for profit, so the owner's own profit signal provides weak feedback against it. Third, when delegation capability accumulates with use and depreciates with disuse, the model admits multiple steady states, and a myopic owner whose business remained small for long enough converges to the low-delegation steady state. Fourth, in a recursive hierarchy the returns to delegation compound across tiers, so a uniform under-delegation culture imposes losses that grow geometrically with the depth of a business group. An accounting identity yields an observable diagnostic: sustained growth without neglected decisions requires the number of decisions the owner no longer makes to grow faster than the portfolio itself. Use of generative AI. The research question and the main ideas of this paper are the author's. Generative AI tools (Claude, Anthropic; ChatGPT, OpenAI) were used extensively in formalizing the model and the proofs, computing the numerical examples, preparing the references, typesetting, and writing the text; see the Declaration on the Use of Generative AI in the paper. Version 1.1 (18 October 2026). Added a declaration on the use of generative AI. Corrected Proposition 3 (part 4), Proposition 5 and Proposition 6 and their proofs, the discussion of group depth in Section 7.2 and prediction P5, the description of Bloom et al. (2013) and Langer (1975), and a rounding error in Table 1. Added Figure 1, updated the literature review and references, and revised the exposition. JEL codes: D23, D91, L22, L25, M12, G34. Online version: zackha.com/research/decisions-not-made