The KPI Spiral: Why Replacing Metrics Fails and How Architectural Separation of Loops Halts Degradation (An Agent-Based Model and Institutional Solution)

Classic accounts of KPI dysfunctions (Campbell’s law, Goodhart’s law) describe the fact of metric corruption but do not explain the mechanism by which a productive organization irreversibly degrades into total simulation. Using an agent-based model, this paper demonstrates that this process is a progressive narrowing of the agents’ feasible decision space, driven by rational managerial interventions. Each new requirement—output KPIs, process regulations, KPIs on compliance—cuts away productive strategies until simulation becomes the only feasible choice, not because it is profitable but because no alternatives remain. When cognitive adaptation (endogenous preference change) is introduced, the model reproduces the “bottom” regime—complete loss of productive activity while subjective well-being recovers—as well as a moderate imitation equilibrium with gradual erosion of intrinsic motivation. In the second part, we propose an architectural solution: functional separation into a “Gateway” loop (routine tasks under KPIs) and an “Office” loop (KPI-free productive work), with a mechanism that calibrates Gateway thresholds based on Office success. Experiments show that the intervention stops the KPI spiral, restores productivity and decision space, but the cognitive degradation of Gateway agents persists, indicating that additional measures are needed to recover agency. The architectural principle is externally validated through the spontaneous evolution of customer service systems (separation into chatbots and human operators), which confirms both the effectiveness of loop separation and the risk of new metrics re-emerging.