When coordination is avoidable: A monotonicity analysis of organizational tasks

Area of Science:

• Artificial Intelligence
• Distributed Systems Theory
• Workflow Optimization

Background:

• Organizations invest heavily in coordination, but its necessity for task correctness is often undetermined.
• In multiagent AI systems, coordination costs can be substantial and measurable, sometimes exceeding task costs.
• Distributed systems theory offers a criterion: nonmonotonic task specifications necessitate coordination.

Purpose of the Study:

• To develop a decision rule for determining when coordination is essential for task correctness.
• To map Thompson's interdependence taxonomy to the nonmonotonicity criterion.
• To quantify unnecessary coordination spending in real-world workflows and tasks.

Main Methods:

• Formalized a correspondence between interdependence types and the nonmonotonicity criterion using a bridge theorem.
• Applied the derived decision rule to 65 American Productivity & Quality Center (APQC) workflows.
• Analyzed 13,417 Occupational Information Network (O*NET) tasks using a calibrated large language model (LLM).
• Illustrated the findings in multiagent AI simulations.

Main Results:

• 74% of analyzed workflows were found to be monotonic, not requiring coordination for correctness.
• 42% of O*NET tasks were identified as monotonic.
• This suggests that 24% to 57% of current coordination expenditures may be unnecessary for ensuring correctness.

Conclusions:

• A precise, theory-based rule can identify tasks requiring coordination, moving beyond general practice.
• Significant potential exists for reducing coordination costs in multiagent AI and organizational workflows.
• Optimizing coordination based on task monotonicity can lead to substantial efficiency gains.