AI strategic embeddedness and the contingent value of artificial intelligence in organisations
TL;DR
This paper develops the construct of AI strategic embeddedness, defined as the configuration of four placements through which AI systems become constitutive of an organisation’s attention structures, decision routines, work arrangements and learning processes rather than adjacent to them.
Abstract
Firms making comparable investments in artificial intelligence can realise very different strategic returns. Existing research explains AI capability, managerial cognition, human-AI work design and governance largely in separate streams, leaving unresolved how AI’s position inside organisational decision architecture conditions the value it produces. Why does AI change what some organisations are able to decide and do, while remaining a productivity tool at the edge of others? This paper develops the construct of AI strategic embeddedness, defined as the configuration of four placements through which AI systems become constitutive of an organisation’s attention structures, decision routines, work arrangements and learning processes rather than adjacent to them. Drawing on dynamic capabilities, the attention-based view and research on human-AI complementarity, we specify four mechanisms through which embeddedness produces value. These are attention reallocation, routine reconfiguration, complementarity in judgement and recursive learning. We then theorise four features that existing accounts miss. The attention mechanism expands the signals an organisation sees while confining them to the region its models represent. Governance conditions that value through an adequacy threshold and a burden gradient, so value is lower below a minimum standard of governance adequacy, and returns to further procedural burden above it diminish and eventually turn negative. Embeddedness contains a corroding loop in which sustained reliance erodes the human expertise that the complementarity mechanism requires, which we call the capability-dependence paradox. And embeddedness built on widely shared general-purpose models compresses the dispersion of strategic issue framing between firms, while raising absolute decision quality only where the models are well fitted to the task, so private value can rise without relative advantage. Ten numbered propositions, with Proposition 8 divided into two linked predictions, and a research agenda follow from the framework. The last of them concerns how domain-level embeddedness aggregates to the firm.