Responsible Artificial Intelligence in Enterprise Customer Relationship Management Systems: An Accountability Transfer Taxonomy and Research Agenda

Abstract

Responsible AI research can be organized by principle, by regulatory risk tier, by stage of the development lifecycle, or by technique. Each of these lenses is useful, but none indicates the point at which a system displaces the authority of a human decision-maker. This review proposes a different taxonomy that distinguishes, for customer-facing enterprise systems, five loci of accountability transfer: perception, framing, option, execution and commitment. Through perception, the system determines what a human attends to. Through framing, it determines how the human interprets it. Through option, it limits the choice set. Through execution, it performs a reversible action. Through commitment, it performs an action with customer-visible or binding effect. Twenty-four peer-reviewed articles published from 2020 onwards were synthesized against this taxonomy, and coverage across it was uneven. Option, execution and commitment are discussed in some detail, commitment most of all because it carries regulatory obligations. Perception and framing are only weakly discussed, or discussed in work focused on other questions. They are, however, the two loci at which customer relationship management automation operates most often, since triage, prioritization, summarisation and sentiment interpretation are far more common than autonomous action. This asymmetry yields four research gaps and a corresponding research agenda. The main limitation is that the taxonomy is analytically derived from the reviewed corpus rather than empirically validated against practitioners' experience of where authority is lost.