The algorithmic third actor in service work

Service research has tended to treat algorithmic management as a single phenomenon, yet the systems standing between service firms and their frontline staff differ in ways that matter for how work is experienced and for the quality of service delivered. This conceptual paper proposes three system properties for distinguishing among them: decision autonomy, whether the decision rule was specified by people or learned from data; outcome consequentiality, the degree to which outputs translate into material outcomes without further human deliberation; and learning adaptivity, whether the rule is revised after deployment. Combined, these generate a typology of eight system types, from minimally agentic tools to continuously learning autonomous partners. Because the three properties are not service-specific, the framework introduces encounter exposure as a moderating condition: the extent to which the managed function bears on work performed in real time, in the customer’s presence, under customer-induced variability. Seven propositions connect configurations to frontline discretion, service recovery capability, customer evaluations, employee trust and strain, and firm governance. Two are ordinal or conditional rather than directional: that consequentiality constrains discretion more than autonomy does, and that adaptivity becomes costly only at high consequentiality.