How do advisory and transactional AI agents differ in function?

Bilgihan, Ostinelli, Lorenz and Zhang, in a Psychology & Marketing editorial, propose a definition of "AI agents" (AIAs) built on three distinguishing features: "autonomy, adaptive learning and memory, and tool-enabled connectivity." On top of this definition they introduce a typology distinguishing "advisory AIAs, which recommend options, from transactional AIAs, which execute decisions on the consumer's behalf." The editorial frames this as filling a gap it says marketing scholarship currently lacks: "a shared definition of what constitutes an AIA and a theoretical framework adequate to capture its effects on consumer behavior."

The typology sits inside a three-stage process framework the authors propose for consumer decision-making with AIAs — consumer prompting, AIA processing, and consumer response — which they embed in two feedback loops: a relational capital loop (trust and delegation building across repeated interactions) and an AI learning loop (the agent updating its internal representations over time). The authors support the framework with a systematic mapping of AIA-focused articles in leading marketing journals from 2020–2026, reporting that existing work is "heavily concentrated on consumer responses to AI recommendations while largely neglecting prompting behavior, agent cognition, privacy, transparency, and relational dynamics" — the gap their framework is designed to address.

This typology reads alongside What UX principles do workplace users want in AI agents?: both are design-space frameworks proposing how to categorize agent behavior for a specific domain (workplace UX versus marketing/consumer decisions), and both treat delegation and control as central axes. The "autonomy" feature in Bilgihan and colleagues' definition also maps onto Does machine agency exist on a spectrum rather than binary?; their advisory/transactional split is effectively a coarser, domain-specific version of that finer five-level spectrum, with transactional AIAs corresponding to the more cooperative end. The typology is also the structural basis for the companion note Do AI shopping agents reshape what consumers actually want?, which draws its paradigm-shift argument directly from the transactional-agent category and the two feedback loops defined here.

What the excerpt does not establish is any validation of the typology or framework against real consumer behavior — this is a conceptual and definitional contribution, proposed in an editorial, and the available text is the publisher's abstract only. The systematic mapping finding (that prompting behavior, agent cognition, privacy, transparency, and relational dynamics are under-studied) describes a gap in the existing literature, not a claim about how consumers actually behave with advisory or transactional agents; the authors present the framework as a research agenda for others to test, not as an empirically grounded model.