Joseph Ollier
Frontiers in AI: From Generative to Agentic Recommender Systems
Die Forschung untersucht, wie agentische KI-Empfehlungssysteme Entscheidungen nicht nur unterstützen, sondern auch eigenständig in reale Handlungen überführen können.
Abstract
The integration of artificial intelligence (AI) in financial services is transforming consumer decision-making, particularly in the insurance sector, where competitive markets ensure new innovations such as AI are at the forefront of successful business practice (Bauer et al., 2024). While Generative AI has been estimated to generate over $50 billion in annual economic benefits and increase insurance industry revenues by 15% (Bain & Company), to date, mass investments in AI technologies have failed to deliver corresponding changes in profitability, a term practitioners call the GenAI paradox (McKinsey). The advent of Agentic AI, defined as systems integrating multiple specialized agents in order to complete a goal in the real world, however, represents a paradigm shift in the capabilities of AI, closing the loop from user intents to real-world action.
One domain where Agentic AI shows much promise are Recommender Systems (RS), defined as computational frameworks to infer and suggest items most relevant to users’ needs. RS have a long history in simplifying user search and decision-making, particularly when integrated into chatbots, and regularly feature the newest AI developments due to competitive markets and strong potential for profits: Consider, for example, that 35% of total sales at Amazon (IBM) and 80% of content watched on Netflix (Gomez-Uribe and Hunt, 2015) arise from RS recommendations directly. While RS powered by traditional AI (e.g., machine-learning based collaborative filtering) and Generative AI (e.g., large language models) can predict most likely items and generate rationale for suggestions, Agentic AI extends these abilities by also allowing the user to delegate a real-world action to the system, such as insurance policy purchase, representing the newest frontier in AI’s evolution.
In the insurance industry specifically, the implications of applying Agentic RS are huge. Tight-profit margins and the desire to compete on more than price alone have meant mastery of RS technology is a key predictor of firm success, for example, by facilitating cross-selling or greater personalization (Pathak et al., 2010). Outside the insurance industry, the implications of semi- or fully autonomizing the process of search, recommendations, and real-world behaviors clearly has wider, transformational potential in a variety of societally relevant contexts such as healthcare (e.g., medication monitoring and purchase), education (e.g., tailoring learning activities and sending students feedback) and SME management (e.g., suggesting new suppliers and intitating contracts).
For Agentic RS to meet this challenge and unlock its transformative potential, however, key research questions remain open: for users, who may be hesitant to delegate decision making, understanding what factors encourage delegation and psychological processes governing this is key. For firms, where Agentic RS implementation utilizes time and monetary resources, understanding how Agentic RS generates business value and when not to apply Agentic AI solutions is equally important. Lastly, for society, due to the potential of cognitive offloading, overreliance on AI systems, and rogue actors, potential dark sides of delegation (e.g., acceptance of inferior products) must be examined.
die arbeiten im 2026
Die folgenden Beispiele geben Einblick in die Bandbreite bisher unterstützter Vorhaben. Von Arbeits- und Organisationspsychologie über Cyber Risks bis zu Innovation, Cleantech und Wissensspillovers: Die Förderbeispiele verdeutlichen, wie vielseitig die unterstützten Themen sind.
Nils Löwhagen
Beyond Classical Limits: Quantum as an Emerging Technology in Operations Management
Wie und wann Quantentechnologien von wissenschaftlicher Vision zu praktischem Mehrwert in Unternehmen, Management und Politik werden können.
Joseph Ollier
Frontiers in AI: From Generative to Agentic Recommender Systems
Die Forschung untersucht, wie agentische KI-Empfehlungssysteme Entscheidungen nicht nur unterstützen, sondern auch eigenständig in reale Handlungen überführen können.
Melanie Baumgartner
Feel It, Name It | Bodily Sensation Mapping – A Scalable VR Application for Accessible Mental Healthcare
Ein VR-basiertes, körperorientiertes Tool soll Menschen dabei helfen, Emotionen besser wahrzunehmen, zu benennen und zu regulieren – als zugängliche Ergänzung in der psychischen Gesundheitsversorgung.
Maurizio Daniele
A Unified Framework for Real-Time Macroeconomic Reanalysis and Forecasting via Probabilistic Graph Neural Networks
Das Projekt entwickelt ein neues, graphbasiertes KI-Modell, um wirtschaftliche Entwicklungen in Echtzeit konsistenter, genauer und mit Unsicherheiten prognostizieren zu können.
Förderung beantragen
fragen sie uns an
Sie möchten ein förderungswürdiges Vorhaben einreichen? Hier finden Sie die Voraussetzungen, den Ablauf und alle wichtigen Informationen zur Antragstellung.
Über Uns
Tradition mit Zukunft
SMC‑S steht in einer langen Tradition: Aus einer 1929 gegründeten Initiative zur Stärkung betriebswissenschaftlicher Kompetenzen an der ETH Zürich ist eine Organisation entstanden, die bis heute Forschung, Ausbildung und Wissenstransfer fördert. Heute verstehen wir uns als aktive Plattform für Austausch, Förderung und verantwortungsvolles Unternehmertum.