ODC Webinar
Research in Progress on Organizational Decision Structures
Thursday, March 27th, 2025 at 3PM CET/10am EST
In this ODC "Research in Progress" (RiP) webinar, organized and moderated by Franziska Lauenstein (Kuehne Logistics University), two junior scholars – Nelberto Nicolas Qunito (UCL) and Charles Wan (Center for Collective Learning, Corvinus Institute for Advanced Studies) - working on research Organizational Decision Structures, present their working papers on “Aggregating Human and Algorithmic Perspectives” and “Negotiation as Search: Rethinking Impasse and Strategies for their Resolution”. The presentations will be followed by a commentary by Michael Christensen (University of Southern Denmark). The webinar will end with feedback/comments from the audience. Please see below for the papers’ abstracts and the participants’ bios.
Franziska Lauenstein [Organizer and Moderator]
Franziska is an Assistant Professor at the Department of Management at Kuehne Logistics University (KLU). Her research lies at the intersection of behavioral strategy and organization design. Franziska studies how changes in organization design influence individual level decisions and aggregated results. Before joining KLU, Franziska was an Assistant Professor in the Strategic Organization Design unit at University of Southern Denmark.
Nelberto Nicholas Quinto [Presenter]
Presentation Title: Negotiation as Search: Rethinking Impasse and Strategies for their Resolution (with Hart Posen, Joshua Becker, Jon Atwell)
Abstract: Why do negotiations fail to reach agreement even when mutually beneficial outcomes exist? Traditional explanations focus on negotiators’ cognitive and psychological limitations. However, we propose that impasses also arise from structural features of a negotiation shaping how boundedly rational negotiators search for agreement. While prior research indicates that structural elements can constrain the bargaining zone (i.e., the set of offers that lead to agreements), it remains unclear why negotiators sometimes fail to find this zone when it exists and what search behaviors drive these failures. This limits existing theory’s ability to explain and effectively resolve impasses. To address this problem, we develop an agent-based model and conduct computational experiments on standard negotiation cases and algorithmically generated scenarios, yielding several counterintuitive findings. First, our model shows that negotiators can fail to reach agreement due to the computational complexity of negotiations, suggesting that traditional interventions—such as promoting prosociality, cooperation, and information sharing—may be ineffective if search behavior remains unchanged. We show that ideal strategies depend on negotiation time constraints. Second, contrary to existing theory, we find that the difficulty of reaching agreement depends not on the size of the bargaining zone but on how hard it is to find that zone. Third, while extreme, self-maximizing first offers conventionally predict value claiming but increase impasse risk, certain search behaviors can potentially neutralize both effects. We conclude by discussing how our theoretical framework informs future work by supporting a new empirical paradigm for studying complex negotiation dynamics, including multi-party and AI-integrated negotiations.
Short Bio: Nelberto Nicholas Quinto, or "Sam" for short, is a 2nd-year PhD student in Organizational Behavior at the UCL School of Management. He is a collective intelligence researcher who studies negotiation groups and human-AI collaboration in creative and conflict management contexts. His work employs a diverse set of methods, including agent-based modeling, experiments, and inductive qualitative approaches. Sam is currently interested in integrating generative AI as agents in computational experiments and welcomes opportunities to connect with scholars working in this emerging area. Before joining UCL, he worked as a data scientist at BNP Paribas and at an AI think tank, both based in Paris.
Charles Wan [Presenter]
Presentation Title: Aggregating Human and Algorithmic Perspectives (with Helge Klapper and Ting Li)
Abstract: Previous work on human-AI collaboration has investigated two different modes of aggregating the opinions of humans and AI. Firstly, in a sequential process, AI makes a recommendation, which a human then either accepts or rejects. This requires confidence thresholds for deciding when human judgment is required and whether the human should reject AI’s recommendation. Secondly, AI and a human can simultaneously make decisions, which are aggregated using confidence- or expertise-weighted votes. While both sequential and concurrent decision aggregation structures have been extensively studied in the organizational information processing literature, a key respect in which human-AI collaboration differs from human-human collaboration is that humans and AI are agents who have different cognitive architectures and whose ground truths may diverge significantly. This suggests that organization design may benefit from some form of alignment gained through learning. We develop an agent-based model where the AI agent generates output that is closer to the organization-relevant ground truth than the human agent. Our experiments show that a sequential structure with learned confidence thresholds outperforms a simultaneous confidence-weighted voting structure in accuracy when the two agents are sufficiently dissimilar. The outperformance is especially significant when confidence is noisy. However, when the two agents are not sufficiently dissimilar or when there is organization-wide distrust in either the human agent or the AI agent, “tuning” the confidence thresholds of the sequential decision structure provides no or even negative added value compared to a simple confidence-weighted vote. We make the following contributions. Firstly, we show that confidence thresholds can be explicit parameters of organization design. Secondly, we show how the dissimilarity of agents can affect organization design, specifically contrasting a sequential structure with learned confidence thresholds and a confidence-weighted voting structure.
Short Bio: Charles is a postdoctoral research fellow at the Center for Collective Learning, Corvinus Institute for Advanced Studies. He was a PhD candidate at the Rotterdam School of Management, Erasmus University. Prior to his academic turn, Charles had worked as a commodities trader. His current research interests are organizational learning and decision-making, complexity, and collective intelligence.
Michael Christensen [Discussant]
Michael Christensen is Associate Professor at the department of Business & Management at the University of Southern Denmark. His primary research interests concern the interplay between organizational structure and cognitive processes, learning, bounded rationality, decision making, and complex structures/dynamics in general. Michael holds a PhD degree in Physics (SDU) and published in natural science outlets like Physical Review D/E and Journal of Computational Physics. He has experience from working in the scientific software industry for over a decade.
Two of his publications are very much related to this session’s topic of Organizational Decision Structures. The first was published in Management Science in 2010 and investigates “Design of decision-making organizations” and changes how we think about aggregating decisions in organizations. More recently, his paper on “Context and Aggregation: An Experimental Study of Bias and Discrimination in Organizational Decisions” advances how individuals adapt in these structures.
Registration closes 26th Mar, 2025 at 9 am (eastern time)
Hope you will be able to join us!