ai negotiation skills study

from ivizlab.sfu.ca From thesis: Embodied Conversational AI for Women's Negotiation Training: A Regulatory-Focus-Informed Simulation and Evaluation Qianying (Charlotte) Hou, ( Supervisor Steve DiPaola) Women in the workplace face a negotiation double bind: assertive self-advocacy risks social backlash, while cautious approaches produce economic and career disadvantages. Under this pressure, women in these contexts adopt prevention-oriented behavioral defaults that undermine negotiation effectiveness. Traditional training builds declarative knowledge but cannot recalibrate the fast, automatic responses that govern real-time performance — and conventional role-play reproduces the very social visibility that activates prevention-oriented behavior in the first place. This thesis designs, implements, and evaluates an AI-mediated negotiation training system grounded in Social Role Theory, Regulatory Focus Theory, Dual-Process Theory, and Experiential Learning. The system's core component is an AI-driven Embodied Conversational Agent (AI-ECA) — a photorealistic, real-time rendered character capable of naturalistic dialogue, emotionally expressive facial animation, and synchronized speech. The AI-ECA is paired with a retrieval-augmented domain knowledge base and a theory-derived evaluation framework targeting prevention-oriented communication patterns, creating a structured, repeatable, low-stakes rehearsal environment for women in the workplace. A participant evaluation study (N = 48) finds the system to be socially credible at the level of dialogue and professional interaction, diagnostically accurate in identifying real workplace communication patterns, and perceived as actionable in its structured feedback. The thesis also identifies two key boundary conditions for future development: the structural limits of transcript-based evaluation for capturing the full communicative complexity of negotiation, and the gap between perceived feedback quality and behavioral change intention under observational rather than direct-interaction conditions. The thesis provides the theoretical, architectural, and empirical foundation for direct-interaction studies capable of testing durable behavioral change. ivizlab.sfu.ca