Brave new procurement deals: An experimental study of how generative artificial intelligence reshapes buyer–supplier negotiations

The technological breakthrough of artificial intelligence (AI) is impacting buyer-supplier negotiations, which are increasingly moving toward human-to-machine negotiations using AI-based chatbots. While the first AI-powered negotiation solutions are currently being used by procurement professionals to negotiate for non-critical spend items, which is an example of structural influence, the behavioral influence of AI-based chatbots (i.e., on negotiation approach) remains unknown. It is unclear in which behavioral settings these chatbots deliver value to the buying firm in terms of economic, psychological, and relational outcomes. To fill this gap, we conduct three experiments in buyer–supplier negotiation settings, two in a lab-setting with undergraduate business students and one online experiment with professional negotiators. In our interactive simulations, participants play the role of the supplier, while a ChatGPT-based custom-trained chatbot acts as the buyer. We find that when the chatbot deploys a competitive, as compared to a collaborative, negotiation approach, it will achieve a higher price discount, better payment terms, and a quicker negotiation. However, suppliers trust a collaboratively prompted, as compared to a competitively prompted, chatbot more and demonstrate higher outcome satisfaction, as well as a stronger desire for future interaction. A text analysis of the chat interactions indicates a higher level of similarity when a competitively prompted chatbot is employed, which implies that suppliers also use more insistent and intimidating language, thereby matching the chatbot's negotiation approach to a greater degree. While the negotiation approach is a significant influencing factor, we do not find significant evidence that item type, in our case non-critical or bottleneck, matters, which indicates that AI-based chatbots can be effective in various buyer–supplier settings.