Market-Based Collective Intelligence in Enterprise 2.0 Decision Making

Organizations increasingly use social computing to harness the collective intelligence of their employees, partners, and customers in collective problem-solving and decision making. This trend towards Enterprise 2.0 can be defined as ”the use of emergent social software platforms by organizations in pursuit of their goals” [McAfee 2009, p. 73]. One Enterprise 2.0 benefit is collective intelligence [McAfee 2009, p. 130], e.g. the use of prediction markets to elicit and aggregate information from a crowd and form a group prediction [Surowiecki 2005; Malone et al. 2010]. Prediction markets incentivize information revelation and frequently produce more accurate predictions than other information aggregation mechanisms [Berg et al. 2008; Ledyard et al. 2009; Teschner et al. 2011; Bennouri et al. 2011]. Given these properties, it is not surprising that many organizations adopt prediction markets as emergent social software platforms in crowdsourcing information. Enterprise prediction markets are, for example, used to forecast product development outcomes, demand, company and industry news [Cowgill et al. 2009], to forecast sales [Gillen et al. 2012], and to support project management [Ortner 1998]. In a recent survey among 2,609 entrepreneurs and employees, 9% reported that their company’s set of social computing tools includes prediction markets [Bughin et al. 2013]. Today, ”practice is still far ahead of theory in the field of collective intelligence [...] with the notable exception of prediction markets (for which a large body of work helps us understand what works and why)” [Bonabeau 2009, p. 50]. Using this large body of work might seem promising – it is, however, intricate as this work typically relies on the assumption that either prediction is an end by itself (there is no subsequent decision) or that market participants are decision-agnostic. On the contrary, contemporary use of prediction markets as Enterprise 2.0 decision support tools typically involves market participants with a vested interest in the decision. As market participants (employees) are directly affected by the decision taken, they have incentives to manipulate market prices and the subsequent decision. In literature these markets are sometimes referred to as decision markets [Hanson 1999]. Hence, simply carrying over knowledge on prediction markets to the design of such decision markets might be premature. To back this hypothesis, we build on theory and evidence on prediction markets, decision markets, and market manipulation, and test the key theoretical predictions in a lab experiment. Our data suggest that indeed, people who have a stake in the decision manipulate the market and partially corrupt information aggregation. More precisely, we show how different designs of the principal’s decision rule affect manipulation, information aggregation, and decision quality. In a nutshell, we find that the common practice is better than random but worse than an alternative design.1