How can businesses tap into the collective wisdom of the crowd? Don Tapscott and Anthony Williams explored mass collaboration in Wikinomics as well as Macrowikinomics. Josh Bernoff and Charlene Li focused on how digital social technologies facilitate consumer-corporate collaboration in their book, Groundswell: Winning in World Transformed by Social Technologies.
A undated research paper (2011?) from Carnegie Mellon  analyzes participation at IdeaStorm.com (@IdeaStorm), launched by Dell in 2007 as a way to tap suggestions (“idea generation”) from its customers.
IdeaStorm, essentially a digital suggestion box that is open to more than employees, is structured like most community web sites: first you set up an account, then you either vote or comment on existing ideas or post new suggestions.
The paper includes eye-glazing equations and economic assumptions of rational utility functions. But the key take-away rests on common sense: businesses first need to listen (ie, read those suggestions) and then provide meaningful and timely feedback. The feedback loop is an essential part of the learning process, which the authors argue makes such a user-generated content space more valuable over time.
[T]hrough experience and learning, those customers who are ‘bad’ at coming up with new ideas (marginal idea contributors) recognize their inabilities and may reduce the number of ideas they propose over time and eventually stop generating new ideas. In contrast, ‘good’ new product idea generators (good idea contributors)2 will be encouraged to continue to provide new product ideas.
When Dell doesn’t respond, or responds very late, the authors observe (more common sense) that individuals are “disincentivized.”
That’s all well and good. But it’s this recommendation that’s troubling (emphasis added):
Our policy simulations indicate that Dell can accelerate the filtering out of marginal idea contributors by providing more precise cost signals. Under another policy experiment, we find that actively responding to all unimplemented ideas will adversely affect the filtering process because marginal idea-contributors who would drop out under the current policy will stay longer under the new policy. As a result, the firm would end up with more low potential ideas. In other words, the firm is better off when it selectively responds to ideas.
[W]e find that in the left plot, the curve labeled “All” is below the curve representing current policy everywhere, indicating that if the firm improves the response time and response rate, it completely removes the disincentive, and the firm is worse off because it receives more ideas with significantly lower potential. Therefore, the firm should strategically select the ideas to which it responds. It is easier to implement the “Differentiate Ideas” strategy because all the firm needs to do is to look at the votes and respond to the ideas for which the log of votes is above average. Furthermore, this strategy leads to the submission of only slightly more ideas, and it outperforms the “All” strategy in terms of the potential of the ideas.
OK. I get that you don’t want to encourage crackpots.
But at the same time, if you only encourage people who have ideas that clearly match current business goals and products, and you do this solely by judging popularity based on community voting, then you are by design limiting yourself to incremental, sustaining innovation (Clayton Christensen). And yet. Lots of media reports credit IdeaStorm with Dell’s decision to sell machines pre-installed with Linux.
The paper looks only at data from 2007 and 2008, so be careful how you project the results. It’s not a definitive analysis of how Dell is using the tool or how effective (however defined) the tool might be:
[W]e only use the data from the initiation of IdeaStorm.com to September 2008. During October 2008, a large number of material changes were made to the initiative, and therefore, we restrict our attention to data prior to these changes.
The research paper does not explore how Dell picks the employees who monitor IdeaStorm or the process it was using for responding. Or how Dell decides which ideas merit a response. A more comprehensive examination of the site seems to be in order if it is to be used as a model (pro or con) for crowdsourcing innovative ideas.
Regarding Bayesnian learning as a theoretical foundation – not going there. Ditto the question of how do you reward good ideas.
 Crowdsourcing New Product Ideas under Consumer Learning (pdf) [local copy]
Authors: Yan Huang, Param Vir Singh, Kannan Srinivasan