Laserlike

I have met with several startups over the past few months that have developed “wisdom of crowds” products.  After explaining how the masses will come together to bring clarity to something (the future price of a stock, the likely winner of a college hoops game, the quality of a product or service), they move on to their “viral distribution” model — that is, they won’t need to do any marketing because the same people providing the wisdom will invite all of their friends to the site.

Oh, Wise is Wise, and Viral is Viral, and never the twain shall meet.

In order to understand why mixing knowledge aggregation products with viral distribution can be a counterproductive combination, let’s review the mechanics of a wisdom of crowds system.  In order to do so, let us develop a hypothetical product for illustrative purposes.

Where In The World is Oussama Bin Laden.com (aka UBL.com).

The purpose of UBL.com is to leverage the wisdom of the masses to help the good guys find public enemy #1.  A user will get a random section of a global map and will then be asked to put a pin in the spot where she believes UBL is hiding out.  That’s it.  That’s the product.

If the starting point is truly random, then people who have no idea about the location of UBL will produce random guesses which will have minimal overlap and cancel out.  But those individuals with bits and pieces of knowledge will together paint a picture of his location, even if none of them individually actually knows with any level of certainty.  For example, someone who grew up in Afghanistan may know that a particular area of the Afghanistan-Pakistan border has an extensive cave system that is hard to see from the sky and hard to reach by foot.  Special Forces commandos who have explored a number of areas over the past few years may know where UBL is *not* located, increasing the signal of where he might be hiding.  You get the idea.

Now let’s say that Conan O’Brien does a review of UBL.com just after launch which accounts for 90% of UBL.com’s traffic.  Conan tells everyone that he’s pretty sure UBL is hiding out at the U.S. Embassy in Stockholm.  There is a good chance that UBL.com will experience a non-random number of pins on the U.S. Embassy distorting the accuracy of our guess about the whereabouts of UBL.

Similarly, viral marketing schemes risk launching information cascades — if a large chunk of the user base was introduced to UBL.com by friends, the diversity of input is almost certainly diminished.  Whether this is explicit (user invites a friend with the note “I think UBL is at the U.S. Embassy in Stockholm”) or implicit (offline discussion with respect to UBL’s whereabouts or even biases within certain social groups) it is clearly counterproductive to the goal of accuracy.

Knowledge aggregation systems benefit greatly from diverse inputs.  If accuracy is your goal, you are better off developing antiviral features than viral ones.

The importance of diversity.

Diversity increases the fidelity of inputs, at least with respect to knowledge aggregation.  One of the key strengths of the American system is that we have both distributed governance systems (capitalism, democracy) AND a diverse population.  In addition to the many benefits of importing hungry, bright talent, immigration offers the continuous injection of diverse points of view.   Diversity is just as critical to the efficacy of wisdom of crowd businesses online.  So how can we get diverse inputs?

Enter Google.

You have heard the stories about how a few thousand contributors account for the large percentage of edits on Wikipedia.  But the diversity of the inputs needn’t be measured on a byte-weighted basis.  The vast majority of Wikipedia’s 78.2MM U.S. monthly unique visitors arrive via search (Google in particular, I would guess).  The semi-random distribution of users [with respect to the user] to a particular Wikipedia page would tend to limit the risk of a particular bias [if the consumer disagrees with the content, he can simply make the change].  In fact, many of the best wisdom of crowd business are continuously reviewed by a stream of semi-randomly distributed “reviewers” thanks to Google.

Even if you have a small community of creators, if the vast majority of users find your site through third party search [and there are mechanisms for them to improve or challenge the content], you will go a long way to increasing the diversity of your inputs.

You may also want to track the inputs for a particular product by type to see how [and potentially remove] certain biases may impact the system.  In our UBL example, you may review the data from the perspective of all inputs, the best guess of military personell, by country based on IP address, etc.

There is a fundamental tension between building great community products where people invite friends and building a product that aims to assemble diverse inputs to paint a collective picture.  Understanding that tension will go a long way to helping you build a better product.