I've been playing a lot lately with a music service called blip.fm (@mikegil), where one can "virtually DJ" songs stored by others to create a playlist. When trying to find songs based on a keyword or theme, I'll often search based on that word (for example, "Saturday").
Blip.fm doesn't have anything like faceted search, and very little metadata associated with the songs, so the search results may return anything from songs with "Saturday" in the title to bands with the word "Saturday" in their name. Pretty sloppy, right? Lousy user experience, right?
Perhaps it's the unique nature of music, but the imprecision of blip.fm's search actually makes my searches fun -- I often find some little pearls that I didn't expect and frankly wasn't looking for, but am happy to encounter. I've discovered songs and artists I never would have found otherwise.
Commentators often lament the "niche culture" and "death of serendipity," those precious moments where you find the unexpectedly fascinating article on page 2 of the Metro section instead of the one article you googled and found on Boston.com, so this imprecise search has been a breath of fresh air.
However (and this underscores points we make all the time about the difference between enterprise data and consumer data), I'm not seeking serendipity when I'm searching for my firm's institutional knowledge. I need one authoritative data point, not ten decent ones. This requirement frequently means insertion of a human into the process to broker the flow of knowledge.
Am I going about it wrong? Is there a place for serendipity, for Amazon-esque "Customers Who Bought This Item Also Bought..." suggestion, or is precision the most important? Is the distinction a work/fun one? An enterprise/consumer one?
There's some great reading on this point -- I especially recommend the work that Dan Keldsen and Carl Frappaolo have done on behalf of AIIM in the area of findability.