Not tuning the Google Search Appliance

  • 1-Jun-2006

As Enterprise Search Report readers know, one of the drawbacks of Google's search appliance is that it cannot be "tuned" the way you can with most enterprise search products. Now along comes a Google to say, in effect: "not a bug, but a feature." First, kudos to the Google enterprise team for sharing their thinking about their product (Know any Autonomy bloggers? Me neither.), and for constantly emphasizing the primacy of usability. But I have to disagree with the main thrust of Google's post. Yes, tuning is a pain, but so is creating (and maintaining!) "best bets" -- what Google calls "Keymatches" -- their suggested alternative to tuning. Many contemporary search systems have simple, web-based tuning interfaces so that adjustments become a matter of configuration, rather than scripting. If everybody's information was the same, search would be easy. But everybody's information is different, so tuning (and yes, re-tuning when new repositories come online) becomes important. Google's assertion that their algorithms' inherent superiority obviates the need to expose their black box is patently wrong. The fact is, all search systems offer a brew of different algorithms that work better or worse in combination against different sets of heterogeneous enterprise data. It's impossible to know a priori how well one system will work with your collections until you test it -- and then tune it. Like much in life, searchers don't know "good" results until they see them. In search, simple is hard. Simple in the enterprise takes hard work. Test and re-test before you buy.