I saw a talk by interface guru Alan Cooper last month, who presented a well thought-out process for designing an interface and feature set, then astounded the audience at the end, responding to a question about usability testing by stating, "We don't test."

Now true, anyone who's worked on a site and gone through focus groups (especially if the client is on the dark side of the glass with you) knows the dangers of testing, namely that someone (you, the client, or another focus group attendee) will glom on to a few tidbits that came up that they happen to agree with, creating 'refinement by anecdote' and generally killing off usability for the many by tailoring to the few, where 'the few' is a moving target, in the end creating a big venn diagram of usability that includes nobody.

Is there an alternative? I won't claim to have the answer (and, humbly, I would venture that Mr. Cooper doesn't either), but it might be a good first step to admit that your site might not be the site for everyone. Think of how auto manufacturers design cars, or how they focus test for them. They are tailoring a specific model to a specific mindset, and it's okay to throw out audience segments in order to be the best of breed for one particular segment, be it a sports car enthusiast, or the impulse buyer.

Outpost.com does a pretty good job of this, with zero shipping or tax, anything you want can show up the next day for less than it would cost to go get it locally.

Anyhow, the point here is that by trying to grab the biggest market share, many fledgeling dotcoms are grabbing the thinnest market share, who's only reason for shopping preference is offline advertising or a sock puppet.

We can do better than this.