The best thing about a data-driven approach is that it gives room to be wrong. And we're proven wrong all the time: basically every meeting, there's some measured data where we go "that can't be right" - and then we either discuss ways in which we might've misinterpreted it or say "Okay, assuming this is right, then what do we do?" It sounds idiotic, but really we're (humanity in general) the idiots.
Malcolm Gladwell had an interesting podcast about people's ability/inability to express what they really mean and feel. He uses the Aeron chair as an example of a product that was despised/considered ugly but eventually embraced for it breakthrough functionality/aesthetic.
I can see how data-backed design decisions can produced locally optimized results, the "ideal" shade of blue, logo size, typography, etc.. But I wonder if on the whole, all those optimizations work together optimally, as a wholistic experience?
[edit] Gladwell's Podcast: http://itc.conversationsnetwork.org/shows/detail230.html
So, it looks like the designer didn't know.
I'd contend that's because most designers don't optimize for performance, they optimize for aesthetic. When those two are in conflict which should win?
It sounds like Bowman thinks the latter, while Google thinks the former. As a stockholder I'm thankful that's the company's attitude.