Personally, I think the best way to figure out what your customers want is to do their job. Even if it's just on a trial, couple-of-weeks basis. Just make sure you try to do it to the best of your ability and actually seek out best-of-breed existing solutions.
People differ greatly in their needs, but most people doing the same thing run into the same problems. So if you're a financial software startup, try to make money with your investments. If you're a gaming startup, play lots of games. If you're a dating or social networking startup, try to get laid. If you're an e-mail startup, manage your company through e-mail. If you're a bugtracking software startup, put lots of bugs in your product. ;-)
Really, just a variant on "eat your own dogfood", except you can start imagining yourself as a dog before the dogfood actually exists.
I'm never again considering what people say they want without a mountain of salt. All too often, what people say they want ends up not being used at all, not even by those who requested the change or addition. Users don't seem to be magically smarter than the developers in my experience. Design by committee has never worked so well for me. Especially in such a flexible medium as the internet.
But, releasing early is a good way to measure what people will actually use.