Some of the best advice I've been given on this is to look at how the potential customer is already solving the problem today. If they're just ignoring the problem altogether, then they're not going to spend any money on you to solve it. If they're spending considerable time and/or money working around or manually solving the problem (maybe by working weekends, or buying a whole team of vendors, or outsourcing to Mechanical Turk type stuff), and you can solve the problem for them for less money and/or time, then it's a feature worth shipping.

It can give false negatives, especially with future looking and platform-type work, but it's a great heuristic for weeding out useless feature work