I've only ever worked on something that solved a personal problem. I just go ahead and build something rudimentary/MVP that solves my problem. That MVP could still include manual bits, be sitting on a free subdomain (subdomains from Google App Engine), not have a proper logo (just use text for the branding), etc.

Once I get rudimentary version, I can then put it out there in public to validate it. Maybe do a Show HN or do a 'low budget' Ad on Facebook/Twitter. This is what I did for my current project, https://nocommandline.com (I did a Show HN).

I feel that software needs a 'trial' version to be able to gauge whether people will actually use it or not.