Phase I requirements are typically (but not always) that you have to produce a report that you did feasibility research on the problem. Sometimes a working prototype is the Phase I deliverable. Usually Phase II is where the working prototype is and Phase III is a delivered working system (though for larger projects, Phase III is just the prototype or improvements to Phase II's prototype).
Typical payouts for the phases:
Phase I - 75-100K
Phase II - 750K
Phase III - 2 mil
Most of these projects are challenging enough that for 75K, you're not going to be able to deliver much more than a report. Once you factor in overhead, that's about 4-6 man-months.
I agree with you wholeheartedly, though, that it is greatly taken advantage of -- on a very large scale, and the relationship between companies and granting Program Managers is a big, big deal.
There are definitely companies that play the "we'll do nearly anything" open-ended engineering game and pay themselves using Phase I's.
I have seem some legitimately great work come out of NSF SBIRs, which are similar, but quite a different game in many ways from military SBIRs.
I worked for a company writing military SBIRs for 10 months. Worst job of my life, probably. It was also mind-blowing how OK with all of this that most people of all levels of that chain were.
EDIT: formatting, minor content