I've worked at a startup (not really a startup anymore) for 6 years that does software for the P & C insurance industry, and that sort of software is just an entirely different ballgame to develop, implement, and sell. Yes, their existing systems are terrible, but they work(-ish), they've been building them for 20 years so they do a ton of things, and they have anywhere from 5 to 30 different systems that all need to be pieced together. In addition, those systems are often really core to their business: changing things requires huge investments in terms of integration work and re-training, and such decisions are not made lightly. In addition, they're still not (yet) comfortable with losing control of their data. That's changing, but most big non-tech businesses just aren't going to put, say, their insurance policies or their medical records in some third-party website's hands where they don't have direct access to the database. That part's probably going to change eventually, though.
The other problem is that building these systems is hard: they're huge in terms of functionality (a claims system just does a hell of a lot and has a ton of different screens, rules, and tables), so they take a long time to build, and they require an enormous amount of industry-specific knowledge that takes a long time to build up. Feedback cycles are also hard because of the deployment/integration model; you can't just throw something out there, have people try it out, and quickly iterate. All that also means it's expensive to build the software; it takes a decent number of people a really long time to build something that's even vaguely adequate for replacing their existing, 20-years-of-development COBOL mainframe system. Combined with the general niche market you're talking about, that means you need to get paid a lot for every customer, which also tends to preclude any sort of hosted service-based model; the number of customers is too low and the R & D costs are too high, so the economies of scale just aren't there. For whatever reason, businesses just tend to see services as cost centers to be minimized and value them differently than software purchases.
All that, though, means that 1) their existing systems suck and 2) hardly anyone smart is working on making them better, which is one reason why our company is able to be successful. So if you can round up a bunch of smart developers, some good business analyst-types who can delve into the corporate world and discover what they really need built, and some persistent sales people that can handle the agonizing, 12-24 month sales cycles, there's a whole lot of money to be made there. But it's a very, very different world than your standard web 2.0 app, and usability really doesn't factor much into the purchase decisions: their existing systems are so bad and so hard to maintain, the dev process is so brutal, and the competition is so poor that the most important thing is to build a system that works and is flexible in the right ways, and usability is more of a secondary concern. If there wasn't such a high barrier to entry, you might actually see products compete on usability instead of just on functionality.
If you don't want to go the full-on enterprise route, you can certainly have success chopping off parts of their business that are hugely common across business and industries, like salesforce.com, google apps, or wiki services. But I don't think that approach will really work for core systems that are highly industry-specific.