1) Very few customers wanted to upload their data to the web. Even less were allowed by law. 2) Most had integration problems - i.e. they couldn't actually access their data because it was in disparate systems. 3) Most people didn't actually use the information for anything, it was simply to provide them with "evidence" that they were right before looking at the charts. As a statistician this makes me sad, but it was the way things were at least when we investigated it.
Over time we gradually realised why BI was priced so high; it needs system integration alongside it. Now it turns out that we may have been wrong (gooddata.com are doing a good job as far as I can tell) but I still think it's the tip of the iceberg in terms of potential market.
My advice: target a niche with a huge problem where the customer is unable to get at their data at the moment. Choose the niche which has 1-2 big vendors of data collection systems with 80+% combined market share and provide the solution to the problem. And charge.
I'll also be following up with another story soon about the when and why the final decision was made (the story above mostly talks about how they got started and mistakes that were made, not the specific reasons they shut down).