I particularly liked your point about building features that no-one used - I'm sure it was kinda hard to watch all that hard work go to waste.
You might have covered this in your talk, but which features did you add that you could have omitted? Without having that usage data early on, is there any way you could have done this differently?
Your slides are awesome by the way :)
Keep in mind that it really all depends on your goals, and my goals aren't to create a huge app that generates millions but to build something that's pleasant to use and supports me to the point that I can continue to work on it.
That said, I think that we probably should have invested in improving and polishing our existing features before adding source control and email integration. However, I'm sure those have generated a handful new customers that otherwise probably wouldn't have used Sifter. I think they fall into the categories of things that people need when they're evaluating software but that in reality they really don't end up using.
Really, there's so many variables that it's hard to contribute any level of success or failure to one decision. I'm really pretty happy with everything that we have or haven't done to this point. It's just a bit disappointing to pour effort into a feature and find out that it's only being used 2-3% of the time.