Shidan was mentioning that they might be interested in a predictive dialer application. It's a fairly straightforward system, you have a statistical algorithm that tries to predict how many operators will become free over the next X seconds (taking currently length of each call into account), another algorithm figures out how many connections will go through for a given number of calls, you initiate the calls (tracking the percentage that get through for the second algorithm), you pass the callers to the next available operator (or a parking IVR) and track the duration of the call to feed into the first algorithm.

You can tweak the two algorithms considerably:

* short versus long calls (if the call goes longer than X seconds it has a very different expected lifetime than a simple average),
* tracking individual operator averages statistics to calculate call length
* take time of day into account

but that's the basics. The system only works well if you've got enough operators to make playing the statistics reasonable, but if you've got enough, you can make it work.

Would want to connect it to a web monitoring system that would let the operator see the connected user's information, track time on the call, all that lovely stuff (I'm assuming this would be a livepage in Nevow). Anyway, not sure I want to get into the phone survey market just yet, but I must admit, it seems like it would be a fun little project to produce.