In my book, the source that spawns the actual conversation with the customer is the one that gets the most weight -- the stuff that happens before and after is really inconsequential (as far as naming/identifying a "source") -- I'm talking to my customer after they interacted with X.
From there, I'm not interested in giving "Credit" for a particular sale (or lost sale)-- but I am very much interested in tracking patterns: was this particular interaction a fluke? Am I seeing a lot of interactions in the same pattern? If I "play," can I increase these interactions? How do these interactions compare to Y? Does playing with Y affect X?
So for those out there saying that this information is useless, I say this: make the best use of the actual data you can track by understanding what the data can actually mean.
And then just keep spending all your money on radio and TV! LOL!!