Collaborations almost never start with a cold email in my experience. All of my collaborations have involved people from my former lab, people I met at conference, or an obvious opportunity where I had the data and they had the analysis tool. Anecdote aside, I think you have to have something the other side will obviously want: a dataset, a method, a tool; access to a site or species or specimen; something besides an idea. Lots of people have ideas. Lots of people even have the same idea. But in order to start a research relationship with someone you don't know, you've probably got to have something they want and vice-versa.