So you want to talk at this conference?
I regularly review talk submissions for tech conferences, and here's a list of what I'm mostly looking for when deciding to accept or reject a talk.
Other reviewers might be looking for different things - this is just my own criteria.
My first question is always are you going to get people interested in your stuff. Are you a dynamic speaker who keeps people on their toes, or the kind of person that delivers their talk seated at a desk. I've seen the latter happen, and it's not pretty! For me, a brilliant speaker gets a slot almost every time, also because they usually know which topics will raise people's interest.
Unless you're famous already, the best way to convince me that you're a good speaker is to point to a video of one of your talks. And I also need to know why you think you're qualified to deliver this talk.
Then, I'm looking for a topic that will add value to the conference. Promoting your product or company might not add much value, whereas a talk that will open people's minds and maybe save them hours of work in their practice is a guaranteed winner. Signs of a value-adding topic are pointers to concrete achievements using the techniques presented in the talk.
The quality of the submission comes next, especially if I don't know the speaker. Someone who's unable to present their ideas clearly in a talk submission is unlikely to present them clearly at the conference. Or maybe they're a misunderstood genius, you should also look for those but they are rare. A concise submission that packs lots of useful information about what's going to be delivered at the talk is a good promise of success.
Last but not least, original and inspiring ideas get lots of bonus points from me. Being able to predict the abstract's contents from the title is usually a bad sign, except if it's a talk for beginners. We don't need conferences to exchange information today, that's supposed to happen on the Web. Talks should be inspiring, maybe teasers to convince people to look at your value-adding stuff, but not rehash information that's found elsewhere.
Update: I forgot to mention the movie trailer thing: a talk abstract is a lot like a movie trailer, if you feel you've seen all the good parts of the movie after watching the trailer, it's not a good sign. Similarly, a good talk abstract leaves me with the impression that there's much more to discover in the talk, compared to what's mentioned in the abstract.