When I think about, in my data programming related work, I'd say about 5% is doing analysis or executing statistical routines. And 95% of my time is spent on finding, cleaning, and properly normalizing data. This applies to whether you're a solo researcher or Facebook...think about it: Facebook is a pretty good website, but what it excels better at than just about anyone is being a platform to collect personal data in a way that...well, causes you to quite willingly give it your personal data.
There was a presentation where Peter Norvig pointed out a data routine in which someone had implemented with a naive Bayesian classifier with a comment saying that they'd think of something better...and years later, no one realized it was still a todo. Norvig said something like "You don't have to be very smart when you have a lot of data"
It's called data munging. Good short article on dataspora about it a while back:
1. Your boss says something vague
2. You think very hard on how to move the needle
3. Where’s the data?
4. What’s in this dataset?
5. What’s all the f#$#$ crap in the data?
6. Clean the data
7. Run some off-the-shelf data mining algorithm
8. ...
9. Productionize, act on the insight
10. Rinse, repeat
[1] http://www.slideshare.net/Hadoop_Summit/scaling-big-data-min...