The problem is that "impact on innovation" starts usually already with a big discussion on how to measure innovation.

The "easy" way is to take very accessible data points such as the number of patent applications. But you can imagine just by looking at the patent system related discussions around here on hn that this is a very "discussable" subject.

I wasn't communicating well, what I'm on about is that knowledge flow in large organisations is not always (or even often) through 'official' channels. I'm thinking of Etienne Wenger's work on situated learning, especially the insurance office example in Lave and Wenger's book.

It struck me that datasets from email, and from sharepoint like systems might help people identify the 'betweeners' in their organisation.

Yes, people do this - it is called Social Network Analysis and has been around a few decades. It is well introduced in http://www.amazon.com/gp/aw/d/1449306462 . There is actually a thorough example of analyzing an organization's "social leaders" and arranging departments to take advantage of them.

I read an article about using IBM's Watson to do such a thing about a year ago but I can't seem to find it right now. The gist was much like Watson can be used to find how data and information moved in e-discovery, the same principles could be used to determine which employees were driving innovation and had the most influence in the company.