Dion Hinchcliffe (emphasis added by me):

But most organizations already understand that spreadsheets, presentation files, e-mails, word processing documents, and private databases are where much of the valuable institutional information is.  While centralized "big IT" systems do a lot of routine record keeping, the heart and soul of an organization in the form of corporate strategies, product development plans, project notes, key performance metrics, and so on is really kept in e-mail folders and user's directories.  And while some of it must remain under strict control, particularly in public companies, much of it is unnessarily — and usually to a fault — hidden, unreused, and unexploited.

I like this quote, as it ties into the "datasphere" concept we've been tossing around, which says that the way to a semantic web is not through data formats but through the HTML world we're already familiar with:

Discoverability isn't an afterthought, it's the core. 
Google and other search engines made the Web usable.  The enterprise has not caught up, largely because most enterprise information doesn't allow a hyperlink structure, and links aren't encouraged very much when it does.  McAfee recommends setting up blog and wiki directories as well as good enterprise search based on link ranking (which is what Google does to make the right information come up in the first few pages of search results.)  Enterprise 2.0 tools should also extract folksonomies and other structural information (from microformats and XML tags) into discoverability mechanisms like tags lists and clouds, making user organization schemes obvious, public, and emergent.  One easy trap to fall into is to assume your existing enterprise search will do the job.  It probably won't, so be sure that it's well integrated into your Enterprise 2.0 effort, perhaps by offering a blog or wiki search option. Provide your own search engine in the tools only if you must.