Indexing Word documents for plain search or natural language search is fairly easy, there are many packages commercial and open that will do this. I think what you are asking for is a way to identify concepts and relationships between segments of a document, say paragraphs that have 50+ words. I have only personally seen one software package that will do this, IBM's Content Discovery Server, formally iPhrase. The idea is that you can try to define "buckets" of concepts, and this software can sort incoming segments of text into those buckets. The downside is that it won't readily identify concepts outside of your defined set.

IBM also has donated an open source method for indexing concepts also, hosted at the Apache UIMA site. I haven't looked at this too closely, but it probably wouldn't be easy or fast to set up in a production system and have something useful.

Your best bet is probably a search engine that will index Word documents and perform some basic natural language search techniques, like word variants, spell checking and stemming. When your sales guys want other documents to look at, search through the previous ones with the same keywords and present snippets of the documents in the results so they can quickly find what they are looking for. Generally a commercial natural language search engine will perform better than any free one.