I like the Dr. House/diagnostic medicine heuristic the best: you need to do something when you've squeezed all the possible intelligence out of the data you already have.

Clay Bodine is making the equivalent simpler distinction in domains where more data = implementation, such as product launching. But I prefer the more general answer based on information and feedback. Implementation is a special case, you provide others with something (a product) that is likely to change their behavior in ways that give you data.

Ultimately implementation is really about getting more data. For domains where "implementation" isn't really distinct from reflecting, such as writing, communicating or marketing (it's all thinking and words), diagnostics, detective work, investigations etc., the real distinction lies in data. These days I do no implementation of anything, and I try to stay sharply sensitive to my data needs.

In business, the ultimate form of data is money. When you make cartloads of money, you are getting cartloads of data. Smart money carries more bits of information per incoming dollar than dumb money. The relation between information and money (and entropy) is a big interest of mine these days.

Dr. House was inspired by Sherlock Holmes of course, and you might recall why Sherlock's brother Mycroft didn't do much detecting: he was unwilling to get out of his armchair when data collection was needed. On the other hand, he was the invisible hand pulling strings in areas where data could flow easily to him, such as deep government affairs.