We still find that differing keyword tools can give you very different results.   Keyword Discovery, for instance, gives (on occasion) millions of searches for a keyword that Overture says nobody has searched on.   Overture also often makes no distinction between singular and plural keywords or (often) the order of the keywords (surely this is precisely what marketers want) so that "uk flowers" will give you no result at all but instead give you the results for the singular re-ordered keyword "flower uk", are they selling the four individual keywords "uk flowers"; "uk flower"; "flowers uk"; "flower uk" all on the basis of the results of the four combined?

OR if you were to put (in October 2007) the two seperate keywords “ring tone” in the Overture's UK Keyword Suggestion Tool (see Tool Two below) its Keyword Assistant will tell you there were over two million searches (2,022565) for the combined term “ringtone” (searched during January 2007).  

On the other hand if you put in the term “tone” you will be told that there were searches for the seperated keywords “real tone” “wolfe tone” “true tone” “free tone” “two tone” but none for the term “ring tone”
(searched during May 2007)

  I don't need to tell you how massive these differences are and how completely inexcusable this is if the smaller figures are more accurate or if both “ringtone” and “ring tone” are being sold on the basis of their combined searches.

It is difficult to understand what is going on with Overtures algorithm.   For instance, when you search (in October 2007) for the keywords "womens car insurance" in their UK Keyword Assistant you are told that that there were 420,841 searches made during May 2007 not for your keywords but for "woman car insurance".

But then when you actually search for "woman car insurance" you are told that that there were 420,841 searches made in the month for "car insurance womens".   It is clear that there never was a single search term that was searched 420,841 times and that this figure is produced by combining at least two (my guess many more) search terms.

The best policy is probably to only trust keyword frequencies that are universally consistent and don't completely trust the figures from the agency that is selling services based on them.