This question came to my private inbox but I thought it better to answer it here.
I hear a great many people complain that PPC "doesn't work" but in 99% of cases it is just not being used right. Just because you don't know how to work something doesn't mean it "doesn't work".
Adwords and Analytics together are a powerful combo where you can root out the niches that can radically reduce your CPC. So while you might lose money by targeting 'Keyword X' in every search you might make money if you target 'Keyword X' to New York residents using ipads between the hours of 10pm and 6am at weekends.
But how do you know that? At first you don't so you are going to have to throw some money at Adwords to build up enough data in order to draw conclusions. The most common mistake is to draw conclusions on too little data.
A general rule of thumb is that n=30. You must have 30 pieces of data to draw a conclusion. In this case you must have 30 visitors who came to your site via PPC from New York, using an ipad between the hours of 10pm and 6am at weekends.
Personally I regard n=30 as too low. The lower your number the more 'degrees of error' there are in your conclusion. You see these every day in political polls - "Party X is supported by 40% of the population" and then in small writing somewhere "+/- 3%"
That "+/- 3%" is the degree of error. The polling company did not ask everyone in the country, they usually ask 1,000 people which is enough to give them a fairly small (3%) degree of error.
Same goes when you analyse your PPC and hence why I go for n=50. More means more accuracy.
Now when you think about it - 50 ipad wielding New Yorkers in the middle of the night + 50 Android tablet owners in lunching in LA and all the other combinations that might occur, you are going to need a lot of clicks to draw proper conclusions, thousands.
If a basic search for a keyword is requiring $5 CPC that's thousands times $5. Adword Planner will do some of the geographical filtering but the rest can only be done once you have made Google richer.
But we are not there yet!
We also need to consider if those Apple loving New Yorkers are only performing better because your landing page appeals to them and our tablet owners in LA might perform even better if they had an alternative landing page.
So say your first cut cost you $50,000. Then you need to run the test again with different landing pages and different adverts and analyse that. Something like this:
For keyword X
- Advert A -> Landing Page A
- Advert B -> Landing Page A
- Advert B -> Landing Page B
- Advert A -> Landing Page B
Now you can see it is pretty easy to clock up $200,000 just to gather the data you need and you still don't know if it will reveal any promising niches or you are going to have to introduce Advert C and Landing Page C to find out more.
But this is what the big names do. They invest heavily and then save substantially in the long term. By doing so they also create a barrier to entry for newcomers who don't have that sort of budget to start out.
So why is PPC so expensive sometimes - because there are too many novices running about in there, burning up all their money on untargeted campaigns and then complaining it doesn't work ... and Google isn't complaining!