What I can see working out is google setting some kind of conversion:cost metric. ie. if you average 1 action per 1,000 displays pay $10/action; if you average 1 action per 100,000 displays pay $500. It will be a challenge determining the initial conversion ratio. But by common logic - it might take 10,000 clicks for someone to buy a Merc online and Merc won't mind paying Goog $500 for it. Problem arises when a $40 product takes 2000 displays to get a sale.
Well Google can always put unproven sites on the backburner, sending them very little traffic. 10,000 clicks to sell a Merc is a good point, though, it would take forever. Maybe they could do a backburner system but you can pay to jump the queue? Almost defeats the purpose.
What I don't understand is why they aren't letting these sit side by side Adsense and Google.com ads. It wouldn't be hard to figure out a profit per display metric, although it would need one or two order of magnitudes more displays to calculate. Maybe it will once it's out of beta?
It surprises me that they didn't hook this up to Checkout first (http://news.ycombinator.com/comments?id=2087). It would have been much simpler for webmasters that already use Checkout to get started, and for Goolge to track.
So would it be possible for say, Amazon.com to define their action as the purchase of some esoteric food product? They'd get the benefit of all the page views, but would only have to pay the ransom on the 0.01% of users who buy Amazon milk. It seems hard to address the issue without a lot of human review, for each advertiser.
Actually, it can work on the same principle as PPC, except the $/click takes time to work out (you have to send traffic to the site to determine how much revenue it gets you). Analagous to your example is Amazon biding $0.00001 per click - they would simply be outbid by sites willing to pay more for the keywords. The only difference is that with PPC there's a minimum - I think it's 5 cents, though it might have increased to 10 - but you can define an action as an extremely infrequent event. But that difference only comes into play when you've got no competition, and if you've got no competition for a keyword, it's extremely unlikely that it's going to be a lucrative one.
The key point is that Google figures out how much it makes by sending people to a site, and ranks the sites based on that metric. So just like Adwords, there's no getting around the fact that paying Google more (by bidding higher, or on more frequent events) will get you better placement, and therefore more traffic.
The human review would certainly cause scalability problems. Amazon.com already does PPA with their book referal program. For instance, they pay their ad publishers only when their users purchase a particular book through their site.
How is this different from affliate? It's all CPA (or PPA, if that's what they've decided to call it). I guess in this case you're not paying based on the actual value of the sale, you're paying based on what you bid as the actual value of the sale...
I see this as Google getting into the affilate marketplace, and integrating the adwords buying (which many many many affiliate marketers use) and affiliate payout into one financial transaction.
Hmmn, thinking about this more, it's actually cutting out the middlemen in a vast number of affliate transactions-- those that are entirely reliant on adwords buys instead of eyeballs coming in through a particular publisher. Now you can buy affiliate clicks on Google's entire network, instead of cutting individual deals with lots of sites and dealing with the administrative overhead of managing affiliates.
Google recently revealed that their click fraud is less than 0.02 per cent so it is barely a problem anymore. This means that for every ten thousand CPC, less than two are possible click frauds. Moreover, Google has algorithms that automatically compensate adwords users based on the percentage click fraud.
Ads pay for Google's multibillion dollar business. They are introducing Pay-Per-Action in order to diversify their income channels, not because Adwords is failing.
It wouldn't be a problem if that 0.02% was evenly distributed across all advertisers, but I really doubt that is the case and it's the edge cases that make the news, cause lawsuits and damage Google's reputation.
An example I'm personally familiar with, at www.iamfacingforeclosure.com a few visitors banded together and started clicking on the ads over and over again. Sure enough, the guys Adsense was canned, and that couldn't have happened with PPA.
What I can see working out is google setting some kind of conversion:cost metric. ie. if you average 1 action per 1,000 displays pay $10/action; if you average 1 action per 100,000 displays pay $500. It will be a challenge determining the initial conversion ratio. But by common logic - it might take 10,000 clicks for someone to buy a Merc online and Merc won't mind paying Goog $500 for it. Problem arises when a $40 product takes 2000 displays to get a sale.
Lots of setting standards, way I see it.