"Cost Per Whatever" to displace CPC and CPA

Google Checkout and experiments in "cost-per-action" (CPA) advertising bring Google several steps closer to a "pay-per-click-you-like" future.

What's that, you ask?

Well, take CPC (cost-per-click) and CPA advertising as points on a continuous curve: CPC pays for every (or every 'legitimate') click, CPA only pays for clicks that result in sales. But in fact there are as many kinds of clicks as there are clickers, and every one has a different value. Further, this value is only evident at some time after the click is delivered. Was it a shallow click -- a confused customer or fraudster? Was it a valid lead that left contact info but may not purchase for weeks? Was it a small sale? A large sale? A sale that ultimately resulted in a return?

With that insight, we can take a radical leap to a new model for pricing clicks: pay whatever you feel like, and only after you've decided what the click was worth to you. Pay nothing for clicks you don't like, after you've had time to evaluate them. (No more anguished appeals to PPC advertisers for fraud investigations!)

"Huh? Wha? It'll never work!" you might protest. People will lie about what clicks are worth; they'll game the system; it's too complicated.

But think it through, and you'll see there's a better alignment of incentives, and less opportunity for fraud and gaming, than with the current discontinuous and fixed-price CPC and CPA systems.

Could an advertiser game Google, by underpaying for valuable clicks? Only at threat of having those exact same kinds of clicks dry up in future periods. If you want more of any particular click, you have to pony up. Conversely, by refusing to pay for worthless clicks of any kind -- by whatever criteria makes sense for your business, not magic secret 'fraud-detecting' algorithms deep in Google's bowels -- you'll send Google the best feedback possible about what clicks (and ad inventory/impressions) you want. Google won't be guessing what clicks are bad with algorithms; they'll be knowing because customers will have told them directly.

Can a fraudster bilk advertisers? Only if they manage to simulate valuable clicktrails through target sites -- a much harder problem against diverse real businesses than generating simple clicks. And any business fearing fraud has the ultimate countermeasure completely in their own hands: just clam up and only pay for real, verified, unreturned sales.

Is it too complicated? Small and unsophisticated advertisers can still pursue degenerate strategies: pay for every click, or pay for only sales. They can gradually vary their payments as their sophistication grows, or use free or subsidized tools like Google Analytics and Google Checkout to improve their pay-strategy. And yet, they'll still benefit from the info spillover of the sophisticated buyers. Google will learn from the 'pro' customers, who use powerful click-valuation software, which clicks are truly valuable. Further, it is in Google's self-interest to even send 'dumb' CPC buyers just enough true value to make their ad spend worthwhile (or else lose the customer).

And Google? Such a system takes them even further towards being a giant and dominant click-routing brain, sending impressions to wherever they have the greatest expected click-value return, based on past payment patterns. As I noted last year in a post introducing an earlier variant of this idea:

Google is uniquely positioned to implement and benefit from such a policy. They already rank AdWords ads not by raw bid, but by actual profitability over time. They already reserve for themself overwhelming placement discretion, and keep raw performance, price, and payout rates to themselves. All they ask of advertisers is, "are you getting sufficient value for value paid?" -- and that'd be the core proposition at the heart of a pay-per-click-you-like system, too. (Pay no attention to the half-million computers behind the curtain.)

The patterns of unpaid-for clicks would become just another massive and valuable data stream for Google to mine, along with web crawls, query logs, toolbar data, per-browser/per-IP address web-trail info, and whatever other tricks they have up their sleeves. Their core competence is discovering patterns that reveal implied value in giant dynamic datasets, and there's market power in having the biggest dataset.

So the day Google adopts this kind of policy, click fraud ceases being a major problem for them and starts to be a giant club they can use to beat off their smaller competitors. Who else will have as large and detailed a map of which clicks are wanted? And once you've been with Google for a while, and fed it months of data about the clicks you like and don't, switching to any other advertiser would involve a big cost and efficiency hit while the patterns are relearned.

Call it cost-per-whatever (CPW). Let every click find its own value! But watch out for the proprietary click-intelligence monopoly it could create.

(Previous entries on this topic: Killing click-fraud (and the competition) with one stone [March 14, 2005]; Google Analytics & the "pay-per-click-you-like" future [November 14, 2005] .)

Tags: , , , , , , , , , ,

great thought piece gordon. enjoyed the read, and the thinking behind it.

particularly like the way you position CPC / CPA as points on the 'pricing possibility curve'. the efficient frontier for internet advertising ;)

don't know if we've ever met before, but might be fun to chat sometime, perhaps at an upcoming internet conf or event. maybe see you around...

- dave mcclure
Post a Comment