Also, you're kind of at the advertiser's mercy on how reliable they are about reporting conversions, right? I can't see how the ad network could audit conversions with certainty, unless they were also processing the payments.

I would be very interested in hearing how it's done if I'm wrong, though.

I suppose it's kind of the same with an affiliate program, but on an ad network with thousands of participants, it seems much more uncertain.

We partner with the major affiliate networks to use the conversion tracking they have, We also do regular test transactions to make sure that it works.

> I can't see how the ad network could audit conversions with certainty, unless they were also processing the payments.

A ping script, from the ad network, on the transaction completion page is all that's needed. Pretty straight forward.

No, I meant in such a way where a successful conversion couldn't be hidden by the advertiser. They could very easily serve up a completion page with the ad network script missing half the time.

This is known as "shaving" among affiliate marketers and it mostly doesn't matter. E.g. when you are choosing among 10 offers to monetize your traffic, you split test and keep the one that brings you most money, regardless of whether advertisers are honest or not.

We try to do this automatically. If a product attracts clicks but doesn't convert then we disable it temporarily, to give other products a chance to be shown in their spot.

So if the retailer is cheating, then their products should not be shown very often - and products that do convert should be shown in their place.

You still require honesty from the advertiser. Whats to stop the advertiser omitting the script on every nth completion page to reduce their outgoings? Maybe omitting it for high value orders as test orders are likely to be low value, whitelisting the ad networks known ip ranges etc.

> You still require honesty from the advertiser.

Yes. And full-circle dishonesty is inherent in upstart/third-tier ad networks (not necessarily the network, but their advertising clients typically).

But in the semi-perfect world of advertisers really trying to sell product, and ad networks with code optimized to make that happen, it works well. If product "A" from retailer "X" is converting well, the system should promote that over products that aren't converting -- and the only way to automatically know is the action-complete ping.

But yes, deception is one of the primary reasons content publishers avoid CPA networks.

Test it, measure your revenue per 1000 and dump it if doesn't work for you. There's plenty of competition in this space.

Isn't this similar to Skimlinks i.e. you don't get paid by them until conversion?

Yes, they are also CPA.

We've also got something similar to them, in that we can rewrite links on the fly, converting them into revenue-earning links, if they match one of our merchants.

Yep, Skimlinks also do the link conversion as well.

And so does VigLink, and arguably at greater scale...it takes scale to do this well. VigLink claims to have over 30,000 merchants.

You are right, it does take scale to do this well. I'm CEO at Skimlinks, so although a little biased, I can assure you Skimlinks is at the same if not greater scale. Our uptime and speed is the best in the industry, as you can see from our public pingdom page http://api-status.skimlinks.com/554403/2012/10 And all this is while processing 230million clicks a month. Even Hurricane Sandy didn't affect our uptime. Our CTO wrote a great piece about it here: http://blog.skimlinks.com/2012/10/25/the-importance-of-globa...

Yes.

We track the clicks and if a product shown on an advert is getting lots of clicks but zero conversions, we swap it out and show alternatives instead.

So we try to make sure that only stuff that will convert is shown.