Optimizing SEM vs. Affiliate Channel Investment: Amazon Giveth, Amazon Taketh Away

Last week Amazon announced that it would stop paying affiliates that drive traffic via paid search.

This is worth paying attention to and understanding better.  Amazon is the biggest online retailer.  Affiliates (through its Associates Program, not to be confused with its third-party sellers) are a big part of its marketing program.  And, driving scale / lower costs by encouraging and supporting (paradoxically) the growth of seeming competitors is a fundamental part of Amazon's "Wheel of Growth" strategy (see Scott Wingo's excellent analysis on Seeking Alpha, including the Jeff Bezos napkin diagram that beautifully captures the idea.)

I didn't understand Amazon's decision, so I asked Rob Schmults, CMO at Smart Destinations, for his take.  His answer was an enlightening window into the big leagues of multi-channel optimization in ecommerce, and with his permission I've shared our exchange (emphasis mine).

Cesar:

Why would Amazon do this if they pay affiliates on a CPA basis?  If an affiliate uses a paid search unit to drive traffic to Amazon vs. putting the marginal dollar into a better site, it shouldn't matter to Amazon, because the affiliate bears all the risk on that traffic-driving investment.  Could it be that there's an overall systemic side effect, which is that if everyone spends on SEM, the overall affiliate ecosystem's investment in value-added content helping to sell Amazon stuff is depressed?

Rob:

I think Amazon is doing it because they want to clear out the pile-on around terms they are buying directly themselves. Most companies forbid affiliates from bidding on their "branded terms" as you know (i.e., the name of the company and its branded terms -- to be clear: NOT the names of other brands it may happen to sell). Those terms tend to have an effective acquisition cost well below affiliate fees, so any traffic from those terms going through an affiliate costs more that if they had come through the company's own paid link.

But the affiliates also drive up the cost of that paid link by bidding on the term. Assuming Amazon has purely economic motives for its decision, then it must have decided it has enough keywords that it can buy profitably where the acquisition cost is already below the affiliate cost OR will be once the affiliates are no longer bidding on the terms.

One other economic consideration is double paying. Most companies pay affiliate on ANY sales to someone who came through that affiliate for a period of time (30, 60, even 90 days). And most companies don't carve out sales to that customer where they may have come through another program in the meantime (PPC, e-mail, online ad). In those cases where the affiliate visit came first, but another program was the last click before purchase, it might be argued that you are double paying for that customer.

Here's an example:

I visit your site and come through an affiliate on March 15. I don't buy but I get cookied and you have a 60 program. Sometime before mid-May I click one of your PPC terms and then buy from you. Not only did you have to pay the PPC fee as always, but now you owe the affiliate 8% or whatever.

Because of the economics of PPC, nobody complains when PPC gets clicked first for no sale, and then the affiliate is the second trip that converts. But then the premise of PPC is PPC, not pay for performance!

Anyway, cutting out the PPC riding affiliates would probably eliminate the biggest source of double payments. The bigger affiliates tend to loan you their customers rather than pick off ones opportunistically.

Cesar:

Hey this is really great -- thanks for taking the time to explain it!  I guess at this scale of keyword purchases, etc. the volumes really move the market so optimization matters lots.   One interesting question is whether the time window issue might presage the emergence of a two-tier/staged payment model where affiliates get paid for the click if there's no buy , and later for the buy if it happens in the time window?

Rob:

Interesting thought. I think there could be value in it. One challenge is that it’s a pain to track double payments in an actionable way. You can do the analysis to identify, but not sure how many companies can do it in a systematic way that would be necessary to either stop double paying or go to a variable rate. Maybe not as hard as it seems – just needs someone to put a concerted effort to it. There are real dollars there.

Rob's emphasis on actionability here is of course the crucial part, especially in these resource constrained ways.  One alternative to the two-tier payment structure I suggested would be to shorten the duration of the time window during which you would credit an affiliate for a sale.  Pushing this once step further, you might vary that by product category, to match purchase decision cycles for different products.  On more complex sales with longer cycles (cars), it might be appropriate to retain a longer window.  For less complex decisions (books), it might make sense to go much shorter.

Regardless, thank you Rob for your clear and thorough explanation!