Last October I wrote a post about how Amazon uses data from sellers like us to gain a competitive advantage. You can pretty much sum up the entire post with this promo image of Amazon’s that I “amended” to reflect our experiences:
We’ve since stopped selling on Amazon, but I’ve continued to receive emails about the post. Once such email was from a graduate student, Baojun Jiang of Carnegie Mellon, who was writing a thesis on the topic of topic. I think it’s absolutely awesome that someone would want to delve deeper into Amazon’s awkward relationship with their sellers. He asked for my permission to cite my post, which I of course gave him, and we exchanged several more emails about the details of our experiences with Tastefully Driven.
The paper was finally published online last week. The math was waaay over my head (it’s been quite a while since I took calculus and differential equations, and even then I’m not sure if I’d “get it”), but I nonetheless found the paper very interesting to read, especially considering that this potentially controversial topic doesn’t seem to garner very much attention.
Here is an excerpt from the paper, sort of a “guest post” by Baojun and his partners:
While millions of products are sold on its retail platform, Amazon.com itself stocks and sells only a small fraction of them. Most of these products are sold by independent sellers, who pay Amazon a fee for each unit sold. Empirical evidence clearly suggests that Amazon is likely to sell the high-demand products and leave the long-tail products for independent sellers to offer. However, for “mid-tail” products, those that it cannot classify with certainty as either high-volume products or low-volume products, Amazon’s strategy is less clear. While Amazon may let independent sellers offer such a mid-tail product, it may be tempted to offer the product directly, especially if the product shows the promise to become a bestseller. Amazon’s “cherry picking” of the successful products, however, gives an independent seller the incentive to hide any high demand by lowering his sales with a reduced pre-sale service level unobserved by Amazon.
A closer examination of product sales on Amazon’s platform reveals an interesting fact – Amazon indeed sells a disproportionately large number of products with high demand. For example, though Amazon directly sells only 7% of all electronics products listed on its website, it sells 64 of the top 100 bestsellers in that category. Digging a little deeper, for the “Digital SLRs” camera subcategory (with 928 products listed), we find that the percent of products sold by Amazon decreases sharply as we go down the list of bestsellers—Amazon carries 16 of the top 20 bestsellers, but only 5 of the products with sales ranks from 150 to 250
For a mid-tail product whose sales potential is not readily obvious, Amazon can initially let the independent seller sell it, track the early sales of the product, and then decide whether or not to offer the product directly. And therein lies the inherent risk faced by a midtail independent seller: If the product sells well, Amazon can observe this (since it processes all sales orders on its website) and is likely to enter and offer the product itself. In doing so, Amazon can eviscerate the independent seller’s market, driving him out of existence. [Amazon’s] agreement with independent sellers allows it to terminate any seller at any time without notice for any reason. But Amazon does not actually have to take such an extreme action. After it starts selling the product directly, Amazon can boost its own sales in various ways. For instance, it can display its own offering very prominently, and given its advantages in scale and not having to pay its own sales fee it typically offers lower prices along with free shipping, etc.
Anecdotal evidence from popular discussion blogs suggests that Amazon indeed does this kind of cherry picking of relatively high-demand products, and that once Amazon starts selling a product directly, independent sellers essentially lose most if not all of their sales to Amazon.
This creates a dilemma for the high-demand independent seller. He may make more profits early on by selling many units of the product, but if he sells too many units, Amazon will learn that this product is worth selling directly, and the seller will lose substantial future sales. Thus, if the seller has a relatively high-demand product, he has an incentive to reduce his unit sales, perhaps through a lower pre-sale service level to the potential buyers. For instance, he may carry a lower inventory level and periodically post out-of-stock notices. He can also devote less time and resources to dealing with consumers’ inquiries about his product or related post-sale services (e.g., he may answer inquiries less conscientiously and with a longer time lag, hence losing some potential sales). Such pre-sale interactions with consumers typically occur outside Amazon’s sales platform and cannot be directly observed by Amazon. Moreover, with hundreds of thousands of independent sellers, Amazon may find it too costly to monitor even the somewhat observable aspects of seller services. Hence, Amazon may face a demand-learning problem for mid-tail products—if it observes not-so-high unit sales for the seller’s product, it may not be able to infer whether or not the product has the potential for high-enough sales to warrant direct selling, because the observed not-so-high unit sales may be due to either a not-so-popular product or a popular product but not-so-good seller services/efforts.
Therefore, the mid-tail products on online retail platforms give rise to an interesting market in which the independent seller benefits from selling on the platform, but he may also be in competition with the platform provider itself. We find that if Amazon’s ex ante belief is that a product has high demand with a high-enough probability, it will set a large fee such that only the high-demand seller chooses to sell on Amazon’s platform. Therefore, Amazon will be able to identify the high-demand product and sell it directly in the later period. But if the probability of high demand is small, then Amazon will charge a low fee such that both types of sellers will sell on the platform. This, however, allows a high-demand seller to mask himself as a low-demand seller by under-investing in service. As a result, Amazon is unable to learn the seller’s true type; in the first period, both types of sellers will have the low-type seller’s first-best sales (conditional on Amazon’s fee), and in the second period, the high-type seller will choose his own first best service level and price. In this case, Amazon will actually make a higher expected profit if it commits ex ante to not selling the product in the future.
Furthermore, we find that both types of sellers may make higher profits if Amazon keeps its entry option than if it forgoes it. This is because if the probability of high demand is not large, the possibility of the low-demand seller will deter Amazon’s entry, while the possibility of a high-demand seller mimicking a low-demand seller results in a lower fee charged by Amazon. In addition, we find that Amazon’s threat of entry may actually reduce consumer surplus in the early period though it increases consumer surplus in the future period. This is because in the case of a high-demand seller, the lost demand in the first period due to his reduced service level to mimic the low-demand seller more than offsets the consumers’ benefit from the seller’s lowered price.