As a company, we're obviously very firmly in the pro-RTB camp. There are a number of reasons why this should be the case, including:
- Maximize the value of an impression - this means that all sources of demand could be compared concurrently for a given auction, meaning whoever would bid the most in an eligible bid would be the winner;
- Minimize latency - this compares to the previous system of ad passbacks, where each network would have the chance to fill at a specific price in successive redirects, rather than all at once (this is related to the value maximization question); and
- Consolidated ad quality - if there's only one auction (which, admittedly, isn't quite the case today, but it's the theory), then the publisher would only need to input their various settings regarding floors, eligible creatives, etc in one place.
It is important to compare RTB not to the ideal case of what could be, but the previous system that existed, when judging its merits. As discussed above, prior to RTB, publishers would generally cascade ad calls in ad networks, allowing them to buy impressions for at least a certain price, and then pass back to the next network if not. This often meant there would be several hops in the chain - causing material latency and discrepancy issues - as well as failing to capture the absolute maximum that would be spent for an impression. It created numerous arbitrage opportunities, as ad networks would jockey to get the right place in the waterfall where they could potentially buy cheap impressions and resell them another way for profit.
RTB came along 10-ish years ago and changed a lot of this behavior very quickly. The auction did a good, but not perfect job of maximizing yield and was quickly embraced as the next generation. Below we discuss certain drawbacks of RTB - or potentially different ways of thinking about monetization.
Real-time bidding is fundamentally about maximizing the value of a given impression. There is no concept about the lifetime value of a user in RTB. As a result, for a given impression, the highest price will generally win. In the most extreme example, a pop-up could win because it would pay the highest. This would be by far the most annoying for a user and ultimately drive consumers away. But this concept isn't factored in. There are various approximations, like segmenting users based on the referral source - for example, users from social platforms that are less likely to be loyal may be eligible for more intrusive ads, while those that are loyal readers maybe be in another segment that does not see interstitials. Other approximations include ad quality rules that set up parameters like "only show 1 interstitial per user per day." This certainly gets closer. But fundamentally there is a separation of concerns - RTB is only about maximizing the valuation based on eligible bids. It is not about determining what those concerns could or should be - and an unsophisticated publisher may not have the wherewithal to figure out these more advanced setups.
The fundamental mechanism of RTB is currently the 2nd-price auction (though arguably first-price would be better). The goal of a 2nd price auction is to ensure a fair price is paid by incentivizing the participants to bid honestly. But this outcome requires certain assumptions, including that all the bids are actually submitted and that it is an isolated auction independent of any others. Both are false in the case of RTB. As the number of DSPs shrinks and the demand is consolidated into ever fewer platforms, each platform actually conducts its own auction before submitting its winner to the exchange. One could imagine an example where there are 10,000 sources of demand only bidding into The Trade Desk (TTD) and Google's DoubleClick Bid Manager (DBM). Both TTD and DBM would each submit 1 bid, so only 2 bids would be submitted on behalf of 10,000 buyers - meaning the 2nd price in the exchange would not be the actual second price of all buyers and what they would bid - just the 2nd price of two different bidders. Further, DSPs use the results of prior auctions to inform their future bidding. If they know they can win a similar auction - based on the domain, device and user characteristics - with 80% certainty at a certain price, and 70% certainty at a much lower price, they might be that might lower price because they don't need to win all the time and would rather pay much less. The DSP is thus not bidding what the impression is truly "worth" to them, but rather some alternate number.
Comparison to the adserver
Up until very recently (this is solved to some degree by header bidding), RTB was wholly separate from the publisher's directly sold inventory. In the past, the publisher would sell some version of a guaranteed campaign - allowing an advertiser to have high confidence it would get a certain number and type of impressions - for a high priority, before RTB could compete. These campaigns would almost always be trafficked in DoubleClick for Publishers (DFP). DFP is a very full-functioned platform that does all sorts of forecasting, prioritization, etc that publishers have grown to rely on. So RTB could only get a crack at the auction if the publisher's guaranteed campaigns had served to completion (often there are expensive make-goods if they don't deliver in full). In certain cases, however, RTB demand would be willing to pay significantly more for impressions than the IO campaigns in DFP. In the classic version, this wouldn't be allowed to serve. Google solved this by allowed Google demand to compete in DFP when campaigns were pacing sufficiently well that there was a high likelihood they could deliver. This was largely considered unfair - and header bidding came out to enable RTB to bid on every impression and compete directly against DFP. That said, there is now the challenge (I think...) that RTB via header bidding won't necessarily turn off if it's unlikely that the guaranteed campaigns won't deliver.
Commoditization of inventory
One of the big complaints about RTB was that it would commoditize publisher inventory. Buyers would buy only looking for the audiences they wanted, and not necessarily care about the publishers that they ran on. To a large extent, this is exactly what happened. RTB was then beset by fraud and viewability concerns, which were largely addressed through whitelists of publishers and measurement vendors. This created a situation where RTB buyers were now buying whatever they wanted on the sufficiently good publishers. Prices went up, but not a ton. Often the very best inventory was held back by publishers as they sought to sell direct, guaranteed deals. But in chasing improved monetization, publishers enabled header bidding across their inventory. RTB through header bidding has created an interesting unwinding of premium publisher inventory. Now header bidding allows all inventory to flow through to the exchanges - meaning that what was once only available directly can be purchased with increasing confidence and likelihood via the exchange.