“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is written by Bram Woolcott, product manager at TripleLift.
Supply path optimization (SPO) is about finding the most efficient path to inventory, but there is no consensus about what efficiency actually is.
Demand-side platforms (DSPs) simply optimize for different things. Some look at dollars spent per queries per second, meaning how much the DSP spends for the impressions it sees from a source. Another DSP may optimize for win rate, or how often the DSP wins for each bid it submits.
There are dozens of permutations, each of which result in a different set of impressions being sent to a DSP, thus yielding different outcomes for every participant in the value chain.
Considering and understanding the different human and machine factors that contribute to these permutations are critical for marketers when evaluating the most optimal paths to supply. While marketers have specific objectives for their campaigns, the way those objectives factor into the DSP strategies may be entirely at odds.
The concept of how SPO decisions are made are still fairly opaque, and understanding which strategies are in play for each individual impression can be quite difficult. A better holistic understanding of how these decisions are evaluated throughout the supply chain can help marketers more effectively accomplish their goals.
Looking just at win rate, two otherwise identical exchanges may choose completely divergent strategies. One may send to a particular DSP only the impressions that it expects will receive the fewest bids from other DSPs. Those impressions may be the least attractive but also the most likely to be won because of low bid density.
Meanwhile, another exchange may choose only the “best” impressions, meaning those that are expected to receive the most bids but that will also be hardest to win.
Another exchange may route impressions via an ad server’s server-to-server platform. This has twin benefits. First, it would avoid sending requests for sponsorship impressions. Second, it would include the winning header-bidding bid bucket as a floor to beat, allowing DSPs to avoid responding when they don’t have a bid above this floor. Bids above this floor would be much more likely to win the impression as the floor price is set by the highest bid from the combination of all other eligible demand. Both factors would increase win rates without increasing winnable impressions.
While all three strategies achieve the DSP’s goal of reducing the number of impressions it receives, none are directly related to the marketer’s objective.
The first sends impressions that are less likely to be highly valued by the marketer.
The second sends impressions that will have highly competitive internal DSP auctions, after which the impressions may be easier to win. But there will be a false signal: The win rate will imply that the marketer can win more impressions. The marketer can indeed win more but only in cases where it wins the internal auction, which it will often lose. In other words, optimizing to a win rate that depends on winning the DSP internal auction will actually limit reach, because it focuses on auctions that are less easily won.
The third strategy adds an extra fee without necessarily improving the impressions that can actually be won. The floor price from the header bidding auction price or last look would also lose relevance with consolidated first-price auctions, which may drastically change the SPO mechanisms that favor these paths. This dynamic will be closely monitored by publishers as they work on strategies, flooring or otherwise to ensure their inventory is properly valued.
Test, test, test
Ultimately, a marketer wants to buy the right audience – determined either by context, audience/user data or something else – at the right price. Marketers should divide the audience into equally and randomly distributed, non-overlapping buckets, then analyze what percent of the actual audience can be purchased through various exchanges, on different publishers and at different price points.
While there is a degree of operational work to get to a place of testing this distributed division of audience buckets, it would be in a marketer’s best interest to get a holistic understanding of where their ad dollars are flowing and which paths are the most effective. Without undertaking this analysis, the end-to-end life of an impression is a fairly nebulous concept, and the effectiveness of each bid is hard to quantify.
This methodology allows marketers to skip questions that ultimately aren’t germane to the determination of which path gets them their target audience at the best price, the most often.