Real time auctions have opened a great opportunity for advertisers to reach their target audience. The general ease of setup, scale, and targeting allow for marketers to advertise in ways they could not do before. One tricky part of bidding, however, is how much does one bid? Bid too low and the quality of the domains we can advertise on suffer, and we may not be able to scale. Bid too high, and efficiency and reach suffer, and we will not hit as many people as we otherwise could. We are going to look at some examples in display advertising across different domains, and how dynamic bidding can help make marketing dollars work more efficiently. It also takes the guesswork out of picking a CPM and avoids some of the negative consequences of bidding a flat CPM (Cost per Thousand Impressions).
One notion is higher CPM results in bidding and winning on higher quality domains (where we will use domain as a generic term for a particular website/app/channel). While it is true that the win rate (successful bids / total bids) will increase on higher quality domains, we also win at an even higher rate on lesser quality domains. This may not be intuitive, but just raising CPM means we win more across all domains and win at an even higher rate for lower quality domains. I will illustrate three real world examples of this and show how the win rate varies across all domains as CPM changes.
(Example of an upper-tier display website)
In this example, as we bid below $5 CPM, we win very little, but around $5 dollars then the win rate starts to climb as we bid more. If we bid $5 CPM on this domain, we would win approximately 6% of the time.
(Example of a mid-tier display website)
This domain is more linear and we win more often at lower bid prices. It does also start to climb more quickly at around $5. If we bid $5 CPM on this domain, we would win approximately 16% of the time.
(Example of a budget-tier display website)
This domain is even more linear, and we win more often at lower bid prices. If we bid $5 CPM on this domain, we would win approximately 41% of the time.
If we take all our domains for display and model each domain, and then weight together those domains via the amount of traffic we can get a “model of models” that shows how win rate varies on average across all domains.
If we bid $5 CPM on average, we would win approximately 30% of the time across all domains.
As we can see, bidding a flat $5 CPM results in very different behavior across domains. While it is true we win more premium domains at $5 compared to $4, the win rate for other domains will increase as well. This impacts the number of impressions that we can ultimately afford with a given budget. As we increase CPM the reach of our campaign can suffer, so choosing the correct flat CPM value is difficult.
So where does that leave us? We have a campaign, we have a budget in mind, but what should we bid? If we put in a higher CPM we will win more premium inventory, but also overpay for inventory that we could have won for less. We believe that a better way to think about this problem is with Martin dynamic bidding. Instead of setting a flat CPM, we put in a max CPM. We want to bid dynamically such that we can achieve a win rate across domains that allows us to reach our audience, as well as increase the number of impressions and people reached, while not overpaying for domains that we could have won at a lower price. This allows us to maximize our budget and get the most impressions for your dollar while still hitting the audience we are targeting.
In conclusion, in real time auctions domains behave differently from one another. While there are other factors that affect win rate, domain has a strong relationship to price. On average, the domain explains around 60% of the variance in win rate due to CPM. If we take those models and weight them together by the frequency of each domain, we can also get a sense of the win rate on average across all domains. This allows us to use a dynamic bidding strategy. We can vary the win rate to pace given the inventory available and the budget to maximize the number of impressions that a campaign can reach while controlling for the quality of the domains that are getting bid upon. Compared to a flat CPM strategy this often will result in an increase in the number of impressions served and people reached, while also delivering a better-quality mix of domains for the same CPM. This strategy is available in the Martin DSP platform to maximize the campaign’s potential and is just one way Martin is measurably better.
Jason Taylor
Lead Engineer Research & Analytics
