RTB campaigns actually perform well when executed correctly. But many assumptions and hypotheses about RTB 1.0 are simply incorrect. Here are six reasons:
1. Cookies are far shorter lived than thought. The RTB industry structure -- separating data from optimization, and lower-purchase-funnel retargeting from upper-purchase-funnel prospecting -- assumes that cookies are reasonably long-lived. In reality, the half-life of an average third-party cookie is about three days, and one-third of all cookies last less than an hour. Buyers can err by either focusing on only a population subset with stable cookies or buying lists of cookies that likely won't be found again. Stale cookie lists degrade effectiveness and efficiency; real-time bidding requires real-time data.
2. Clicks are a poor metric for display advertising. Reusing search metrics for RTB is asking for trouble. We've found that the clicker profiles for thousands of advertisers of products ranging from car insurance to plane tickets to men's pants to groceries are almost exactly the same. In general, the profile of converters is the exact opposite of clickers, with studies showing that single-digit percentages of online users ever click on ads.
3. Prospecting and retargeting are not separate activities. Advertisers have been encouraged to split marketing objectives into upper and lower funnel, a rational practice if cookies were reasonably stable. But over a couple of days a customer can appear as two or more different cookies because of cookie half-life. The advertising that drove the person to the website, the single hardest task in advertising, won't be properly attributed because cookie churn has broken the causal link. Mistaken attribution causes advertisers to over-invest in retargeting and under-invest in finding new customers.
4. Data are necessary but not sufficient. Along with whom to target (the data or cookie list), it is crucial to know when (purchase funnel management), where (context and placement), how often (frequency) and how much (auction strategy) to pay. Separating the data from the algorithmic optimization and bidding makes the dynamic optimization required in RTB impossible. Comparing the performance of campaigns with these elements integrated to those that combine independent components reveals that integrated campaigns produce a two to seven times lower cost per action (CPA).
5. Data volume matters. RTB still requires petabytes of data, for two reasons. First, data freshness -- less data means more dated insights and greater sensitivity to cookie deletion. Second, targeting models are only as good as the amount and quality of data against which they train. Not all look-alike models are created equal. The amount of model training differentiates the world-class from the mediocre.
6. Machines beat people. The data required for optimal display targeting decisions are incredibly large, and impression volume has exploded. The data Quantcast applies to targeting decisions -- 10 billion impression auctions each day -- is the equivalent of 200 million filing cabinets of paper. Processing and applying that much information is no job for humans and spreadsheets. Use machine-learning techniques to make tactical buying decisions, and let people focus on strategic campaign planning.
An integrated approach to RTB -- data, algorithm and bidding -- across the entire purchase funnel is essential to creating a coordinated set of targeting tactics. The industry's not there yet, but each day moves closer to fulfilling the promise of RTB and targeting -- from RTB 1.0 to RTB 2.0.