The digital advertising industry is in the midst of a structural change brought about by years of scrutiny around privacy and the impact of addressable messaging. Government policies like GDPR and CCPA and technology increasing limitations on the use of cookies and device IDs have pushed the advertising ecosystem to start thinking of ways to keep its promise of efficiency and effectiveness without showing all the fundamental signals of modern advertising.
Since the release of the IDFA with iOS 7, Apple has allowed users to opt out of being tracked by advertisers. Since then, the percentage of users choosing to be unidentified by advertisers has increased to almost 35% in 2020, according to recent data released by Singular.
As an important clarification, being unidentified doesn’t mean that an individual is unreachable or unattributable. Let me explain why:
In Smadex, we receive more than 1 million requests (or offers) per second from ad exchanges to show an ad to a user. This request comes with metadata that we use to determine if that impression is of value for the campaign. These metadata could be details such as the type of inventory; contextual information like date, time, or location; a unique ID like IDFA, or even whether or not that user is trackable. In this case, we would use the IDFA to determine if we’ve known this user from a previously shown impression or if the user is part of a data cluster that we are interested in based on third-party data or our own database.
So, although many media buying networks rely on their own databases of IDs to target users, Smadex uses machine learning algorithms to learn how each signal affects the results of the campaign, including its trackability. This is why limited ad tracking (LAT) users are known entities for Smadex.
10 billion Limited Ad Tracking Impression Requests Later
In the last 12 months, Smadex has seen more than 10 billion Limited Ad Tracking impression requests. Because of the way Smadex works, those impressions weren’t discarded but instead were saved. All requests were analyzed, combining all available signals and evaluating the relative value of each to each advertiser’s goals.
Each combination of two or more variables receives a score according to platform historical and campaign-specific results.
Let’s see an example. We get a Limited Ad Tracking impression from an IPhone XR with iOS 14 on the weather forecasting app Wunderground, at 3PM, in New York. It has a score of 0,7643 based on impressions from equal or similar combinations of signals. That score will be used to evaluate if that impression is of value to the campaign and how much are we willing to bid for it.
The same process is used to analyze an impression opportunity from a known user, based on its IDFA or AAID, and its combination of signals that include an iPhone 11, at 2:23 AM, for a 300×250 banner on Word with Friends, that could have a score of 0,45.
The outcome of this analysis is the cornerstone of how Smadex works: when giving space to our algorithm to learn continuously, we don’t trust only what we think we know from known users, and our platform has proven that it beats performance results on most occasions.
In order to illustrate this, a double-entry matrix, where we can visualize how the combination of two dimensions correlates with the probability that the campaign will result in positive outcomes.
Many ad networks have historically relied on pools of users to re-engage with them based on those users’ engagement with previous campaigns, leaving LAT users out of their targeting scope.
The industry predicts a massive migration to Limited Ad Tracking by iOS users, so we expect to see an increase in CPM for previous LAT users, and a decrease in CPM for previously retargeted users, creating a balance in the pricing of the ecosystem.
Our data science and product teams are always researching new ways to find efficiencies in our media buying processes. If you want to discuss ideas or learn more about how Smadex works, feel free to reach out to me directly or contact us.
