This week’s guest blog post is contributed by Jonathan Seidner, Senior Director, Engineering, Oracle Data Cloud.
Because of the nature of probabilistic device maps, it is possible for them to indicate a predicted likelihood that any two matched devices are being used by the same individual consumer.
As opposed to a simple listing of binary (yes/no) device matches, as provided by all deterministic device maps and most probabilistic device maps, the inclusion of match likelihood figures provides marketers with a percentage score indicating the probability that two devices are matched to a single user.
This allows the marketer to tweak the matching thresholds and/or bidding algorithms differently for each application.
Beyond selecting a device map based on its particular trade-off between precision and recall for one or more desired applications, a device map that provides match likelihood figures for all indicated matches enables marketers and analysts to fine-tune how the device map is applied for each application.
One example of using device-match likelihood figures is setting a threshold for CPC retargeting campaigns: a marketer could program his CPC retargeting campaigns to show ads to additional devices of previous site visitors whenever the match likelihood is over 50%, while not risking the bids on matches with a device-match confidence less than this threshold.
Another example is in the area of cross-device conversion attribution, where marketing analysts generally want to ensure that resulting metrics are highly accurate.
In this type of application, the analyst could decide to attribute cross-device conversions only when the device-match likelihood is greater than 85% or 90%.
Taking this concept one step further, marketers can actually utilize the match likelihood figures within their algorithms to customize application parameters to individual devices.
In other words, device-match probability scores can be used as granular signals for optimizing programmatic treatment of specific devices.
For example, a programmatic retargeter can optimize campaign ROI by tying bid amounts to device-match confidence scores: the more likely that additional devices belong to a particular known user, the higher the bids can be to show ads on those additional devices.
The results of this approach are generally highly effective campaigns at a significantly lower overall cost.
The availability of predicted match likelihoods, a “feature” of good probabilistic device maps, allows marketers and analysts to tweak how the device map is used in each use case, for optimum business performance.
For this reason, it is usually very valuable to select a device map that contains numeric scores for every included device match.
Image: ESB Basic/Shutterstock