Cross Device Tracking – solution approaches and metrics for tracking

Cross Device Tracking – solution approaches and metrics for tracking

At a time when many users no longer have just one device with which to explore the Internet, marketers have to learn to work across platforms. An overview.

Multiscreen, second screen effect, pixels, personas, matches – the list of terms that revolve around so-called cross-device marketing is getting longer every week. Ultimately it's just about solving one problem:

When users are on different devices and some purchasing processes take place across multiple platforms – how can marketers keep track of things and publishers be compensated “fairly” according to their contribution?

Without wanting to go into details, it quickly becomes clear: a cross-device tracking solution is needed that is able to track users across multiple stations and devices and thus assign any sales and leads to the appropriate affiliates.

Approaches for cross-device tracking

First of all, there are two general concepts when it comes to cross-device tracking:

Deterministic: This approach simply means that users must be logged in to a website or app for reliable tracking. The user account is tracked. The best-known example is certainly Facebook. If you log in here via smartphone/desktop and do not explicitly log out of one of the devices, all actions are tracked across the various devices.

In short: There is no algorithm used here, a direct connection between the user and the device is required.

Probabilistic: As the name suggests, this method is a probability-based approach. From various data such as IP, device, location, browser and a large amount of other information, an algorithm calculates the probability that a specific user and various devices belong together. We're obviously fishing in the dark here - but the approach allows a cross-device campaign to be run, even without very large amounts of data. Not necessarily an exact science.

It quickly becomes clear: As soon as you have reached a certain number of users and collected the appropriate data on them, the deterministic approach is to be preferred - if you can afford it. However, with security concerns becoming more and more common and users increasingly using ad blockers, collecting data is not necessarily becoming easier - so the probabilistic approach also has its justification, it depends on the campaign and target group.

Tracking metrics

As with all forms of online marketing, the following applies here: in order to be able to at least partially predict/measure the success of a campaign, you need the right data. To do this, however, you have to be clear about what each measurement actually means and for what type of campaign it is of interest.

Match Rate: A term that divides opinions. What at first glance probably describes a relationship between possible and actual matches (assignment of user to device) turns out to be relatively undefined in practice. Nevertheless, the term is used often enough without being explained in more detail.

Accuracy: Accuracy describes what most people imagine by “match rate”: a ratio between possible and actual matches. However, taking into account the potential “non-matches” in a graph (see below). Since, depending on the number, the probability that the user and the device do NOT belong together is significantly higher than the other way around, but this probability is included in the calculation, “Accuracy” is unfortunately not really useful as a measurement. There are exceptions.

Precision: Precision describes what most advertisers actually expect from a graph: A correctly described ratio between predicted and actual matches, where the number of matches is divided by the number of all matches plus non-matches:
Formula precision
Recall: Says something about the quality of a graph by comparing the number of all actual matches in a given region with the number of matches recognized as positive in the graph. In short: How meaningful is your own database?
Recall formula

Cross-Device Graph: The database in which all matches, regardless of the method used to determine them, are recorded. The size of this graph often determines which tracking method is used. As already mentioned above, once the number of users reaches a certain number, the deterministic approach becomes more and more accurate. Pixel: Similar to other methods, an “invisible” pixel is used to collect data such as device, IP, browser, language, etc. in order to compare them afterwards and find possible matches.

Of course, cross-device tracking providers can manipulate these metrics through different weights and priorities for their customers. Depending on the campaign goal, accuracy, recall or even precision may be more in focus. (e.g. sales vs. branding).