6 Things You Need to Know About Fingerprinting for Mobile Attribution [Cheat Sheet]

1) Definition
Fingerprinting, AKA device recognition, is an identification method used in mobile advertising to match a click to an install for the purpose of attribution. It uses uses publicly available parameters (i.e. device name, device type, OS version, platform, IP address, carrier, to name just a few), to form a digital fingerprint ID that statistically matches specific device attributes.
2) Fall-back method
Fingerprinting is a probabilistic model and therefore is not 100% accurate. That’s why it’s only used when deterministic identifiers like Google Play Referrer and device identifiers such as IDFA or Google Advertising ID are not available for attribution (e.g. when the click comes from the mobile web, or when the data is not passed by an ad network).
3) Highly accurate in short term
The longer the time gap between a click and an install, the greater the chance that the user would change a setting in his/her device, thereby creating a new fingerprint (this is especially true with IPs as mobile users constantly change their location). That’s why the attribution window is short, usually 24 hours. In app install campaigns, the vast majority of clicks, installs and first app opens occur within 1-2 hours, in which case the fingerprint is extremely accurate.
4) Used primarily in iOS
As the referrer method is not available in iOS (and almost always available in Android via Google Play), fingerprinting is used more often on Apple devices. With only one deterministic option to work with, there’s a greater likelihood to use the fallback method.
5) Anonymous
A mobile fingerprint is privacy compliant and does not include and personally identifiable information (PII).
6) Fingerprinting 2.0: Machine learning
Going beyond standard parameters and into the realm of big data can take the accuracy of fingerprinting to the next level. The main method to execute this is by having algorithms train on deterministic data to inform probabilistic data. For example, we can learn from clicks that were matched using a device ID and clicks that didn’t match but were probable candidates. However, the right learnings can only be drawn when there is scale. After all, it’s called big data for a reason.