The Fallback Method for Matching Installs to Clicks Using Known ID Parameters
This identification method uses publicly available parameters such as device name, type, OS version, platform, IP address, and carrier to form a digital fingerprint ID that can be analyzed and attributed. If any of these attributes are changed (e.g., a user upgrades the OS version or changes IP addresses), a new fingerprint is created and the old one is no longer usable.
Based on statistical probability, fingerprinting is not 100% accurate and therefore used as a fallback method when other identifiers are not available. But within its attribution window of up to 24 hours it is very accurate. Since the vast majority of users click and then install an app within hours, this window is usually enough. Beyond this short attribution window, its accuracy level drops as users are more likely to change device parameters.
To increase the accuracy of fingerprinting, the length of the lookback window is set based on the population of mobile devices with the same IP address. A “uniqueness” rating can then be assigned to a particular IP address to help increase the accuracy of the attribution. For example, if there are two people (at home) sharing the same IP address, the IP uniqueness rating would be quite high. And the lookback window would be longer. If there are 10,000 people who share the same IP address (in a public place), the uniqueness rating is much lower. You might either set a very short lookback window or disregard this population of devices altogether.
To optimize fingerprinting, your attribution window should be dynamic, allowing you to grow, shrink or close the window based on the population size of devices using the same IP address.