Probabilistic attribution

Probabilistic attribution is a mobile attribution technique sometimes used on Android devices in conjunction with deterministic methods in order to recognize a mobile device, laptop, or browser device. Deterministic methods, such as cookies on the web or advertising identifiers like the GAID on Android, use a unique identifier associated with a device, whereas probabilistic methods rely on collecting behavioral and other user agent data and then attempting to match it with other records.

Since probabilistic attribution relies on trying to match device data with a user database, this technique is typically less accurate and is used as a fallback when a deterministic identifier is unavailable. In addition, the data collected for probabilistic attribution, such as the user’s IP address, changes much more frequently than a device identifier, which means that probabilistic attribution has a shorter attribution lookback period than deterministic attribution. Specifically, the accuracy of correctly identifying a device decays exponentially after 24 hours, so this is the typical cutoff period for the probabilistic attribution lookback window.

As of iOS 14.5, Apple has banned probabilistic attribution on iPhone and iPad.

One of the primary use cases of probabilistic attribution is in mobile web environments since deterministic device identifiers are generally only available in mobile app environments. For example, if an advertiser is running a web-based campaign that is targeting mobile devices and the Google Install Referrer isn’t available, probabilistic attribution could be used.

In addition to web-based mobile ad campaigns, probabilistic attribution can often be used for tracking email campaigns that may be opened on a mobile device, or organic downloads from a mobile landing page.

Ultimately, the goal of probabilistic attribution is to have a fallback in place if a deterministic identifier is not available, ensuring that advertisers can still determine the performance and ROAS of their marketing campaigns. Even if probabilistic attribution isn’t 100% accurate, this additional attribution data still gives advertisers an advantage over those that solely rely on deterministic attribution.

As AlgoLift highlights, this data is critical in predicting LTV and distributing ad budgets accordingly:

Later evaluations of ROAS for particular campaigns or channels should incorporate more matured behavioural data, as opposed to operating on just an early predicted LTV bucket or conversion event. This is where probabilistic attribution becomes critical — so that advertisers can properly distribute their most up to date revenue projections amongst campaigns and channels.

In addition to mobile attribution on specific devices that lack a unique identifier, probabilistic attribution is valuable for cross-device tracking and attribution. Cross-device attribution allows advertisers to understand how their marketing efforts are impacting their customer’s journey, which is often a fragmented user experience involving multiple platforms, devices, and campaigns.

Since probabilistic attribution is less reliable than deterministically matching devices, Singular only uses this technique as a fallback method if the GAID or the Google Install referrer is unavailable. As discussed in our probabilistic attribution FAQ, Singular can collect publicly available data points from HTTP headers including:

• IP address
• Platform
• OS name and version
• User agent
• Timestamp

Singular uses the tracking URL that advertisers use in their campaigns in order to collect the above information. This information is then stored in a database and referenced when attempting to match a device with a specific user action, such as an app install.

Again, post iOS 14.5, probabilistic attribution is not permitted on iOS devices. We have put in new controls that prevent any non-compliant attribution in these situations.

In order to support mobile advertisers using Singular’s platform, we have implemented several requirements to ensure probabilistic attribution is performed in a privacy-preserving manner. The first step to ensure these requirements are met is to reach out to your Singular partner manager or email support@singular.net and they will assist you.