ExchangeWire columnist Gareth Davies, AdBrain, CEO and co-founder, explains: 'WTF are statistical IDs, and why should I care?'

In an increasingly mobile-first world, the proliferation of mobile device adoption presents a major challenge to digital marketers who traditionally relied upon desktop cookie tracking to target and reach their wired audiences. While the cookie isn’t dead just yet, it’s certainly on its last legs as a new battle is being fought by tech goliaths and upstarts alike in a bid to better understand multi-screen consumers. The prize; a holistic audience profile. This way we can connect mobile users with other devices, to create a single cross-device audience ID, and in return more relevant and impactful ads for consumers, not to mention a greater share of the digital ad dollar.

Solving this problem requires overcoming massive technical feats, both for deterministic data owners such as Google, Facebook et al to tie their vast behavioral data sets together and for statistical technology vendors to build broader cross-device profiles from first, second and third party data via artificial intelligence and machine learning algorithms.

To better understand what all of this means and why we should care, let’s dig a little deeper into both approaches.

Deterministic attribution

Deterministic attribution is an accurate audience identification model that uses rich troves of first party data collected from user logins. The tech goliaths have this by the data-center boatload. Google, Facebook, Amazon, Rakuten et al have plentiful device IDs, behavioral, location, and other key data sets collected through individual user logins. When a user is logged into their respective accounts across mobile, tablet, and desktop it creates an accurate, real-time picture of that individual user across devices. The single sign-on is therefore a complete solution for mapping users across mobile, tablet, and desktop making cross-device targeting a theoretical reality, today.

Trouble in paradise

The ability to map user devices via first party data is extremely valuable but this solution is far from perfect. Deterministic attribution operates exclusively within walled gardens. Great when you’re delivering ads within any of the tech giants’ walls but the limitations aren’t so rosy. Audience identification is restricted to a single tech platform and that specific audience cannot be identified – transparency is at best partial, at worst nonexistent. There are important data ownership implications here too; plugging in first party data into platforms you have no control over and the fragmentation of audience data across platforms. Not to mention syncing and buying your audiences across a multitude of isolated platforms, negating the operational efficiencies that programmatic advertising offers.

Statistical identification

Statistical, or probabilistic, identification is a form of device recognition technology allowing advertisers to identify both mobile and multi-screen audiences in the absence of cookies or other deterministic data.

Delivered through JavaScript tags at the impression and click level, these technologies capture a wide range of common, non-personally identifiable device data points. The model relies on key data points such as Apple’s IDFA and Google’s AndroidID serving as mobile anchors alongside rich location, environmental and behavioral data to feed algorithms that graph the probability of a user being connected to device 1, 2 and 3. This model is fundamentally different than deterministic attribution in that no PII is ever collected and the process of collection is not necessarily controlled by any one ad-tech platform.

Open works

It’s important to highlight the open nature of this identification methodology, and the scale it offers advertisers to reach and control cross-device audience delivery. Given that statistical identification is a technology-driven methodology and not a black-box solution, inevitably there are numerous ad technology providers independently working on better ways to target mobile audiences. It’s generally accepted that statistical identification has an accuracy of 60-90% – not perfect, but bound to improve with further innovation, creating a strong argument for greater collaboration across vendors, particularly around implementing an open industry standard for statistical cross-device identification.

What this means for marketers

Deterministic attribution and statistical identification are two credible approaches to the multi-screen challenge, but neither would be possible without a key ingredient – data. Marketers need to identify and reach their audiences at scale while retaining control of the buying process. Working with deterministic models usually requires sacrificing scale and control of data, while statistically modelling requires that one is comfortable with a degree of uncertainty.

Ultimately we have a way to go before a clear winner emerges, but perhaps a more nuanced approach is yet to come, where the advertiser marries user opt-in data with scalable, media-agnostic statistical tools to make highly educated guesses that work. For this to happen advertisers need to take control of their valuable data, whilst using open, privacy safe statistical identification models to deliver scale. Done right, the whole ecosystem emerges triumphant.