Introduction to Deterministic and Probabilistic Identification

Reports predict that each of us will have five internet-enabled devices by 2017. For advertisers trying to reach audiences in a measurable, targeted way that’s a hell of a lot of fragmentation to overcome.

Frequency (the number of times a user sees an ad) is an important part of brand campaigns, so advertisers need to be able to match multiple devices to the same user. There’s also a huge opportunity for advertisers to deliver ads that compliment and evolve alongside the device experience a user is having at a particular point in their day.

However, it’s easier said than done. The two main avenues for identification, or device matching, are deterministic and probabilistic identification. Here’s a summary of what they are, how well they work and what the strengths and weaknesses of each are.


What is it?

Deterministic identification uses personally identifiable first-party data to recognise unique users across multiple devices. The data is known to services users have signed up to, and made available to advertisers via anonymised IDs specific to those services.

Services with scale are best positioned to realize deterministic identification and deliver advertising against it, the best examples being Facebook (Facebook User ID), Google (Google Advertising ID) and Apple (Apple Advertising Identifier).

How accurate is it?



It works, at least within the walled-gardens of the Google, Facebook and Apple ecosystems. Advertisers can execute campaigns that reach a unique user at different points in their daily device usage. There’s also positives for media companies here, who see the benefits of premium CPMs delivered through these closed advertising platforms.


This accuracy of user recognition is based on personally identifiable information, which presents privacy issues. Another problem is that while deterministic identification is great for advertisers executing campaigns within the ecosystems above, campaigns across multiple platforms require individual execution. This is a pain.


What is it?

Probabilistic identification combines a number of anonymous pieces of non-personally identifiable information (PII) to recognise a unique user across multiple devices. These datapoints (see below) are fed into algorithms that output a percentage probability that a user matches an existing unique profile.

Parameters in probabilistic device recognition:

  • screen resolution

  • device type

  • operating system

  • data points such as Apple’s IDFA and Google’s AndroidID serving as mobile anchors

  • rich location

  • environmental

  • behavioral/browsing data

How accurate is it?



Unlike deterministic identification it’s a way of mapping complex multi-device user behaviors without exposing personally identifiable information (PII). It also has, unlike deterministic solutions from Google and Facebook, the potential to serve as an open framework for advertisers to identify and reach anonymous users across multiple platforms and properties.


Despite progress, that confidence-level isn’t quite high enough yet to convince advertisers that it’s going to deliver a return on investment. The special-sauce unique to each vendor in this sector is typically a closely-guarded secret, so there’s no unified way to audit methodologies across providers.

Ultimately, for a media company looking to work with a provider to deliver probabilistic targeting for brands this method of user identification hasn’t been proven to deliver an uplift in CPMs yet.

All in all

Deterministic identification is the more mature of the two methods, both in terms of technology and investment from advertisers. However, it is inherently flawed. Probabilistic sounds great in theory, but incredibly hard to deliver. So we have two solutions, neither of which are entirely satisfactory.

Even so, we can expect to see rapid advancements over the coming 12 months as the industry puts more resource into tackling multi-screen advertising.

Find out more about: