What is Identity Resolution?

Identity resolution sits at the core of modern digital engagement, allowing organisations to create accurate, unified customer profiles that span behaviour and transactions across multiple channels and devices. It’s also commonly referred to as cross-device identification and underpins the Single Customer View.

It is the bedrock of important data-related use cases such as collection and storage, audience and segment building, and personalised, multichannel customer engagement.

Why is identity resolution important?

In action, identity resolution combines multiple identifiers to a) pinpoint an individual customer as they interact with you in multiple ways and across various channels and b) attribute this behaviour to a single profile.

In other words, it’s how an organisation knows that a person who has bought on several occasions from our online store is the same person who has recently downloaded the app.  

Or, to give another example, it’s a way of unifying a customer’s behaviour pre and post-login event or account creation.

Without identity resolution, you would have two separate customer profiles in each of these instances, likely stored in two silos. This leads to disjointed engagement campaigns and ultimately limits your ability to meaningfully connect and influence. The elimination of incomplete, duplicated profiles is, therefore, one of identity resolutions’ primary benefits.

As the volume of devices, channels and data points increases, the need for accurate customer profiles becomes more urgent. Spend on identity resolution is predicted to reach $2.6 billion in the US alone by 2022.

Identity Resolution: Probabilistic vs Deterministic Matching

The history to date of identity resolution can be viewed as the progression from educated guesses to certainty. 

The origins of identity resolution as we know it today go back some 25 years or more. In those days, the goal was to identify households and individuals within the same household. 

It relied on a hierarchy of information like house number, postcode, town and county. It was an example of what we now call probabilistic or “fuzzy” matching. It couldn’t identify individuals with complete certainty, instead it arrived at likely conclusions.

This method still exists today, with weighted scores attached to the various kinds of information and potential identifiers. Each piece of data is given its own statistical value. So you might decide that an IP address, for example, carries very little weight because multiple people may use the same device to browse the internet and make purchases.

Within the probabilistic method, a “certainty threshold” is set on a case by case basis. So the system must be 93% sure, for example, before the match is made. Each matching element, or piece of data, found increases the statistical likelihood that the system has identified the same individual in multiple places.

What is deterministic matching?

This method seeks to eliminate all doubt and risk that an inaccurate match has been made.

When you use deterministic matching you are reliant on high-quality, accurate data being fed into the system. It ideally is fueled by the collection of personal data points such as government-issued IDs (social security/driver’s licence/passport) or other pieces of information that are unique to an individual.

FYI: An email address doesn’t fit these criteria because it is possible that two people, a married couple, for example, might have a shared account.

Of course, one potential issue is simply getting access to this kind of 100% individually unique data. In most implementations, more common data points like email address, phone number, date of birth etc are each matched individually, contributing to an overall confidence score when taken together.

As part of their overall customer engagement strategy, more and more brands are gatekeeping content behind a login, or incentivising account creation, in order to facilitate deterministic matching and sure up their identity resolution.

Which is best?

At Xtremepush, we champion deterministic matching, powered by first-party data given with explicit consent by the consumer.

Whilst a fuzzy, “fingers-crossed” methodology may have its use for finding lookalike audiences for targeted advertising, it is not suited to the goal of personalised, one-to-one marketing. 

Deterministic matching gives you the confidence to tailor relevant experiences and campaigns to each of your customers no matter where you want to reach them.

Identity resolution 101: Key elements

What is a customer profile?

Let’s think beyond the old-fashioned CRM version of customer profile, which is very limited in scope compared to what modern Customer Data Platforms are capable of doing. You can think of a profile as a big folder of data: events, attributes, subscriptions, identities, attribution info, device info. 

What is an Identity Graph?

An Identity Graph (or ID Graph) is a database which stores identifiers related to an individual profile. The Identity graph itself doesn’t store any of the user data, such as demographic information or behaviour, it only stores identifiers needed to be able to pinpoint the user. The main purpose of ID Graph is to assign incoming data to the correct Profile.

Types of Identifiers

These identifiers could be anything from usernames to email, phone, cookies and even offline identifiers like loyalty card numbers.

Merge protection rules

Think of these as your safeguards in the battle against inaccurate matching. Let’s look at a real-world example. There are many brands that run in-store campaigns encouraging consumers to sign-up and create accounts. The same device (iPad, laptop etc) will be used by various people to submit data throughout the day. So now you have multiple events and pieces of data all linked to the same device ID. Merge protection rules stop these all being associated (or merged) with one profile.

Is Identity Resolution compliant with privacy laws?

The short answer is yes, but it’s worth diving a little deeper into it. 

The first thing to say is that probabilistic matching is entering some murky territory as it often uses IP addresses, device IDs, advertising identifiers (read more on the upcoming iOS release and its impact). For a long time, these have not been strictly classed as P.I.I. (personally identifiable information). They could therefore be collected without the consumer’s consent.

This is changing, however, and before long it’s likely that their collection will require explicit permission. This will drastically reduce availability.

As we’ve mentioned, we’ve always recommended a deterministic method, fueled by openly solicited and shared data. This is an example of privacy by design, where compliance is built into the operational bones.