Unifying Customer Data: Identity Resolution

In the quest to unify customer data, there are no decisions more consequential than picking which attributes indicate identity. At Actable, we advocate for a practical, pragmatic approach to selecting identifiers.  In this essay, we’ll share some of what we’ve learned helping our clients go through this process. 

The Customer vs The Individual 
Many a LinkedIn advertisement promises to help organizations create “a single view of the customer.” This is a compelling pitch. It is easy to understand and is a useful vehicle for communicating a high-level concept: centralized audience management. 

Actable believes in the utility of the centralized audience management concept. We also believe that this concept should be balanced with a nuanced approach to defining what constitutes a customer identity. In other words, do not necessarily conflate a single customer with a single individual real-life person. It is better to permit a single person to have multiple customer identities, in some circumstances. This sounds counterintuitive, but consider the following circumstance. 

At a previous job, I was responsible for purchasing Amazon gift codes en masse, for the purpose of fulfilling survey incentives. To do this code buying, I created an Amazon customer account with my work email address. With this account, this is all I ever did: buy gift codes.  

However, I also bought other things from Amazon. These other things had nothing to do with work: electronics, home goods, and so on. I had a different Amazon for this purchasing, and that account was associated with my personal email address.  

I have no doubt that Amazon has technology that enables them to know (or guess) that those two accounts were one person. And I know that there are solutions out there that can enable other enterprises to do the same.   

This is impressive and useful in some contexts, but for the purposes of marketing, I ask this question: Why do we care to know? 

My work code-buying account had little in the way of relevance to my personal account. There was no browsing behavior that would be useful for insights to inform my personal account. Likewise, if Amazon were to make product recommendations based on my work account purchasing behavior to my personal account, they would be off the mark. 

Further, consider what my experience would be, to see such recommendations. It would betray that Amazon was doing something behind the scenes, to link my two identities. In this increasingly privacy-conscious atmosphere, that may well have an adverse effect.  

An over-resolved profile can also present other technical challenges. If Amazon wanted to send me marketing communication, which address should they use? There might be a need to establish hierarchical rules to determine which channels use which IDs, and when. All of this adds overhead to execution. 

Rather than look at multiple accounts as a challenge, it is better to interpret it as a signal which entreats us to permit our customers to have multiple identities.    

Simplicity 
Every additional way we have to know who our customers are may be a key to unlocking more data. But adding more identifiers into the mix adds complexity. This complexity manifests in additional time to value: more complicated data modeling, increased data processing costs, and more opportunities for the creation of edge cases which may gum up execution. 

Each system in your marketing universe has its own internal IDs . An ESP may have one or two, and the same goes for nearly any other channel tool. A web page may have several sets of cookies to cover anonymous and authenticated sessions. CRM systems will also have customer identifiers, and sometimes they can even change over time.  

Trying to keep track of the relationships between all of these systems can be challenging, if every one of those system-specific values are in the mix. Try to pick the minimum number necessary to do efficient segmentation. Accept some of the limitations this might impose, in service of a flexible and easy-to-use identity model that puts activation closer at hand. 

Strong 1:1 
Pick one strong, non-mutating 1:1 identifier. Mutating identifiers can be problematic, because most enterprise channel tools will have a requirement to manage relational data. This relational data can be difficult to manage, if the keys that are used to join that data changes. 

Some common choices here may be email address, or an account identifier. If no such values exist, it might be a good idea to create one.

The Big Picture 
Unifying customer data should be a means to an end, rather than an end itself.  Keep an eye on actionability and utility. Have the business cases defined, and use them as a north star to guide an identity resolution strategy. There will always be potentially ambiguous relationships, conflicts, missing records and other problems. That is the nature of collecting any data. Solve for the common use cases, which will realize the biggest return on investment. Refine your models over time, and don’t get caught up in solving for edge cases with limited value.