Identity Resolution
- Jonathan Seow
- Oct 1, 2024
- 3 min read
One User, Many Identities
Imagine a user named Jane who visits the e-commerce website of a hypothetical shoe brand called ActiveKicks.
Like most users, Jane doesn’t register for an account. She browses the website, clicks on a few shoes that interest her—Shoe A, Shoe B, and Shoe C—and then leaves without adding anything to the cart.
Since the website doesn’t know who Jane is, she’s an anonymous user.
The next day, Jane revisits the website, creates an account, and purchases Shoe C. This time, Jane is an actual user of ActiveKicks.
From the site’s perspective, it appears that two different user profiles interacted with it—one viewed three shoes, while the other purchased one. However, we know for a fact that they are the same person!
How can we merge these two user profiles to create a single profile for Jane, combining her anonymous browsing history with her recent purchase records?
What is Identity Resolution?
A Customer Data Platform (CDP) supports identity resolution that stitches multiple user profiles into a unified profile.
Identity resolution helps you better understand user activities across all platforms and devices. Marketers can track how users move through the sales funnel and optimize end-to-end customer journeys.
In Jane's case, a unified user profile reveals her interest in Shoe A and B. With this information, you can implement targeted marketing strategies to incentivize her to complete the purchase.
In summary, identity resolutions increase marketing effectiveness and provide consistent user experiences across all channels!
Deterministic vs Probabilistic
Different CDP vendors offer unique methods of identity resolution. They can be either deterministic or probabilistic.
A deterministic approach relies on your business data as the source of truth for user identities. You need to provide records of user identities, such as emails, anonymous session IDs, and user IDs.
A CDP generates an internal ID for each group of user identities. You can think of the ID as a unique user profile identifier.
The deterministic approach provides certainty about each user's actions. For instance, 28018769 points to Jane, who viewed three shoes and bought one.
However, the results of a deterministic approach depend on the quality of your user identity data. If the data is inaccurate or incomplete, it can reduce the accuracy of identity resolutions.
Meanwhile, some vendors offer a probabilistic approach, using algorithms and statistical modeling to predict the likelihood that different identities belong to the same person.
This approach is flexible because it can tolerate incomplete user identities, but it may lead to inaccuracies in user profiles due to false positives or negatives.
A Dive into Deterministic Resolution
Let’s walk through a scenario using the deterministic approach.
When Jane visits ActiveKicks' website, she's assigned an anonymous session ID. This session ID links to an internal CDP ID, which tracks her activity — like viewing three pairs of shoes.
When Jane creates a new account, she's assigned a user ID. Her shoe purchase relates to this user ID, which links to a different CDP ID.
Both profiles are created in real time. To unify them accurately, the CDP must ingest user identities from your business to understand the relationship between users and anonymous IDs.
The merging process depends on profile priority. For instance, user IDs can be prioritized over anonymous IDs, meaning anonymous profiles will be merged into the corresponding user profiles.
Conclusion
Identity resolution is a valuable function in a CDP, allowing businesses to understand the behaviors of their target audiences across many platforms and channels.
However, building reliable identity resolution strategies is not straightforward. Technical complexity aside, close collaboration between clients and CDP vendors is needed to create high-quality user profiles.
🫵🏻 If you’re interested in leveraging a CDP to enhance your data ecosystem, feel free to reach out to discuss tailored solutions for your use cases.
Comments