TDWI Upside - Where Data Means Business

Customer Data Profiling: Persona Disambiguation and Behavioral Analytics (Part 3 of 4)

When multiple customers use one account, it can cause problems for your predictive models. Use behavioral analytics to separate the personas and improve your customer service.

In spite of a company’s best intentions to maintain a unique profile associated with each of its customers, it is not unusual to provide products and services to multiple individuals via the same set of delivery channels.

In our previous article we shared some examples, such as a shared Netflix viewing account used to stream different content to multiple user devices (TV receivers, computers, and mobile devices) within the same household, or several organizations sharing a single e-commerce account to take advantage of bulk order discounts.

After determining that there are multiple “personas” associated with a customer account, you can begin to distinguish the activities and transactions so you can disambiguate the personas. By disambiguating who is who, you can isolate the activities and transactions performed by each unique individual associated with the customer account and overcome the issues that skew customer profiling and lead to imprecise predictive models.

The most obvious way to differentiate the different personas associated with a single customer account is to directly reach out and ask the customer if there are multiple users associated with the account.

Companies such as Netflix have taken this approach. Simply changing the user interface to allow an individual to select “who’s watching,” manage profiles, or quickly add a new profile indicates not only awareness of personas but also that the company is interested in catering its recommendations by persona.

Alternatively, you can attempt to disambiguate personas using behavioral analytics. Although multiple individuals may be using the same channel, you might use the ways that each individual performs their interactions to identify characteristics and expose behaviors to use for differentiation.

For example, we can look at the following characteristics of customer behaviors to see if there are any correlations and patterns for segmentation.

Time: At what time of day was the interaction performed? Different individuals may be engaged at different times of the day. Children might be playing a video game in the afternoon, but late-night gaming is more likely to be an adult.

Location: Where was the interaction initiated? This might refer to a physical location -- derived from a mobile device’s GPS capability or inferred location from an IP address. Geographical coordinates not only place the individual but also determine if the individual is moving or not.

Alternatively, this might refer to a location within the context of the product or service, such as making an in-game purchase from a particular location within the game or the location within a video stream that the individual pauses and subsequently restarts.

Channel: What type of device is used to perform the action? Is it through a computer, a game console, a smart TV, a smartphone, or a different type of mobile device? What is the IP address associated with the device? In many cases, the unique identifier of the device (such as a mobile phone serial ID) is sufficient to identify a single individual.

Intent: What is the reason for the transaction? For example, is the individual searching for a particular product in the online catalog, adding an item to a shopping cart, or checking out and paying for a set of items?

Content: What are the characteristics of the products or services being delivered? In the streaming content example, this would examine the metadata associated with the video (such as genre, length, quality of the video, or any known celebrities featured). For the e-commerce example, this would identify characteristics of the products in the shopping cart (such as product type, intended age, or cost).

Personas can be configured in relation to the attributes associated with the transactions. This allows you to differentiate between a persona that is active between 3:00 and 6:00 p.m. on weekday afternoons streaming children’s videos through an Xbox, a persona streaming dramatic videos between 9:30 and 11:00 p.m. through a smart TV, and a persona streaming horror movies on a mobile phone between 2:00 and 4:00 in the morning.

Essentially, these (and potentially other) facets of the interactions become attributes of the personas. Often these characteristics are sufficient to identify the usage patterns necessary for persona disambiguation.

We continue our examination of persona disambiguation in Part 4 of this series.


Articles in this series on customer data profiling:

Part 1: The Precision Problem

Part 2: Problems with Personas

Part 3: Persona Disambiguation and Behavioral Analytics

Part 4: Persona Analytics

About the Author

David Loshin is a recognized thought leader in the areas of data quality and governance, master data management, and business intelligence. David is a prolific author regarding BI best practices via the expert channel at BeyeNETWORK and numerous books on BI and data quality. His valuable MDM insights can be found in his book, Master Data Management, which has been endorsed by data management industry leaders.


TDWI Membership

Accelerate Your Projects,
and Your Career

TDWI Members have access to exclusive research reports, publications, communities and training.

Individual, Student, & Team memberships available.