Academic Partnership: Q&A with Dr. Longxiu Tian on consumer predictive behavior data

December 13, 2022

AS WE PARTNER WITH LEADING LEADING ACADEMIC INSTITUTIONS TO CREATE INDUSTRY-MOVING RESEARCH, there’s no better way to build an innovative community than to connect the deep knowledge of our academic partners with the next generation of forward-thinkers at up-and-coming technology companies. It’s key, as Equifax’s CEO Mark Begor says, to look outward, not inward in bringing solutions to customers.

Driving innovation and creating transformative solutions is a main focus for several Equifax programs, including Equifax Accelerate and the Academic Partnerships program. 

Dr. Longxiu Tian is the Assistant Professor of Marketing at the University of North Carolina (UNC) Kenan-Flagler Business School. Tian studies how consumers respond to marketing activities, particularly when making choices in uncertain environments, to better inform firms’ actions for customer relationship management.

[RELATED: Equifax Launches Second Annual Developer Challenge and Accelerator Program]

Tell us about your work with the Equifax Accelerate Program. 

Through talking to my MBA students and stakeholders at UNC, I became really interested in this whole entrepreneur side of the business world that's related to marketing - especially selling directly to consumers and helping consumers make better choices.

The opportunity with the Accelerate Program came up as a part of my work with the Academic Partnerships program. This is the second year Equifax hosted the Accelerate Program, which is all about helping technology companies of all sizes tap into the Equifax information ecosystem to build transformative, data-driven solutions and also just to network with each other. 

Given my background of working with Equifax data, and also thinking broadly about where Equifax data falls within what I call the “taxonomy of marketing data”, my role was to talk with these companies, understand where they are in terms of their product lifecycle or startup development cycle, and help them recognize the value of different types of data for their marketing needs.

Can you tell us a bit more about how certain data fits into the taxonomy of marketing data?

In the universe of marketing and marketing analytics, customer data can be broken down into four broad categories:

  1. Identifiers: Who the customers are, where they live, their email, and a phone number for their SMS. These are what you would call the classic segmentation variables that companies capture on their customers. And in some ways these are the variables that are prevalent but not as predictive, not as informative of future preferences of customers. And that's what market results really care about.
  2. Engagement Data: When do customers log in? Do they click into your emails? Generally, how they are interacting with your app and website.
  3. Relationship Data: What's the phone plan that they’re on? What package do they use? What services have they acquired from you? 
  4. Perception Data: How do they like your product? The kind of reviews that they provide online. 

So, where does Equifax data cut across this taxonomy? Equifax data can augment this customer data and help inform future preferences and behavior.  Say a company wants to make a  recommendation on a credit card for their customer. In that space, what the company would love to know are the kind of trade-offs that their customers specifically, and consumers broadly, are making in their credit decisions. What kind of loans have they considered? What kind of credit cards have they considered? What is their credit score right now? That's going to greatly inform a company in terms of the credit card that can be recommended to consumers and that's the perspective that Equifax data can provide, based on the company’s permissible purpose.

What sorts of conversations did you have with the participating companies in the Accelerate Program?

There were structured interactions as well as unstructured interactions. There were also specific “office hours” that were hosted for these companies. Afterwards, there were breakout sessions, and actually, that's where some of the most interesting conversations occurred. 

For example, with one company called Vacay HQ, they are positioning themselves as the travel rewards specialists. As they are thinking about their product and how they want to position themselves to investors, VacayHQ was interested to know the value of the insights they’re generating from their customers and how to convey to investors the value of what they're doing over and above just making a recommendation on the next credit card that gives the most amount of points.

Ultimately, what VCs [venture capitalists] are looking for is not only how you serve your customers but the value of the data that you create as well.

Back to your Academic Partnerships role and your collaboration with Equifax data scientists. One area of collaboration involves Bayesian machine learning; can you explain more about Bayesian machine learning and how it blends new and old ideas?

Bayesian machine learning is all about flexibly understanding the complex relationship between customer choices and behavior with firm decisions, without needing to limit ourselves to the kind of traditional models that you see in economics and statistics that often are constrained in terms of data types or assumptions about the underlying data generating process. At the same time, Bayesian machine learning retains model inference, such as the ability to quantify our uncertainty about an input factor on its impact on an outcome variable. If you will, modern machinery with old school explainability.

One of the techniques that I use is called a variational autoencoder (VAE) and that originates from computer vision and natural language processing. VAE has seen a lot of success in those domains, including the recent text-to-image data fusion models based on Stable Diffusion, which is a type of hierarchical VAE. There is active research applying VAE to the kind of marketing data that we see, such as those four types of taxonomy of marketing as well as outcome variables like purchase, acquisition and retention. Unfortunately, you can't just take a model from computer vision and just use it on marketing data. My research has involved developing new VAE models that enable data fusion on marketing data, such as customer surveys with customer relationship management databases. 

Another technique that is popular in Bayesian machine learning is called Gaussian process modeling. That comes from signal processing but it’s been applied in marketing to great success to understand nonlinear time trends. This also has a lot of relevance to understanding how credit scores evolve at the individual consumer level, how these latent time trends vary across customers, and in turn, how lenders and other financial institutions can use this information to better serve their customers, as well as in areas such as fraud and anomaly detection.

Overall, it’s been very exciting over the course of my partnership with Equifax, working with the data scientists here to apply Bayesian machine learning techniques to better understand consumers’ credit and financial decisions. The scale and richness of Equifax data are allowing us to uncover insights and patterns that are unparalleled in the world of data analytics. My hope is that through this research partnership, we can find innovative ways to better address consumer needs and help them better navigate financial decision making.

For information on the Accelerate Program with Equifax, CLICK HERE.