AI Powers Identity-Driven Marketing Innovation at TransUnio

Jeff Mosier:

Hi, I'm Jeff Mosier with the Infosys Knowledge Institute. I'm here with Marc Vermut, vice president of Marketing Solutions Knowledge Lab at TransUnion. We're here to talk about AI and marketing. Marc, thanks for joining us today.

Marc Vermut:

Really happy to be here.

Jeff Mosier:

Great. So tell us a little bit about how AI is changing how TransUnion operates your marketing organization.

Marc Vermut:

The perspective I'm going to bring is less about how TransUnion is marketing from a B2B basis, and more about how the Marketing Solutions organization within TransUnion is working with our clients and partners, and how AI is impacting how we work with them.

As background, TransUnion's Marketing Solutions organization is focused on helping marketers, agencies and media companies understand who consumers are from an identity perspective, how to build audiences to engage with those consumers, or serve those consumers ads and content that's relevant and meaningful and impactful to them, have insight about who those customers are, consumers, reach them, and then measure if it's impactful. And if you think across that chain of marketing, who consumers are, who your customers are, what do you need to know about them, how do we create audiences that are sizable but alike, reach them with advertising and experiences, and then know if it works, AI is impacting across the board all of those different stages of how brands are marketing to consumers, just at a high level.

Jeff Mosier:

Obviously, you guys have had access to a lot of data and had very sophisticated marketing for a long time in Marketing Solutions. What are some of the areas where AI has been the most transformative? Where have you seen the biggest changes?

Marc Vermut:

So I think there are two things to think about. First is, today, where people are really feeling AI is the explosion of large language models and generative AI. But we think of AI in much broader concepts of using data and science and math in order to leverage large models to help make predictions and decisions. And so, we've been using machine learning for a number of years around our identity resolution. So we have, in the US, an online/offline graph that allows us to understand who consumers are, and there's a large machine learning model behind that. And we also do measurement to understand, at a granular real-time level, how effective marketing is at driving consumer behavior, and that's also been machine learned.

Where we see changes today is from a couple of different perspectives. One is around workflows. So if you think about GenAI, how are people using AI to influence the workflows that they have to do marketing and to do their jobs? And we're slowly seeing changes. TransUnion, our foundation is as a credit reporting agency, so our use of data and the data that we have is highly regulated, very confidential and proprietary, and we have to take strong, good care of it. So the motions that we're making in leveraging AI ourselves and for our clients is conservative. So we've been doing exploration and research, and where the data's less confidential, we're exploring more how does GenAI enable our workflows to work.

The biggest impact that we're seeing is around the data. So we've done explorations with how do you use GenAI chat interfaces on top of corpuses of data to find out information more quickly. So we're looking at audience discovery. So we have all of these attributes about consumers, what they're interested in, where you can find them, what they tend to buy, and thousands and thousands and thousands of different audiences and attributes. How does a marketer quickly identify an audience that's relevant to their campaign and their campaign objectives? So we're exploring the use of an interface that makes that process faster and easier by using generative AI, interactive chat, and to do that, what we're realizing is you actually have to have a clean data foundation. And, for example, part of the responsibility of my team is to generate the knowledge, the content, the documentation and the learning for both our internal teams and our clients to leverage our marketing technology.

And we've been exploring putting LLMs on top of that information to enable people to more quickly find the information they need to accomplish the tasks they have or learn about the data that we offer. And what we've found is that, A, it's not at a point where it's not hallucinating all the time, and B, you need to prepare the data, you need to put ontology on top of it, a taxonomy, so that it can better interact and engage with the data that you want to make available to the end user. So we are focused on the preparation of the data, the ontology of it, to then improve workflows and put interfaces on top of it, and I think that's where the marketplace is today. You're starting to see also agentic creation, and what we find there is it's not going to be one agent to rule them all, it's going to be individual specific agents trained on a specific task with a narrow set of data, and then an orchestrating or supervising agent on top of that.

Jeff Mosier:

How far have y'all gotten in your use of or experimentation with agentic AI?

Marc Vermut:

We're early stages. We're experimenting with it. We're doing two things. One, we're providing our data to our clients who are inputting that into their agentic AI. So we work with an agency that's leveraging our identity and attribute and audience data to build a more interactive chat-based audience discovery platform. And we're also building our own to do that as well. So we're identifying specific tasks and building agents on top of the data. We are building an interface to enable easier optimizations. So if you think about the fact that we have this marketing measurement capability for brands, they're very big and deep in terms of the granularity and the dimensions that we're measuring, and so we're exploring putting an agent on top of that to direct an optimization and provide feedback on how it's operating. So we're exploring from an agentic perspective and layering that.

Our focus today really is the ontology, building the data in a way that is credible and usable for agents and in workflow, building the workflow, so layering the UI and the interface on top of it to enable the interactive chat, and then implementing the agents. So it's still early days. Because of the challenges around data and hallucinations and expected performance, we want to make sure that it's ready for our clients to use. Where we're really focused on innovating outside of that is around measurement, which isn't necessarily generative AI, but we're creating Bayesian knowledge graphs to integrate multiple forms of measurement to enable a brand to have a single source of truth at any level, not just online video, not just Hulu, but specific types of programs or campaigns and how it performs.

Today, brands use multiple measurement approaches to understand how their marketing's performing, and they'll give you slightly different answers, because it's at different levels with different measurement types at different dates. And so, the advancements that we've made with AI is to use a Bayesian knowledge graph to take all those points of measurement, put an ontology, a semantic framework around it, and leverage AI models to generate a weighted and organized, a single measurement, that's consistent with all of the different measurement for how any of your marketing will perform to allow a brand to then optimize across that knowledge. We've been spending considerable time building that capability, and we're now rolling it out to a number of our clients on an early basis.

Jeff Mosier:

And I'm assuming that measurement will continue to evolve as the demands change on what you're trying to measure and the different abilities that you have with AI.

Marc Vermut:

Well, no, exactly. The framework we're creating is flexible, so it's updated on an ongoing basis with the latest measurement, you're able to incorporate different types of measurement into that system. And the use cases for how clients will want to take that measurement and then optimize and plan against it can continually evolve because of that flexibility. And then, ultimately, what I would expect to happen is that math and that AI will be underneath agents and an interface that will also leverage AI to let less sophisticated, less data science or analytics personnel query the data and develop plans in a more straightforward, plain English way than they might today.

Jeff Mosier:

You mentioned hallucinations and some of the issues you have with data and it being in the right format or structured correctly to be able to be used effectively. What are some of the other challenges you're facing?

Marc Vermut:

When GenAI launched and large language models launched, in the first few months, there was the whole idea of prompt engineers, you need people to have to query the AI in order to generate the best answers. And I think if you move away from the explosive idea of a whole new category of job, there's actually something to it.

What we're finding as we leverage LLMs and GenAI is that you need to provide a structured way for people to answer questions within the way that the model that sits on top of the data is going to actually respond. So we're finding the most effective way to leverage it is to create prompts that answer the basic questions that you would like, and then people can start interfacing and interacting with the UI underneath it. So I think what's really important, given that we know that sometimes there are going to be hallucinations, is that you actually work with the model that you've built to test it. You run QA, you identify the best prompts that are going to answer the initial questions that you expect a user might have from it, and we're using that with our own internal systems across TU, as well as the engines that we're building for clients.

So I think in the beginning, as people are exploring it, especially when you're working with performance and marketing data underneath, is that you guide the experience to make sure that what comes out is reliable and useful.

Jeff Mosier:

What do you think is going to be the next big breakthrough for you in AI and your capabilities and marketing? What will it be able to do that you can't do now?

Marc Vermut:

I think it's going to be workflow, and I think it's going to be the use of agents. If you think about how we work today, there are a number of different steps and data sets that a marketer needs to work through. So what's their goal? Their goal is to create a campaign, to put it in the right places, that will reach the right people, who will then respond to that message. So you have to move from an identity data set with attributes to building an audience, to aligning the message and making the audience available with the media partners or the channels where you're going to find those people, and then figure out if it works. That's a lot of different steps to take for an individual user. So we built a platform that enables you to do all of those things in one place, so that's a breakthrough for us.

The next step is to make the process of all of those use cases with all of those different tools more seamless for the marketers who leverage those capabilities. And I think that that's going to be the breakthrough that we're going to see over the next few years, is that the multiple steps that take different types of people will be made a bit more seamless and straightforward in the background, but you're always going to need people involved, it needs to be supervised, to make sure that what's coming out of each stage makes sense relative to a marketer's goals. But having to do the work should be a lot easier than it is today.

Jeff Mosier:

What do you see as the long-term implications for AI and marketing? How is it going to change it more broadly?

Marc Vermut:

I think it will speed the amount of time that's required, or reduce the amount of time that's required, to make decisions about campaigns, to make decisions about marketing investments, to make decisions about plans, which should allow companies to be more nimble in how they operate and actually close the feedback loop. Today, it takes a while to build a campaign, launch a campaign, measure that campaign, and then optimize against it and plan for the future. I think you'll see faster cycles in marketing in a way that's manageable and coherent.

Companies will be able to go with more versions of creative, will be able to go with different types of messaging, and iterate on what's effective more quickly, because there'll be a stronger engine in the back to analyze it. And the people who interface with it can do it in a more plain English, common sense way than having to analyze data, aggregate those insights, and then translate it into a way that people are able to understand it, because today, you have data scientists, data engineers, business strategists, business analysts, creatives, strategic planners. I think the ability to take those insights and get from data, activation, performance, to what do I do next, will collapse.

Jeff Mosier:

What is one piece of advice that you would give to others who are seeking to integrate AI into marketing and change how they operate?

Marc Vermut:

Explore. Play with it personally so you understand it. Get your data in order. Connect it. So one of the strengths of what TransUnion does is we're able to bring a lot of disparate types of data sets together in a trusted, connected and governed way. Ultimately, if you can't connect the different data points together to consumers so that you have a complete view of who a consumer is, where you can reach them, if you've reached them, and if they've done anything after being engaged with, anything that sits on top of that from an analytic perspective, it's not going to be accurate. So get your data house in order, build the ontology and the semantics that will improve the performance of the AI that you leverage in your marketing process. So I would say play and get your data house in order.

Jeff Mosier:

Well, this has been fantastic. Thank you so much for your insights, Marc. This has been fascinating. I've really enjoyed it.

Marc Vermut:

I really appreciate the opportunity to sit down and talk about it, and I'm looking forward to seeing what you hear across the board and learn.