AI, Privacy, and the Future of DTC: Walking the Tightrope of Innovation
David: Everybody is talking about the potential of AI. Let's start a little bit differently. What disappoints you about AI?
Luk: It's a bit of a tough question. There's been a lot of hype. I don't want to say “disappoint.” I'm excited by it, but we're still trying to figure it out. Where exactly is it going to take us? Given my regulatory background, I do sometimes wonder about how it's being used. Is it being used effectively? Is it being used in a way that doesn't breach certain protocols we currently have from a governance or a security perspective? And then lastly, the big thing is scale. How are we going to scale that in a way that is effective but also very responsible?
David And I think, how is the pharmaceutical industry going to understand this and apply it? In our work, we both touch on regulatory matters a lot. How are the regulatory folks going to be able to process this, other than at the surface level?
How do you feel about the assertion that AI is nothing but math? Is there any art in that sauce, or is it just math?
Luk: It's a bit of both. Yes, it's all backed up in math. But at the same time, there's a creativity in how we're taking these ideas and really pushing insights.
David: Can you share any ways in which IQVIA is looking to apply AI?
Luk: IQVIA has been in the AI space for a long time. The group that I work with has been doing large language models for 15-20 years. We actually have a system that we built before we were really framing it as a large language model. It's called a BERT model we use for regulatory submissions. We’re considering how to use large language models, and agentic AI, which are basically large language models that are specifically trained for a decision task.
Here's a question for you, David. I hear the term “omnichannel.” I'm assuming omnichannel means all kinds of different data from different areas. How do you actually coordinate so many different kinds of data?
David: Omnichannel is a confusing term, unfortunately, because there's a lot of definitions of it. Many people, especially in my circle, refer to omnichannel as meaning across digital channels. So that could be doing ads on websites versus ads on social media, versus email pushes or that kind of stuff. When we talk about omnichannel marketing, we're taking something learned in one channel and applying it to another channel.
When we create a consumer audience, which we must do for DTC advertising, we are concerned about certain regulatory issues and specifically, consumer re-identification. In order to stay compliant with current data privacy laws, we go through de-identification to anonymize the data. Then we use probabilistic methods of creating an audience that introduce noise that make it hard to re-identify any individual. A lot of academic work has shown that even if you de-identify a person, if you add in enough other data sets, you raise the chance that you can re-identify that person. So that is a legal and regulatory concern as well.
Here's another question for you. I've heard the phrase “defensible AI,” and I don't know what that means. Can you tell me what defensible AI means?
Luk: My team and I came up with the term probably about a year ago to describe some of the work that we're doing. Across the board, we want to make sure the AIs that we build are defensible. And I mean it from a very broad perspective. Not just from a regulatory or responsible use perspective. It's got to be defensible among three groups. First, your internal group, of course, then to your clients and then the regulators. All of these require different conversations about different aspects.
David: If you were at a manufacturer and someone came to you with an AI-based marketing program, what additional new kinds of questions would you ask of that vendor? Let’s say the use case is creating consumer audiences.
Luk: The more sensitive it is, the more closely it's related to patient safety and patient outcomes. It comes down to responsible application, but also in terms relevant to consumer audiences. From an AI perspective, obviously data quality is important, but also, if you don’t have consent, then how are you using this data? What's appropriate? You can document all that, but you also need to have strong, robust guardrails around it. I would ask, how do you validate this? How do you ensure it is responsible and trustworthy?
Hear the full conversation on how the intersection of AI, DTC consumer marketing, and data privacy is reshaping how healthcare brands engage with their audiences, and how pharma marketers that embrace innovation will be best positioned to lead in this new era.
