What Brands Need to Know About Generative AI
Google Cloud didn’t develop its Vertex AI platform in a vacuum. According to vice president Carrie Tharp, the company talked with “everybody from luxury down to everyday goods” businesses to understand the challenges and where technology like generative AI could benefit the most.
The former head of retail and consumer at GC, Tharp is applying her considerable retail experience from stints at The Neiman Marcus Group, Bergdorf Goodman, Fossil and others in her expanded position leading strategic industries for Google’s cloud business.
More from WWD
“For me, all of [our] announcements are very important to retail and consumer goods,” she explained in an interview with WWD. “It matters that Google and other scaled retailers will start infusing these experiences because that will set the bar for the rest of the industry.”
Given her lens on the technology and its specific applicability in retail, Tharp can see the opportunity and challenges for the sector. WWD spoke with the vice president to learn more about her perspective and the advice she would offer brands during this pivotal period, as the generative AI movement gets underway.
WWD: It seems like there’s a race to adopt AI, particularly generative AI, and that has retailers and brands weighing multiple ways to apply it to their businesses. What are they looking at and, perhaps more importantly, what are they not thinking about that they should be?
Carrie Tharp: This technology is nascent and early. AI had been slowly infusing its way into the business…[but] generative AI is not the direct equivalent to AI. It is a new discipline, so it will require new testing protocols, new and advanced thinking about responsible AI.
If you think about old customer service chatbots, they were programmed. They said exactly what you wanted them to say, based on the prompts they were given. In generative [AI], it’s a creative process, and so you can have things coming out that are not fully factually accurate. We’re creating capabilities that can help handle that, but you want to make sure it’s adhering to your brand standards, etc.
Generative AI is not a solution for every hard problem you ever had, without limitation. So just like all technology, it has to be developed; we have to learn the change management around it, and manage those new disciplines. If you asked an enterprise today, how are you going to A-B test generative AI, which is not the same deterministic flow in previous AI? They don’t really have the answer to that, and so senior leaders need to be thinking about, how do I pace my organization into this change, such that I’m moving fast enough — and I don’t get left behind or lose market share — but I’m not moving so fast that I actually have missteps that impact my brand perception or brand interactions with customers.
WWD: How do you draw the distinction between previous iterations of AI and new forms like generative AI, in terms of the data and an organization’s approach?
C.T.: The boring fundamentals of preparing a data foundation had to change. Generative AI doesn’t allow you to skip steps. It allows you to do things you couldn’t do before — have it look at a picture and tell you product attributes.
Before, either humans had to do it or you just didn’t have that detail. As a leader, there was never a dollar that was misspent on building that data foundation. [Generative AI] allows you to tap into unstructured data in a way that you couldn’t before. But how do you want to use that data?
If I were to translate that into a leader’s guide to AI…I think you’d have to ask yourself a couple of questions. One, it comes back to, what’s the problem statement? What are you actually solving for? How to use generative AI will not be the same for everyone. They still have to go back to the fundamentals of what the brand stands for. You could use generative AI to simplify, streamline, automate back-office processes that allow you to reduce your cost base. But you could also use it to enhance your differentiators.
So if you truly believe customer service is your thing, or styling, etc., how do you use an AI assist in a way that adds to what the humans and the ambassadors in your organization do? Then determine your approach after that.
WWD: If they’re unsure about the tech, is it wise to stick to back-end processes and not trot it out in front of customers?
C.T.: You have to understand your level of tolerance for experimentation versus control.
At the end of the day, these are all still businesses, so you need to understand if you’re willing to put nascent technology directed to consumer interactions, or are you going to start with back-office process like marketing optimization, so your organization can come up the learning curve? Do you have people in your organization that know how to do prompt engineering, for example?
If you look at the risk-versus-reward matrix, you may choose to start in certain capability areas first, so you can get comfortable that your own organization is coming up that curve. You can release, unleash the power of generative AI when you have the process and protocols to manage that, [while] still managing the brand and all those experiences, the way you would any experience you’re releasing.
So whether somebody is contemplating a pop-up shop, or a new app, those things go through a lot of testing, before they scale. And those fundamentals don’t change, either.
WWD: In the past, I’ve heard from leaders at legacy brands and department stores with generations’ worth of customer data who are wondering what to do with it. When the question came up in a panel a few years ago, the tech expert urged her to start from scratch. It was just too big a job.
C.T.: That problem could feel absolutely overwhelming. And sometimes [brands] wouldn’t pull in data sources the organization had because they were unstructured. They didn’t have enough human resources to take unstructured data and turn it into something structured and usable.
In generative [AI], when you think of large language models — it can read the data, synthesize it and understand what the insight or takeaway was, or how that information should be used for another purpose. So in that sense, and I don’t want to call it a shortcut, but it’s allowing you to tap into all of the data in your corpus of knowledge as a corporation and use that data effectively. You’ll hear the conversation that data is the new asset of all these organizations and the future of e-commerce, and all these things that retailers focus on centers around data. Generative AI now lets you access [that].
It’s like a human brain. It’s allowing you to access more of that enterprise knowledge than you ever had before and actually put it to work. I think that’s some of where the hype and excitement comes from. You can take creative information, social media insights, sentiment, visual — that’s a huge one.
In the past, an image was an image, and you couldn’t extract the data and necessarily use it. The perfect example is recommendations.
The old recommendations engine only had the product attributes you gave it, and so it could only discern a relationship based on the information it had. It might have been completely missing the mark, if you were missing the most important product attributes of the apparel that it was looking at. That’s why you saw people going through this phase of, like, “I need more information,” because the AI was only as good as what you’d tell it. Machine learning could only see patterns in the kind of attributions that you’re giving it.
Now, you have more tools to say I can go have generative AI look at that image and discern more information about it that might be the determiner of why that item is so popular. It might not be the basic information a human previously decided was important. The efficacy of those recommendations, or whatever you’re feeding that into, could be inventory planning and assortment choices, and all of that gets better in a way that you couldn’t access before.
WWD: This new generation of AI touches a lot of areas, such as chatbots capable of more human-like interactions and AI image generation. It sounds great, but what are the risks? One apparel brand triggered backlash over using AI to show diversity, as opposed to hiring diverse fashion models. Will we see more issues like that?
C.T.: This goes back to my point of “determine the problem, determine your approach and then pick the technology.” Authentic is authentic is authentic. If you are not a brand that is focused on diversity and inclusion holistically, and then you use generative AI to just impose different models onto your images, that’s an example where brand history would tell us that that’s probably not going to work.
Generative AI solves the technical problems…but you have to line up the rest of your business process and plan around that. When we say responsible AI, there’s a lot of technical meaning around that, of how you use AI in a way that is inclusive and safe and all of these things. [It’s about] responsible use of technology to get to the right outcome. Don’t use it as a shortcut and skip what would have been a classic step you needed to do in your overall brand communication or overall kind of business process. Because that’s where I think you’ll find yourself in hot water.
When you come back to the tools and technology, that’s where you have to ask the questions: Am I picking the right platform? Am I picking the right partner to have thought through the change management and the total process that wrapped around us? Because any technology use for technology’s sake or as a shortcut is going to potentially lead to risk in the business.
WWD: What else should brands consider when choosing an AI platform partner?
C.T.: Will this platform or software provider that I’m working with, will they scale to the scale I need to operate at? Is there security and protection? Like, is your data protected? When you put your enterprise data up against a model, whether it’s image-related or a large language model, is it taking that data and then understanding things about your customer or your brand that you don’t want part of public information?
We built our tools to basically allow you to tune with your own data, that don’t become a part of the Google model. So thinking through all those things are kind of new dimensions. You have to rethink some of your evaluation criteria and your change management and process control around it, or we will see more stores that have problems [or] didn’t think through all of the different avenues of impact or how it would be received.
WWD: There’s this whole idea of sampling, borrowing, copying and remixing creative content, because AI makes it so easy — like, here’s what “Pulp Fiction” would look like if Wes Anderson made it. But isn’t that a thorny intellectual property issue? What’s to stop others from ripping off others’ designs?
C.T.: This falls in the bucket of responsible AI. We’ve seen new tools come up in photography moving to digital — like, you can look back through the years of the challenges created by technology. In this sense, the whole industry collectively has to adjust.
I think we’re right at that beginning phase with generative AI. We think it opens up many avenues for creative interaction, productivity and how creatives work. But you did call out some of those pitfalls.
Google is supportive of regulation, and we are focused on responsible AI. [But] at the end of the day, there still is IP, the brand voice and copy matters, product design, all of those things will have to be infused into how those things are protected in a systematic way across the tools and capabilities.
That’s where, as we’re building this platform [Vertex AI], we’re focusing on ensuring we put those things in our tools — such that when you’re working with Google, you know that you’re not using somebody else’s product or creativity in a way that you shouldn’t or is too referential. So that’s one where all the nuance of exactly how it works is literally being defined as we speak. But I think is one of those types of things that will rapidly have to come up the curve.
I do also think from our perspective, and what we’re sharing with brands…during this early emergence of technology [ways] to take that position of trusted content. We’re talking with media players, as an example, about how to become a trusted platform. So if somebody’s watching something, how do they know it’s actually not a deep fake? How do you start endorsing and verifying content? Because we will go through this dynamic that people want to see or actually hear the original, not just versions of it.
Best of WWD
