Summary
In this episode of the AI Engineering podcast Praveen Gujar, Director of Product at LinkedIn, talks about the applications of generative AI in digital advertising. He highlights the key areas of digital advertising, including audience targeting, content creation, and ROI measurement, and delves into how generative AI is revolutionizing these aspects. Praveen shares successful case studies of generative AI in digital advertising, including campaigns by Heinz, the Barbie movie, and Maggi, and discusses the potential pitfalls and risks associated with AI-powered tools. He concludes with insights into the future of generative AI in digital advertising, highlighting the importance of cultural transformation and the synergy between human creativity and AI.
Announcements
Parting Question
The intro and outro music is from Hitman's Lovesong feat. Paola Graziano by The Freak Fandango Orchestra/CC BY-SA 3.0
In this episode of the AI Engineering podcast Praveen Gujar, Director of Product at LinkedIn, talks about the applications of generative AI in digital advertising. He highlights the key areas of digital advertising, including audience targeting, content creation, and ROI measurement, and delves into how generative AI is revolutionizing these aspects. Praveen shares successful case studies of generative AI in digital advertising, including campaigns by Heinz, the Barbie movie, and Maggi, and discusses the potential pitfalls and risks associated with AI-powered tools. He concludes with insights into the future of generative AI in digital advertising, highlighting the importance of cultural transformation and the synergy between human creativity and AI.
Announcements
- Hello and welcome to the AI Engineering Podcast, your guide to the fast-moving world of building scalable and maintainable AI systems
- Your host is Tobias Macey and today I'm interviewing Praveen Gujar about the applications of generative AI in digital advertising
- Introduction
- How did you get involved in machine learning?
- Can you start by defining "digital advertising" for the scope of this conversation?
- What are the key elements/characteristics/goals of digital avertising?
- In the world before generative AI, what did a typical end-to-end advertising campaign workflow look like?
- What are the stages of that workflow where generative AI are proving to be most useful?
- How do the current limitations of generative AI (e.g. hallucinations, non-determinism) impact the ways in which they can be used?
- What are the stages of that workflow where generative AI are proving to be most useful?
- What are the technological and organizational systems that need to be implemented to effectively apply generative AI in public-facing applications that are so closely tied to brand/company image?
- What are the elements of user education/expectation setting that are necessary when working with marketing/advertising personnel to help avoid damage to the brands?
- What are some examples of applications for generative AI in digital advertising that have gone well?
- Any that have gone wrong?
- What are the most interesting, innovative, or unexpected ways that you have seen generative AI used in digital advertising?
- What are the most interesting, unexpected, or challenging lessons that you have learned while working on digital advertising applications of generative AI?
- When is generative AI the wrong choice?
- What are your future predictions for the use of generative AI in dgital advertising?
Parting Question
- From your perspective, what is the biggest barrier to adoption of machine learning today?
- Thank you for listening! Don't forget to check out our other shows. The Data Engineering Podcast covers the latest on modern data management. Podcast.__init__ covers the Python language, its community, and the innovative ways it is being used.
- Visit the site to subscribe to the show, sign up for the mailing list, and read the show notes.
- If you've learned something or tried out a project from the show then tell us about it! Email hosts@aiengineeringpodcast.com with your story.
- To help other people find the show please leave a review on iTunes and tell your friends and co-workers.
The intro and outro music is from Hitman's Lovesong feat. Paola Graziano by The Freak Fandango Orchestra/CC BY-SA 3.0
[00:00:05]
Tobias Macey:
Hello, and welcome to the AI Engineering podcast, your guide to the fast moving world of building scalable and maintainable AI systems. Your host is Tobias Macy. And today, I'm interviewing Praveen Gujar about the applications of generative AI for digital advertising. So, Praveen, can you start by introducing yourself? Thanks, Tobias. Thanks for having me again here. This is,
[00:00:32] Praveen Gujar:
exciting to talk about generated AI, which is pretty much the household name these days. Even my mom knows about generated AI these days. So to begin with, my name is Praveen Gujar. I'm a director of product at LinkedIn, and, my expertise, involves digital advertising cloud, and, all of these areas are really powered by AI and generated AI systems as well. Over the course of last 15 years, I have been instrumental in building large scale enterprise products, in companies like LinkedIn, Amazon, as well as, Twitter. And, yeah, I think not just products but also, multibillion dollar businesses in these larger tech companies. So, yeah, looking forward to talking about generative AI,
[00:01:17] Tobias Macey:
and its influence on digital advertising today. And do you remember how you first got started working in machine learning and AI?
[00:01:24] Praveen Gujar:
Yeah. Of course. I think as many of us in the tech industry know that machine learning and AI have have been here for decades right now. But they are they were just not household names, be it recommendation that are surfacing your ecommerce website, recommendations that are coming out, of your famous streaming service as well. These are all powered by, AIs and machine learnings. What, what has become recently is they had a lot more household name because of the success of OpenAI and the chat GPDs. As I said, my experience involves in 2 areas. Right? Cloud and with the overlap of our data products and also digital advertising.
For, decades, these areas have been, powered by machine learning and area algorithms. If you take an example in data products, be it your anomaly detection, products, right, where you detect, you know, when you detect anomalies within your business metrics, these are all, governed by, sophisticated AI algorithms. Or if you're basically building an experimentation platform, how to pace an experiment and how to, increase your targeting audience is all governed by AI algorithms as well. Similarly, in digital ad tech, I think AI and machine learning have been there for tickets as well, where areas like targeting, predictive analytics, serving, bidding and optimizations, and pacing optimization. These are all governed by complex AI algorithms because these are not something that inherently human beings can do that well as AIs.
So as being in this industry, I've been associated with the annual area for more than a decade right now. And, yeah, I think, now that it's a house household name, I, I see that popularity rising.
[00:03:09] Tobias Macey:
And for the purpose of this conversation, we're focusing on the application of these newer AI models in the context of digital advertising, and I'm wondering if you can just give some clarifying definition about what digital advertising is and some of the avenues that it encompasses.
[00:03:26] Praveen Gujar:
Yeah. Let's, take a simple, example or even describe this what digital advertising is first in the scope. Right? Traditionally, how advertising started was print media and, linear TV, just TV, what it used to be called, before, right, where you used to see advertisements of 30 seconds in 1 year or print ads on your newspapers and magazines as well. They suffered from, a few things. One, they are very generic. It was something that was not personalized for an individual. And measuring, impact or ROI, return on investment, all of these were really difficult, as well. With, advent of Internet and digital advertising started to become a new medium for advertisers to promote their brand and generate leads as well. And with that, they were able to like, what the digital advertising's, promise was to solve these two problem. It is to personalize the message that you have for your target audience in the right channel. And second one was to provide you tools and the metrics to measure ROI of your investment as well. So if you break it down, digital advertising has a few key areas.
One is targeting your audience, reaching them. Second one is reaching them with specific message at the time they are most active, and in the channel, they are most active. And when when we talk about channel today, you have, proliferation of devices that you use. You use TV. You use mobile. You use desktops. Reaching them in the right channel, in the social media or email or anything else is very relevant as well. So that's the score. That's the promise of digital advertising. 3rd, basically, is what I talked about measurement. Right? Giving more granular metrics in terms of engagement and clicks, conversion, and leads. All those things becomes very critical for a marketer to measure ROI. So, this is another aspect of digital advertising that is, critical for the marketer. So these are the scope. But in today's discussion, I think really focusing on content generation and content strategy using Gen AI and LLMs is, the key thing here. That will be the scope of our conversation today. And in digital advertising, you mentioned a number of these different mediums, but what is typically the overarching goal of a particular ad campaign or a particular advertisement?
Yeah. That's a good question. If you are a marketer, you are tasked with, certain goals. Typically, let's just simplify it to 4 different goals. One is brand awareness. How do you reach to audience that may not, be brand, that the brand is not only reaching, increase the reach, increase the retention, of brand brand as well in, these users' mind. That's one of the key codes that the marketer will have as well. And you can see this as, like, a funnel as well. That's the top of the funnel. The second is to basically engage with the audience at the right time and right channel as well. I talked about how we as users consume so many different medium to the email, social media, over a mobile desktop, or even, connected TV as well, Netflix and Hulu's of the world. So how do you reach them with the right message, to show that they can you can increase engagement with your messages. That's a key goal as well.
And you're making the user move down the funnel. The third aspect would be to generate lead after this. That means, basically, you are have a potential buyer who are interested in your brand and in your product now, and you generate leads that can then be handed over to sales to follow-up or that can be handed over to your systems for conversion. The 4th but not the least is to convert these leads into actual customers that you, who buy your products as well. So this is, typically what a marketer really focuses on to make a potential buyer or a user, drive from the brand awareness to convert them to buy their products and services.
[00:07:17] Tobias Macey:
That's typically what the marketer's goal is. Taking a step backward in time to a few years ago before generative AI became this massive force across a number of different industries and use cases, what did a typical end to end advertising campaign workflow look like, and who are the people that needed to be involved to ensure that the entire sequence was executed successfully?
[00:07:43] Praveen Gujar:
Yeah. That's an awesome question. Let's also take it a little further back as well, not just in AI, but how AI, before, even the, more ubiquitous use of AI as well. So I think, there are 4 key stages, in running a marketing campaign. Let's just summarize first, and then we can talk about them individually. Number 1, is planning, research and planning. Second 1 is, content strategy and content creation. 3rd one is campaign creation and serving of the campaigns. The 4th one is measurement and optimization. So you measure the outcome and then you tweak the campaigns. So let's take a, first step. I talked about market research and planning.
These are different individuals who basically specialize in analyzing what the market trends are, who the specific target audience for your, brands are, and how what is the right way to basically reach them as well. These are typically planners or market researchers who do this work as well. The second piece of the puzzle is content creation. So the creative directors in your, agencies are in your in house marketing team. Basically, go take that information about the market research to have a content strategy and generate multiple copies of content, through different mediums. It may be images, videos, emails, similar, form factors as well. And these are typically the creative directors, as I said. The third one is, like, a marketer, what we term as a marketer, who brings them together. And along with the planner works on what is the right channel to reach this audience and what is it and along with the creative director, what is the right copy or what is the right creative to use for these channels as well. They stitch everything together, have a campaign created, and serve the campaign or launch the campaign to the target audience. Last but not the least, either marketer or analyst in your house, basically and try to understand the performance of the, campaign and optimize or fine tune the campaign to for a desired result as well. So you can imagine here, there are multiple players in, who are working together. So there are obviously key challenges.
A tight collaboration, with these individuals are very key. They often don't speak the same languages. So, a marketer typically plays a very key role in bringing all them together, stitching everything together, and making key decisions. And these are all the problems that, AI and now Gen AI are solving to an extent. We are released to a world where a marketer alone can launch a single campaign with a single click of a button, where Gen AI can actually read through your product catalog, for example, and come up with ideas about, what products to basically what campaigns to launch specific to a product, even generate copies of what the messaging should be that the marketer can tweak. And then with single click of button, by using back in the AI algorithms, you can do auto bidding, so that, you have higher propensity of winning the ads and then basically serving to your target audience. And all these things are fully automated in a way that optimized and continuously learned so that you can, continuously improve and deliver higher ROI to your brand as well. So, yeah, this is, where the journey was and how it is transformed with the advent of Gen AI and also lot more sophistication in, AI over a decade plus. In those stages
[00:11:09] Tobias Macey:
of the workflow, you mentioned a few where generative AI is starting to be incorporated. Obviously, the temptation is, oh, I've got this magical tool. It will do everything I want, and so you want to apply it everywhere possible. And I'm curious if there are any components of that workflow or any pieces of the digital advertising life cycle that are best suited to the use of these generative AI algorithms and some of the cases where the limitations of generative AI prevent it from being a good application, particularly in the case where you have the problem around hallucinations, invalid information, nondeterminism?
[00:11:47] Praveen Gujar:
Yeah. That's a good question. Let's look at from both, what's working and what's not working, angle. Let's take what's, really generative AI, very useful for at this point. There are multiple places where, generative AI plays a very critical role, but I'll, list 3 key ones. So if you take a step back, what LLMs and gen AI can do a lot better is basically, synthesize a large amount of data and, identify patterns and then use those pattern to generate, content as well. And if you directly apply these capabilities to digital advertising, 3 key areas can be influenced quite a bit. One is your audience segmentation. So So it's critical. Right? Every single marketer worries about, the number one thing is who is my target audience. They have to be as precise as possible, as, as segmented as possible so that they can actually reach the right audience with the right message as well. So that's really where, understanding your previous camp like, understanding your data from your previous campaigns, how audiences have reacted to campaigns.
LLMs and can play a very critical role in identifying the right segments, clustering the segments in, suitable way so that the marketer can actually make meaningful sense out of it and have different strategies for different audience segments. So that's number 1. Right? Audience and segmentation there. Number 2 is content strategy and content creation as well. Generative AI is really excels at generative content. Like, synthesizing this data to have a strategy on what is the right kind of content for what channel and what format is really key. And, whether it's basically transformer based models, DALL E, text to speech, text to image converter models and everything can play a very critical role in generating the content.
So that's this, third one. And second 4 and sorry. 2nd one. The third one is, personalization of ads and dynamic serving of this. So now we are bringing these 2 together, like the audience segmentation and content creation. So you can now decide at scale which is the right, ad to solve for a while, creative to solve for an individual in the context of what they are seeing in the ads as well. And that's really where Gen AI excels in that personalization and dynamically creating the content and serving the content at the time of, ad serving, so that, in the content that the user is seeing is more personalized, more relevant in the context, and more relevant to the channel that they are viewing as well. So what I what I mean by that so if you are basically having it in the social media, the right form factor, over Instagram, for example, matters. So being with that form factor creative is important. So all these, things are key that, Gen AI is now powering and excelling it as well. Things can only improve from here. Now that the other side of the equation, right, what does, it suffer from as well? I think you rightfully pointed out about the hallucination and nondeterminism nature of it. Right? And if you take these 3 same three bucket, let's talk about audience segmentation.
So I think the key challenges there would be specifically if you have minimal data or if you're not augmenting, the foundational models with your own brand specific data, audience segmentation may not be accurate or may not be accurate enough for you to run meaningful campaigns. That's the hallucination effect. The non deterministic effort is that segmentation may change from campaign to campaigns, even if your goals are same. That's basically, one of the side effect. The same applies to creatives or content generation, as well. I think misrepresentation, biases in content created are very key, aspects that can damage your brand. That's where human intervention is very, very key, before a content basically get released. And that's the that aspect is today, we are not 100% there when serving these dynamically created, dynamically generated ad at the time of serving to, users.
This is because of all the challenges that we face with hallucination, non domestic nature, a human feedback, or an, a supplementary AI, like a region, RLS, is necessary, basically, to fine tune these models to make sure that, they produce more accurate, content, audience segments.
[00:16:07] Tobias Macey:
From the technological and organizational systems perspective, obviously, you can't go from, I have my marketing team and my engineering team and my analysis team, and they're all working together on these digital campaigns directly to, I have one marketer and a bunch of magical AI tools and everything is great. So I'm curious, what are the systems and technologies that need to be implemented and organized and planned around to be able to effectively bring these generative AI tools into that context
[00:16:53] Praveen Gujar:
Yeah. I Yeah. I think, the future is, still evolving as you can imagine, but I think we can talk in terms of what it is, today. Right? 1st and foremost, I think, organizations, big or small, needs to adapt culturally as well when it comes to the influence of Gen AI in marketing capabilities, whether you have a in house, marketing team that is doing all the marketing, act operations for your brand, or you are collaborating with an agency, Having that cultural transformation to leverage, Genai or AI based tools to be more effective at your job is very key as well. And that cultural transformation can be most efficiently brought on top top down, making sure that the organization structures and everything are structured in a way that it actually helps that, culture to thrive as well. The second, I think, organizational, structure is very key as well. This probably is a little bit more relevant to large organizations.
When you have an in house marketing team and who also collaborate with the, external agencies, embedding the right talent, within these teams is very important. For example, having a data scientist or an AI engineer embedded within the marketing team is very key for success because they can lot more closely collaborate with the marketer. They understand their pain point a lot better. They understand the pain point of serving the campaign a lot better, as well. So then they can fine tune the experience for the marketer so that organizational transformation is our structure is very important for successful as well. And, last but not the least, I think here, the brand identity and brand image is very key. So having right checks and balances, to ensure that anything that the Gen AI content created, at least today, has a human review, which is very important in most cases. Right? It's not like you're done 1,000 and millions of, campaigns. So human intervention in reviewing these campaigns and making sure that the creatives are contained that you're actually generating, adherence to the brand guidelines, brand identity, the tone, look, feel, all those things are very key, to make sure that you have the right checks and balances before these ad campaigns are live. So they produce a desired result and not an outcome that is unwanted for the brand. The other aspect,
[00:19:19] Tobias Macey:
bringing these tools into a context in which they are going to be producing content or maybe even directly interacting with end users and customers obviously has a certain amount element of risk to it that also has a potential for a big reward if the advertising campaign is successful. What are the aspects of education and expectation setting that are necessary when you're working with those marketing and advertising personnel to make sure that they understand the risk factors and some of the ways that they can mitigate those potential error situations and the types of monitoring that they need to do to make sure that those generative AI tools are behaving themselves?
[00:20:02] Praveen Gujar:
Yeah. I think, that's a great question, and we see this on a regular basis. And this is not really limited to digital ad tech. Right? Wherever AI is used. One of the aspects that I very closely work is on the trust side as well. Right? And it's very relevant in that space as well. So, number 1, I think, training plays a very key role in which, educating marketer and the human reviewers who are custodian here for brand identity and image on the capabilities of the AI and the limitations of the AI as well. So when they understand this very fully, they can actually make meaningful decisions when they have to make a a call on a creative going live or not. That's 1.
That's number 1. Number 2 is, continuous learning is very important here. As AIs evolve as well, their output, evolves as well. As your argument models, the foundational model, especially with more brand aware data, as well, they, the output changes as well. So, these marketers and the human reviewers should keep pace with those changing needs. So training them continuously, making sure that they are in line with the the technological advancements, become very critical as well. 3rd, I think providing tools for especially for the reviewers is very key with clear areas, they can they have to focus on, so that they can actually make, informed decisions.
What we measure basically here is, like, the decision quality of these reviewers. Basically, take a sample of, decisions that they have actually made and try to understand how they are how they were made and how many of them were in line with the expectations versus not is very key. So providing them tools and, as well on top of, education to make those decisions time and again, with higher accuracy is important. So I think, to summarize, education, continuous learning, tools to basically make informed decisions are very essential to ensure the launch of your campaigns that are powered by Gen AI is meaningful to your users.
[00:22:10] Tobias Macey:
To that point of the evolution of the models, the evolution of the tools, they are definitely moving very quickly. What is a typical time frame or life cycle of a given campaign, and what are some of the ways that the length of that campaign factors into the ways that you go about the selection of which models and which tool chains to use and how the evolution of the tool chains maybe prompts the time frame in which you want to keep a certain campaign alive.
[00:22:42] Praveen Gujar:
Yeah. I think, like, these days, these campaigns are lot like, the model's a lot more effective for campaigns that are shorter duration than long running, campaigns because the models are continuously learning. And, also, especially in the initial days, their output basically may significantly vary. So what do brands typically do? These are campaign they run campaigns that are typically 1 to 2 weeks, is one category. 1 month to 3 months is second category. And the 3rd category is long running campaigns that are basically, maybe sometimes go for 6 months or a year. I think what we are seeing evidence wise is basically that we are not more effective for shorter campaign durations, so that you can actually quickly learn experiment from, different content generated through Gen AI, whether it's ad copy, whether it's images that are generating, or whether it's the look and feel of the creative that are being generated as well, and then experiment, with that to take it to the next level. That's how that's where a lot more effectiveness is. I think, for a long running campaign that can go for 6 to, 9 months, we we as marketers still rely on lot more of human intervention than for shorter campaigns.
[00:23:53] Tobias Macey:
In terms of concrete examples, I'm wondering if there are any campaigns that relied on generative AI that you've seen that went particularly well and some of the strategies that they used to be able to manage the AI capabilities in that campaign context effectively.
[00:24:12] Praveen Gujar:
Yeah. I think and let's talk about three examples that I'm particularly a fan of. But before that, I think this is something that I wanted to call out and how critical brand awareness is for a, for a brand and marketer. Right? I saw, an ad on, a a reel actually on Instagram today where it was an ad which was 20 years, 30 years back where McDonald's, there was, like, the ad that McDonald's published was an individual goes to different countries where he can't speak the native languages, and they don't understand anything, there he says.
But when he asked for McDonald's, a direction to McDonald's, every single individual points them to it. So that basically, what McDonald's wants to highlight here is the brand awareness and brand recall and the brand image with so many different, culturally diverse, individuals in different countries. So this is a key thing. And if you I gave this context because the first case case study that we are gonna talk about is a great example of it. So when it comes to, catch up, what comes to your mind? Heinz. There you go. Right? So Heinz wanted to basically use this, as a great, as a great tool to further enhance their brand awareness. So they launched a campaign where they, give simple prompts to charge a video on any based models, like ketchup bottle, ketchup bottle with x. And a lot of the images that basically came out looked very similar to what Heinz, would package their ketchup in. And that ad became extremely vital on social media, even trickling challenges, with, individuals where, they could basically give their own prompts, to generate Heinz look alike brand, bottles as well. That basically shows the power of how Gen AI can be used for creating more brand awareness, especially when you have a very established brand. You can basically go instantly viral. That's a great use case that I have seen, recently.
The second one, is, a metal, like, the Barbie movie that actually came out. So how do you generate more awareness about the movie? Right? It's basically you go viral in social media. So for that, they teamed up with a agency of theirs to create a generative AI based tool where an individual can upload a photo. And, is it if that's the right word. What I mean, basically, is those portraits basically then gets converted into a Barbie background and feel and everything, so that you you you feel like you are in the Barbie land. That, ad campaign became extremely powerful, driving a lot of the traffic to the CA generated tool where, individuals would, upload their photos to create their own, Barbie images and basically post it back on social media. Again, a a great example of how generative AI enabled them to fast track these solutions, launch it, and go viral, as well.
Last but not the least, the Next layer, which is a very popular consumer brand, has, similar to what, we see here Topgeman. They have a a product called Maggi, which is in Asia Pacific. And they used the interactive capability of generative AI to, launch a campaign where the interactive campaign would ask for individuals or users to provide the ingredients they want to use in, in the Maggie, and they would, generate a specific recipe based on the ingredients you actually provide. You can, understand how brand is connecting with the individuals. It's so fun for the individuals, that uniqueness that, that, generates basically a lot more traction with the individuals, building more brand image as well. So these are great examples of how GIN AI has helped brands to building their increase their brand image as well as, connect with their users in more fun manner.
[00:28:12] Tobias Macey:
And on the Converse, obviously, there have been numerous examples in recent memory of cases where different brands or companies took some AI powered tool, put it on the Internet, and everything went wrong. And I'm wondering if there are any methods of failure that you've seen in this context of digital advertising campaigns and ways to potentially guard against or prevent those situations?
[00:28:40] Praveen Gujar:
Yeah. I think, there are plenty of them. So let's, refrain from using brands here and specific examples, but let me categorize 3 areas where we have seen, a lot of these issues. Right? It's basically, one of the top thing that happens is when, your campaigns are not in line with your brand wise and image web. So, if you are Coca Cola, you have a clear brand image or a red bubbly, sauna drink that, resonates with every individual. But if your campaign is seeing otherwise, that basically damages your, brand, reputation. So, brand wise, inconsistencies is really key, and that's bay if it's the key reason for failures, I would say.
Number 2 is, transparency. Right? I think, there is a general way of, increasing transparency, about AI, about, any data that's being used to basically, generate and launch campaigns. And, if the users feel like there are AI generated images, but, the brand is not being transparent about that, that's an area where the brands have issued backlash from the users. 3rd but not the least is obvious. I think this should be obvious for all of us is in bias in the content, whether it's basically bias to a particular ethnicity, where our our, a segment of population or a a particular, anything like that can basically, inevitably create a lot of a problem for the brand. And these are the top three categories where use of Gen AI, when not done right, with not no clear human intuition or no clear argumentation of data to make sure that, your adherence to that, no checks and balances can go horribly wrong.
And, yeah, I think, this is something that every branch should be aware of and, try to minimize.
[00:30:35] Tobias Macey:
In your work in this space, working with brands, working with these generative AI tools, what are some of the most interesting or unexpected or challenging lessons that you've learned in this context?
[00:30:47] Praveen Gujar:
Yeah. I think, the challenging, thing basically is, number one thing, I was recently presenting in one of the marketing conference. And number one thing that I basically emphasized again and again, which has been a very key learning, we shouldn't think about, Gen AI or AI basically replacing human beings, but how, humans and AI can coexist and complement each other. What we have seen again and again with evidence is a combination of AI systems plus human creativity is a clear winner. I think this is the case where 1 plus 1 is 3 rather than 2. That's been one of the biggest learning, and biggest ever like, I've seen multiple evidence of this.
I think that's one of the reasons why we say that, the human creativity is key to come up with, creative ideas, then, basically, AI can, make it at scale, take it up to scale, make the mundane job lot more easier, for the human beings as well so they can focus more on strategy. I think, the McDonald's ad that I just talked about is a great example, that the creative thinking of creating this ad ad is what is human really good at at this point, before AGI becomes a reality, at least. But if you extend it to, like, when used with, JennyI, you can actually productionize these kinds of an ad a lot faster, can create different variants of this ads for specific audience segments much faster as well. So that's where you have a clear winner, and we have seen that, again and again in evidence.
[00:32:32] Tobias Macey:
For people who are working in this space or thinking about how to change the way that they're approaching their advertising campaigns? What are the cases where generative AI is just the wrong choice?
[00:32:45] Praveen Gujar:
Yeah. I think let's, start with a few things here. Right? I don't know if this is, true in this day and age, but if you are new to digital advertising, that means basically you don't have, data to basically argument the foundational models with your own brand specific information. In those cases, it's definitely not the best choice, for example. I don't really struggle to think about what are these use cases where brands have never really used it. Maybe if you are a startup, this is the first advertisement you're basically reading, then, yeah, I would rather rely on human intervention, here than using Geni based models to create it because you'd have very less information to augment your own brand data, for these foundational models to make sure that these are not more customized for your brand image. Right? I think that's one clear case where you have, you have challenges. The second one is, if you are specifically in extremely sensitive industry, and also targeting to extremely sensitive audience. This is where I would say it may not make sense as well. For example, if you're targeting a fat loss bill, for example, brands would want to be really, really careful about how they're going about the generative AI, usage in these campaigns, as well.
And I think the, the third aspect is what you last, asked as well, like, long running campaigns that you wanna basically experiment for a longer run. At this time, Trinity AI may not be the right fit, for those kinds of, long running campaigns where you want more interested in getting market sentiment in a long running campaign as well. So those are 3 areas, I would say. The way I may not be the best fit at this point.
[00:34:31] Tobias Macey:
As you continue to keep an eye on this space, work with different brands, work with these AI tools, what are the predictions that you have for future applications of generative AI for digital advertising?
[00:34:44] Praveen Gujar:
Yeah. As I, look forward, I think that 3 key things that are really interesting to me, and I'm actually even writing a a paper about this that we soon published. Number 1, basically, is multimodal, content, that are trained with the multimodal models and are in rich format as well. So if you look at, connected TV is a very emerging space in digital ad tech. After years of subscription only model, even Netflix has gravitated towards, multiple reasons to it, but, one of the reasons also is the success of connected TV. So how can we basically, generate TV ready content using multimodal models? We are not there yet. That's one of the things that I'm really, excited about. I think we all might have seen, the demo videos of Swara.
There are imperfections in that still, but I think that's one of the areas where we can make significant advancements so that, these, generated become content can also be shown on a big screen like a television. Number 2, immersive, ads. These are something that I'm super excited about now that specifically Apple is in the AR and VR kind of, mindset. I think, like, brands can really engage with, users. Like, imagine, you're trying to basically shop for a, soft shop for a table lamp, and you receive an ad that helps you to basically understand what kind of a table lamp fits on your, tables. Like, I think some of this actually happens today. If you're in Amazon.com, they enable you out in the Wayfair.
They enable you to look what the product is, but imagine this is dynamically generated ad to a user who can interact with the ad in an immersive environment as well. 3rd is basically interactive ads, where you ask you provide prompts to, the users. And based on the prompt, you basically customize the ads that are served to the users as well. It's not only fun and engaging to the users, but also you make it a lot more relevant to the users as well. I think these are the capabilities that, without Jenny I, we couldn't have imagined in such a short duration. So I think these three areas are where I'm really excited about, and looking forward to how marketers will use Evergrow, like, these models to basically create these kinds of, creative contents.
[00:37:04] Tobias Macey:
Are there any other aspects of this space of generative AI, its application to digital advertising, and some of the risks and strategies involved that we didn't discuss yet that you'd like to cover before we close out the show?
[00:37:18] Praveen Gujar:
Yeah. I think, one thing that basically I will emphasize, I talked about this a little bit, but, I think every change basically has a lot of resistance in an organization as well. And if you are a brand with huge, in house marketing team, they will be, hesitant to adapt to these, to begin with. A cultural change is very important. I cannot, you know, emphasize that enough as a medium to basically make transformative changes in your organization as well. Whether we like it or not, this tide is gonna basically take us all, and adapt into it in the right manner is very key.
This is very specific to, like, the mobile desktop, era as well and how teams and how organizations have to team up by that as well and who didn't were left behind. The second basically is, human, and AI. Like, synergy is very key. I think, when it comes to brand image, management, when it comes to, making sure that, you minimize the biases introduced us, today. Our overall success, I think human plus AI system synergy is very key. I wouldn't basically that that way is second key things as well. I think that, the last piece is data clean cleanliness and standardization is equally important, I would say. These foundational models are trained with such a huge amount of data, but you have to argument them with, your own brand specific data. For that, you have to have standardized and clean data in your ecosystem, so you can train these models with your own data and making sure that the content are generally more brand relevant, more in line with what you want, the message to should be. So I would say these are the 3 key things as, like, key things to remember, towards the end.
[00:39:09] Tobias Macey:
Alright. Well, for anybody who wants to get in touch with you and follow along with the work that you're doing, I'll have you add your preferred contact information to the show notes. And as the final question, I'd like to get your perspective on what you see as being the biggest gap in the tooling or technology that's available for AI engineering today.
[00:39:26] Praveen Gujar:
I think, I may basically answer this slightly differently. Instead of focusing on the tools or AI, I think we talked quite a bit about the gaps and how to improve them, how we are actually working on certain models to improve them as well. The biggest barrier I see is cultural transformation. I think that's really the key thing. Like, I've, I I talked about this marketing conference that I was really, part of. I could see every marketer had hesitation about using Gen AI, what does it mean for their own job. Every creative director worried about what it does.
I think, that's, unless the organization go through a cultural transformation of adapting and embracing Gen AI, which is, at this point, I would say, inevitable. I think the success of, the brand, when it comes to our marketing, campaign will be limited. I think that's the the cultural aspect is, I would say, is the biggest barrier, right now.
[00:40:31] Tobias Macey:
Alright. Well, thank you very much for taking the time today to join me and share your experience and perspective of the ways that generative AI is being used in these digital marketing campaigns. It's definitely a very interesting application of these technologies. It's definitely great to hear some of the ways that they can go right and wrong. So I appreciate the time you've taken today, and I hope you enjoy the rest of your day. Yeah. Thank you so much. Thanks for having me,
[00:40:58] Praveen Gujar:
here. It was exciting to talk about the influence of Gen AI in the digital advertising space. I only I can only say that the future is really bright, and I'm super excited about that.
[00:41:14] Tobias Macey:
Thank you for listening, and don't forget to check out our other shows, the Data Engineering Podcast, which covers the latest in modern data management, and podcasts.init, which covers the Python language, its community, and the innovative ways it is being used. You can visit the site at the machine learning podcast.com to subscribe to the show, sign up for the mailing list, and read the show notes. And if you've learned something or tried out a project from the show, then tell us about it. Email hosts at themachinelearningpodcast.com with your story. To help other people find the show, please leave a review on Apple Podcasts and tell your friends and coworkers.
Hello, and welcome to the AI Engineering podcast, your guide to the fast moving world of building scalable and maintainable AI systems. Your host is Tobias Macy. And today, I'm interviewing Praveen Gujar about the applications of generative AI for digital advertising. So, Praveen, can you start by introducing yourself? Thanks, Tobias. Thanks for having me again here. This is,
[00:00:32] Praveen Gujar:
exciting to talk about generated AI, which is pretty much the household name these days. Even my mom knows about generated AI these days. So to begin with, my name is Praveen Gujar. I'm a director of product at LinkedIn, and, my expertise, involves digital advertising cloud, and, all of these areas are really powered by AI and generated AI systems as well. Over the course of last 15 years, I have been instrumental in building large scale enterprise products, in companies like LinkedIn, Amazon, as well as, Twitter. And, yeah, I think not just products but also, multibillion dollar businesses in these larger tech companies. So, yeah, looking forward to talking about generative AI,
[00:01:17] Tobias Macey:
and its influence on digital advertising today. And do you remember how you first got started working in machine learning and AI?
[00:01:24] Praveen Gujar:
Yeah. Of course. I think as many of us in the tech industry know that machine learning and AI have have been here for decades right now. But they are they were just not household names, be it recommendation that are surfacing your ecommerce website, recommendations that are coming out, of your famous streaming service as well. These are all powered by, AIs and machine learnings. What, what has become recently is they had a lot more household name because of the success of OpenAI and the chat GPDs. As I said, my experience involves in 2 areas. Right? Cloud and with the overlap of our data products and also digital advertising.
For, decades, these areas have been, powered by machine learning and area algorithms. If you take an example in data products, be it your anomaly detection, products, right, where you detect, you know, when you detect anomalies within your business metrics, these are all, governed by, sophisticated AI algorithms. Or if you're basically building an experimentation platform, how to pace an experiment and how to, increase your targeting audience is all governed by AI algorithms as well. Similarly, in digital ad tech, I think AI and machine learning have been there for tickets as well, where areas like targeting, predictive analytics, serving, bidding and optimizations, and pacing optimization. These are all governed by complex AI algorithms because these are not something that inherently human beings can do that well as AIs.
So as being in this industry, I've been associated with the annual area for more than a decade right now. And, yeah, I think, now that it's a house household name, I, I see that popularity rising.
[00:03:09] Tobias Macey:
And for the purpose of this conversation, we're focusing on the application of these newer AI models in the context of digital advertising, and I'm wondering if you can just give some clarifying definition about what digital advertising is and some of the avenues that it encompasses.
[00:03:26] Praveen Gujar:
Yeah. Let's, take a simple, example or even describe this what digital advertising is first in the scope. Right? Traditionally, how advertising started was print media and, linear TV, just TV, what it used to be called, before, right, where you used to see advertisements of 30 seconds in 1 year or print ads on your newspapers and magazines as well. They suffered from, a few things. One, they are very generic. It was something that was not personalized for an individual. And measuring, impact or ROI, return on investment, all of these were really difficult, as well. With, advent of Internet and digital advertising started to become a new medium for advertisers to promote their brand and generate leads as well. And with that, they were able to like, what the digital advertising's, promise was to solve these two problem. It is to personalize the message that you have for your target audience in the right channel. And second one was to provide you tools and the metrics to measure ROI of your investment as well. So if you break it down, digital advertising has a few key areas.
One is targeting your audience, reaching them. Second one is reaching them with specific message at the time they are most active, and in the channel, they are most active. And when when we talk about channel today, you have, proliferation of devices that you use. You use TV. You use mobile. You use desktops. Reaching them in the right channel, in the social media or email or anything else is very relevant as well. So that's the score. That's the promise of digital advertising. 3rd, basically, is what I talked about measurement. Right? Giving more granular metrics in terms of engagement and clicks, conversion, and leads. All those things becomes very critical for a marketer to measure ROI. So, this is another aspect of digital advertising that is, critical for the marketer. So these are the scope. But in today's discussion, I think really focusing on content generation and content strategy using Gen AI and LLMs is, the key thing here. That will be the scope of our conversation today. And in digital advertising, you mentioned a number of these different mediums, but what is typically the overarching goal of a particular ad campaign or a particular advertisement?
Yeah. That's a good question. If you are a marketer, you are tasked with, certain goals. Typically, let's just simplify it to 4 different goals. One is brand awareness. How do you reach to audience that may not, be brand, that the brand is not only reaching, increase the reach, increase the retention, of brand brand as well in, these users' mind. That's one of the key codes that the marketer will have as well. And you can see this as, like, a funnel as well. That's the top of the funnel. The second is to basically engage with the audience at the right time and right channel as well. I talked about how we as users consume so many different medium to the email, social media, over a mobile desktop, or even, connected TV as well, Netflix and Hulu's of the world. So how do you reach them with the right message, to show that they can you can increase engagement with your messages. That's a key goal as well.
And you're making the user move down the funnel. The third aspect would be to generate lead after this. That means, basically, you are have a potential buyer who are interested in your brand and in your product now, and you generate leads that can then be handed over to sales to follow-up or that can be handed over to your systems for conversion. The 4th but not the least is to convert these leads into actual customers that you, who buy your products as well. So this is, typically what a marketer really focuses on to make a potential buyer or a user, drive from the brand awareness to convert them to buy their products and services.
[00:07:17] Tobias Macey:
That's typically what the marketer's goal is. Taking a step backward in time to a few years ago before generative AI became this massive force across a number of different industries and use cases, what did a typical end to end advertising campaign workflow look like, and who are the people that needed to be involved to ensure that the entire sequence was executed successfully?
[00:07:43] Praveen Gujar:
Yeah. That's an awesome question. Let's also take it a little further back as well, not just in AI, but how AI, before, even the, more ubiquitous use of AI as well. So I think, there are 4 key stages, in running a marketing campaign. Let's just summarize first, and then we can talk about them individually. Number 1, is planning, research and planning. Second 1 is, content strategy and content creation. 3rd one is campaign creation and serving of the campaigns. The 4th one is measurement and optimization. So you measure the outcome and then you tweak the campaigns. So let's take a, first step. I talked about market research and planning.
These are different individuals who basically specialize in analyzing what the market trends are, who the specific target audience for your, brands are, and how what is the right way to basically reach them as well. These are typically planners or market researchers who do this work as well. The second piece of the puzzle is content creation. So the creative directors in your, agencies are in your in house marketing team. Basically, go take that information about the market research to have a content strategy and generate multiple copies of content, through different mediums. It may be images, videos, emails, similar, form factors as well. And these are typically the creative directors, as I said. The third one is, like, a marketer, what we term as a marketer, who brings them together. And along with the planner works on what is the right channel to reach this audience and what is it and along with the creative director, what is the right copy or what is the right creative to use for these channels as well. They stitch everything together, have a campaign created, and serve the campaign or launch the campaign to the target audience. Last but not the least, either marketer or analyst in your house, basically and try to understand the performance of the, campaign and optimize or fine tune the campaign to for a desired result as well. So you can imagine here, there are multiple players in, who are working together. So there are obviously key challenges.
A tight collaboration, with these individuals are very key. They often don't speak the same languages. So, a marketer typically plays a very key role in bringing all them together, stitching everything together, and making key decisions. And these are all the problems that, AI and now Gen AI are solving to an extent. We are released to a world where a marketer alone can launch a single campaign with a single click of a button, where Gen AI can actually read through your product catalog, for example, and come up with ideas about, what products to basically what campaigns to launch specific to a product, even generate copies of what the messaging should be that the marketer can tweak. And then with single click of button, by using back in the AI algorithms, you can do auto bidding, so that, you have higher propensity of winning the ads and then basically serving to your target audience. And all these things are fully automated in a way that optimized and continuously learned so that you can, continuously improve and deliver higher ROI to your brand as well. So, yeah, this is, where the journey was and how it is transformed with the advent of Gen AI and also lot more sophistication in, AI over a decade plus. In those stages
[00:11:09] Tobias Macey:
of the workflow, you mentioned a few where generative AI is starting to be incorporated. Obviously, the temptation is, oh, I've got this magical tool. It will do everything I want, and so you want to apply it everywhere possible. And I'm curious if there are any components of that workflow or any pieces of the digital advertising life cycle that are best suited to the use of these generative AI algorithms and some of the cases where the limitations of generative AI prevent it from being a good application, particularly in the case where you have the problem around hallucinations, invalid information, nondeterminism?
[00:11:47] Praveen Gujar:
Yeah. That's a good question. Let's look at from both, what's working and what's not working, angle. Let's take what's, really generative AI, very useful for at this point. There are multiple places where, generative AI plays a very critical role, but I'll, list 3 key ones. So if you take a step back, what LLMs and gen AI can do a lot better is basically, synthesize a large amount of data and, identify patterns and then use those pattern to generate, content as well. And if you directly apply these capabilities to digital advertising, 3 key areas can be influenced quite a bit. One is your audience segmentation. So So it's critical. Right? Every single marketer worries about, the number one thing is who is my target audience. They have to be as precise as possible, as, as segmented as possible so that they can actually reach the right audience with the right message as well. So that's really where, understanding your previous camp like, understanding your data from your previous campaigns, how audiences have reacted to campaigns.
LLMs and can play a very critical role in identifying the right segments, clustering the segments in, suitable way so that the marketer can actually make meaningful sense out of it and have different strategies for different audience segments. So that's number 1. Right? Audience and segmentation there. Number 2 is content strategy and content creation as well. Generative AI is really excels at generative content. Like, synthesizing this data to have a strategy on what is the right kind of content for what channel and what format is really key. And, whether it's basically transformer based models, DALL E, text to speech, text to image converter models and everything can play a very critical role in generating the content.
So that's this, third one. And second 4 and sorry. 2nd one. The third one is, personalization of ads and dynamic serving of this. So now we are bringing these 2 together, like the audience segmentation and content creation. So you can now decide at scale which is the right, ad to solve for a while, creative to solve for an individual in the context of what they are seeing in the ads as well. And that's really where Gen AI excels in that personalization and dynamically creating the content and serving the content at the time of, ad serving, so that, in the content that the user is seeing is more personalized, more relevant in the context, and more relevant to the channel that they are viewing as well. So what I what I mean by that so if you are basically having it in the social media, the right form factor, over Instagram, for example, matters. So being with that form factor creative is important. So all these, things are key that, Gen AI is now powering and excelling it as well. Things can only improve from here. Now that the other side of the equation, right, what does, it suffer from as well? I think you rightfully pointed out about the hallucination and nondeterminism nature of it. Right? And if you take these 3 same three bucket, let's talk about audience segmentation.
So I think the key challenges there would be specifically if you have minimal data or if you're not augmenting, the foundational models with your own brand specific data, audience segmentation may not be accurate or may not be accurate enough for you to run meaningful campaigns. That's the hallucination effect. The non deterministic effort is that segmentation may change from campaign to campaigns, even if your goals are same. That's basically, one of the side effect. The same applies to creatives or content generation, as well. I think misrepresentation, biases in content created are very key, aspects that can damage your brand. That's where human intervention is very, very key, before a content basically get released. And that's the that aspect is today, we are not 100% there when serving these dynamically created, dynamically generated ad at the time of serving to, users.
This is because of all the challenges that we face with hallucination, non domestic nature, a human feedback, or an, a supplementary AI, like a region, RLS, is necessary, basically, to fine tune these models to make sure that, they produce more accurate, content, audience segments.
[00:16:07] Tobias Macey:
From the technological and organizational systems perspective, obviously, you can't go from, I have my marketing team and my engineering team and my analysis team, and they're all working together on these digital campaigns directly to, I have one marketer and a bunch of magical AI tools and everything is great. So I'm curious, what are the systems and technologies that need to be implemented and organized and planned around to be able to effectively bring these generative AI tools into that context
[00:16:53] Praveen Gujar:
Yeah. I Yeah. I think, the future is, still evolving as you can imagine, but I think we can talk in terms of what it is, today. Right? 1st and foremost, I think, organizations, big or small, needs to adapt culturally as well when it comes to the influence of Gen AI in marketing capabilities, whether you have a in house, marketing team that is doing all the marketing, act operations for your brand, or you are collaborating with an agency, Having that cultural transformation to leverage, Genai or AI based tools to be more effective at your job is very key as well. And that cultural transformation can be most efficiently brought on top top down, making sure that the organization structures and everything are structured in a way that it actually helps that, culture to thrive as well. The second, I think, organizational, structure is very key as well. This probably is a little bit more relevant to large organizations.
When you have an in house marketing team and who also collaborate with the, external agencies, embedding the right talent, within these teams is very important. For example, having a data scientist or an AI engineer embedded within the marketing team is very key for success because they can lot more closely collaborate with the marketer. They understand their pain point a lot better. They understand the pain point of serving the campaign a lot better, as well. So then they can fine tune the experience for the marketer so that organizational transformation is our structure is very important for successful as well. And, last but not the least, I think here, the brand identity and brand image is very key. So having right checks and balances, to ensure that anything that the Gen AI content created, at least today, has a human review, which is very important in most cases. Right? It's not like you're done 1,000 and millions of, campaigns. So human intervention in reviewing these campaigns and making sure that the creatives are contained that you're actually generating, adherence to the brand guidelines, brand identity, the tone, look, feel, all those things are very key, to make sure that you have the right checks and balances before these ad campaigns are live. So they produce a desired result and not an outcome that is unwanted for the brand. The other aspect,
[00:19:19] Tobias Macey:
bringing these tools into a context in which they are going to be producing content or maybe even directly interacting with end users and customers obviously has a certain amount element of risk to it that also has a potential for a big reward if the advertising campaign is successful. What are the aspects of education and expectation setting that are necessary when you're working with those marketing and advertising personnel to make sure that they understand the risk factors and some of the ways that they can mitigate those potential error situations and the types of monitoring that they need to do to make sure that those generative AI tools are behaving themselves?
[00:20:02] Praveen Gujar:
Yeah. I think, that's a great question, and we see this on a regular basis. And this is not really limited to digital ad tech. Right? Wherever AI is used. One of the aspects that I very closely work is on the trust side as well. Right? And it's very relevant in that space as well. So, number 1, I think, training plays a very key role in which, educating marketer and the human reviewers who are custodian here for brand identity and image on the capabilities of the AI and the limitations of the AI as well. So when they understand this very fully, they can actually make meaningful decisions when they have to make a a call on a creative going live or not. That's 1.
That's number 1. Number 2 is, continuous learning is very important here. As AIs evolve as well, their output, evolves as well. As your argument models, the foundational model, especially with more brand aware data, as well, they, the output changes as well. So, these marketers and the human reviewers should keep pace with those changing needs. So training them continuously, making sure that they are in line with the the technological advancements, become very critical as well. 3rd, I think providing tools for especially for the reviewers is very key with clear areas, they can they have to focus on, so that they can actually make, informed decisions.
What we measure basically here is, like, the decision quality of these reviewers. Basically, take a sample of, decisions that they have actually made and try to understand how they are how they were made and how many of them were in line with the expectations versus not is very key. So providing them tools and, as well on top of, education to make those decisions time and again, with higher accuracy is important. So I think, to summarize, education, continuous learning, tools to basically make informed decisions are very essential to ensure the launch of your campaigns that are powered by Gen AI is meaningful to your users.
[00:22:10] Tobias Macey:
To that point of the evolution of the models, the evolution of the tools, they are definitely moving very quickly. What is a typical time frame or life cycle of a given campaign, and what are some of the ways that the length of that campaign factors into the ways that you go about the selection of which models and which tool chains to use and how the evolution of the tool chains maybe prompts the time frame in which you want to keep a certain campaign alive.
[00:22:42] Praveen Gujar:
Yeah. I think, like, these days, these campaigns are lot like, the model's a lot more effective for campaigns that are shorter duration than long running, campaigns because the models are continuously learning. And, also, especially in the initial days, their output basically may significantly vary. So what do brands typically do? These are campaign they run campaigns that are typically 1 to 2 weeks, is one category. 1 month to 3 months is second category. And the 3rd category is long running campaigns that are basically, maybe sometimes go for 6 months or a year. I think what we are seeing evidence wise is basically that we are not more effective for shorter campaign durations, so that you can actually quickly learn experiment from, different content generated through Gen AI, whether it's ad copy, whether it's images that are generating, or whether it's the look and feel of the creative that are being generated as well, and then experiment, with that to take it to the next level. That's how that's where a lot more effectiveness is. I think, for a long running campaign that can go for 6 to, 9 months, we we as marketers still rely on lot more of human intervention than for shorter campaigns.
[00:23:53] Tobias Macey:
In terms of concrete examples, I'm wondering if there are any campaigns that relied on generative AI that you've seen that went particularly well and some of the strategies that they used to be able to manage the AI capabilities in that campaign context effectively.
[00:24:12] Praveen Gujar:
Yeah. I think and let's talk about three examples that I'm particularly a fan of. But before that, I think this is something that I wanted to call out and how critical brand awareness is for a, for a brand and marketer. Right? I saw, an ad on, a a reel actually on Instagram today where it was an ad which was 20 years, 30 years back where McDonald's, there was, like, the ad that McDonald's published was an individual goes to different countries where he can't speak the native languages, and they don't understand anything, there he says.
But when he asked for McDonald's, a direction to McDonald's, every single individual points them to it. So that basically, what McDonald's wants to highlight here is the brand awareness and brand recall and the brand image with so many different, culturally diverse, individuals in different countries. So this is a key thing. And if you I gave this context because the first case case study that we are gonna talk about is a great example of it. So when it comes to, catch up, what comes to your mind? Heinz. There you go. Right? So Heinz wanted to basically use this, as a great, as a great tool to further enhance their brand awareness. So they launched a campaign where they, give simple prompts to charge a video on any based models, like ketchup bottle, ketchup bottle with x. And a lot of the images that basically came out looked very similar to what Heinz, would package their ketchup in. And that ad became extremely vital on social media, even trickling challenges, with, individuals where, they could basically give their own prompts, to generate Heinz look alike brand, bottles as well. That basically shows the power of how Gen AI can be used for creating more brand awareness, especially when you have a very established brand. You can basically go instantly viral. That's a great use case that I have seen, recently.
The second one, is, a metal, like, the Barbie movie that actually came out. So how do you generate more awareness about the movie? Right? It's basically you go viral in social media. So for that, they teamed up with a agency of theirs to create a generative AI based tool where an individual can upload a photo. And, is it if that's the right word. What I mean, basically, is those portraits basically then gets converted into a Barbie background and feel and everything, so that you you you feel like you are in the Barbie land. That, ad campaign became extremely powerful, driving a lot of the traffic to the CA generated tool where, individuals would, upload their photos to create their own, Barbie images and basically post it back on social media. Again, a a great example of how generative AI enabled them to fast track these solutions, launch it, and go viral, as well.
Last but not the least, the Next layer, which is a very popular consumer brand, has, similar to what, we see here Topgeman. They have a a product called Maggi, which is in Asia Pacific. And they used the interactive capability of generative AI to, launch a campaign where the interactive campaign would ask for individuals or users to provide the ingredients they want to use in, in the Maggie, and they would, generate a specific recipe based on the ingredients you actually provide. You can, understand how brand is connecting with the individuals. It's so fun for the individuals, that uniqueness that, that, generates basically a lot more traction with the individuals, building more brand image as well. So these are great examples of how GIN AI has helped brands to building their increase their brand image as well as, connect with their users in more fun manner.
[00:28:12] Tobias Macey:
And on the Converse, obviously, there have been numerous examples in recent memory of cases where different brands or companies took some AI powered tool, put it on the Internet, and everything went wrong. And I'm wondering if there are any methods of failure that you've seen in this context of digital advertising campaigns and ways to potentially guard against or prevent those situations?
[00:28:40] Praveen Gujar:
Yeah. I think, there are plenty of them. So let's, refrain from using brands here and specific examples, but let me categorize 3 areas where we have seen, a lot of these issues. Right? It's basically, one of the top thing that happens is when, your campaigns are not in line with your brand wise and image web. So, if you are Coca Cola, you have a clear brand image or a red bubbly, sauna drink that, resonates with every individual. But if your campaign is seeing otherwise, that basically damages your, brand, reputation. So, brand wise, inconsistencies is really key, and that's bay if it's the key reason for failures, I would say.
Number 2 is, transparency. Right? I think, there is a general way of, increasing transparency, about AI, about, any data that's being used to basically, generate and launch campaigns. And, if the users feel like there are AI generated images, but, the brand is not being transparent about that, that's an area where the brands have issued backlash from the users. 3rd but not the least is obvious. I think this should be obvious for all of us is in bias in the content, whether it's basically bias to a particular ethnicity, where our our, a segment of population or a a particular, anything like that can basically, inevitably create a lot of a problem for the brand. And these are the top three categories where use of Gen AI, when not done right, with not no clear human intuition or no clear argumentation of data to make sure that, your adherence to that, no checks and balances can go horribly wrong.
And, yeah, I think, this is something that every branch should be aware of and, try to minimize.
[00:30:35] Tobias Macey:
In your work in this space, working with brands, working with these generative AI tools, what are some of the most interesting or unexpected or challenging lessons that you've learned in this context?
[00:30:47] Praveen Gujar:
Yeah. I think, the challenging, thing basically is, number one thing, I was recently presenting in one of the marketing conference. And number one thing that I basically emphasized again and again, which has been a very key learning, we shouldn't think about, Gen AI or AI basically replacing human beings, but how, humans and AI can coexist and complement each other. What we have seen again and again with evidence is a combination of AI systems plus human creativity is a clear winner. I think this is the case where 1 plus 1 is 3 rather than 2. That's been one of the biggest learning, and biggest ever like, I've seen multiple evidence of this.
I think that's one of the reasons why we say that, the human creativity is key to come up with, creative ideas, then, basically, AI can, make it at scale, take it up to scale, make the mundane job lot more easier, for the human beings as well so they can focus more on strategy. I think, the McDonald's ad that I just talked about is a great example, that the creative thinking of creating this ad ad is what is human really good at at this point, before AGI becomes a reality, at least. But if you extend it to, like, when used with, JennyI, you can actually productionize these kinds of an ad a lot faster, can create different variants of this ads for specific audience segments much faster as well. So that's where you have a clear winner, and we have seen that, again and again in evidence.
[00:32:32] Tobias Macey:
For people who are working in this space or thinking about how to change the way that they're approaching their advertising campaigns? What are the cases where generative AI is just the wrong choice?
[00:32:45] Praveen Gujar:
Yeah. I think let's, start with a few things here. Right? I don't know if this is, true in this day and age, but if you are new to digital advertising, that means basically you don't have, data to basically argument the foundational models with your own brand specific information. In those cases, it's definitely not the best choice, for example. I don't really struggle to think about what are these use cases where brands have never really used it. Maybe if you are a startup, this is the first advertisement you're basically reading, then, yeah, I would rather rely on human intervention, here than using Geni based models to create it because you'd have very less information to augment your own brand data, for these foundational models to make sure that these are not more customized for your brand image. Right? I think that's one clear case where you have, you have challenges. The second one is, if you are specifically in extremely sensitive industry, and also targeting to extremely sensitive audience. This is where I would say it may not make sense as well. For example, if you're targeting a fat loss bill, for example, brands would want to be really, really careful about how they're going about the generative AI, usage in these campaigns, as well.
And I think the, the third aspect is what you last, asked as well, like, long running campaigns that you wanna basically experiment for a longer run. At this time, Trinity AI may not be the right fit, for those kinds of, long running campaigns where you want more interested in getting market sentiment in a long running campaign as well. So those are 3 areas, I would say. The way I may not be the best fit at this point.
[00:34:31] Tobias Macey:
As you continue to keep an eye on this space, work with different brands, work with these AI tools, what are the predictions that you have for future applications of generative AI for digital advertising?
[00:34:44] Praveen Gujar:
Yeah. As I, look forward, I think that 3 key things that are really interesting to me, and I'm actually even writing a a paper about this that we soon published. Number 1, basically, is multimodal, content, that are trained with the multimodal models and are in rich format as well. So if you look at, connected TV is a very emerging space in digital ad tech. After years of subscription only model, even Netflix has gravitated towards, multiple reasons to it, but, one of the reasons also is the success of connected TV. So how can we basically, generate TV ready content using multimodal models? We are not there yet. That's one of the things that I'm really, excited about. I think we all might have seen, the demo videos of Swara.
There are imperfections in that still, but I think that's one of the areas where we can make significant advancements so that, these, generated become content can also be shown on a big screen like a television. Number 2, immersive, ads. These are something that I'm super excited about now that specifically Apple is in the AR and VR kind of, mindset. I think, like, brands can really engage with, users. Like, imagine, you're trying to basically shop for a, soft shop for a table lamp, and you receive an ad that helps you to basically understand what kind of a table lamp fits on your, tables. Like, I think some of this actually happens today. If you're in Amazon.com, they enable you out in the Wayfair.
They enable you to look what the product is, but imagine this is dynamically generated ad to a user who can interact with the ad in an immersive environment as well. 3rd is basically interactive ads, where you ask you provide prompts to, the users. And based on the prompt, you basically customize the ads that are served to the users as well. It's not only fun and engaging to the users, but also you make it a lot more relevant to the users as well. I think these are the capabilities that, without Jenny I, we couldn't have imagined in such a short duration. So I think these three areas are where I'm really excited about, and looking forward to how marketers will use Evergrow, like, these models to basically create these kinds of, creative contents.
[00:37:04] Tobias Macey:
Are there any other aspects of this space of generative AI, its application to digital advertising, and some of the risks and strategies involved that we didn't discuss yet that you'd like to cover before we close out the show?
[00:37:18] Praveen Gujar:
Yeah. I think, one thing that basically I will emphasize, I talked about this a little bit, but, I think every change basically has a lot of resistance in an organization as well. And if you are a brand with huge, in house marketing team, they will be, hesitant to adapt to these, to begin with. A cultural change is very important. I cannot, you know, emphasize that enough as a medium to basically make transformative changes in your organization as well. Whether we like it or not, this tide is gonna basically take us all, and adapt into it in the right manner is very key.
This is very specific to, like, the mobile desktop, era as well and how teams and how organizations have to team up by that as well and who didn't were left behind. The second basically is, human, and AI. Like, synergy is very key. I think, when it comes to brand image, management, when it comes to, making sure that, you minimize the biases introduced us, today. Our overall success, I think human plus AI system synergy is very key. I wouldn't basically that that way is second key things as well. I think that, the last piece is data clean cleanliness and standardization is equally important, I would say. These foundational models are trained with such a huge amount of data, but you have to argument them with, your own brand specific data. For that, you have to have standardized and clean data in your ecosystem, so you can train these models with your own data and making sure that the content are generally more brand relevant, more in line with what you want, the message to should be. So I would say these are the 3 key things as, like, key things to remember, towards the end.
[00:39:09] Tobias Macey:
Alright. Well, for anybody who wants to get in touch with you and follow along with the work that you're doing, I'll have you add your preferred contact information to the show notes. And as the final question, I'd like to get your perspective on what you see as being the biggest gap in the tooling or technology that's available for AI engineering today.
[00:39:26] Praveen Gujar:
I think, I may basically answer this slightly differently. Instead of focusing on the tools or AI, I think we talked quite a bit about the gaps and how to improve them, how we are actually working on certain models to improve them as well. The biggest barrier I see is cultural transformation. I think that's really the key thing. Like, I've, I I talked about this marketing conference that I was really, part of. I could see every marketer had hesitation about using Gen AI, what does it mean for their own job. Every creative director worried about what it does.
I think, that's, unless the organization go through a cultural transformation of adapting and embracing Gen AI, which is, at this point, I would say, inevitable. I think the success of, the brand, when it comes to our marketing, campaign will be limited. I think that's the the cultural aspect is, I would say, is the biggest barrier, right now.
[00:40:31] Tobias Macey:
Alright. Well, thank you very much for taking the time today to join me and share your experience and perspective of the ways that generative AI is being used in these digital marketing campaigns. It's definitely a very interesting application of these technologies. It's definitely great to hear some of the ways that they can go right and wrong. So I appreciate the time you've taken today, and I hope you enjoy the rest of your day. Yeah. Thank you so much. Thanks for having me,
[00:40:58] Praveen Gujar:
here. It was exciting to talk about the influence of Gen AI in the digital advertising space. I only I can only say that the future is really bright, and I'm super excited about that.
[00:41:14] Tobias Macey:
Thank you for listening, and don't forget to check out our other shows, the Data Engineering Podcast, which covers the latest in modern data management, and podcasts.init, which covers the Python language, its community, and the innovative ways it is being used. You can visit the site at the machine learning podcast.com to subscribe to the show, sign up for the mailing list, and read the show notes. And if you've learned something or tried out a project from the show, then tell us about it. Email hosts at themachinelearningpodcast.com with your story. To help other people find the show, please leave a review on Apple Podcasts and tell your friends and coworkers.
Introduction and Guest Introduction
Praveen's Journey into AI and Machine Learning
Defining Digital Advertising and Its Evolution
Goals of Digital Advertising Campaigns
Traditional Advertising Workflow vs. AI Integration
Generative AI in Digital Advertising: Benefits and Limitations
Implementing Generative AI in Marketing Teams
Risk Management and Education for AI Tools
Campaign Life Cycles and AI Tool Selection
Successful AI-Powered Campaigns
Failures and Risks in AI-Powered Campaigns
Lessons Learned in AI-Driven Advertising
When Generative AI is Not the Right Choice
Future Predictions for AI in Digital Advertising
Final Thoughts and Cultural Transformation
