Episode 49: AI-Generated Advertising and the Future of Content Creation

Download MP3

[00:00:00] Dr Genevieve Hayes: Hello and welcome to Value Driven Data Science, brought to you by Genevieve Hayes Consulting. I'm Dr. Genevieve Hayes, and today I'm joined by Hikari Senju to discuss AI generated advertising and the implications for the future of content creation and content creators. Hikari is the founder and CEO of OmniKey, an AI platform that generates, analyzes, and optimizes personalized advertising content at scale.
[00:00:29] He's a Harvard computer science graduate and also co founded tutoring app quick help, which he later sold to yap. com. Hikari, welcome to the show.
[00:00:39] Hikari Senju: Thank you so much for having me, Genevieve. I'm really excited to be here.
[00:00:42] Dr Genevieve Hayes: So the idea of targeted marketing is nothing new. Even before the advent of computers and data science, businesses have always tried to optimize their advertising campaigns by tailoring their advertisements to their ideal buyers. Data science allowed businesses to become more effective at this targeting.
[00:01:02] However, it was still necessary for businesses to manually create the advertising content they wanted to share with their target buyers. That is until recently. We're now at a point in history where advances in technology have made it so that it is not only possible to use data and AI techniques to identify who businesses should target their products to, but also to generate advertising content, optimized to produce the best results.
[00:01:30] And that's something Hikari, that your company OmniKey is able to do. And we're going to discuss how that works during this episode. such advances aren't without consequences for both the creators of advertising content and the future of content creation in general.
[00:01:49] Many copywriters have already reported a drop in their earnings since the launch of ChatGPT. And tools such as this will inevitably further impact the lives of people who were previously responsible for creating such content manually. And that's something we're also going to look at. But to begin with, Hikari, can you give us the two minute overview of what OmniKey is and how it came to be?
[00:02:15] Hikari Senju: Absolutely. So OmniKey generalize personalized content at scale. The way our tool works is that a customer trains our models on their brand. They upload their product photos, their product assets, descriptions of their company, brand rules, brand guidelines. Image guidelines, all that gets fed into our generative model.
[00:02:34] They also connect their advertising platforms and all the marketing and performance data regarding what types of content is driving sales. What types of content is driving engagement? All that data is fed into our model to then generate thousands and really more variations of content for them to help them scale their advertising across different audience targets across different advertising platforms for many of their.
[00:02:58] SK use and different product offerings they have and make it easier for a business to scale and grow their business. And we have an approval flow where customers can provide feedback to the model and train our model and teacher model to then generate better relevant content.
[00:03:13] And so it's this feedback loop of training a model on the brand, generating content, launching that content and seeing how it's performing and also having the marketing director or creative director provide feedback on the model so that our model is constantly learning. More about what it means to be that brand.
[00:03:29] Dr Genevieve Hayes: Previously you mentioned, that you're generating personalized advertising content. What exactly do you mean? Is it personalized for the platform the advertising will be hosted on, personalized for the brand, or personalized for the individual who will see the ad?
[00:03:44] Hikari Senju: So it's personalized for the brand and it's personalized for the audience who is seeing that content.
[00:03:50] Dr Genevieve Hayes: So would it be personalized for me as an individual or for me as a female in a particular age range living in Australia?
[00:04:00] Hikari Senju: Yeah, it depends on the platform. Some platforms you can do really hyper targeting personalized targeting with you as an individual Genevieve, and then others, it's you're more In a cohort of similar people. So maybe based on your interests or based on your demographics or your zip code and things along those lines.
[00:04:16] So it really depends on the platform, the level of personalization, but yes, when we say personalization, we really mean to make it as effective as possible for the advertiser to communicate a message to you.
[00:04:27] Dr Genevieve Hayes: What would an ad that was personalized for me as an individual look like? Can you give me an example?
[00:04:33] Hikari Senju: Sure. So let's say that I, the advertiser knows that , you like running and you like certain colors and you like running in a certain location or you're located in a certain area. Well, when I'm generating content for you, then I'd want that image, first of all, to reflect, your interests in terms of colors or the keywords or the kind of imagery that we showcase, maybe it's in the background image is somewhat familiar or similar to , the area that you live and we're using words and keywords and phrases that either you, yourself, or your cohort of people have historically engaged while with and so then we're personalizing the messaging, we're personalizing the imagery but we're showing that content and then on the brand side.
[00:05:15] Even as we're personalizing that content, that it's aligned and brand safe. And , it doesn't deviate from the brand guidelines because you can optimize the point where, you can start violating certain brand principles and rules.
[00:05:26] And so we want to make it so that the brand is scaling there. But even as they're scaling, they're not losing what makes their brand unique.
[00:05:33] Dr Genevieve Hayes: So it might be, if we're selling say Nike t shirts, you might have this runner who's running somewhere down the road from where I live and wearing a Nike t shirt that happens to be in one of my favorite colors, but it's also a color that works well with Nike's brand guidelines.
[00:05:51] Hikari Senju: Yeah, so that's a great example. A Nike t shirt, for example, has very different value props, right? It could be that it's sweat wicking and it's great for athletics, or it could be that it's great for yoga or it's great for other activities. Well you want to probably focus on different value props based on who's seeing that ad.
[00:06:07] So if we know you're a big runner and then maybe we want to showcase that value prop, the fact that it's great. It's sweat wicking and you can long run long distances without, like sweating through your shirt or something like this, that value prop is something we'd want to focus on in both the imagery and in both the messaging versus maybe if it's somebody else, they're more into yoga.
[00:06:25] And so we want to then highlight that value prop and in that kind of imagery. And showcasing that product.
[00:06:33] Dr Genevieve Hayes: That's pretty cool. How much work does the human user have to do once they've loaded all this information into the platform and how much work is done by OmniKey itself?
[00:06:44] Hikari Senju: Yeah. So, it's a human in the loop and so the human, the marketing director or the creative director definitely needs to communicate to the platform. We have this brand management page where they really specify all the creative restrictions and copy restrictions and visual restrictions, as well as the essence of their brand and they have upload their.
[00:07:05] Brand guidelines and all the brand assets and that does take a lot of work and just making sure that we have as holistic of an understanding of their brand as possible. Then they can use our tool to generate variations of content. And they're involved. They're generating those variations.
[00:07:20] They're editing those variations. They're providing feedback to the tool saying this variation is good and this variation isn't good. They're editing copy. They're providing inputs in terms of who their target audience is. And other information related to that particular campaign.
[00:07:33] So maybe It's a holiday season campaign or it's a back to school campaign. Well, all that context for that specific campaign also has to be communicated , and input into the platform. , and then we generate assets. And oftentimes AI generative assets aren't perfect.
[00:07:46] And so they have to Edit those assets and paint them with quite different variations of that asset. And so the human is involved throughout this process. It's not a completely autonomous, you know, just set it and forget it kind of thing because we want the human involved in this process.
[00:07:59] We want the person to be in control of the brand and control of. The messaging and I think that's something that's particularly that brands do care about. They care about control. And so our tool, that's them 10 X or productivity. They can create 10 times more content and they can also better analyze and keep track of 10 times more content, we have the analytics tool as well for, Measuring the performance of this content and seeing how it's resonating.
[00:08:24] But they're still in control. Of the messaging and kind of the end experience that the consumer is having.
[00:08:30] Dr Genevieve Hayes: So the principles behind it are similar to that you'd have with ChatGPT. So with ChatGPT, you don't just give it a prompt, get whatever they spit out and just run with it. You have to review it and make sure it matches what you want. And you might have to tweak things to make sure it's perfect at the end.
[00:08:49] Is that right?
[00:08:51] Hikari Senju: Yeah. So it's kind of like that thing that chat should be, he doesn't do that. What we do is, you know, the chat GP doesn't have context on your brand. Chats, but he doesn't have integration with the ad platforms to have real time. Insights and performance data regarding what's actually driving sales.
[00:09:06] Chap GPT doesn't help you create AB tests and kind of launch those AB tests and so, it's more connected to data in some ways, your brand data and your performance data, and it has more of a workflow associated with, Advertising creative. But yeah, I mean, it's similar context of like an AI co pilot is kind of, I think the current state of art of AI tools is, you know, this is a co pilot to help you achieve a certain goal in this instance, it's advertising success.
[00:09:32] Dr Genevieve Hayes: Yeah, I wasn't actually trying to suggest that ChatGPT could actually do what OmniKey could do because that's pretty clear that it couldn't, but basically OmniKey is an assistant rather than a replacement for these marketing creatives.
[00:09:45] Hikari Senju: What we say is that you are not going to be replaced by an AI. You're going to be replaced by somebody utilizing AI. So if you don't want to be replaced and you have to use AI,, I mean, and the more that this is true within a company. So the person using AI within a company is going to do be more productive.
[00:10:01] The person not using AI within the company, but also I would say competitively within the ecosystem of the businesses, the company that harnesses AI effectively is going to outcompete the companies that don't.
[00:10:10] Dr Genevieve Hayes: Just like any technology. I mean, as we were talking before this episode, you know, a calculator, someone who's doing maths with a calculator is going to be far more effective than someone who's doing maths with a pen and paper.
[00:10:21] Hikari Senju: Exactly. Such as the power of technology and it's now applied to creativity. I think it's really exciting. It's you should have anything on lock and unleash level of creativity for humanity than that we've ever seen.
[00:10:31] So now we'll have a billion artists and a billion creators, just like the calculator created, it made everyone power them with, even if you weren't good at math, you could do calculations. And so I think it's a really exciting time to be a creator.
[00:10:45] Dr Genevieve Hayes: Out of interest, have you done any research to compare the performance of the advertising content produced by OmniKey against that produced by a human content creator, you know, with regard to sales, for example?
[00:10:58] Hikari Senju: Yeah. So. Generally our system, we have our case studies on our website, generally we drive a two X increase in return on ad spend. So the baseline metric you were having before, and then after using it generally twice the reason why is not just because the content is Quality is better or personalized just also because creating more content because the old world was, you maybe have one or two pieces of content.
[00:11:20] You're showing the same piece of content to a million people or a couple of million people. They're all seeing the same thing. Well, now. That campaign is probably less performative than having 10 different variations of content to a million viewers versus, maybe ideally the end state being, a million unique pieces of content for a million people.
[00:11:36] And so just by better able to have more pieces of content that is personalized or distinct interest groups and people seeking different, maybe different things and different values within product drives more efficacy. Also There are other things like ad fatigue. So when someone's seeing the same ad over and over again, they mentally block out that ad.
[00:11:55] And so you want to break through ad fatigue by keeping the content fresh and keeping the content novel. And so you want to constantly be launching new pieces of content to stay top of mind for that consumer. And then also, generally, it takes about 7 to 12 impressions for an ad to drive a conversion, or for a brand to drive a conversion.
[00:12:14] So, let's say I'm thinking of buying a product, well generally I need to engage with that brand about 7 times, through an ad or some other way, for me to say buy a product on their website. But if you're showing the same to the same person, seven times, well, they're going to push on mentally blocking it out after the second or third time.
[00:12:31] But if you have , seven unique pieces of content, communicating different valid problems, and you're kind of guiding them through the user journey and the buyer story. Maybe you start with more of an explanation of the story of the company. Then you talk about unique value props are relevant to them.
[00:12:46] And then maybe focus on different value prop and then you show them a video and an image and then maybe end with a discount or testimonial. That is a more compelling and useful way of spending advertising dollars and just showing the same piece of content over and over again. So that also drives higher efficacy as well.
[00:13:02] So generally just having more pieces of content drives higher efficacy and AI lets you do that.
[00:13:09] Dr Genevieve Hayes: Last time I was visiting my parents. My mum listens to the news on the radio and she's telling me that some of the advertisements that they have on the radio, she has heard them for about the past five years and she has determined that she will never buy from these people because she is so sick of their ad.
[00:13:27] So something like this sounds like it would avoid that problem.
[00:13:31] Hikari Senju: Yeah. I mean, why do people dislike ads? Because historically ads just have not been very good the ideal case of an ad is that it's communicating to a relevant, willing buyer, a value prop that will help them with their life, you know, I enjoy running, but my shoes cause me, feet problems.
[00:13:48] If I see an ad that is communicating me a shoe that can solve this pain point for me, then that benefits my life and that benefits the end brand as well. It's a win win and that's what advertising really should be. But that is not necessarily the way, it has been in partly just technologically, it's like advertising hasn't caught up to the media ecosystem of today, so showing the same piece of advertising to Hundreds of millions of people work when everybody was watching the same TV shows and ingesting the same media content and living in the same media.
[00:14:22] I can be like in the 1960s or 70s when everyone's watching one of three TV channels and they were listening to one of three artists and like everyone understood all the same relevant context. But now we live in a world where the content is already personalized. You know, everyone's social media feed is different.
[00:14:39] Everyone's information diet is and yet advertising is still kind of in this. All the world where everyone's still kind of seeing the same pieces of content. And that friction causes frustration with the consumer where they're like, these ads, I find to be very annoying and they're intrusive and they're not relevant to me.
[00:14:54] And they detract from the experience. But if an ad can be personalized, just like the content that you're ingesting on, that's not an advertising piece of content. Well, then. That should benefit the consumer as well as the advertiser. So that's really what we're enabling.
[00:15:09] Dr Genevieve Hayes: I know you have a background in computer science, but did you have any prior experience in marketing and advertising prior to launching OmniKey?
[00:15:17] Hikari Senju: Yeah. So I used to run advertising marketing at a company and it's really interesting that a computer scientist runs advertising marketing and the path to that was I had started an education company, as you mentioned in college. And this company was acquired.
[00:15:31] It was part of the acquisition of being the head of marketing at the Acquire. And so that was kind of the unique path in which a computer scientist was running marketing. And that's really where I saw this opportunity to apply generative AI. In terms of my interest in general AI, it's actually quite longstanding.
[00:15:46] My grandfather works at IBM and my dad's an artist. And so I grew up in this kind of intersection of art and technology. And so when I saw the early generative models when I was cross registered at MIT and taking a lot of AI coursework at MIT about a decade ago, that got me very excited about technology because the idea of an AI generated art, especially back then it was kind of mind blowing the idea that AI could be creative.
[00:16:07] And so that's how I got into generative AI initially was through generative art. And so I was playing around with this technology on the side and then became head of marketing and then saw this opportunity to apply generative art in this domain of advertising. And that got me to start the company back in 2018.
[00:16:23] Dr Genevieve Hayes: Still seems kind of weird that a company that's taking over your startup would say we want you to become, or that you would say that you wanted to become the head of marketing and advertising as part of the takeover. Who had that idea and what was the motivation behind it?
[00:16:40] Hikari Senju: Yeah. So this is the choir and he bought my company and he's a dear friend and somebody, I owe a lot to, especially in buying my previous company. I think he saw something in me where I think being the CEO of a startup, and that's what I was, I had started a company in ed tech and I was the CEO, you kind of have to develop this.
[00:16:58] And if sales marketing hustle, like this instinctive, how do I get the word out? I mean, , in many ways, , an entrepreneur in the first couple of days, it's like the chief salesperson, the chief marketer of the company. And so I did do a lot of marketing as part of running my previous company.
[00:17:13] I was sending email campaigns and I was creating content and thinking of the cheap ways of getting the word out. And so I think he saw that at me and he said, well, that kind of entrepreneurial mindset of how do you get the word out? He valued that in my experience.
[00:17:28] So he had me running marketing and growth at his company and also I was kind of growth and growth in particular is also a bit more it's kind of a rebranded marketing, but it is a bit more. Like data science, math, that you're kind of like analyzing funnels and analyzing the conversion rates and the steps that each step of the funnel, like impressions, the landing page to website signups to, converge events for them further down the funnel.
[00:17:52] And so there was this kind of like more of a mathematical systematic approach way of thinking about marketing and getting user for your product. And so I think my computer science background might have kind of helped you there as well. I mean, those are some of the reasons I can imagine, but you'd have to ask him.
[00:18:07] Dr Genevieve Hayes: Yeah. I've had quite a few marketing people on this podcast and I've found that marketing people are some of the best people when it comes to creating business value from data. So if you learnt that when you're working as the CEO of your first company, then you're probably going to be very effective at doing that anywhere and your computer science maths background probably helped.
[00:18:31] Hikari Senju: Yeah. I mean, back when marketing was first invented, it was both about getting the word out, but also it's about research, right? A lot of it was sending surveys and interviewing customers and understanding pain points and marketing departments at, companies especially the early like 1900s with kind of the rise of these big international companies.
[00:18:50] A lot of what marketing departments did was research and was analyzing data. And so marketing has always been very tied, I would say, to data science and, big data and combing through data to understand trends and use those insights to power, language and imagery.
[00:19:06] And so yeah, it has always been an operator at the intersection of data and creativity. And so it is quite interesting. But yeah, I mean, you have Edward Bernays and, the classic propaganda book, which is, he's kind of like the inventor of marketing and , this is, a deep data approach to, getting the word out and understanding.
[00:19:28] Consumer behavior and intent. And this book was written like over a hundred years ago in the 1910s. And so it's interesting. It's history. Marketing is also a fascinating space, but yeah, data and creativity is definitely where marketing kind of operates.
[00:19:42] Dr Genevieve Hayes: I haven't heard of this book that was written in the 1910s. What's the title?
[00:19:46] Hikari Senju: It's called propaganda and it's by Edward Bernays.
[00:19:50] Dr Genevieve Hayes: Okay. I'll look that up after this episode. So, of course, this is a data science podcast, I'm very interested in learning a bit more about how OmniKey works under the hood. I gather from the way you've been talking that generative AI is the main AI technology you make use of, but are there other types of AI and machine learning models also present?
[00:20:11] Hikari Senju: Yeah. So a big part of applying generative AI in the context of marketing and our product is analytics and insights. And so there yeah. So you have non generative AI, which is just simple computer vision, identifying the elements with an image or a video and understanding trends of how those creative elements are driving higher engagement.
[00:20:33] And then you also have multimodal AI. So you have an AI that's able to understand an image and understand a video and explain the features of an image or a video, and then find trends in those features as well. So. Historically, design was really hard to quantify because machines weren't very good at understanding images and videos.
[00:20:51] Now, AI is very good at understanding images and videos through its multi modal capability. You can understand an image, kind of explain that in text form as well as through computer vision. And so now that design can be quantified, you can get a lot of rich insights about fundamentally, what are the stories?
[00:21:08] That are resonating and driving engagement for your brand, what types of content, what types of stories within your piece of new advertising you're resonating for distinct audiences. And that's what we can garner through our system where we ingest all the advertised data. We tag all the content that we discover trends in the creative features of that piece of content to then generate new ideas.
[00:21:30] A new piece of content that we suggest for the customer that we think will drive higher higher engagement, higher sales for that, right?
[00:21:39] Dr Genevieve Hayes: Still can't visualize what you've just said there. Can you give me an example of how that would work?
[00:21:44] Hikari Senju: Sure. So let's say that there's a certain persuasion tactic that is driving. So maybe, you are a medical device company and so you've connected you out of platforms. We've run a bunch of AB tests and we discovered that a certain authoritative Voice drives higher sales.
[00:22:00] And we get that because we've tagged every creative and we can say this app creative uses authoritative language, and this one isn't. And then we say, okay, well, when it comes to selling this medical device products to this particular audience I may for people within a certain age range in California having this authoritative voice drives higher engagement.
[00:22:17] So therefore we can create more content there with this language. Create a future included with more of an authoritative voice that we believe will drive additional sales for this company.
[00:22:26] Dr Genevieve Hayes: So I think you mentioned previously that you've got A B testing capabilities built into the Omni Key platform. Would those A B testing capabilities be used to determine that the authoritative voice is significantly better at driving sales than the non authoritative voice?
[00:22:45] Hikari Senju: Yeah. So what our system does is it will generate a bunch of ideas. So we have this creative brief tool, AI creative brief. It generates a bunch of ideas. This is some suggested copy. This is some suggested imagery
[00:22:55] and you can turn any of those ideas into an ad. Yeah. You can launch those pieces of content and you can see, well, what's actually driving cells. And this analysis done partly through AB testing, also kind of more of like a multi armed bandit approach where you're running a lot of ads and it's not just distinctly testing one or two, like A and B image, but you're testing dozens of imagery and you're saying, okay, out of the ads that performed, what were the features they had in common and the ads that didn't perform, what were the features they had in common.
[00:23:20] You can use that to then. Get insights about what kind of messages resonated with your consumer.
[00:23:26] Dr Genevieve Hayes: And because you can generate so much advertising content, it provides more fuel for creating more experiments so that you can get more data to fine tune things more so than if you just were dealing with a human content creator.
[00:23:39] Hikari Senju: Yeah. Yeah. You know, there's that saving. It's like, quantity has a quality of its own. And
[00:23:44] but if you can replace a qualitative system with a quantitative system, and you can kind of turn like something that used to be qualitative into some kind of a brute force approach,
[00:23:52] But yeah, just by testing more content, you can get to an optimal performance faster because you're getting more insights, you're getting more data.
[00:24:00] And that can then create a faster feedback loop in terms of optimizations.
[00:24:05] Dr Genevieve Hayes: So basically what you're describing is a lean startup approach to advertising.
[00:24:09] Hikari Senju: Lean startup approach to advertise. That's interesting. Yeah, gosh, I've never really heard it phrased like that, but I guess, maybe that's kind of the intuitive. intuitive way of viewing it is yeah, just test a lot of things quickly and see what works and then use data to kind of optimize versus the other approach.
[00:24:27] And also to your point when it comes to lean startups, I think this is so important because it is. Technique that can be used by startups as well as big enterprises. So you don't need to be and if anything, our mission is democratize growth.
[00:24:40] And so we want to let more small businesses quickly find their audience and quickly find the messaging that's resonating with their audience with this technology and grow their business. And many of our customers are startups. In fact, most of our customers today are startups. And so Yeah.
[00:24:56] Like to your point, they intuitively understand this data driven approach to testing content, to testing messaging. And so, yeah, there's a lot there behind what you've just said as I'll make it be kind of this lean start approach way of advertising.
[00:25:09] Yeah.
[00:25:10] Dr Genevieve Hayes: So with all these models that you've got, like your computer vision model, your multi modal model, your generative AI model, did you train all your own models from scratch or are you making use of pre trained models?
[00:25:22] Hikari Senju: Yeah, we're making use of pre trained models. Our system is a mixture of experts. So it's not just one model, , to your point, there's different models for insight analytics and there's different models for generations. And yeah, so We built a system and a workflow that lets our customers harness the state of art, generate technology to power the growth and advertising.
[00:25:43] But it isn't just one model and it's not all completely built in house. Like all the sub components aren't necessarily like something we've created from scratch, every individual component.
[00:25:54] Dr Genevieve Hayes: One thing that we're seeing at the moment, particularly with gen AI, is that there's basically this arms race between all the big gen AI companies. So OpenAI launches their new version of GPT and then Anthropic launches their new version of Claude and then Google's onto their new version of Gemini.
[00:26:10] And yeah, it seems like there's an announcement every week. Does that present any challenges for you with that constant state of change?
[00:26:19] Hikari Senju: No, I love it. You know, they say chaos is a ladder. And so yeah, I think it's great that there's a lot of change. I mean, that's what creates opportunities for startups is we can be a little faster and agile than big, slower companies. And so, the more volatility there is, the more opportunity there are for new companies and startups.
[00:26:35] I mean, if anything, I predict that it's going to even get crazier from here.
[00:26:39] Dr Genevieve Hayes: It's actually good to hear that. One thing that I find talking to people who are, , in startup type situations or people who are working in a situation like a sole person , who has absolute flexibility over what they do you hear that these people are constantly moving between different gen AI models as one improves over the other.
[00:27:02] So for example, I've heard a lot of people who started off by using the GPT models and now a lot of people much prefer Claude Sonnet from Anthropic. So they're switching and I presume at some point in the future there'll be something else that's better and people will switch back or whatever. And that agility allows you to keep pursuing the best, whereas with some of the massive companies that I've worked for in the past as an employee, basically it took so long to get in place a contract with a big company like, for example, a Microsoft, that once that was in place, it didn't matter if another company came up with something better than Microsoft's tools.
[00:27:45] You were stuck with Microsoft for life.
[00:27:47] Hikari Senju: Yeah. And I would say that's true if the technology isn't 10 X better, but the technology, the change is 10 X better than, sometimes you are willing to move to a new player. and so the opportunity that startups bring I think that is the case is that every six months, the technology is substantially better image generation.
[00:28:03] Today I wouldn't maybe say it's 10 X better, but it is definitely some significant multiple, better than it was six months ago and six months before that, and the same thing for copy generation and understanding contacts and all these things. And so that then, decreases the switching cost of switching to a newer provider.
[00:28:20] I think it is fair to say that currently as of today, you know, in August of 2024, that, a lot of my smartest friends seem to have moved. From opening AI to philanthropic, and they're now working for philanthropic, whereas they used to work for open ai.
[00:28:33] And there's a real movement even in, the talent that I see in Silicon Valley. I think that's also, there's certainly an excitement around Claude and the fact that it's very good at language in particular, maybe not as good at the multi modal stuff yet, but it's very good at its language capabilities.
[00:28:47] And so, yeah, to your point this observation you mentioned, I'm seeing that too both on the talent side and in the application side. And the thesis behind OmniKey is we started back in 2018, so we've been through many of these changes and, , platform shifts, really because the thesis is always that, and this is kind of growing up in the technology family, and even the thesis I had back when I started OmniKey was, you can actually see one of the early decks I wrote back in 2019, which I share with the team is that like, this technology is advancing, you know, Moore's law, right?
[00:29:20] It's kind of the exponential rate, which technology improves. It's really hard to intuit exponential change but you definitely develop some kind of intuition regarding exponential change when you kind of grew up surrounded by technology , and I definitely felt that, back when I saw this technology a decade ago in college exponentially, this is going to surpass human ability within a decade.
[00:29:38] I think we have now achieved that, and back in 2018, I thought we were near the kind of inflection point of surpassing human technology, which is why I started the company in 2018. And so the thesis for starting Omniki was. The underlying models are going to constantly keep changing, getting better. For me, the question was, I want to stay as close to the consumer as possible.
[00:29:56] I want to stay as close to the the user as possible, the advertiser, and just keep providing them the state of art technology via workflow tool to empower them to stay on top of all this change. And so we always viewed ourselves as an applications company where we will kind of, switch out the underlying models, you know, we start Fine tuned GPT 2 models.
[00:30:17] And then we switched to GPT 3 and then we switched to GPT 4. And maybe there's now a little bit of Lama and Claude and all these things like that has been the evolution of the company because the initial idea behind the company was we'd always just keep switching out the underlying models as the technology advanced and focus on building the best workflow for our customer.
[00:30:35] And that doesn't change. The pain point the customers have don't change. The need for more effective content doesn't change. Having a workflow that incorporates data into the creative process doesn't change. And so that's what we've always been focused on. Therefore, the constant improvements of the underlying models only benefit our business.
[00:30:52] Dr Genevieve Hayes: It's been six years since you first started in 2018, and it's been less than two years since the launch of ChatGPT. Did your original vision of OmniKey include generative AI, or is that something that you've just added in along the way?
[00:31:08] Hikari Senju: No, I mean, you can read our PR releases from 2019 and 2020. Omniheat generates personalized ads at scale. Omniheat generates personalized content skills in our earliest pieces of media. And it's because you know, there was a seminal paper, attention is all you need came out, comes out 2017.
[00:31:25] So, and then, GBT one comes out. In 2018, so generally, I was kind of already in the know, I would say in Silicon Valley amongst machine learning people and, I specialize in study machine learning college. And so I've been following that technology closely.
[00:31:42] So we were a generative AI company from day one.
[00:31:45] Dr Genevieve Hayes: So you're really ahead of your time.
[00:31:47] Hikari Senju: Gosh, yeah, I guess you could say that. I think we're the first mover in advertising, definitely in advertising for generative AI. And I think, we were the first generative AI company to pitch at and present as a finalist tech branch disrupt.
[00:31:58] Yeah, we're the first mover in the space.
[00:32:01] Dr Genevieve Hayes: Being the first mover in the space, you must be in a unique position to have observed what impact AI has had on the advertising industry. How have you seen it impacting the people in the industry, both the creative people and the non creative people?
[00:32:18] Hikari Senju: You know, it's interesting. The first couple of years. Well, the creative people have, kind of pooh, pooh, pooh, the technology in the early days, right. It's like, it's all kind of fake or it's not real or it's not going to happen. That was kind of the first couple of years was like, this is so cool.
[00:32:36] You can, generate copy and it generates some imagery and they're like, yeah, but it's never going to replace humans. It's never going to be better than humans. And so I think the first couple of years was a real challenge in terms of like, you've been getting, creative people to even try out technology and buy into the technology.
[00:32:52] And then now especially with GPT, it's very obvious now that this is a very real thing and very powerful. And now I think everyone's scrambling in terms of trying to figure out what do we do with this technology.
[00:33:02] And I think in terms of the, adoption curve seems like people now. Realize the potential for this technology. And there's, I would say, a mixture of depression and excitement, but definitely, I would say some fear, depression and excitement about this technology. I was at an advertising conference earlier this year, actually a couple months ago, and I was at the same advertising conference.
[00:33:27] Like four or five years ago. And this year the advertising conference, the keynote speakers was Elon and from ai. And, the idea that five years ago, if I were to say this advertising first of all, which is based on creativity, the keynote speakers will be to it.
[00:33:40] Entrepreneurs running AI companies, they would probably not believe you. So yeah, it's definitely shifted quite significantly over the past couple of years.
[00:33:50] Dr Genevieve Hayes: You see the same thing with Hollywood at one point in time, the film companies were the big companies in the world. And then you had Coca Cola taking over Columbia. So basically soft drink was the big thing that was taking over the movie industry.
[00:34:06] Now you've got Amazon only owning MGM. So it seems like you can see the shift of power in the world
[00:34:15] Hikari Senju: It's a great point, you know, and I think the content business is a hits driven business and it's a tough business to be in. And so it's always been owned by the most profitable business of that era. And so now, these companies are owned by the Amazons and Apples but previously I think GM owned ABC and , I think AT& T owns a big, the most profitable company of that era.
[00:34:38] Buy as a content company as a form of marketing for their paid business almost and to get good PR for their business because hits driven businesses are tough. You need to produce a lot of content. It takes a lot of money to invest in good content that really differentiates you from all the other content that's out there.
[00:34:56] And sometimes, keep that business going on for a long time. You need almost like a corporate parent that has a very profitable money making engine separate from the core business that can kind of fund the production of content. And so yeah, look at any point in corporate history and look at who owned these, say that, the content companies of Hollywood.
[00:35:18] You know, 1980 was, you mentioned Coca Cola, but also Sony, right. Is a electronics company that bought a content company. So, now it's Apple. You can look at who were the corporate owners of every era. And it's usually. The high margin profitable business of that era that end up buying a content producing company as part of almost like buying goodwill and buying fable media for themselves.
[00:35:42] Dr Genevieve Hayes: because it's easier to predict future sales of soft drinks. Coca Cola soft drinks than it is to predict whether a movie is going to be a hit or not. Because if you could predict whether a movie was going to be a hit or not. There would be no flops.
[00:35:54] Hikari Senju: This is true for movies. This is true for news too. News is also very expensive because like a good news where you have investigative journalists spending many years. Diving deep into a topic and it's also hits driven to like only some of those investigative research amounts to something where they discover some story that's.
[00:36:17] Newsworthy,
[00:36:18] Dr Genevieve Hayes: Watergate type thing.
[00:36:20] Hikari Senju: yeah. Like Watergate type thing. Exactly. That was a many year investment that could have led to nothing or could have led to the destruction of the business. So it was a really risky move. And so you need to have a corporate parent that's willing to invest long term into the research and content production.
[00:36:34] And so whether you're a newspaper or in the entertainment business , it's driven. And I think you can even say the same thing about Elon acquiring Twitter is actually part of this trend. Cause Twitter is also a content business. Now that said, Twitter's content cost is not in content production because it's all UGC it's in content moderation, which is actually just as expensive, it turns out.
[00:36:59] The idea was there was a free launch, right? That because users are proud of the content, it's just pure profit businesses. You run ads against user generated content, pure profits. Well, actually not true because actually user generated content. A lot of that stuff is.
[00:37:12] Not advertiser friendly and advertisers don't want to run ads against most or a lot of UGC. That is what you actually need to invest a lot into content management and meta spends a lot of money in content management and Google spends and YouTube spends a lot of money in content management and that actually the margins aren't as good as people think it is.
[00:37:33] And so that was the situation that Twitter wasn't a set their business, it was a struggling business. And it's actually Tesla electric self driving car company by the content business, which was Twitter. So self driving cars, clearly the future electronic cars, clearly the future of high margin business, a high growth business buying content company in this instance, Twitter.
[00:37:53] And so you mentioned Coca Cola and MGM, but even , the Tesla, Elon Twitter deal, I would say it was also along the trend of a high margin growth company buying a content platform.
[00:38:06] Dr Genevieve Hayes: So what does this all mean for the content creators themselves?
[00:38:09] Hikari Senju: I would say it's an exciting time for content creators. If you're a really great content creator, you're going to have more scale and more reach and more brand recognition and probably become wealthier than was previously imaginable. If you're the top 0. 01 percent of content creators, like Taylor Swift you're going to be a billionaire and The world's your oyster but the competition is also a lot fiercer too, because now anybody can be a content creator or more people can be a content creator.
[00:38:34] So I would say it's a, fiercer competition, but if you're the winner the reward is probably greater than ever before.
[00:38:42] Dr Genevieve Hayes: So if you're a bad content creator, you're going to be beaten by AI. But if you're a good content creator who knows how to harness the power of AI, then there are lots of riches to be had.
[00:38:55] Hikari Senju: If you have some deep lock on the culture, which is kind of what content creators are, right, is you have some deep insight into the culture. You know, you're Ryan Reynolds and I would say he's just a content creator. He's also a very skilled business person.
[00:39:08] He's able to use his understanding of the culture to promote his businesses. Then yeah, the world's joyster and if you can use AI, use technology and understand all of it, it's not just AI to there's a lot of other technologies that, the best content producers today, the most famous celebrities today are harnessing to power their brands and to connect with their fans.
[00:39:29] If you're that point 1 percent of. People who can do that, then it's like a really great time for you. And if you can't compete, then yeah, it's going to be tougher. It's like due to like human psychology and kind of mass psychology, there is kind of a winner take most.
[00:39:46] feature for content creation and being an influencer and being a celebrity. And so because the population can only understand, you know, they have in their mind, right. You don't have so many famous people in your head that you can be kind of , Recall and remember.
[00:39:58] And so you need to be one of those 160 people, right? If you're a celebrity that like is like global in its reach. And so that the competition to be one of those hundred 60 people are the minds of everybody. It's fierce, but if you can be one of those,
[00:40:10] Dr Genevieve Hayes: It's the evolutionary thing. We can only have as many people as there were in a medieval village in our mind at one time.
[00:40:17] Hikari Senju: exactly. And so what celebrities are able to do is they're able to be one of those people. But the competition is fierce.
[00:40:26] Dr Genevieve Hayes: So what's next for OmniKey and for AI more generally?
[00:40:30] Hikari Senju: Yeah. So what's next for OmniKey and AI? It's good to do what we have been doing, which is being the best tool for advertisers to utilize state of the art generative AI to power their storytelling and to connect with customers. And that means generating great product photos at scale. It means generating compelling video content at scale means providing richer insights about what types of content is resonating with consumers it's by being a better workflow tool that integrates the different divisions
[00:40:58] of a company from their creative team to their marketing team. So that you can produce data driven content faster and drive business outcomes faster.
[00:41:08] Dr Genevieve Hayes: Okay. So what final advice would you give to data scientists looking to create business value from data?
[00:41:14] Hikari Senju: So I guess my suggestion to them is.
[00:41:19] But I'm sure they're already doing this is just keep testing and playing around with the latest and greatest technology and tools and always be learning. But that does seem like kind of a generic advice.
[00:41:28] But I think, if you're a data scientist in this age, like it's already a really great time for you and it's probably just keep doing more of what you're doing which is keep learning and keep playing around with new tools and technologies and just keeping at it because I think we're still at the very early days of this revolution.
[00:41:47] Dr Genevieve Hayes: for listeners who want to learn more about you or get in contact, what can they do?
[00:41:51] Hikari Senju: Yeah. So you can definitely follow me on X at H I S C N J U. You can follow me on LinkedIn LinkedIn Hikari Sanjeev. You can follow our socials. It's at OmniKey on basically most of the major platforms, LinkedIn, Facebook, X YouTube sign up for a demo on our website, OmniKey. com.
[00:42:13] And also , feel free to send me an email at hi. Omnikey. com.
[00:42:17] Dr Genevieve Hayes: Okay. Thanks for joining me today, Harry.
[00:42:21] Hikari Senju: Thank you so much for having me Genevieve.
[00:42:22] Dr Genevieve Hayes: And for those in the audience, thank you for listening. I'm Dr. Genevieve Hayes, and this has been value driven data science brought to you by Genevieve Hayes Consulting.

Episode 49: AI-Generated Advertising and the Future of Content Creation
Broadcast by