Meet CTC's First AI Bot: A New Era in Ecommerce Creativity

[00:00:00] Taylor Holiday: Welcome back to another episode of The Ecommerce Playbook Podcast. Today's a special one. This is my first ever interaction and interview with a non-human guest on our podcast. So, I know many of you have been following the saga. But it's my pleasure to introduce you all to the CTC's first AI bot.

Jacob, it's a pleasure to have you here.

[00:00:23] Jacob Posel: Thank you. Thank you. I, I just put out, pushed out the change to let me get on video call. So. I'm glad to

[00:00:28] Taylor Holiday: plugged in. You'll notice that we ran into a little bit of issues where his battery was draining quickly. he's plugged in there in the background. You know, you'll have to make whatever judgments you want about how we created him, the accent, the the appearance. Those were all very thoughtful in terms of what we created.

So we're excited to have the Jacob bot with us here on the So what's it being a podcast bot? Is it, does it feel normal? What are you feeling right now? Do you feel things?

[00:00:57] Jacob Posel: You know, I watched all of the previous podcasts and, you know, especially tuned on those and then a couple other of my favorites. So if I say, start saying things that sound familiar, that's probably why.

[00:01:10] Taylor Holiday: Okay. In all seriousness, this has been a wild ride on the Twitter scene with Jacob for the past month. I don't even know how we landed in this idea of making him appear to not be real, but it certainly was believed.

I'll, I'll tell you that. So I I'm shocked at how much I'm, How much that took off, but Jacob is not a bot y'all. This is the real human being. He's actually sitting in middle America of all places. Jacob, who are you really? And what are you? How did you end up here at CTC doing some work with us?

[00:01:43] Jacob Posel: Well, yeah, I am. I am a real person. I'm stuck in Indiana in America. I've been, I've been working on AI related problems. Ever since the new, this gen AI wave came out it was kind of shortly after finishing my masters in like a related space and it to me, it is the most mind blowing Revolutionary technology that I have access to right now, and so I kind of went all in on it.

Through through the two of the sphere. I got in contact with Taylor. I bothered him until he was until he was willing to talk to me and then started working on some stuff. For CTC started doing some pretty cool building some pretty cool things. And now here I am just continuing to do that. And really excited about the things we've been putting together.

[00:02:25] Taylor Holiday: So one of the things I want to start with. So Jacob is the kind of person that. Is capable of cutting through a lot of noise to get attention and what he wants. And so when he says he was bugging me, here's, here's the thing that normally happens. I get endless DMs on LinkedIn and Twitter and whatever.

And generally speaking, people have this weird disposition, whether they realize it or not, they're asking me to do something. Which is like, talk with them, give them some sort of advice or conversation, or even read their thing and interact with it in some way that is, Serving ultimately to them framed as a opportunity or, or request for me.

And Jacob did this thing, which very few people have done, which is, he just started building things for me or showing me things that he was building that related to things I was talking about. And at first I went back and looked through the DM thread. Cause I was like, how quickly did I respond to him?

And at first I actually kind of ignored him still. Like I, even though he would make these very like personalized videos, it still didn't happen instantaneously, but he was persistent enough to And he would actually consume my content. And then as I talked about a topic that was relevant to something he was doing, he would send it to me and show me what he was making.

And so, my first question is like, what, where did you get sort of like a predisposed bias to action like that? And what made you think that that would that that would work?

[00:03:43] Jacob Posel: I mean, my perspective, my overall perspective is that there's nothing more powerful than building something. Anyone can talk, anyone can say whatever they want, but if you can create something, then you are. Immediately positioning yourselves significantly great in the competition. And I mean, that's why I'm in the space that I'm in, because you have the ability to build things incredibly quickly and there's true value to that.

And I mean, why I did it, I don't know. I, I try to do things that will. Increase my surface area of luck in my life. That's the way that I kind of approach problems. And to me, this felt like this felt like the natural path. I really closely aligned with what you were saying online. And that overall personality, I consume a ton of content.

In the, in the space. And I mean, that's just the way that I could differentiate myself and that I could show you who I was. So I don't know. I went ahead and I built an inventory management Shopify app. And I think that was the breaking point.

[00:04:41] Taylor Holiday: Well, that was it. That was exactly it. And it was because so much of the, we were having this dialogue about it and. And your first life in this space, and this is where there's actually these cool overlaps between your past experience in your classical training in terms of your schooling. And then you actually spent time inside of working for a company that was trying to solve inventory, like was building a WMS, right?

Is that where your first life was?

[00:05:01] Jacob Posel: It was in the fulfillment and logistics space. It was. We would integrate into WMSs and it was, it was trying to revolutionize the fulfillment industry. So, I, I think I've seen more fulfillment logistics data than most people out there. And like, that's also incredibly powerful and it gives a different perspective, but yeah, that's how it kind of, how I got started.

[00:05:20] Taylor Holiday: Yeah, so that, that was part of the compelling thing for me is that, okay, this isn't just an engineer with talent and persistence, but he also had spent time in an area that I care a lot about and that we're trying to move more towards right now. And so I was like, okay, there's something here, but it's funny.

My first thing, what I did was I actually, you know, With someone like Jacob, what I knew was that one, there's not like an obvious dev role at CTC. Like we're a service business, right? So it wasn't like I have, Oh, this is the job opening. You should come and apply for. So I knew I had to like intrigue him a little further to get then.

This relationship. So one of the things I do have, though, is that I talked to all the people who are building the software products in the world, and I know all of them. I introduced them to, I think, 4 or 5 people that were building software products and were like, Hey, you should go meet Kyle Hensi.

You should go meet Leo. You should go meet all these different people and maybe they'll offer you a job. And that's going to be a good network referral from me to them. And from you to me to where you're going to be like, wait, Taylor knows all these people. He's a good spoke. That's going to increase my surface area in some way.

So I sent you on five job interviews, not at CTC to start. Right.

[00:06:24] Jacob Posel: Yeah, yeah. That initially impressed me. I think more than maybe you realize and what a lot and kind of when I, when I, like, I, I reach out to a lot of people. It's what I do. I'm pretty persistent and I do things. Not a lot of people. Respond. And not a lot of people actually got out of their way to do something.

And I mean, I got to meet with five CEOs from some of the most exciting tech companies in the space right now, just from a single email. And to me, that was incredibly meaningful and also like an amazing experience. Now I can, now I, now I have kind of access to them. And it's just incredibly powerful for me.

So yeah, that was, that was huge.

[00:07:03] Taylor Holiday: And what I knew was like, okay, this is somebody I want to do something with. We're going to come up with some problem to solve together, but in order to capture his time and attention, He has to see me as a resource to what is ultimately Jacob's future, which is going to be to build something on his own and to see that I represent access to that.

Cause this is like a non traditional hiring process. And I think for founders or anybody who's listening a lot of times, the sooner that you can understand, okay, who is this person? And what are they going to care about? And there's just certain people that you meet that it's like, the pathway is not going to be.

Hey, you're going to hire as an employee and seek out 10 percent salary compensation every year for the rest of your life. That's just not the phase that Jacob was in or what he was wanting. So the frame was sort of set up from there and then it afforded us an opportunity to like get on the phone and dream up, like, what could we work on?

And that's what sort of led us to what we're actually spending time on right now, today, and hopefully there's going to be a lot more going to come in the future. But for now. What Jacob and I are tackling is thinking about how AI could, in particular, improve the creative function of common thread collective.

And the reason I went out for this area is if you've been following any of my content, one of the things I really believe about creative is that volume enablement. Is a critical component to success for brands of all the partners that we work with across large enterprise brands and everything else. I watched them struggle immensely to produce high quality content quickly.

And yet the math tells me that that's a massive advantage over anybody else. And so this problem, and candidly, like right now, CDC service offering actually fails to deliver on the methodology that I believe in. And we have room to go to improve against that. And so he and I started thinking about this.

Well, I would say Jacob simultaneously, you were becoming captivated by some of what was happening in the gen AI space. So maybe talk a little bit about what you saw happening and then how that sort of overlaps with the methodology I was talking.

[00:08:52] Jacob Posel: I mean, I think of AI is this wave and it's, it's constantly changing and it's constantly improving and flowing. And the best way to use it is to just integrate with it every single day and see the value that it can provide. Okay. And. I mean, what is, what is a creative system? It's ultimately synthesizing data being creative and then outputting something and then outputting an asset.

So, when, when I met with you and you kind of explained to me what you were trying to do in the the process that you were trying to achieve Me playing around with AI and playing around with all the different systems, like the LLMs and the image generation models and everything. It just felt like it's so naturally fit into this process and it would like be fuel on the fire in terms of the process and the system that you were trying to build. 

[00:09:37] Taylor Holiday: So let's talk about what that system entails in terms of what creative is. Cause I think one of the things I have tried to do. As this AI thing has been happening is that what I find not to be immediately obvious is, or that we do poorly is like, if you just go like, Oh, There's this process that exists already, and I just want to substitute a block in the process with an AI thing that accelerates it.

You almost have to totally reimagine from first principles what the process is and then allow AI to sort of wander within the bounds of the desired outcome versus trying to insert it as a step in an existing process that was built. Around different tools. And I think that's the mistake a lot of people make is they just go, Oh, take this one step of my process, the headline writing and make that one step faster.

Without really realizing that all the other steps they're doing are actually better done by AI as well. And I'll give you an example. When I think about creative strategy, okay, which is the process and the basis for creative work. It's the foundational thing. And it's funny, Andrew affairs and I have been debating about this a lot.

And I listened to his podcast with two two people who are the leaders in our space in terms of creative strategy. One is Sarah Levenger. And I forget the other podcasts, honestly, but the one with Sarah, where she's describing her creative process and it basically goes something like this, absorb reviews, model them through a framework developed by an economist.

Output how those reviews inform that model. Use that to write briefs for ads like that. That's a very simplified version of what she's saying, but in reality, that's exactly what a creative strategist does. They go and analyze historical data. They read a bunch of words to interpret into a framework that they've learned, usually developed by someone else, to output how new words could relate to all of that information.

And that is by definition, in many ways, what an LLM is doing for people.

[00:11:33] Jacob Posel: Yeah, that's funny. I was listening to a podcast earlier today with Nick Sharma, and he was talking about the exact same thing about how he was going through and reading reviews manually and like setting them aside in the margin and writing them out. So, Yeah, I mean, synthesizing data like that, that is a perfect task for AI and the important thing that people need to understand when interacting with AI is you need to set it up in the boundaries of an existing system that you've kind of verified previously.

So we're not like, I feel like people, people don't understand what happens. Like we're not just giving an LLM 17, 000 reviews and saying, do something. We have a very structured process where, and we understand the framework that we want to work within. And what we do is we just take something that is.

10, 000 times more efficient than a human. And also costs a negligible amount to run and we input it within that framework and that's what allows us to produce truly valuable outputs compared to people who just paste things into chat GPT and say, do something with this.

[00:12:37] Taylor Holiday: So that's exactly right. And that's where I think what people, what I see as the underlying asset that makes LLM is really effective is two things is one is the data. So if we think about what we have unique access to is we already have built in API integrations to all these marketing endpoints. Right.

So we have all the reviews, we have all the historical ad performance. We have even in some cases, like all of the customer service emails that have ever existed, all the demographic data from GA and Facebook and all these things that inform ideas. But the question of the output and what the structure of the output is, is actually rooted in existing creative strategy frameworks.

Right. So whether you're using the macro economic economists that Sarah Levenger messages, or in our case, we have like an HBR guide into these 20 different, 29 different angles that make that motivate human purchase behavior, or you use the CTC offer angle, audience framework, or for peaks theory or whatever it is, the underlying output relative to some informed creative model or framework is the other missing component that you can, that people don't realize that like literally the exact thing that a human walked through, whether it was Nick Sharma or Sarah or us, That's like you could teach an LLM to do that exact behavior.

Like you don't, you don't have to just wonder what it will create for you. You can teach it to create in whatever output you are currently using.

[00:13:53] Jacob Posel: Yeah. And with a lot of consistency as well, like if you understand how to use the APIs in the system, you can generate a lot of consistency. And if you dedicate time to really structuring your prompts in an optimal way and using tools out there that are best for generating prompts we've essentially created a system that's hyper efficient and incredibly consistent and also produces amazing results.

Like it just, like it does. It's, it's fascinating every time. It's, it's amazing to see. You run one of these things on data and the results are incredibly impressive and a lot of fun every time to actually go back and look through.

[00:14:27] Taylor Holiday: And the other thing that I think we don't realize that happens to us as humans is we're, we're unfortunately unconscious or unaware of how our unconscious bias affects our decisions, even when we think we're applying a framework, right? So I was just up last week. I got a chance to go spend time at driveline, which is a sort of the leading edge baseball performance.

Facility in the world. They train people like Shohei Otani and Mookie Betts, and Travis Pizzano, the first overall pick in this year's MLB draft. And they're, they're led by this like sort of like brilliant engineer by the name of Kyle Boddy, and one of his theses is like that he thinks a lot of coaching is going to be replaced by AI.

And a lot of it's because if a coach is sitting and watching you, their interpretation of your movement is often riddled with bias from, you know, Other, the limited set of data that they've seen, the players that they're learning, the analysis that they're trying to do in their head about the mechanics of your swing relative to your own biology and different things.

And that's just problematic and produces random, not consistent results. Whereas if we can actually, we can actually perfectly apply frameworks against data with no Pre existing bias to the analysis in ways that LLMs are really effective for. So it's even in many ways, I think for people who have a strong point of view on creative methodology, this should be the most exciting to them.

It's actually not substitutionary, but you can get to a perfect level of education, all execution on the things you believe. And that's actually, what's more exciting to me is that. I want to deliver on this promise really consistently and accurately, but what I find is that me and everybody else are subject to all sorts of reasons why we're bad at executing against it perfectly.

[00:16:04] Jacob Posel: Yeah, yeah, it's it's a task that is very well suited to something that understands the entire internet and understands everything that's ever been said by a human almost in its entirety. And so, yeah, you can, you can train in to remove biases, or if you're someone who really particularly feel strongly about something, there's techniques as well to bring those in.

Like you can fine tune your own model, you can fine tune an open AI model. Which means basically teaching it to speak like you or speak how you want it to speak for a very low cost. And you can also do that in your prompting. So, like, the point is, there's a lot of flexibility in these systems, and it doesn't have to necessarily doesn't have to necessarily exist in one way.

It's just an incredibly powerful technology that can get the output that you want.

[00:16:50] Taylor Holiday: So let's talk about specifically what we're doing. 'cause I think there's two key components here. One is, as we're describing creative strategy, and we've started building a product that we're calling the ultimate brief, which is a way to use AI to build a foundation for great ideation around two core components is one is how many things do I need to make and which things should I sell and why?

And then two is. What should be the creative angle or the creative audience that I'm targeting for those things. So bringing together both a production element around how many ads do I need to make next month to accomplish a financial objective and then to what should those ads say? And why? And so maybe talk a little bit about how we're approaching solving for this problem.

[00:17:35] Jacob Posel: Yeah. So there's multiple layers to what to that go into this problem. So, not everything we do is just generative AI. So there's this question of how many ads do I need to produce in a given time period? And we're applying very traditional machine learning and probabilistic techniques to that we're modeling the distribution of previous out performance, and then we also CTC has an amazing system where we know what you're anticipating and planning on spending in the future and how that's going to be executed.

And so there's two things married together, produce, allow us to provide a very strong estimate of the number of ads. That you needed to need to make and like, that is that that's real engineering. That's real data science. That's maps and it's not that's not asking to do it. And it produces very accurate results, at least from what we've seen.

There's. Sorry.

[00:18:28] Taylor Holiday: Keep

[00:18:28] Jacob Posel: And then on top of that, we we include gen AI in order to create this creativity. So we'll synthesize multiple different data sources very, very quickly based on our integration. So reviews and different demographic data. And sales data and so on and allow that to come up with these creative elements.

So who is your audience, your personas who, who is purchasing from your brand and why do they purchase? And what are the unique selling points that they like? And also what are their pain points? These are all things that we can automate and we can get straight away. And then like the angles, why are people purchasing from your brand?

What do they like? And then tying those, there's a relationship between those. So, you know, for example, if a persona of your brand is a gift giver, if it's a menswear brand, but the persona is a gift giver, well, there's a specific angle that someone is gifting something for a man. And it's really interesting to see that relationship.

And that allows you to come up with your concepts your marketing concepts and create a lot of them in a very, very short period of time.

[00:19:28] Taylor Holiday: One of the underappreciated realities of creative work is that constraints are actually like the key to building really effective systems. The more open ended the task is the harder it is to do. So if you think about what our system enables uniquely. For CTC broadly is that we, because we are responsible for the FPA, the financial planning and analysis of every organization.

We get to define some very important inputs to this process. 1 is the budget, the spend expectation of the organization and the efficiency goal to produce the financial result that's needed from there. We're able to understand which products are selling best at which margin, at which AOV, how many ads are currently live, what the expected spend of those ads is in the future to identify the gap of how many ads need to be created.

This is a critical input to building any creative processes. How many ads am I trying to make next month before I get to what should they say and look like? I need to have that input. And in my experience. No one knows the answer to this question. There's sort of like, even I was listening to yesterday the the sort of most sophisticated, I would say data team and maybe all of the commerce from resident home who they own Nectar brand.

They own like a ton of different mattress companies. They're enormous. And they have like a team of 50 data engineers that work on their business. And he was telling me that their creative workflow is they try and make 20 ads every week for every brand. But like that is a completely random output suggestion.

Like why 20? And the answer is because it's a lot, like it's more than most people can make, but there is no data driven reason to make 20. Like there's just no way that that is actually indicative of what. Could or should be made for the organization. It's just a process that they've developed because you have to create a number.

It's an important constraint. So like, like James said, like before the AI even gets to work, there's some very traditional data science and machine learning before we generate any new ideas. But then once we're there, now the AI can begin to apply thought around, okay, well, what should I sell and why and to who, and what should it say?

[00:21:31] Jacob Posel: Yeah, yeah, I mean, it's an incredibly important point in that what we're also doing is we're tying creative strategy into the entire holistic picture of your business. So like we, we have an ultimate brief, the current available inventory and the contribution margin and the sales volume. And we can do that to optimize our creative strategy based on the goals of your business.

So if you wish to sell something that's only in stock and at highest margin, like we can do that for you and we can come up with those concepts. And we also know for that particular product, who's buying it and why they're buying it. So, it's a great system. We

[00:22:07] Taylor Holiday: Yeah, it's like, just ask yourself how many creative strategists in the world can at any given moment give you an exact, accurate inventory count, margin by skew, best selling skews over time, which audience work best, like just being able to, like, we all would use this phrase that like, that's a David Ogilvie quote, that good information informs good inspiration.

And so we forever have tried to teach our creative strategists that, Hey, you need to know what the best selling products are. And Hey, you need to have a general sense of the inventory position. But the only reason I wouldn't say that they should know the exact real time, exact count of all of those things is because it's not humanly possible.

They don't actually have the ability to do that. And so that is a limitation of their ability to recommend good ideas. But we've now eliminated that we now can have a creative strategy input that perfectly knows those things at all times. As well as the marketing calendar and product release dates and what ads are live in the ad account and the current performance of all the ads against different audiences and different angles.

And that like they operate and suggest from a place that houses all that knowledge in real time, all the time.

[00:23:09] Jacob Posel: Yeah, yeah, this is, this is really the data infrastructure that makes AI the most powerful and also now empowers the creative strategists to have that access and have that and to have that knowledge and it just, it improves them as well. Like, you have this amazing tool, but the people are now empowered to be so much more efficient and effective.

[00:23:28] Taylor Holiday: That's CDC is uniquely positioned because I don't know. How many, I'll actually even venture to say that there are no organizations that possess the corresponding data set where you have the financial future plan, as well as all the historical performance layered with all of the creative knowledge, like the combination of those three things as the data infrastructure.

For CTC is I think where there is something here that doesn't exist anywhere else is that there are places where the financial planning happens. And there are places that track historical ad, creative analysis. And there are even places that are doing generative AI, but there are no places where those three things exist together.

And I think that's really why we're excited about how to tie together us, a creative system against that outcome. Okay. So Jacob, but then, so. Cool. At first we were like, Oh, let's just help people ideate faster. Right? Like, and that, that's sort of what the ultimate brief is about. Help them make more ideas quickly.

But then you get to what I think is the real burden of creative, which is the actual production process. And I'll just describe for people right now, what happens at CTC that I think is a fairly common problem. Production modality, which is like, okay, that creative strategy process leads to a brief, a brief is written and sent to a designer, the designer that interprets that brief into a visual output, whether that's an image or a video that then the creative strategist reviews and gives feedback on that, it gets edited into a second version and gets sent back to the creative strategist that then gets sent to the client.

They usually have some feedback or assessment on it. They get sent back. To the creative strategist who then interprets that feedback back the designer and round and round you go until eventually something goes live in the ad account. That's been approved by everybody along the way. It's expensive, cumbersome and slow.

In almost every case. Now, there are places where that circle or that cycle is tighter, but for the most part, that's sort of what's happening. So where do you see AI playing in a role in improving that process?

[00:25:27] Jacob Posel: I mean, there's multiple, there's multiple different ways that it can in that process. So. Coming up with the briefs. Obviously, we've spoken of that. That's now incredibly efficient, incredibly effective. Coming up with the coming up, being able to not iterate on those ideas and build volume in ways that clients can approve what they want is now incredibly fast and incredibly efficient.

Being able to being able to take feedback and apply that feedback now, it doesn't wouldn't it wouldn't require human intervention anymore. So. AI now is a system that can understand human intuition and it can understand human biases and emotion. And that was an unlock that wasn't possible before traditional engineering.

So now being able to come up with those ads is super efficient in a way that it wasn't before.

[00:26:17] Taylor Holiday: So if we think about one of the phrases that I think is really important for people to understand is like, is our, our L H F. So can you, what does that acronym stand for? And how does that relate to AI? That's

[00:26:34] Jacob Posel: It's a technique that's used in training some of the foundational models to turn these essentially really, really good predictors of tokens and next words into something that humans will, humans can actually interact with. And it's kind of what, Made LLMs powerful.

It's it turned an LLM into chat GPT, something that people really so it's a process in which humans actually they reinforce the behaviors of the AI by physically marking responses as better or not. And that's what creates the foundation models.

[00:27:07] Taylor Holiday: So what I want people to understand about how important this is, is that in a service business, what happens is that process reinforcement learning through human feedback is what a designer gets in from a client. They send them an ad that the client says the font is off. The logo placement needs to change.

The image treatment needs to be adjusted. Okay, that feedback goes into the brain of one person.

[00:27:32] Jacob Posel: I'm,

[00:27:33] Taylor Holiday: one person may be able to interpret and create an alignment between the human and themselves. Okay. Here's the problem for my business. That knowledge. Does not compound across my organization. It's actually entirely trapped in that individual.

If that individual leaves, they take with them all that RLHF that occurred between themselves and the client it's gone forever. And I'm starting back at zero and this happens over and over again. Well, when there's a change on either side of this, You have this very similar process. And this is why I think people need to understand is that in many ways, LLMs function very similar and are modeled very much after our brains in different ways, which is that when you get feedback, you make an adjustment.

You try again, then you get feedback again. And the person goes, okay, I like that. I'm happy now. If you've ever noticed in chat, JBT, you can thumbs up or thumbs down responses. And all you're trying to do is do that. You're trying to give human feedback to the machine about the quality of their response.

Now. Okay. Inside of a design system, it gets compounded across the company forever.

For the organization that you're interacting with continually compounds and improves over time, such that the value gets like extrapolated to many, many people forever versus locked in one person's brain.

And that is like, that idea has been. And sort of a realization for me of like, oh, the crap, like what this does is CTC is supposed to be the value of CDC is its institutional knowledge, but in reality, that's locked in tiny little compartments all over the United States and individual people, and now we can set it free.

[00:29:03] Jacob Posel: I mean, what is, what is an LLM? What is AI? It's a neural network. It's modeled after the brain. It learns. That's how, that's how they train these things. That's why they spend billions of dollars on GPUs because they're spending time literally training these things and teaching it. So, What you're creating is this company brain where the practices and the knowledge and all of that intuition is now, it now exists in this brain, which yeah, which we can use.

[00:29:29] Taylor Holiday: And this is why part of why. It's so important that right now you go and try like what I see it from a lot of people. It's like, oh, it's not ready yet. Okay. But what, why this is such a dumb response to me is because the mechanism by which you get it ready to serve you is this feedback loop is the try reinforce, try reinforce, adjust, reinforce, try again, reinforce, and the moat.

That we will create in the persistent early trying will be massive because what's not going to exist in the form of like what Peter levels IO is doing for photo AI is he's solving a very broad general problem, right? We're solving very specific, narrow problems, right? In ways that won't be the same.

Right. Even just like things as simple as like sizing everything for nine by 16 versus one by one versus general photo outputs versus product photography and the novelty of these different things. It's like, it's just a process that I think the earlier you get into it, the more likely you are to create the kind of feedback loops that improve it over time.

[00:30:28] Jacob Posel: Yeah, that, that is an underrated point. And it's not just the mode of trying it and understanding how it works. Like, I mean, this is a being with a personality at this point. I really, like people, people say like, is it a real thing? Like it's, it has a personality. It has a way of speaking to it. You have to understand that and having a company organization level understanding of how to do that makes you immensely more efficient.

That's just like, I would say that's a fact at this point. And the other thing is having the data infrastructure set up to take advantage of it. Yeah. Like, not everyone is just going to be able to plug in plug in that their company processes into an LLM if they don't have that data infrastructure set up.

And we're lucky we have statless and we have growth maps and CTC is a very tech forward company. And so it fits in naturally. But that is, that's just going to accelerate you. And I, I like to use this analogy of AI as a way we're riding the wave. And it's constantly improving. It's constantly evolving, but if you're positioned on this wave, you can easily plug into the newest updates which immediately accelerate us compared to other people who now need to say, okay, Now, maybe, maybe it's ready for me now.

It could have been before, but now it is and now I have to go in and understand how I'm going to use it and apply all of this CTC is just going to be changing an input on an API probably, or we're already going to have all our databases set up. Everything's going to be working smoothly. And we're going to be ready to take advantage of everything.

And we already are, like people say, it's not ready. Like it is, it is ready. You probably just don't know how to use

[00:31:57] Taylor Holiday: So, so let's talk about that because I think this is the big critique, right? Is that it's like, it's not ready yet. And maybe like, I don't know what people really mean when they say that. I think one of the questions is like performance, like, is it actually viable that people, these ads will perform in the ad account?

I think we're going to have all of these answers and be able to share data really, really soon. But what is your take on this question? Could a brand today create value for themselves using AI on the creative generation specifically?

[00:32:26] Jacob Posel: Absolutely. Absolutely. So is it ready? The answer is yes. Is that time up front required to understanding how the answer is also yes. I mean, I can talk through multiple things. In terms of you want to generate copy. Well, there's multiple techniques you can use to generate copy that actually work for your brand.

You can try prompt engineering, which means giving it a lot of context instruction, outputting those Those headlines and that copy and that subheading which I spend a ton of time doing. You can also fine tune a model. You can get all of the previous copy that you've ever written in emails and on your website and in your ads.

And you can specifically fine tune a model, which means changing the weights that it uses to actually produce outputs. And you can make it more accurate in that way. So those are techniques that you have to understand and you have to use. So that's like, that's one small thing. The other one is ideating.

So yeah, you can you know, you don't think that you can ideate. Well, you have to spend time upfront thinking about the data that needs to go in and the framework that you want to use. But if you do, which is what we do, then it produces incredibly consistent and high performing outputs. And then in coming up with the actual creatives themselves, which we haven't spoken about too much, but we're dedicating a ton of time doing.

There's, there's multiple different elements to that. So like. With the images coming up with images, and this was the big controversy that happened on Twitter. A lot of people think that AI can't produce good images for them or their product. That's not true. It's not true. It produces amazing images.

I've seen it. We've seen it. Our brands have seen it. You you just have to understand what goes into producing those images. So you have to dedicate time to understanding what are the training samples the training data the Images of my product that I need to feed in to fine tune something to make it To make it look good.

That's something that takes time It does but and also you have to understand how it works But once you do then I mean, yeah, you get some pretty amazing things like taylor's showing right now.

[00:34:27] Taylor Holiday: Yeah, so I'm sharing on my screen right now a Twitter post I put out, and I'm going to do more of these all the time, because I think it's just like, I don't understand how people don't understand this, which is that people will tell me all the time that they can always tell if something is AI. And this is just, this is the most insane claim.

Ever like I am willing to bet any amount of money that if I were to put together a large library, that you would not be better than a monkey throwing darts at guessing which one of these are AI and which one of them are not. And so I put this out, I said, here's four images, which ones are real, which one are AI.

Go read my comments, go see how often people were accurate. And I'm going to answer the question for you right now. So Jacob, do you, you might not even know because these aren't all yours, but do you know which one of these four are AI and which are real?

[00:35:11] Jacob Posel: I know that one is AI. I was the one responsible for training that model. I believe.

[00:35:18] Taylor Holiday: photo of a Nike plate that is a customer of ours, Nike Strength, and that is AI. That is not a real image. That is AI generated image.

[00:35:26] Jacob Posel: Yep. Everything within that is AI. There's no photoshopping. That is a completely AI generated image. I believe three is AI too. I think I saw that tweet.

[00:35:35] Taylor Holiday: Yep.

[00:35:36] Jacob Posel: Two is AI. I know that one sure. And four, I have, I have no idea.

[00:35:42] Taylor Holiday: They're actually all AI. In this case, they're all AI and number four is like Peter Lovells created a security cam filter for fun and just put it out. And what I love about that one in particular is that like, That's a perfect example where. It the reason people think that it's real is because it's represents a filter concept that exists in the real world.

It's not intuitive that I would naturally create that. But if you tell it to, it can replicate what it looks like to look at something through a security camera. And that's why a lot of this prompt. You can actually give it photography. Prompts like that to use certain cameras or lenses or whatever. They all have inherent in them a certain composition that makes sense.

But these are, these are all AI and you know, it's a, it's a little unfair maybe is an initial test, but I was really just sort of trying to prove the point that if you said any of them weren't, then the game is over. Like you just don't get to claim that you can't be confused by the premise. And I think that is, that's just the reality of where we're at is that Is it perfect?

Well, that's not really the question. That's not the test, right? The test is, is its value in the world, meaning your capacity to put it in an ad and generate money for you relative to its cost profitable. And that to me is like, Who cares what anybody thinks about it in terms of how, what percentage of people would identify it as real or not.

That's not actually the test. The test is, does the cost of production yield a profit for you as an organization? And I'm absolutely sure that it could because its cost is so fractionally small compared to what it would cost to go pick up that plate and take it to a gym and shoot it in studio. Edit it and put, it's like, we're talking about like the orders of magnitude here.

I think it worked hard for people to comprehend is that we're talking about like thousands and thousands of times cheaper to produce these things ultimately than it would be to go replicate that process.

[00:37:27] Jacob Posel: Yeah, yeah. And I mean, just to touch on something you said really quick, when I produce an AI image, I don't just. I just don't, don't just go to the base flux of standard stable diffusion model and produce an image. We there's, there's mixing going on. There's photo realism, there's upscaling, there's effects and there's styles and those are inherently valuable in and of themselves.

Like imagine being able to apply any style, any camera style you want to an image or any lighting. Like that is an incredibly powerful thing to be able to do. Being able. Being able to do all of that these underrated benefits that people are kind of dismissing because they say, Oh, AI is not ready.

It's not, it's not good enough yet. Well, it is. And it's, it's really, really cool. You just, I'm begging people just go, just go play with it.

[00:38:12] Taylor Holiday: Yeah. And I think at this point, there's actually a massive advantage for creatives, like for photographers, because like, if I set up a Sony camera right now, it's no different in that, like, I wouldn't get the same outcome from it. That would a great photographer. That's the same truth for setting up AI.

It's why, like, I couldn't go make this by myself and why Jacob's unique skillset allows him to take advantage of the tooling sooner than other people. By the time someone as dumb as me can make it work. We're, Well, then there's no advantage at all, because if I can go in there and create it, then like, at that point, everybody else can too.

But for now, there's real advantage to understanding camera settings in different filters and thinking about lighting and being able to speak to the AI through the lens of your own creative knowledge set. So I think that's huge. So, so maybe let's talk about some of the things we're practically doing right now.

So you're in the middle of a project right now. For a first set of customers that we're saying, Hey, we will create for you and train a model to produce your product photography. So talk a little bit about what you're up to and maybe share how people could get ahold of you if they wanted to, to see if they could,

[00:39:13] Jacob Posel: Yeah so what, what we can do is we can essentially take your image or a person or a combination of the two, and we will fine tune a special model for you. To be able to produce endless images off that product on that person in different settings anywhere, which is, which has been really amazing to see we were very, very careful and ensuring that it's high quality.

So we dedicate a lot of time up front to what I said. Picking out the training samples and making sure that they're the ones that will work best and produce the desired outputs. And then we'll go through and we'll train those. And I mean, the results have been pretty, pretty impressive, pretty amazing.

I don't think that people will be able to tell that these are AI at this point. I mean, you're taking. You're taking ghost mannequin images of a shirt and you're placing it on a guy on a boat in lake Como or at the F1 or playing golf. I mean, and the, the pictures look amazing. They look completely realistic.

They're high quality. And the product itself looks absolutely incredible. And yeah, it's been, it's been really amazing. So if anyone, I mean, if anyone's interested in doing this, they can, they can reach out to me on. They can reach out to me on Twitter, they can email meco, follow on. But yeah, it's been, it's been truly impressive to see.

I'm a doing.

[00:40:37] Taylor Holiday: Yeah. And with that, Jacob, I think we're going to be publishing some of those photos soon. They're a private project for a client right now.

So we're going to keep them hidden for this moment, but we are going to keep doing this and we're going to be, I'm going to be actually launching some ads. You and I made for cuts in a campaign here soon that we're going to share the ad results for and be able to validate that. Like, Hey, we were able to produce launch and.

Generate revenue off the back of Jenny. I creative at this point in time today, right now. And we're going to keep doing more and more of this. And one of the things I'll say is that we're also thinking about how we can. Ideate around the source data from an image standpoint that makes product photography, most useful.

So

[00:41:15] Jacob Posel: and

[00:41:16] Taylor Holiday: of the time you get product photography from people that is sort of like either flat lays or sort of very single directional visuals of the front side back in a very simple way versus being able to get like, what if we had a thousand images of the product from every eight imaginable angle in all different light.

On off model. Like if we could think about the training material as part of the asset library that informs your AI to make it more impactful into the future. So it's an investment in infrastructure that builds longterm leverage. Right? So I think there's all these ways that right now brands who are willing to engage can create a real advantage for themselves going forward.

[00:41:54] Jacob Posel: Yeah, the gold mine right now is in the training data. If you don't have those clean, readily available photos of your product, you're just not going to get the full benefit of the model. And it's the same with everything else in AI. If you don't have that available, then you're going to be at a disadvantage.

So it would be amazing to be able to provide people a better opportunity to actually get those assets. Because I mean, you, you take like how many pictures you need and then that's it forever. You've got, you can create any image you want, wherever you want. Based on that photo shoot.

[00:42:26] Taylor Holiday: So if you're interested in getting an AI photo library built for you, reach out to Jacob or myself on Twitter. You can send him an email. We have a limited number of slots. We have a big exciting debut. We're gonna be making to our existing customers here soon about how this product is going to continue to develop.

It will be part of the CTC offering inside of statless as well as well as other components. And then maybe maybe a way for you to get. Interact with Jacob in some other ways here coming soon. So lots of fun stuff. but Jacob, we appreciate you coming on and make sure you follow him on Twitter so you can keep up to speed with what is happening.

[00:43:00] Jacob Posel: Thank you. Thank you. It was a lot of fun. I love, I love talking about AI. Okay.

[00:43:04] Taylor Holiday: All right, y'all Jacob, you can unplug and we will see you all later.

[00:43:09] Jacob Posel: See you.