How Top Brands Are Deploying AI Today | TikTok Operator's Pod

On today's pod I'm joined by a long-time TikTok Shop operator, Sohun Sanka.

Sohun has been a day one player in TikTok Shop on the brand side, software side, and now with this own venture: Clankers - A consultancy working with top brands and agencies to deploy their own AI Operating System.

Together we explore the real use cases of AI inside top brand operations today. How brands are using AI to find the right creators, connecting the dots org-wide to put TikTok content to work everywhere, and why it's mission critical that businesses operate on their own O.S.

Transcript

SPEAKER_00 0:00

Hey, do you want to increase ROAS? It's not as simple as just getting better creative.

SPEAKER_02 0:04

People are raising hundreds of millions of dollars to build products where the customers are AI agents. Only the internal brand teams will have that context. And if they can just connect it to their forecasts and other AI solutions, they will win. They will win in the age of AI.

SPEAKER_00 0:24

Hello, hello, hello. Welcome back, everybody. I am joined by a longtime homie who I think we've done at least three or four different podcasts together at this point, have bantered back on LinkedIn quite a bit. Sohan, who has been in the TikTok shop space since day one, he's been on the brand side. He's worked on the agency side with some folks. He's been in a consulting capacity. He's been with platforms. He's been all over the board and has recently ventured off in his own to start the development and build out of Clankers, which is a name I fell in love with when I saw it. I was really mad I didn't think of it sooner. It's brilliant. And so he's going to give us a breakdown today and how he is working with both brands and agency partners deploying their own internal AI systems and tooling with a general thesis that, and maybe this is where I'll start things in terms of a question here. There's so much unique AI tooling out there right now. There's a different product going viral on X every two to three hours, usually seated by the same seven or eight influencers running big campaigns behind it. And I know personally a feeling that I'm sure many agency operators feel, as well as brand operators feel, which is just like, oh my gosh, this is so cool. I'm going to save it to my Twitter thread. I'm going to copy paste the link, send it to my Slack so I can get back to it this weekend. Inevitably, I never actually end up doing it. Maybe I get to 10% of it. And there's just so much that you're kind of on a hamster wheel trying to keep up with like what's the latest and greatest, where's their new alpha to be captured. Yet from a day-to-day basis, I do find myself falling back to my go-tos. You know, I'm going to use Gemini and Google Notebook for this and more research dynamics. I'm going to spend most of my day inside of Claude and all these projects and agents I have built there. But I still feel, despite everything we've done, that we're still behind in this landscape. So I know there's a lot there, but let me start with that. What's your mission with clankers? What are you setting out to achieve here?

SPEAKER_02 2:45

Yeah, yeah. Thanks for having me. And that's a great intro. So as Jake was saying, I've been on the software side on TikTok shop for about the last two and a half years. And what we've essentially found with the genesis of quad code is that anyone can build all these softwares in-house. And there's too many softwares in the space, actually. Any digital marketer has to manage like 15 softwares. I think their job should just be software handler, not digital marketer, because that's really all they're doing. Is like triple will, North Bee, and Kruva. It's just like all these softwares together, right? But the data isn't shared. People don't actually, people aren't actually able to build one AI solution that custom fits their need between all of them. So what we're helping brands and agencies do is learn how to use the APIs, MCPs, which is model context protocol, allows two softwares or applications to talk and do things with each other. Essentially use these protocols and APIs to build their own custom automations in-house. For example, if I want to create an affiliate health score for TikTok shop and then have Slack Alert surface to each pod that is being managed, like that doesn't exist in any outreach tool because you have to devise your own health score, you have to connect it to your own internal tooling, and then you need context that essentially says, okay, we need it flagged in this case, not that's that case, or we had it flagged like X amount of times over the last month. So let's look into that and do something else here. Like one software itself can't get that deep or that granular. So we're helping brands solve that problem essentially.

SPEAKER_00 4:20

Yeah, I think any operator in our space right now gets uh an urge at least once or every other week to say, screw all this, let me go start an AI consultancy in residential lawn care or roofing or plumbing or health care or finance with local banks. And obviously you've been an operator in this space since day one, but prior to breaking off on your own, you probably had a chance to go look at some sexier market. I know they're not, those aren't sexy industries, but sexier from what your week to week would look like, day-to-day would look like, particularly your schedule around Black Friday and Cyber Monday, that you know, there's no Thanksgivings and in e-com. So I guess I say all of that to ask this. What were the problems? Like surely you're sticking in this lane, not only with your deep knowledge, but because of some of the problems you had seen firsthand. So before we get into all of the wild things that you're deploying right now, what were some of the problems you've seen over the past few years given you've gotten to see in the back end of brands doing 100 million a year in revenue, running insane, complicated creator systems, agencies managing, you know, 100 accounts or more at a time. What were some of the problem sets that you had seen?

SPEAKER_02 5:45

Yeah, I mean, the first problem is that the e-commerce field is super non-technical compared to like what's available in SF, right? So people are still manually doing things like reporting or logging different statuses of their projects or things like that, right? That should have all been automated like maybe seven or eight months ago. Another thing is siloed context for TikTok shops specifically. It's sort of like the growth channel that helps generate demand for Amazon, repurpose assets from there for Meta, like the media buyers and the Amazon managers and the TikTok affiliate managers, they all never talk to each other. But TikTok was explicitly launched by leadership to halo growth for other channels and you know effectively increase the efficiency of your blended cactus TV. But no one's talking to each other, so that's a broken system. So I think that's one thing where it's like you can use AI to just read the data from here and read what people are doing and then just make better decisions moving forward. And the brands who launch TikTok shop and are successful are working with tens of millions of dollars. So that decision making is going to compound in the long run. And then the newest problems we're seeing that are bigger and a bit harder to handle are um agencies and other B2B services losing their enterprise value in the age of more of these outreach tools coming out, more software is allowing brands to become more agenc. Brands have also a little bit of like a negative sentiment about agency right now. Agencies are suffering a bit. And they have no differentiators right now in the TikTok shop space. So all of them are looking to build their own internal tooling, their own data layer to increase their enterprise value and exit, be acquired by a bigger agency or PE, but they don't know where to start. So that's another problem that we're helping them out with. And like an example would be hey, let's say I'm making this up. We have agency A that only does beauty brand marketing on TikTok shop. If they're able to extrapolate the exact patterns between these 15 beauty brands, document them into a database, and then say, hey, this happens when we push this lever, or when I push this lever, affiliate revenue grows by 15%. That becomes a valuable data set that they can tie in into all their strategies and then sell to a private equity firm later, and then they regain that enterprise value that's being diminished by the genesis of quad code and AI being so buildable, right?

SPEAKER_00 8:08

So one of the things to your point of enterprise value, like completely nailed it. I can certainly say too, yeah, it's a big issue in TikTok shop agency world right now, is it's a lot of the same structures, a lot of the same playbooks, which if we're not all doing different things, then we're are we really doing what's best for the ecosystem, which I think is a broader challenge. And then I know one of the things that's been interesting in agency land with TikTok shop is because it's a revenue-generating channel, brands like to squeeze and essentially put on the books that agency fee oftentimes into the PL as well. That's not as common in other channels. I don't know why it's been been applied to to TikTok shop. And this is so different than than Amazon that the Amazon comps like, you know, fail to get there. But one of the problems I've seen universally in TikTok shop lands, and as I speak with Kevin and other kind of operators on a more consistent basis, is there's a lot of there's a lot of good ideas thrown out. And then things that I know like us operators will throw out of like just do this. People will build tooling around or a system or talk about how they did it, but they only go halfway. It's like a it's a very half measure. And I've got a few to click into, if any are worth like double clicking, but I kind of want to explain like how to connect the dots fully here. So, like the first one that I know you recently created some content around and and probably led some product things on is this TikTok creator to meta connection, which gets thrown out by every person at TikTok, all of the agency partners. I know all even throw down on a call. I can tell you, like, that sounds very easy to do. Actually, executing that in practice is a real challenge because of the downstream problems that that exist there, as well as the fact that like even within that stream, you could buy that asset off right off of the creator, or you could run it into partnership ads, but now you need to hook it into a different data system. So if you have any color there, I'd love to know like on specifically that, how are brands and agencies like actually doing that? Because I've heard everybody, everybody just like throws it out there. No one gives an actual five-minute explainer on how to fundamentally do that.

SPEAKER_02 10:32

Yeah, yeah. So I wrote I wrote a piece of content on this recently, and as I was going through it, I just ended up bringing more and more and more tooling and different avenues of communication into it. And I'm like, holy crap, like this could go in like so many different ways. Like, you know, first you have to go and use your outreach tool of choice and then filter the videos by which ones meet your criteria for ads, whether that's hey, this has a 33% hook rate plus, or like maybe it does a thousand GMV and 33% hook rate. So you need to do that, and then you need to also set the cadence of that, whether that's weekly, daily, whether that's a trigger for any creative that meets that. And then you need to come up with a form to get the exact rights to purchase and run that in perpetuity or on a monthly basis per each creator. Now, each creator may also have different rates or different niches or like some sort of different specialty. So usually a boilerplate form cannot work, but you you can try it, and then if you try it and it doesn't work, then someone needs to handle the back and forth communication to get the creator to agree. Another level of complexity is where that communication is handled. Sometimes it might be handled on TikTok, sometimes the creator might be more responsive on their phone number, or they might be managed through an agent on their email. And then you have to track the statuses, conversations, and responses across all these channels for all these creators, and as you get more and more creative, that essentially stacks up more and more and more over time. And even the outreach tools like you know, Kruva Yuka Reacher, they're not able to pull in the statuses of the creators once they leave TikTok shop. So another layer of complexity there. Then you need to get confirmation, signing, and then facilitate a payment to these creators, right? And so now that depends on you know what you use. Do you use PayPal? Okay, how many creators are you paying? Are you gonna do a bulk payment? That's another level of complexity, and then the integration to meta, like Jake was saying, right? You need to get that asset, download the file, and then upload it to your meta ads library, or if you're running partnership ads or whitelisting behind the creator account, that's another level of complexity. So, all that to say, I think the way to solve this with technology is create a process that works for the 80%. Like, don't try to get every single creator who is going to like skew you off in one direction or want this very, very specific thing. Of course, there's exceptions for like if their assets are exceptionally good, but that's what you have human talent to do. Everything else can be automated through one thing that works really well. Like, I would basically say create a formula where it's like hook rates over 33% and assets over a thousand dollars, then send a usage rights form with a boilerplate template and be like, these are our terms. Then once they accept, run them in through, like let's say you send your form through type form, run that type form webhook into PayPal, issue a PayPal payment. Once the PayPal payment goes through, bulk upload to meta ads library, and then run the ad and run that for like 80% of your creatives. And like that's how you would execute that end-to-end. And all you need to do in-house is just get the API keys from every single software and just chuck them in the cloud. Obviously, there's some technical nuance there, but like chuck them into cloud more or less, just voice dump what I said in this transcript and be like, make me a usage rights process. It'll ask you a few questions and it'll just do it, and you just have to speak to it in English, but you can't do that with just one software. So, like, those are the complexities that people have to deal with. And one thing I want to mention earlier, you're saying that people take the idea from an account manager and make it half baked, but that's because the people who are always taking it never have enough context, and the people who are the account managers don't have enough technical expertise to communicate to the technical people and how they would understand it what the process needs to look like. Because every single thing needs to be very specifically communicated, and all your developers are going to be thinking about edge cases, different endpoints, documentation, specific like it needs to be so granular, and and usually account managers and marketers aren't used to getting that specific. So that's also an area where I see things fall apart between non-technical and technical teams.

SPEAKER_00 14:53

I find it interesting that you referenced hook rate. You didn't immediately go to just top line GMV, which is a call out because there's even a myth to assume that, hey, our top videos in TikTok shop ranked by GMV or ranked by ROAS are the ones that will necessarily or directly translate into Meta, along with the fact that the there might be product variants on top of that. There's a whole set of other variables. And not to mention your creative library inside of TikTok shop, you probably have anywhere from a thousand to 10,000 videos in there at a given time. I was just on call with a client earlier and I was explaining look, like if we're only looking at our top 100 creatives and sending that to the meta team who's not kind of looped into this process, well, that's going to be a problem because we could have absolute gold, but it's not seen in the top 100 because it doesn't work on TikTok shop, but because TikTok shop's a different channel, it's a different market, and its ad product is way too simple for the complexity you need to actually reach consumers today. So, like even your call out in there, you cannot do what we just mentioned manually. Because even if I said, hey, sort by click click rate, well, the videos at the top of that are going to be a 50% click through rate because it got two views and one person clicked, you know. Like, so there's there's also a lot of like outliers in that equation. So you do need a system to this. And then even as you're walking through that, for most organizations, we're talking about four to six different points of contact on that chain. That to your point, if they don't have the context, then everything breaks. Because just an example on the payment side, we know this from our past work in the paid influencer world, having to deal with with much bigger budgets. That has to get flagged into finance, which has to verify a contract, which over a certain dollar amount in the state of California, especially, has to automatically generate a W9. Now we need the address of that creator. Where's their communication? Who's directly reaching out to them? Oh, we run our payment through PayPal. They don't have a freaking PayPal. Okay. I guess we have to do Cash App. Okay, that's kind of sketchy, but whatever. Let's just do Cash App. So it, yeah, huge problems throughout. And I love that our entire industry just throws that out as like a one-sentence thing, but then as you just broke down, it's like actually a 12 to 18 step process with like seven or eight different people.

SPEAKER_02 17:30

Exactly. And I I think, and this might ruffle a lot of people the wrong way. I think like marketers just need to expand their scope and like start being so like so siloed in like one thing. Like it's okay if you file a payment. It's okay if you run some ads once in a while, if you're an affiliate manager. Like that's good for you. Like, don't try to delegate things because like things are just not gonna make sense, they're not gonna work. Like, if I was a marketing director or CMO, I would just start like flattening my team immediately and just start having each pod be like a lot more lean so one person could have the context between them. And that's kind of like the prerequisite before I build any automations or AI, even to the clients I work with. Usually I'll tell people like, hey, we can't even build this right now because you guys can't even explain to me how your process works. So, how are we gonna explain this to technology? Like, what does that look like? Does anyone have the process written down on one piece of paper where everyone agrees that this is the process we're doing? And I would say for 80% of the orgs I talk to, no one has that done. Even the agencies whose job it is to be good at ops do not have one place where this is the source of truth ops for our fulfillment process, which is like it blows my mind that like all the like almost all these businesses in e-commerce do not have one place, one source of truth that the organization can reference that this is what we're doing. And I think that's also one of the reasons why e-commerce is like a little bit behind in AI compared to like what's going on in SF. And for your context on what's going on in SF, people are raising hundreds of millions of dollars to build products where the customers are AI agents. So it's literally like, hey, how do we help this AI agent query this result or find this company? How do we help this AI agent gather information? There's a search application just for research so that when you use claw code, it can go and scrape social media, all the third-party resources, and it competes with Google because it's Google for your AI. And like that's how granular like people are getting an SF and they're building products for AI agents, but e-commerce is still hey, how do I use the term? Like, what is claude code? Like I use claude cowork. And I think part of it is because just a lack of ops organization and clarity.

SPEAKER_00 19:52

So if I'm a brand operator right now, all the brand operators I know are are toying with AI quite a bit. But one of the things that's been interesting just with I and I think almost everyone's universally in Claude for the most part at this point for most of their tooling. So I'm gonna ask anyone who's using a different tool for your daily for now. For those that are in Claude quite a bit and have their teams working in, I can tell you like a thought that I consider quite often is, you know, I probably have Claude running four to five hours a day alongside me. And really the only time it's not working was when I'm on calls. Granted, it's taking the information in calls and then doing things with it. I think about that downstream for the rest of our company who are using this for a lot of the things we do on a daily basis, especially on our service side. It is kind of wild that we don't have an internal role. We don't have someone actually managing the architecture of that. I can tell you, we don't have a full-time person on AI engineering. Right now, it's literally a weekly stand-up with Evan and I for 15 minutes where we just pick a new thing and then we build it over the weekends. So operationally, I know this is what you do with clankers, but from just uh an execution standpoint, how should should specifically these brand operators who are also trying to get their team on, how should they be thinking about this from a utilization standpoint? Like, would you suggest, hey, if you're an e-commerce brand and you're not doing X, Y, and Z on a day-to-day basis inside of Cloud, like you're you're you're just missing out. And here's what here's what good actually looks like. I guess that's my bottom line question here. What is like, hey, maybe I'm not at the 1% level like for some of the guys in FSF, but like what what would qualify me to get to good? Because I figure a lot of e commerce brands are like, hey, I don't need to be the best AI driven company in the world, but I I can't be average. I need to be at least in the good category. What qualifies? I would say an excuse. I have some construction going on literally right on the wall next to me. That's how big the AI boom is. They're even able to build, once again, San Fran. That's crazy. I didn't I didn't know they still build things out there.

SPEAKER_02 22:14

Yeah, there's actually a big data center underneath my house that I'm building up, and it's sucking up all the wall. No, I'm just kidding. Um 5G. Exactly. But like basically, the absolute step one is just data hygiene. I think every single entry-level person needs to have a call recorder on their calls, right? They need to be documenting their place in a project management system like ClickUp, Notion, Astana. Like there needs to be good data hygiene. And then after that, the directors need to be able to use AI themselves because they're the ones who are putting together the strategies, the skills. They're ultimately making the call on the different trends in the data. And they're still in the weeds enough where they're setting the direction and analyzing what's working, what's not working. I think when it's just the like C-suite or founders looking down, they're missing too much context because someone also has to QA all the entry-level people's work and be like, okay, here's where things break, here's where we actually need to like change this or that. And like usually the founders or the CEOs never have enough context to do that once the company gets big enough. So those are the basics. And I think that the directors and basically like pod leaders or like anyone who's kind of mid-level should be surfacing new ideas and analyzing different data into some sort of source of truth that then is put towards the founders. And like organizationally, that's where you start is like good data hygiene. We're actually checking and specifying what we want to do. Our strategy is backed by what we're documenting, and then we're surfacing that up more and more to C level to then ultimately make a call. Like that's where I'd start for good AI hygiene and to be like qualified is like good enough. And then from there, it's just like, hey, Claude, you're constantly connected via MCP into all of this and our internal processes, just come up with new ideas for us every week and let's test them. And I think that's like a good enough place to start. It's just like simple MCP connections, reading the processes we write, and then figuring out what to do next. Because like scaling a collagen brand has already been done before. Like AI knows how to do that because it can just go reference the exact path, exit valuation, and data and all the content that that collagen brand has made, and then compare it against your internal data and the efficacy of all your tests and everything you're doing. So that's where I would start. It's not that complicated. It's just getting people to do stuff correctly and put their stuff down on a piece of paper so other people can read it. Like that's pretty much it.

SPEAKER_00 24:51

Yeah. So speaking of on competitive analysis, you know, for folks in TikTok shop space, they're you either using Callo data or Fast Moss, which parses a lot of raw intel. I don't think it tracks certain threads really well. And I don't think you can conduct a true competitive analysis with just a look at that because you you can't even parse the disparity between what of what that creator drove was ad spin, was there a promo running? Like it's tough to do that. I'd seen a blog of yours on competitive analysis, specifically with Claude and your Kruva MCP. What does that look like? Because I especially see from new entrants to TikTok shop an obsession with understanding what their competitors are doing, which I personally believe is kind of overstated because like you're you don't really need to worry about competitors until you break through cold start because cold start's like hard enough. But there is something to be learned there. Yeah, talk to me about competitive analysis and maybe even like the the product side. Like, how do you see deploying that kind of understand actually what competitors are doing?

SPEAKER_02 26:04

Yeah, and I'm happy to share like an anonymized example of like a forecast based off of competitive analysis as well. But yeah, so basically there's Calodata, FastMOS, and those things, but their APIs are like thousands of thousands of dollars. So I just use like Pruva's MCP because it's included in my Ruva plan. And then they also have their own infrastructure where they've scraped and indexed all of the market data, and then you can see the daily trends across creator, product, and brand. And so essentially the next step becomes like, okay, let's find a set of competitors because one competitor on their own, like you're saying, it's hard to attribute what's happening. There's so much stuff we can't see, like ad spend promotions, like what's really causing GMV to spike, right? So find a set of like three to four different competitors, do a comps analysis across them, and look at things like GMV per video, number of active creators, creator attention, month over month GMV, and those will give you insight into how well those competitors are converting, what's really moving the needle, how their affiliate acquisition and efficiency looks like. And that's what you can use as like a baseline for your competitive analysis. And one thing to note is when you're asking AI to do this, you have to look back at like at least a three to four month window to normalize the data. A lot of times people get this process wrong because they're just looking at one moment in time, but the model is not calibrated enough to understand why GMV is spiking that much and it'll lose context. So that's that's pretty much how I do it.

SPEAKER_01 27:41

Let me show you kind of what it looks like as well. Just so and I'm going to do just show it in my browser real quick. Awesome. Yeah, I'm I'm going through a deep dive on this as well next week.

SPEAKER_02 28:01

But yeah, here's an example, like super rough, right? So I'm gonna not show these competitors to kind of the brand's already anonymized, but I don't want people to backtrack, right? It's like even just something as simple as this, where we're actually going in to Kruva's MCP. And for those who aren't familiar with MCP, it's like a claude connector. So it's basically something in Claude that I enable that allows me to read and write data between Kruva, like hey, tell me sales for the last month, or send an automation in Kruva to these creators, right? Something like that. And then an API allows me to pass data in between the two programs. And so if I want to calculate month over-month GMV growth or do some sort of crazy trend analysis or math statistical analysis, I could use an API from Kruva to do this. So let's say we're I'm making this up. Like this is pretty boilerplate, right? But let's say this is the exact way that we're gonna understand how our competitors are doing. GMV per video just lets us know on average how well are our creators converting people who watch their videos. So this is my proxy for conversion rate or click-through rate for any media buyers out there, right? One thing on TikTok shop that's a nuance that most people don't understand is that a select portion of creators drive almost all the revenue. So when we're looking at revenue, we also have to segment revenue by core creator. And I'm just putting this as GMV with over $100, just because when I've been on the software side and looked through the databases, usually around like 1.6% of all 4 million creators in the United States drive over $100, which is crazy. So that also is an anomaly in the data set that you need to call out and adjust for. And again, this will be different for every brand's first party data. But given that I worked at these platforms and I saw the average directly from TikTok's API, I'm gonna use that as a proxy for my model here. So we're looking at videos posted, GMV per video, we're looking at GMV per core creator. This also lets us know like, hey, how well are our community activities working? Is this creator getting enough context and shared learnings from the brand if that's going up or less? Because if you have similar products and all the creators are virtually the same at the top 1%, the differential is going to be in how well the brand supports them and gives them better data and inspiration. So that's what I use to measure that. And then how is this number for revenue driving creators changing as well? That's also a proxy of like how successful our community management and affiliate management tactics are. So what we do is we compare this against four to five other competitors across a time range, and that allows us to understand how well we're doing or where there's room for improvement. And you know, for this brand, the GMV is just a lot lower. But the GMV per video is actually better than this brand, this brand, and this brand, but it ranks like third place, so there's still room for improvement. Videos posted is also the lowest. So I think this is also a phenomenon of like if we get more videos posted, which will increase more GMV, GMV per video will also go down, given the fact that 1.6% of creators also make only over $100, right? So this is all a ton of context that I'm just talking you through right now because that's baked in the back end of my Claude skill here. And what that allows us to do is then project revenue. And we have all of our model assumptions here with different commissions, AOV. We're actually going back and pulling how our active creators are rising over time, what our GMV per active creator looks like over the last three months. And then we have a scenario analysis because as much as I'm feeding so much context into this forecast, TikTok shop GMV is still super hard to predict. So I always ask it to do a base worst and a best case scenario with different ways that GMV could go up. And this is the area that I was talking about earlier. If your community manager or your affiliate manager is not telling your team that is putting together the forecasts what strategies we are trying next, you cannot forecast into the future because the model has no idea what you are going to do. So that's where these come into play, and you can see we have different forecasts going in next month based off of all these inputs over there. So I'm gonna pause here and stop sharing and hand it back to you for a sec, Jake. That is awesome.

SPEAKER_00 32:37

I've got a ton of mental notes. I'm gonna be applying a lot of that. No, I mean, this is brilliant and allows you to operate in this ecosystem with a level of connected sophistication. So much of the early days in this platform have been the consumer journey and the actual execution of what we do is so connected, but you need independent personnel kind of running each arm. Even on the creator side, you sometimes have to have a different person managing like our top creator relationships against like new affiliates because it's it's such a different kind of world. We've even separated that out. Ads is all kind of responsive to the affiliate and the creator side and how they even pace ROI target, daily spend against your monthly budget, how they set up their ad groups, like all of that is directly correlated and following what you do on the creator side, which is connected to the promos, probably done by a different person on your shop ops team who also needs to track to see, hey, that promo led to a lot of sales, but we had a lot of returns with it, or this is what's happening with the conversion rate. And it's just, I think for folks outside of TikTok shop world, they don't have a tendency to look at all of those things together because they they can survive siloed out in other channels. But to the point of your projection model, like it matters drastically what our sampling threshold is over a certain period of time. And there's so many outlier scenarios that that can come to life that then change the projection model. And even we found like, hey, if you want to increase ROAS, it's not as simple as just getting better creative. You sometimes also need to fiddle with the budget, run an aggressive promo to stabilize a ROAS at, you know, a 3.2, and then hope that once you turn off that promo, you can kind of hold to it because you've proven creatives there. There's just like so many secondary things that are connected, and it's such a fight for top-line revenue and trying to maintain some sense of profitability, breakeven, or at least understanding our CAC to LTV ratio. Maybe what I find most interesting too is your data there's only look at TikTok shop, which doesn't even connect in the bigger picture, which adds like, you know, a multiple of complexity to all of this. And that's been, you know, an obsession of mine is I'm starting to see the more traditional CAC to LTV model actually played out in TikTok shop, which is becoming necessary. I'm starting to see bigger brands underwrite value and revenue they drive in Amazon and Shopify, not just calling it the Halo effect, but actually underwriting that to underwrite investments back into the creator side and into the ad side. So all of what you're mentioning here is incredibly valuable because if that data isn't prevalent, and I guess the entire org needs to be in the loop with that data because all those independent decisions have to be working off of the root model. They can't be siloed out.

SPEAKER_02 35:50

Yep, exactly. And the craziest part is like, dude, this is just a skill I can give you guys and transplant, and you guys could make it your own. And I think this is where agencies and you know bigger orgs that with like polished org structures with, right? Because like let's say return on creators takes this and they're like, okay, now we have 25 brands, or like you guys probably have hundreds of brands with historical data and patterns on how revenue moves across your strategies. What if you calibrate your forecast to those patterns? And then what if you segment that down to just having beauty forecasts and health forecasts, right? You guys will have so much more of an accurate forecast than I created here because you can be like, by the way, trends tend to move X percent of the time from beauty brands investing at least 50k a month into samples and ad spend doing this, and then use that as an input, and then you put that into your growth assumptions for next month, and boom, you're great. And then same for Shopify and Amazon. You can be like, hey, here's our average spike in branded search with a conversion rate during this period across this many years of data. This is our growth assumption for our model. And then on Shopify, these are the average number of subscription purchases that happen after someone tries our TikTok shop trial exclusive bundle, and then this is what the customer demographic looks like. And we're activating X number of creators with a persona that appealed to that demographic during this period where people purchase on Shopify. And like, I'll never have that data in context, and neither will any agency, but only the internal brand teams will have that context, and if they can just connect it to their forecasts and other AI solutions, they will win. They will win in the age of AI, but it no one will do it. That is the crazy part. No one will do this.

SPEAKER_00 37:35

Well, this is brilliant. I've now have my plans for the weekend and probably blocking off some days next week. Last but not least, you mentioned you're hosting something next week or presenting something next week, or tell the audience here about that. I'll I'll link below your site and LinkedIn profile uh so they know how to hunt you down as well.

SPEAKER_02 37:57

Yeah, that would be great. Not to promote a webinar on a podcast, but you know, doing a webinar next week with the Kruva team where I'm essentially going through how I built that entire automated PL from scratch. I'm gonna share the skill file, I'm gonna share a recording of the deck, and the Kruva team is also giving free API access for one month so you can build this and try it out for yourself. And then I'm also gonna be giving a free AI consultation to people who R SVP and are interested. So just a ton of value, and all this is being given away for free, which is kind of dumb because I just started my own business, so I should charge for it. But you know, gotta do what you gotta do. So doing that next Monday, going live 8 30 a.m. PST or 11 30 a.m. EST with the crew of guys on like this.

SPEAKER_00 38:47

So and always uh always a pleasure. Appreciate you taking the time.

SPEAKER_02 38:51

Thanks for having me.