The AI Stack Our Team Uses to Manage the Top Brands on TikTok Shop

You're (probably) using AI wrong.

If you're using AI to write content scripts, to reach out to creators, or to design serious marketing strategies... you're using AI to replace the one thing it should be optimizing for:

Human Thinking. 

This week I shared the lean AI Stack our team uses across some of the best performing consumer brands on the planet, and our fundamental philosophy that has led here.

Transcript

SPEAKER_00 0:05

What's up, party people? Welcome to a Jake breakdown. I don't know what we're gonna call these things. I might be live right now. I'm not certain. Probably biggest news. See this wedding ring. I'm married now. So it's time to take long form content seriously. That's what we're gonna do here. I'm gonna try my best to actually look at the camera, but I'm gonna start sharing more and more of our actual insider playbooks, document the things we do internally to run a lot of the top brands on TikTok shop, in TikTok, all those on the leading edge of influencer marketing. Today is gonna be focused on the specific AI stack that we use to run all of our best performing and some of the top brands on TikTok shop. I'm gonna get into some fundamental things here. I'm not gonna prescribe exactly how to set this up. I'm non-technical. Really, you should ultimately set this up for what's gonna be best within your capabilities and YouTube university and other things. Anyway, I'm gonna get into that now. So without further ado, let's just, what are we doing? Let's dive into this. All right, so here's the breakdown. Hopefully, you can still see me on the screen. I'm recording this from Riverside. I have no idea what I'm doing. I'm officially unk in the content game, even though that's the world I come from. It's funny how quickly the world changes. So before we start, I want to add some context here. This is what we're using on a daily basis. There's probably another 10 to 15 to 20 things we're doing behind the scenes that some of our teams are doing that's maybe more specific to running an agency business. Obviously, we have a seven-person now engineering team, and we have our own AI ad tech that we use for all of our brand partners. So I'm not gonna hyperfixate or focus too much on all of that. I'm gonna focus today on the things that are very easy for any consumer brand, regardless of budget, to implement day one and to do so from a non-technical perspective. I'm also not gonna prescribe any specific tools here, softwares, data, any models. Ultimately, I'm gonna show you directionally what you should try to build, and then you're gonna have to figure out the best way to implement that, but all is gonna be easy. And that's what I'll just say is probably one of the problems of AI. A lot of people look to like tools and resources first instead of thinking about what's actually the most valuable thing that they could build, and then going and setting up the infrastructure for that with tooling. There's too much focus on tooling, not enough thinking. Speaking of, that brings me to the first point here. What I see far too often is brands today using AI to write captions, to create videos, to come up with content scripts, to come up with strategies, and essentially to replace thinking. That is not the end game of AI, and that is not where AI can be most valuable in your business. It should not be replacing thinking. Thinking is the thing that you and core operators and team and personnel should be doing, and you should be optimizing everything around that. So there's a general philosophy here. I want you to think about ironically, how can you use AI not to replace thinking, but to enable and to optimize your time spent thinking? Everyone's goal in the AI era should be optimizing the time on your calendar that you get to think, not as a means of replacing what is actually going to separate you from others and make you great. If we throw all of our thinking to AI, you're gonna create a business just like everybody else, given enough time. So that's how we fundamentally theorize and hypothesize with AI. I don't like using AI for AI content generation or captions or doing very human creative things. I like using AI for everything else and letting humans do what humans do best. So this is fundamentally where we believe AI actually wins inside of organizations today. This is for giant publicly held consumer brands we work with, as well as very lean operations when we advise them on what they can implement from our work internally. The first is large data analysis. Uh, humans, we max out at any sort of large data set. We struggle understanding per capita. We struggle understanding percentages. We we struggle with very large math and big numbers. And that's a particular challenge in marketing to just build mental models for things. AI allows you to scale, frankly, math and data analysis without issue. That's really the first and maybe most critical component here. The second is redundant task elimination. This is more obvious. Any repetitive task or decision, things you have to do that are kind of a waste of time that aren't optimizing for your thinking and strategic capacity, AI should work to eliminate that, frankly, and replace it entirely. Three, and this is a very common missed one, but perhaps in the long term, the most valuable is spotting blind spots. Now, one of the challenges is all of these models will often tell you what you want to hear. So there's some things you need to build into that. Other folks have done more long-form content about that. I'm not the expert. You should definitely install all of that. But AI has no ego or emotional bias with these things. You should be using it to spot blind spots and generate information that you're not thinking of. For last but not least, is speed of decision making. I didn't include this one in there, but like we have an executive coach, a trained agent we have internally that's an executive coach that understands all of our long-term objectives and milestones as a business and helps myself and core leadership make decisions more quickly so we don't have to procrastinate on those. Really, this is a game of speed. Decisions hold people back more than execution, I'm a believer in. So speed of decision making is a fourth lane here. All right, let's get into what this actually is. So here's the stack. There's a ton more you could and probably in the near future should be doing with AI. But here are what I believe to be the core and most important things. The first is what we call creator scoring. So every brand out there, especially in TikTok shop, you're working with affiliates, you're working with creators, you're working with paid influencers, you're working with the UC creators, microinfluencers. There's so many different capacities in which you can be working with folks. However, this is a large data analysis problem. You can use any platform on you on Earth, but you're dealing with thousands, tens of thousands, hundreds of thousands, millions of creators who are organized and you just have broad data sets. Trying to do rank order systems or just filtering on that is really minimizing what you can do from a large data analysis model. Now, we have an internal ad tech that we created within ROC that does this at a very insane level that all of our team uses. But there is a way, there is a V1 to what we built before we had a seven-person engineering team. And really, what you can do is design your own scoring model. So give context to your system about what your brand is, what your key messaging is, all those other things, what you look for in a good creator. Do you value sales versus average views per video versus category fit versus engagement rate versus demographic fit? How do you value all of those on a logarithmic scale? Run that into a system, and you can essentially then upload any spreadsheet of creators. It could be 100, 500, 1000 creators and all of their metrics. Pull that from in TikTok shop. There's Caladata and other large resources. There's platform solutions to this. Doesn't matter what you're using. Load that in and have it build a scoring model that's tailored to how you make decisions. So you don't have to do a filter system and look at all of these creators and look at a large repo of data that you're probably not going to fully understand. Have it run a very simple scoring model. This is not that difficult to build whatsoever. Once you have this, you can ultimately define that criteria, and it's going to help you find your ideal creators in the same way you think about finding your ideal customers. So this is the first thing. Any brand that's running TikTok shop affiliates, you're running paid influencer campaigns, you're running UGC campaigns, this is probably the most valuable thing you can build to find more of the right creators. Understand that it's about finding the right creators. It's not about finding just some creators, it's about finding the right creators. So that's the first thing that we use with our own internal system, but you could apply, stop browsing, start scoring. Okay, the second, very connected to this and any active creator campaigns. Everyone hyperfixates on proven content hooks, proven CTAs. They go to look at their competitors on the ads dashboard on the back end or on Caladata to look at their top creators, and they mirror those content scripts one-to-one. And then what they do is they build content briefs, either a document, I've seen Notion pages, I've seen decks, it doesn't matter what it is, and they send a content brief and a content resource guide out to all of those creators. That is totally underutilizing the potential and the reason why you should be working with creators. You should be taking and building a content brief in a rough structure that works for you, and then tailoring that to every single creator you work with so it resonates the most with their audience and is the most tailored and distinguished from all of your other assets. If you're not doing that, you're essentially just looking at creators as creative resources, which is a very expensive way to get creative in today's era. You're not maximizing the value of actually working with an influencer, someone who has influence. They can't be influencers if they are parroting the same video that hundreds of other people are saying and doing the exact same thing. Use them for their influence. So, very simple way to do this. We call this contextual content briefing. Have your system, once you've selected the right creators, have it analyze some of that creator's videos. It could be their top five videos of all time, ranked by sales or views, could be their last 10 videos. That's a very simple way to do it. Have it analyze that with the context of your content brief you created with a brand and have it tailor a specific content brief to that creator. So it opens up with their same hook structure. If they're a creator with a specific content format, it fits your product and brand into that content format where it's actually going to show up and be a very organic and a unique way, not them doing an unboxing video on a page that does daily vlogs. That doesn't make any sense. That's not showing up in the feed or resonating with anyone. You design a very hyper-fixated message around them using their same language, their tonality, their style, tailor content briefs in a contextual way to every single creator you're working with. This is actually probably the most simple to run. It probably runs the most compute though, because you're analyzing videos. But we've done an experiment on this with every single AI model. It wasn't all that expensive. And you're going to see the impact of this almost instantly. You're going to get a huge spread of unique ad creatives, and all your content is going to perform organically way better, and it probably will in ads as well. Okay, speaking of ads, the third thing here is ads analysis. Your ad data has the answer for what you should be doing next. But again, this is a large data problem. For brands we work with in TikTok shop, we have sometimes at times one, two, or three thousand videos loaded in GMV Max. They're spin going to maybe 100, 200, 300 of those videos. If you're in meta, the complexity is way more complicated. You need to be using data to synthesize and analyze what is actually working from a data perspective and from a creative perspective. So your standard analysis, if you're just looking at the top winners, humans cannot analyze more than call it 50 or 100 rows within a spreadsheet, especially with all the data that's given by these ad platforms. And this is where when I talk about spotting blind spots, AI is going to do a great job of spotting blind spots here that you're probably not seeing. You need to build some sort of analysis. This could be a project in Claude, could be a project you train up on this specifically. There's probably platforms for this already out there. I'm not saying have it analyze your ads like the creative. It needs to analyze the creative and the actual ads data. The answer to what next ad you should be working with is not as simple as saying this worked as an ad, this is a strong row ads, duplicate this. There's so much you're not seeing there in their message, in their CTA, and how that creator filmed that video and the style. It's going to pick up on trends and analysis that a human is going to miss. And that's what's going to uncover what you should do next from a creative and ad perspective, and even sometimes ad grouping. Because look, AI understands data way better than humans do. So you need to have a project dedicated to ads analysis. And it can't just be the creative side, it has to be the quantitative side as well. And it can't be as simple as saying, look at our top ads, reverse engineer that into our next content brief. That is too simple of a thing. Use AI for what it is. It's smart, it's intelligent, it's artificial fucking intelligence, right? Use it to map to things that you're not seeing. That's where you're going to get the most value on the ad side. Okay, the fourth we call trend mapping. This is probably the most difficult of the bunch. But if you hooked it into the three former, then building this off of the data that's generated from there can be a lot easier. Essentially, the theory here is there's always emerging opportunities. And I'm gonna get into time to live in a sec, which is something we obsess over. There's always emerging opportunities, but it's about your ability to spot those. So far too often, what I'll see is someone, a brand, go look at their competitors, see what trend they're hopping on to, and then trying to replicate and hop onto that same trend. By the time it's showing up in the ads dashboard, it's too fucking late. It's too late. You have to catch these things far quicker. And the answer is not going to show up in the ads dashboard. The answer is gonna be somewhere else. It's gonna be in organic content, it's gonna be in the comment section, it's gonna be on early conversion data at a very low level before something goes quote unquote viral. Virality happens before you ever see it go viral. That's something very critical to understand. By the time it's viral, it's no longer replicatable. So you need to think about what are the emerging trends, what is the emerging alpha we can capture? Because no different than a stock, by the time the stock is hot, it's too late. It's too late by that time. So you need to reverse engineer correlation components with AI to correlate some of those signals to think about how you jump into it. So that's the problem, is everybody's trying to chase these trends and tap into virality and not look at what actually creates virality. So you need to be looking at actual creator-consumer behavior, category momentum, comment section before it reaches its peak. By the time it's reached its peak, it's gone. So you really need to stop wasting your time looking at products doing a million a month or looking at videos that are at five million views now and $500,000 in sales. By the time it reaches that, there's no alpha to be captured there. You got to spot closer inflection points, closer to zero than closer to their peak, if you will. Okay, last but not least, this is an operational one that even non-consumer brands need to implement. There's a concept with domains called TTL, time to live. How long does your code sit before it goes live? What's the query window? I'm non-technical, I don't know. You need to optimize your organization's TTL. How quickly do you make decisions with your team in calls, in Slack, in conversations, in documents? How long does it take for you to synthesize that, implement it, and prioritize those tasks out to get executed? Because if your typical window is say it takes you four days to come up with an idea and fully execute on that idea, compared to a company that can do so in 48 hours, just two days, you will have be able to run half as many experiments as they do. Now, this seems very stupid. Why should that matter? The more experiments and new things you can launch, generally the better. Even if it's a bad idea, learning is learning, insight is insights. So if you aren't already, you need to implement an ops and a TTL optimization window. What we do here is we have all of our calls internally and with clients recorded. All of our Slack is hooked into our internal organizational cloud database as well as our Google Notebooks LLM. All those meetings get queried out. They don't generate a task queue off of it because if you do that, you're going to get very stupid tasks. They go through an AI synthesis, they go through our internal systems we've created first, then they get prioritized in a task queue, which they go through a prioritization window. And then we even have a taskbot hooked into our Slack. So in addition to this getting assigned to our team and our project managers management system, which we use Notion right now on a per client basis, we also have Taskbot sends our team and personnel, it organizes it, finds the right person who needs to do it, sends them a task, they get it done. They tell Slack it's done, it marks it done, and it lets the other team members know. So if it's a Gantt chart model where you have to do something first in order to get the next thing done, it unlocks that process. So communication doesn't delay anything. So this is a more simple one, but you have to implement this and you have to optimize for your organization's TTL because it's going to optimize the more experiments and tests and things you can ultimately get done. Okay, last but not least, these are the five we run. And if you run these in coordination, they all work to actually support one another. So that's the benefit of building all these systems, especially if you build them in one channel as opposed to spread across a bunch of softwares, is they actually begin to work together because you could assign a task from a client call. And if it knows to run creator scoring, for example, inside of rock, that was one of the tasks. It'll just like go ahead and start doing it. And it'll do the contraband, it'll do the added analysis, it'll apply that thinking almost instantaneously in a really unique way. So again, you can optimize in your organization the time you spend thinking, not the time you spend executing minor tasks. Okay, last but not least, I'm gonna close out with this. This is just our stack, and you don't need our stack. Some of this may be a good fit, some of this may not. The principles and the models here are universal. So you can apply this in ways that are best for your organization. I think our general rule of thumb about thinking is probably the most important thing you can pull from this. And last but not least, urgency. You have to start doing this right now. This is what all the great consumer brand organizations are gonna do, and bigger companies are gonna figure this out sooner or later. Time to move on to all of this is now, but ultimately you gotta design the stack that works best for you. And I hope this helps in doing so. Okay, let's put me front and center to close this out. Thank you for tuning in. If you watched all of this, that's fucking awesome. I'm sure only three to four people actually will. I'm sure they've given up and started implementing this right away. Either comment or DM me on any social channel, Jake Jorseth. There's like no other Jake Jorseth, really, because my last name. DM me on any social channel with what other breakdowns you think would be valuable. Or if you want to like see how this actually looks internally for us, I'm happy to show you some of the nuts and bolts. I'll see you next time. First of many.