How to Sell on TikTok Shop With AI

Matt Edmundson: Welcome to the eCommerce Podcast. My name is Matt Edmundson and it is great to be with you broadcasting, I would say live from Liverpool. I mean, we're having a conversation live, obviously, but you're not hearing this live. But we are broadcasting from our new studio here in Liverpool at our warehouse facility. And so we're still trying to get it dialed in. It's still very messy. If you listen to the last show where I talked about that. Um, but you know, we are getting there slowly but surely. Uh, and it's great to be in the new studio. Uh, same format of the show though. The studio's new. Uh, and the reason we're in a new studio is literally because I moved house like 2 weeks ago. So, um, and I've not had a chance to build a new studio at my new house. So we're using the one we have at our Liverpool warehouse. It's really mad when I think about it, because we've got 3 studios, or we did have. Um, not that you need to know any of this. Um, but we did have 3 studios, and we'll have 3 studios again soon once I get the new house. So I'll have one at the house, we've got one here at our Liverpool, uh, warehouse, and we've got one, uh, in Nottingham. So where we record the So That's Why podcast. Bottom line is you should have a whole bunch of podcast studios, just the way— just the way it is. Just my top tip. Anyway, enough of all that. We, we're not talking about podcasting, we are talking about ecommerce. And so if you're new to the show, a very warm welcome to you. Make sure you like and Subscribe and do all of that good stuff. Like you, I run my own ecommerce business, and I have the great privilege of speaking to guests every week, picking their brains about how I can run my own business better. And you get to listen on to the conversation. So hopefully, you'll become a regular listener. And if you are a regular listener, very warm welcome back. Great to have you join us again. Now, Before we get into the conversation, a little bit of housekeeping. If you haven't done so already, go to the website ecommercepodcast.net. There's a bunch of stuff there that is going to help you if you haven't already signed up. One is the newsletter. We just email out every week the notes from the podcast. So if you don't get a chance to listen to it every week or whatever, at least the notes come to your inbox, which are always great. On those notes is an AI prompt. So if you subscribe to our AI services, you'll get all the transcripts and stuff straight into that as well. You can start to grill the podcast through AI, which is very, very crazy. We've also got on there the cohort group. So if you want to come into like a monthly call with fellow e-commercers from around the world, just talking about e-commerce and thinking about ecommerce and asking questions about ecommerce and sharing insights, um, come do that. They're free, they're on Zoom, um, last about an hour or 2. And I just honestly, one of my favorite times of the month is when I get on these calls, uh, with you guys and just, we all just share what's going on. It's wonderful. And then finally on the website, uh, is our AI system. If you haven't tried that already, go check that out. Anyway, enough of all of that. Let's get on to the show. Welcome to the show, man. How you doing?

Sohun Sanka: Doing great, man. Doing great. Excited to kick off the day with a lovely podcast.

Matt Edmundson: I love how for you, you're kicking off the day and I'm ending mine. Uh, yeah, such is the time zone difference. Whereabouts in the world are you?

Sohun Sanka: Yeah, I'm actually in an Airbnb in Los Angeles right now, traveling, visiting a bunch of people, doing some more podcasts, not cheating on you. Very excited about this, you know. Yeah, yeah, you're totally cheating on us. That's Yeah, just out there. We have an in-person studio co-hosting with another company called Kruva, hosting other people there, but excited for this one.

Matt Edmundson: Fantastic. Of course you are. Of course you are. So for those that don't know you or haven't heard of you because they've not listened to any of the podcasts you're cheating on us with, just give us the brief elevator pitch, who you are, what you do, and why you're here. Yeah.

Sohun Sanka: So, my name's Sohun. I'm the founder of a company called Clankers, a name that took a lot of creativity and time to figure out. And my background is in the influencer affiliate space. Worked at a traditional marketing agency doing media buying, influencer affiliate from Amazon, off-platform to Amazon using traditional tools like Impact, Levanza, Refersion, those guys. And then started to get into the creator commerce space and built up a TikTok Shop division there. Worked at some of the legacy TikTok Shop outreach tools as head of growth employee number one, like Euka and Reacher.

Matt Edmundson: Yeah.

Sohun Sanka: Now working closely with a software called Kruva. And what I saw was that with Claude Code coming out, the way these softwares were going to change, releasing their own public APIs and MCPs, we'll dive into what that means later. Allowing all the marketers to essentially build their own hyper-personalized software and AI workflows in-house. So that's what I'm helping a lot of marketers, brands, agencies do is take these tools around TikTok Shop, influencer affiliate, and learn how to use them productively to make their marketing processes more efficient and impactful. So I'll usually go in, I'll audit an entire TikTok Shop business, I'll go and build custom software and actually put it into a nice GitHub repo and, you know, kind of embed it into their organization. And then I'll teach them how to use that and become more AI-enabled so their business can grow.

Matt Edmundson: Oh, wow. That sounds fun.

Sohun Sanka: Yeah, it's pretty fun. It's pretty up my skill level, you know, younger, uh, 25-year-old just doing Claude and TikTok. So, you know, I can't complain.

Matt Edmundson: It's just funny, isn't it, where the world's got to? It's like, I have kids. um, similar ages to you. And yeah, it's like a few years ago, if they'd have said to me, oh, Dad, my career, I'm basically, I'm just gonna do this thing called TikTok. It's basically a short-form 30-second video thing where people dance, and then I'm gonna mix it with some kind of AI thing. I'd have been like, you are smoking dope, dude. Um, but it turns out actually you're not.

Sohun Sanka: Yeah, I'm not. There's a real business here. You know, it's surprising. Like, I could actually make a living doing this stuff that sounds like complete nonsense to 90% of the world. But even considered for SF, I'm considered an older guy. These 18-year-olds are raising— it's crazy. These SF entrepreneurs will call me Unc, and they're raising $10 million.

Matt Edmundson: Oh, wow.

Sohun Sanka: They're raising $10 million for their business where they're selling to other AI agents. So where I come from, it's just AI agents all day. And I want to take the relevant portion of that and apply it to ecommerce because it's just such a different world. than building emails for AI agents, building landing pages for AI agents. Like, the AI agents are the customers for these businesses. So long way to go in ecommerce. I feel like we're just scratching the surface on AI.

Matt Edmundson: Yeah, I totally agree. Um, so when it comes to— then you're obviously merging AI, you're merging, um, TikTok together. Um, what's the I mean, one, that's pretty niche, but one of the questions I like to ask guests right straight at the start is you've obviously, you're working with e-com companies both in the TikTok space and the AI space. What's the biggest mistake we're all making as far as you can tell and see?

Sohun Sanka: Yeah, the first mistake is really that no one has their processes written down very clearly. So one of the big things when you go from marketing to tech, is marketers sometimes are very ambiguous. They like to talk into like, oh yeah, we're going to increase the efficiency or we're going to do this thing or we're going to test new angles. But it's not like when you go into a technical standpoint, you have to be very specific about the technical, like the angle you're testing plus 3 specific fallbacks plus an example of what good looks like.

Matt Edmundson: Yeah.

Sohun Sanka: And I think a lot of these brands just don't have their processes ironed out and their data even cleaned up. so that AI can learn about their business. And they get very frustrated when they try to plug in things like Claude Code or Codex and they're like, ah, it came garbage. I'm like, dude, you don't even have a folder on what good looks like, your brand colors and stuff. You're not feeding it anything. And so that's the biggest mistake that prevents people from using it in their first go.

Matt Edmundson: That's interesting, what good looks like. Do you have to tell AI what good looks like?

Sohun Sanka: 100%. Otherwise you get what we call AI slop, you know? And I think AI slop is just, It's optimizing for the best usage of its own tokens and what it's been kind of trained on. I mean, you'll see it a lot, especially when you build landing pages. When you see those, you've seen those Claude Code landing pages. Yeah, I've seen boxes, neon colors. They don't even have a proper blog. They're missing multiple headers. The copy is just awful. It's like these AI I don't even know what to call them, but they're like these phrases where it's like, we just killed this, or like, we're fixing this brokenly. It's like these bold claim verbose language that some engineer trained this model on that sounds like nothing like marketing copy. So you definitely need to show it what good looks like. The way I look at it is like when you bring an employee on, people give the employee 90 days to ramp up. with a very meticulous onboarding guide, multiple calls a week, shadowing it. Like, think about how much context a human gets when they're like signing on for a new job, and then they give nothing to AI. They're just like, hey, here's like 2 sentences, do the exact same thing that this guy that had 90 days of onboarding and holding and context in like 1 second. And because you didn't do that in 1 second, Yeah. AI isn't useful. That's what I see a lot of people who are like, I tried AI, it doesn't work.

Matt Edmundson: Yeah. That's really interesting, isn't it? And I guess by answering the question, what does good look like? You're starting to form your own processes because I would venture to say that the reason we don't tell AI what good looks like is we're not entirely sure what good looks like most of the time. Like you say, we use a lot of ambiguous And fluff, a lot of ambiguity and fluff. Excuse me. I think there is, which is ironic, isn't it? In an industry which is so data-driven.

Sohun Sanka: Yeah.

Matt Edmundson: What's your ROAS? What's your CAC? You know, we throw out all these acronyms because we don't actually know what they mean, but we, you know, we ask all these kind of questions because those are important numbers. But actually, what does good look like when it comes to writing? a blog post when it comes to writing product copy? I mean, that's a really deep question straight away. Yeah.

Sohun Sanka: And I think it makes a lot of people scared. I'll say like one of the things that, and I was listening to this podcast with Naval. I don't know if you listen to Naval at all, but he is like, you know, great writer on X, great, you know, tech voice in the space, you know, and he was saying that when people start to like get into complex technology, when they try to talk to it and it makes errors, it makes them feel stupid because they're not getting it right. Right. So there's this innate fear about like, hey, I need to type in very specific commands into AI, and when it doesn't come out right, does that reflect on me or my ability to be specific or give direction? And I really think that's the core skill set behind being able to be a good AI agent handler or AI expert. Is like, how articulate can you be about what you want and what good looks like? And people get scared trying to use this. I find that the reason those who haven't started playing around with it already is because they don't know what they want to do. And when I mean know, I mean bring it all the way down to first principles and explain it to a 5th grader. Like, that's what it means to like truly know something, to be able to explain it in like 10 different ways to 10 different people. Not just to a marketer on like, hey, you know, I'm going to optimize my blended cocktail TV doing blah blah blah. And like, you know, that only resonates with like one very segment of like high-tier marketers, you know. Like, that doesn't like resonate with an AI model that hasn't been trained on all that stuff, you know.

Matt Edmundson: Yeah, yeah, no, that's a fair comment. So how do we, how do we know what good looks like in relation to TikTok? So what are some of the things that we should be Because obviously you're doing this now as an uncle, as a job. What are the sort of the first steps we do to try and understand what good looks like, especially in relation to TikTok? How are you helping companies figure that out? Yeah.

Sohun Sanka: So for me, I've seen so many programs, probably audited like 2,000 affiliate programs at this point. Because I worked in software, I got to see the backend of like over 150 agencies and like, you know, 1,000+ brands, right? I saw every single different creative brief, every message, and I have my own sense on like what the data looks like. Another thing you can do if you want to get more data-driven is take a comparable set of brands in the space and do like a comparables analysis, comps analysis. These tools like, you know, Kruva, they're essentially scraping and indexing every single video, all the KPIs of, and all the products of everything consistently on a daily basis. So you can actually go and pull their performance of like maybe 5 brands and similar GMV and products and be like, hey, here's how I'm pacing month over month against that. And that could be what good looks like. That's a good proxy. The best proxy for marketing agencies is you have first-party data on like 20 to 25 TikTok Shop brands. So you can start benchmarking a good, like, hey, what's a good response rate look like when I launch the beauty brand that does $10,000 a month on TikTok Shop? What's a good sample request rate? How often are my affiliates who post for the first time converting into sales? And compare this rate against the 15 other brands that I'm seeing right now.

Matt Edmundson: Yeah.

Sohun Sanka: They can get very clear On where, what good looks like. And I think that's the advantage that agencies do have with AI that brands really don't.

Matt Edmundson: That's really interesting. Yeah. And so sites like, what was the site called? Kroover? Kroover. Yeah. Okay. Um, so transpose slightly, uh, because I want to get into how you mix AI and TikTok together. What are the big mistakes we're making with TikTok Shop per se right now, or maybe even big mistakes? Like, for those that maybe still haven't started on TikTok Shop, where do we begin?

Sohun Sanka: Yeah. For me, I really look at, and this is outside of these TikTok Shop tools, right? The first thing I look at is what's going on with your Meta ads angles, what's going on with your traditional influencer marketing. If you have a tool like Motion Analytics, I would start going in and reviewing the copy off the highest performing landing pages. Even going back to traditional affiliate, you're buying traffic and you're trying to find the highest EPC, right? And so, you can look at landing pages with very high EPC. You can take the core copy from that and you can put that into a creative testing strategy to give to your TikTok affiliates. And that doesn't require you to be a super AI-focused TikTok Shop. engineer or researcher, you just take what's working on your copy for Meta ads and landing pages and you go and put that into your first test. Maybe you're going through all your reviews on Amazon and you're like, hey, this is what customers have been saying in the last 30 days that they really like about our product. So, let's go take this copy and take those one-liners and prepare them for our affiliates. Let's go find a product. that is very demonstratable on camera that maybe affiliates are already talking about or influencers. And we can use social listening tools to go and find people mentioning products and figure out a sentiment analysis. If there's a lot of creators that are already talking about our product on TikTok, is there anything we can do to reach out to them personally and get them posting so we can really ramp up in the first stage without going through this huge sampling and review process? Those are some preliminary things you can do that aren't really AI forward and aren't really super technical on TikTok Shop.

Matt Edmundson: Yeah, and very helpful too, aren't they? Because you're narrowing down a lot of the errors people make at the start.

Sohun Sanka: Exactly.

Matt Edmundson: Which is always a beautiful thing. Save on the mistakes. Don't screw it up. So let's Let's say then we're, you know, we're doing that. We're going out, we're getting the influencers, we've got a shop set up, and we're starting to see some results. I'm curious how you are using AI in that space. So what, I guess, the start of your journey, how did you, what opportunity did you see there? What was What did AI do for us that we weren't doing for ourselves? And then you think, actually, if we use Claude to do this, you're gonna kill it.

Sohun Sanka: Yeah. I mean, part of it, just speaking for marketing agencies, is you're able to operate so much more efficiently at scale. So historically, the reason why I joined these software tools, just to back it up in the first place, is that every single brand needed to go into one of these influencer affiliate outreach tools, and then reach out to 10,000 creators a day. They would have access to the entire database of TikTok Shop creators. And then you would create highly personalized messaging flows, giving out specific creative briefs for the products you sample, reminding them to post, capturing those creators that performed and moving them into a community like Discord or Lark or WhatsApp or any of those communities, right? That was really the evergreen process. But to actually activate and set that up would take, you know, 2 to 3 hours per, assuming you knew how to do it.

Matt Edmundson: Yeah.

Sohun Sanka: But if you save that first-time process into a Claude skill and you wire it in via MCP, which is called Model Context Protocol, it's just a fancy way of saying, here's how one software talks to another to get it to do stuff. So you can save that script via MCP, and now you have someone that can automatically set up this software for you every single time. It's a clear application of AI. It's like, hey, I have 15 brands that I'm gonna onboard, I only have to set it up once manually and AI can do the other 14, right? So that's like a clear area of arbitrage. Another clear use case that I'm seeing right now is on content analysis and video research. So like the biggest thing with TikTok Shop brands is they need to give feedback to their creators and coach them. A lot of people may not know this, but TikTok affiliates are very different than UGC creators or traditional influencers. a brand portfolio and are used to creating content. They're missing the softer elements of the video, like how to set up your camera correctly, how to actually do proper lighting. And, you know, we really see that matter a lot with TikTok because when someone's scrolling really fast through their feed, if there's like a little bit of bad lighting that's off or the camera is slightly off, there's something in our brain, maybe it's some sort of like taste receptor or judgment. It's like a neuromarketing principle. Doesn't even want us to watch that video for 1 second. And you can put in as much time into scripting, creating the perfect hook, studying your meta ads angles, your highest converting landing pages, but no one's going to see that video because you don't know how to make your video appear very good in the landscape. And so these are things where you can have Gemini go and actually extract that lighting by taking— what it essentially does is It breaks a video down into a bunch of images frame by frame, and then it analyzes each video. The process is called OCR. And so you can actually have it take the first 2 seconds of your video and abstract different lighting concepts and be like, oh, look, you're using a box light here. This is natural lighting. This looks like a vanity mirror. And you can tell affiliates like, hey, the best performing lighting is when you use natural lighting and you're in your bedroom. So here's how you actually do that. And you can show them a diagram of like different lighting angles, times of day that post, and like other lighting tips and hacks. And you understand with setting up a podcast studio and running a podcast how important the lighting and videography is. It can make or break your entire video.

Matt Edmundson: Yeah.

Sohun Sanka: But you don't see people index on it with that much depth in like traditional. And a lot of creators who are just starting, like You know, the barrier is low. You just have to have 1,000 followers and you can have a TikTok Shop. I feel like there's about 4.5 million of them in the United States, but really only 1.6% are making a couple hundred dollars in the last 30 days. So there's a huge education gap to bridge, and AI can help break that down specifically.

Matt Edmundson: That's really interesting. The fact that AI is now analyzing your videos is quite fascinating. Um, and did you say you were using Gemini to do that?

Sohun Sanka: Yeah, Gemini. Um, you could use like 3.5 Pro. Uh, I think they, uh, sunsetted the 2.5 Flash, which is what I used to use. And like the run is like a couple cents per run. Um, but what's great about these like outreach tools, uh, with Kruva specifically, they pass through a full transcript analysis, hooks and selling points and lighting on the video already. So when you're thinking about your costs and doing this across 1,000 videos, you can actually pull their pre-ran analysis and reduce your Gemini cost by about 3 to 5x because you're only analyzing the last 8th of the video to have it analyzed with your proprietary framework. Because the major cost by analyzing a video with AI is generating the transcript, going through frame by frame to find the hook, classify the hook. Hook equals first 2 seconds. There's a visual hook, emotional hook, there's a verbal hook, right? all those processes with individual runs and tokens associated already done through this platform where you have really talented software engineers optimizing those costs versus if you try to go run Gemini on a fresh video and generate all that from scratch, it's, it's just going to cost you a lot in the long run. It's going to make you not want to do this process.

Matt Edmundson: Yeah, that's really interesting. And are they doing it just TikTok or are you getting Instagram, YouTube Shorts, the whole 9 yards now?

Sohun Sanka: Just TikTok Shop for now. Um, you know, they'll probably expand, but the reason TikTok Shop is so interesting for me is because you can see directly attributed sales versus like with these other platforms, you have to use like earned media value, which is like, you know, I think it's kind of BS. Um, and you're essentially like, you know, using these indexes that platforms like Aspire and other people put together where it's like, oh, a view on TikTok is worth 1 cent, you know, but an engagement on Instagram is worth 5 cents of earned media value, and I'm doing the sum across this index, and, you know, I'm using my own CPM on Meta ads as like a benchmark, and I'm like, it doesn't really tell me how well this influencer performed. Like, it just doesn't. It doesn't tell me that this hook is the one that made me the most money, you know? And even on TikTok Shop, I don't know if you know this, but in GMV Max, which is the TikTok Ads Manager, TikTok Shop Ads Manager, you can actually pull the 2-second video view rate and the 6-second video view rate. So you're getting really granular on like, my hook rate or 2-second video view rate is 33%. Let me go and find all the creatives with a 33% plus hook rate from this week and then have Gemini go in and analyze every single aspect of the first 2 seconds of the video and then start to repurpose it into my creative testing pipeline. And think about how much leverage that unlocks for marketers across all their creative. If you take that information back to Meta ads or splice out those hooks and then run them into there, that could reduce your CAC a lot on Meta. You could even test that on Amazon too. We see a lot of people, and this was more of an experiment I was doing in my earlier marketing days at the agency, where we would take these UGC from creative and we would splice them into sponsored display on Amazon and they would outperform. Granted, we'd still have some spliced in A+ content into that video and make it a little polished. But when you like hybridize it a bit more with like a very data-driven approach on like these creators have a higher 2-second video view rate, or these have a higher hold rate, you're able to just get some better arbitrage than that traditional glossy creative.

Matt Edmundson: That's— I can just hear a load of the listeners going, hang on, what? It's just like, You want me to do what? I guess it's, and this is interesting, isn't it? I don't know if you listen to the Modern Wisdom podcast. And Chris was on there talking with some guys about what does AI look like? I think it was in 6 years or something. What does the future look like in 6 years?

Sohun Sanka: Yeah.

Matt Edmundson: And one of the comments from that podcast, which I thought was a really interesting comment, Was, um, one of the guys said that he no longer thinks that frontier AI models are where it's at. Now, when I say frontier AI models, if you don't know what that means, that's where you're talking the likes of Anthropic, ChatGPT, the guys who are creating these massive, um, AI models. They call them frontier models. Um, because frontier models are already Really, really good. And I, and, and the question is, how much better do we actually need them to be? I don't know, probably not a whole lot better than what they are now for most of the tasks that we do, right? And so the real advantage then is going to become going forward is in effect creating the agents that, that connect what you're doing with what the frontier models can do. Right. And which is what you're doing here in the TikTok space, which I think is quite smart. You're connecting AI to TikTok Shops specifically, using the Frontier models to help you. And I guess this is where I think I agree with him in that I see a lot of AI in ecommerce now becoming not just about, can I use AI? Can I get Claude to do what I need it to do? Because I think most of us are getting there pretty quickly. But at the same time, there's still a huge opportunity cost with AI. So there's going to be this interesting sort of swing to getting AI experts. Like we've hired 2 people recently with the sole purpose of doing AI, right? It's just—

Sohun Sanka: Wow.

Matt Edmundson: It's a really interesting shift in our AI In our hiring habits. And so I'm curious where you, I mean, the complexity of it now, this is where I think you need, and I say it's complex in your head, it'll be probably quite simple. You'd always just do this, this, and this. But I think people listening are going, there's so much you can do that we don't know what to do, which is where I think this sort of people who are good at creating agents and keeping those agents up to date are going to do. Wow. Um, in, in the future, what do you— I mean, you've, you've obviously been doing this a while, but I'm, I'm kind of curious, where do you see it? You're, you're creating this agent, you're creating these agents for companies. Where is it going? Where do you, where do you see the next 6 to 12 months specifically for TikTok Shop? Um, that make— would make me think, actually, if I don't do AI here, I'm, I'm probably going to get left behind, maybe.

Sohun Sanka: Yeah. Yeah. So it's a good question. Um, and I've thought on like where the space is gonna go, 'cause like SF has already done it and I'm seeing ecommerce catch up. Um, with, with ecommerce, people are still so behind. They're still not even connecting their TikTok Shop tools into Claude as a connector, which I feel like is a huge miss. Like that's literally what I'm seeing. A lot of the friction when I onboard these brands and agencies, it's like, what, you haven't connected it?

Matt Edmundson: Yeah.

Sohun Sanka: You haven't connected Claude to your software that does all your work. How do you know what's going on in your software? And like, yeah, even agencies who I talk to are still spending like 15 to 17 hours a week on manual reporting, which is crazy. It's crazy.

Matt Edmundson: Yeah.

Sohun Sanka: Because you can just automate that. And I think that's where most of the space is, like 80 to 90% of the space needs to figure out how to click settings, Go to add connector and then add Claude to their tool. Like, 'cause they don't know how to interact with the tool and pull data and build that muscle memory. And I think hopefully in the next 6 to 12 months, every agency and brand is like doing that and getting their hands into that. Because the great part, and this is what I have all my brands and agencies do when they onboard, is that you don't need to learn how to prompt anymore. Like prompt engineering and prompting is kind of a thing of the past. Because you can just use what's called plan mode in Claude Code. And so like a lot of people don't know that this exists, which is crazy because it's just 2 clicks away inside their app. But it's a whole other discussion on SaaS user experience, which I can also get into. But what you do is you essentially click on Claude Code plan mode. And what we find is that these marketers, copywriters, they're actually using Claude Code for daily tasks that don't involve coding because the way that it thinks is just so more sophisticated. So, for those who aren't familiar, what plan mode essentially is, is it's like a built-in assistant that helps you clarify your thoughts. So, if I type in like, hey, I need a report, they're going to be like, oh, here are the 5 brands connected to your account. Which brand do you want? And they're like, okay, here are the main KPIs from your dashboard. Do you want all these or do you want anything else? Okay, how do you want this report to output? Do you want it to be in a spreadsheet? Do you want it to look in a presentation? Okay, how often should this run? Weekly? Do you want it to run as a cron job, which means run autonomously? Or do you want to have something sent to your Slack where you review it before you send it out? Do you want to send this out to a client via email or store it in your Google Drive? So to think about all the components of that, put it into a prompt very specifically, definition of what good looks like inserted into that. You know, the mental load to actually think of that prompt, define it, actually prevents a ton of people from getting started with it in the first place.

Matt Edmundson: Yeah.

Sohun Sanka: 'Cause they're like, I don't have the time for this. I'm really tired today. You know, and those are the thoughts coming through people. But if you just type literally that one sentence, okay, I wanna create this report in Claude, it'll do that legwork for you. And then it really just becomes a back and forth chat. And then what I find is like really what the best people are doing is if they realize they do this process 10 to 15 times a week, they turn it into a skill, which is essentially a recipe or a set of instructions that allows Claude to do this on demand where they'll take that whole report I just said and say, hey, run my weekly report skill. And now I want it to run every Monday without me asking you and then send it to me in Slack. And that's how you start to get more leverage. And that's where I'd love to see the space head over the next 6 months is just like really starting to become more efficient with their tasks.

Matt Edmundson: Yeah. Yeah. And I guess identifying those tasks in itself is a skill, right? Because I think a lot of people would, and not just with TikTok Shop, but just sort of outside that, a lot of people will sit down in front of their machine, Claude's on it and go, right, I know Claude can do some good stuff. What do I want it to help me with? Do you know what I mean? It's that kind of—

Sohun Sanka: It's so weird. And I don't really understand why most people don't do this. Maybe I had a really good training and really good mentors in my life, but the best operators are hyper-conscious of their time. They're doing time studies. They're really putting a dollar amount on their time. And I did this before I was even a founder. I was like, hey, my time is worth, you know, let's say it's worth like $2,000 an hour, being like very, you know, generous. I'm like, is this a $100 an hour task or a $500 an hour task or a $2,000 an hour task? Who do I hire and what things can I do to reduce the amount of $100 an hour tasks that I have on my plate?

Matt Edmundson: Yeah.

Sohun Sanka: And like, really, I'm surprised by the number of director-level people who aren't looking at all the work that's being done across the organization that is tracked in these project management tools and be like, hey, this can be automated. This is a $100 an hour task. This is a $500 an hour task. I'm doing too many of these $500 an hour tasks. I should be doing $1,000 an hour tasks. And even just like taking all the tasks and dumping into Claude and being like, hey, what's a $100 an hour task versus $500 an hour task? Go and pull off Upwork, all the rates of people who could do this and actually organize all my work and tell me, what I should be spending more time on versus what I shouldn't be spending time on. I'm a founder of a 20-person company and like we're trying to hit this revenue goal right now. We're here. Like you could just type that in there and then you should always be analyzing every single task you're doing and figuring out what are the next 3 most important things to get done in the next 2 days. Like if you're not, then you're just operating inefficiently and you're just doing a job. You know, you're not really like growing a business. You know, that's my— philosophy. And like, the first thing that my marketing agency that I onboarded had me do when I was an intern is they had me write SOPs. That was the first thing they do. And typically, like, writing SOPs is not a part of like a junior level person's like training process. I haven't like been in an organization where like anyone is really taught how to write SOPs correctly. I think people just like to say they write SOPs and everyone builds processes, but like No, no one's really training people how to do that and making people conscious of that. And like my first boss out of college was director of operations. He was so specific. He was like, every single task needs to have a verb noun description. If task descriptions are like the name of that task is a verb noun where it's like fill out report. And he's like, if it's longer than that, like going back to the drawing board and reading the title. And then he's like, If the sentence description is longer than 2 to 3 sentences, you should have another doc that links out to the full thing where it's like insert doc template and fill this out. It's so specific about what good looks like and what these SOPs look like where that was the first thing I've ever done in my career. And ever since then, I'm always thinking about how tasks can be broken down into SOPs and their core fundamental roots. And I think that has served me well with AI.

Matt Edmundson: Yeah.

Sohun Sanka: Because you just look at processes that people are doing. You look across call recordings, interviews, and transcripts, and you're looking at it, and I'm like, what's the noun verb here? What's the definition of good? And how do we just take this context and insert it into a machine that's way more intelligent than I am and then start breaking that down? And that's how you figure it out. But people just aren't doing that. Why do you think people aren't doing that, Matt? It's like you, because you've had multiple businesses. podcast brand owner?

Matt Edmundson: I think I'm probably a little bit more nuanced in my thinking. If you're an uncle, I'm a great-granddad, I think. Great-great-great-granddad, maybe. So I think life is nuanced and I think it all depends on who you are. I think if you are younger and grow up with a screen attached to your thumbs, you think very differently to someone who is maybe my generation, Gen X, who grew up on BMX bikes in fields. Do you know what I mean? And it, it, I think there are people who can think very mathematically, very process orientated. And I think there are people which just don't, and they think very creatively, very differently. Um, but I think— For me, in some respects, that's irrelevant. How old you are, how you think, what stage of life you're at, whether the kids are screaming around your legs at the moment, whether the cat's been sick in your slippers or whatever's going on. Right. I think fundamentally it's not knowing where to start. People just don't know where to start. And I say this to our team all the time. If you don't know, just go to Claude and say, I don't know where to start, but this is what's in my head. Where do we begin? And you can do this in plan mode, can't you? And I think, um, I think we, one of the things that we're all slowly starting to realize is actually, if you don't know where to begin, AI is really good at helping you figure out where to begin if you just ask it. You don't need to go with a polished answer.

Sohun Sanka: Yeah.

Matt Edmundson: And I think it's, I think we've approached AI, I suppose, a little bit like we approach employee. I mean, you mentioned, you know, sort of approaching our employees that, um, if I've employed you, the, the, the understanding there is I know what you're supposed to do because I've employed you to do it. Therefore, I have to tell you. Um, whereas I think we can go to AI now and go, I know you're supposed to do something. I just don't know what it is. Let's figure it out because I haven't got a clue. I don't know the answer. Um, and so I, I think it's taken people a little while to get their head around. I don't need to have all the answers. I don't need to know the end from the beginning. What I do need to do, what I do need to know is I need to understand the framework or the process. So let's start with that and I can use AI to help me figure that out.

Sohun Sanka: But don't you think, let's say putting AI aside on a first principles basis, when someone moves from higher education into jobs, things get more ambiguous? You're not following business out of a textbook. You're not using a 4 Ps framework to bring a product to market. And it's just like things get very ambiguous and getting started and taking action kind of is the thing that I've seen at least. people who are very biased towards action, who are just like, even if I don't know what to do, I'm just going to do it. Do you find that there's a generational difference in the characters of people who do that versus not do that? Because for me—

Matt Edmundson: I don't know if it's generational. I think it's very personality-driven. I'm very much— I think there's 2 types of people, right? There's the guy that will get in the car and go, right, we need to get across to the other side of the city. Let's go and figure it out. Right.

Sohun Sanka: Yeah.

Matt Edmundson: As we go. And then there's a person who'll get in the car and want to know exactly where they're going. They'll want to map it out. They'll want to understand where the roadworks are and all that sort of stuff. I'm very much the former, and I think it's very entrepreneurial. It's like, well, let's just do something. Let's just go. And there are people that want to figure everything out before they do start. I don't think either is better than the other. I just think it's a people thing. And I think it's, yeah, I just think it's personality in many ways and what you've been brought up to.

Sohun Sanka: And it's interesting that you say that because like, you know, when it comes to like learning AI, I find that if you give too much documentation, SOPs, and guidance to people, that paired with the fear of feeling stupid about using complex technology prevents people from getting started. Like it's almost paradoxical where it's like, If I give you too much clarity, I increase the mental real estate that I'm taking in your head to get this started versus finding that sweet spot where it's like, hey, just type in this one sentence into plan mode. Just spend 30 minutes typing back and forth with it. That lack of clarity often leads people to become more successful with AI. And I think—

Matt Edmundson: Yeah, it does.

Sohun Sanka: That has been interesting for me to discover because I've gone through the process of education with brands versus agencies is a completely different learning curve. Brands are often like, hey, I just— they get to a point where they're like, if I give them the same instructions as an agency, they're like, can you just do it for me? That's what I always get with brands and agencies. They're like, what do you think about this? I created 3 skills, I'm running cron jobs, I built a software, what do you think? How do we automate this? How do we do—

Matt Edmundson: Yeah, yeah, yeah.

Sohun Sanka: They're really thirsty and hungry to do this. I see such a discrepancy there. Like, you know, from your time, you know, you're a brand owner and you're also like, you know, B2B podcast business owner. Like, do you see any discrepancies between these like psychological patterns between like brands using AI versus like marketing agencies using AI?

Matt Edmundson: Yes, very much so. Because, you know, having done agency for a little while, agency is all about the process. It's all about the framework. Here's my framework. Let me help you figure out how to get your business through that framework to help it grow. You understand those types of things. I think if you're just a sole brand, I don't know of many brand owners that have sat down and gone, what's my framework for this? A lot of it is gut intuition, right? Or circumstantial or providence or whatever the language is that people want to use. But there's It's very much a— if I sit down with a lot of brand owners and I'll go, what is your framework, you know, for taking a product to market? You go, well, they'll go, I don't know. I just, I went down the supermarket and I saw it. I thought it was really cool. I want to put it on the website. And it tends to be a bit more like that where agency can't be like that. It has to justify its fee. And so agency has to have a process, which is its IP.

Sohun Sanka: Yeah.

Matt Edmundson: Just by definition, I would've thought.

Sohun Sanka: So you're basically saying the nature of the business requires you to think about process versus with brand, there's more flexibility and you don't have to think about process as much to get started.

Matt Edmundson: Yeah, because a lot of brands are lifestyle brands, aren't they? And I think people do them as passion projects. And I think people, they love, I don't know, mountaineering. And so they go, right, well, let's just set up a website selling mountaineering equipment. Because it's what their passion is. And they, you know, I come across, um, I'm converting my van at the moment, so you come across a lot of these van conversion videos online. And it's amazing how many of them are, uh, you know, they, they do these videos and they have an online business selling the stuff that they're talking about in the videos, but their passion is the van. And I, I mean, I don't know, I, I could be wrong and I don't want to prejudge, but part of me would be I reckon if I went to that guy and said, right, what's your ecommerce framework for growing your business? They're going to look at me and go, I don't know. I just like making content. We seem to get more people coming along and it, and I get to drive my van out into the country every week and people buy my stuff. And as long as we get good stuff in and I do great reviews, I appreciate that. That is in some respect a bit of a framework, but I don't think they would have thought about it. And I don't think they would have labeled it. That's just what they do.

Sohun Sanka: I do think on TikTok Shops specifically, content volume and just getting after it is huge for success because the thing is with the algorithm, and you hinted at it really clearly, is that the algorithms are getting very, very good for ads and for social media at just matching niche interests in moments where specific people are online. So therefore, the idea, and we saw this, I chatted with someone from TikTok and he was saying China, Turns on its feed with like about some ridiculous number, like 60% shoppable content. Like 60% every TikTok video you see is a shoppable piece of video. And the like in America, like there is a huge like scarcity, like in terms of amount of content you can get. Like there isn't enough content to fuel an algorithm for where people want to turn it up more in America because they're in order to China, right? Um, people just don't have enough content for enough niche interests that exist. So like the thing that I index on right now is like, hey guys, like don't get upset that your first video doesn't blow up on TikTok Shop because it's not. But if you're like the top affiliates are posting 15 to 21 times a week per brand.

Matt Edmundson: That's incredible.

Sohun Sanka: Because they understand, they understand that you have to meet that viewer at the right time with the right sales angle. Like, I look at TikTok Shop as like a little micro sales army. Like, you're onboarding a sales team and you're getting them to just tweak and test their script over and over and over again across as many videos as possible. And the algorithm will meet that viewer at the right interest, you know, because like, you've seen it even on Instagram, you know, if you scroll for like clothes, you're being served clothing ads constantly. And like eventually you will give in. And on TikTok, same thing. If you're just, you know, yeah, varying that creative, people might not even understand why they're buying. Maybe it's a tweak in the lighting, maybe it's a tweak in the hook, maybe they like, you know, spill some ice cream on their pants coincidentally that Friday, it stains, and like they've been served 5 ads prior that week by someone else and they happen to convert. Like you have to reduce the variance as well and let the algorithm carry your interest forward. Like, that's one of the principles, you know, that I believe. And I think where people really mess up is they don't chat and let their creators know that and encourage that sort of positive behavior to get to success. They send a sample out, they do the video, and they're not doing anything after the fact with that context in mind.

Matt Edmundson: Which is interesting, isn't it? Because I, I mean, the general rule of setting up any successful media, whether it's a podcast or TikTok or Instagram or whatever it is, is consistency, is just putting out a shedload of content there. So the same rule in theory would apply to creators who are creating content for your product. You can't just do one video and expect that's going to change your life. You've got to have this consistency of investment, I would have thought.

Sohun Sanka: Well, it's— I think the mindset is like a lot of these people who come on to TikTok Shop are performance marketing oriented, Amazon oriented, where it's like consistency looks like testing different creatives. And then you have an ad mechanism doing all the follow-through. Then you go and basically look at your results, you look at your creative process, you find a new variation, you test it, and the ad mechanism does all the follow-through. On TikTok Shop, the follow-through looks like a human going and talking to creators, educating them, getting them into a community, and sitting down with them and breaking it down and sharing the data. Like, it's a different kind of psychological concept to sit down. Like, imagine if you like, sat down with your PPC algorithm for Meta and you were just like, hey, you did a great job spending on this, but I would spend on this differently. And you sat down for an hour talking to your PPC algorithm. That'd be crazy. People just psychologically don't do that. But that's what you have to do with creators.

Matt Edmundson: Yeah.

Sohun Sanka: Sit down with them like they're a little salesperson, mini algorithm there. Go in-depth on every frame of the piece of content. You have to be hyper-obsessed. And That's what I see with the best founders is like they're sitting down with their top affiliates like they're employees in the company because they are.

Matt Edmundson: Yeah.

Sohun Sanka: Like if you onboarded a sales team, when I onboarded my sales team in SaaS, the first 14 days I have 2 calls with them a day. We're going over scripts. They're shadowing me on calls when I'm doing sales calls, partnership calls and stuff during the day. And at the end of the day, we drill scripts. And like in the first 14 days, I'm making sure they really, really understand our product, our process, our brand, what really makes certain people sell. Give them room to ask questions. Show them what good looks like. I have them bathe on like 20 different sales call recordings and make notes. Like, the amount of hands-on work that I do with the sales team there is crazy. I haven't seen any marketers on the TikTok Shop side do that.

Matt Edmundson: Mm-hmm.

Sohun Sanka: But the best community managers on the TikTok Shop side are doing that. Like, we interviewed, um, Javy Coffee on a podcast about, I think, a month ago, and they had a TikTok Shop 3x their growth from $300K a month to $1 million a month on TikTok Shop. They have 15 community managers in their community responsible for a book of business, like a CSM, and they're making sure that all of those affiliates have a positive experience. They're increasing their GMV, and they're rewarded based off of how much their GMV increases from that book of business.

Matt Edmundson: Yeah.

Sohun Sanka: And I think that really incentivizes a positive feedback loop to take care of your affiliates. And even in the first 30 days, they're actually making sure that affiliate posts 30 times.

Matt Edmundson: Oh, wow.

Sohun Sanka: Once every day in the first 30 days. But if you look at it from a quantitative perspective and consistency perspective, if you post 30 days with a group of community managers that have nurtured a million-dollar-a-month brand, chances are you'll make at least one sale. And chances are people who make a sale versus don't make a sale, are willing to stick with a program and continue with that momentum going forward. So, that's what the top brands are really doing to make sure TikTok Shop is successful.

Matt Edmundson: Yeah. Wow. That's level 2.

Sohun Sanka: It's level 2. I call it level 1 because if you don't, you're just sending off samples into the abyss complaining that it's not profitable and you're sunsetting. It's like a couple hundred thousand dollars waste of time. It's not even a fair shot at TikTok Shop if you aren't doing that.

Matt Edmundson: Yeah.

Sohun Sanka: Level 3, one of my friends, he's a top 100 affiliate in TikTok Shop. This will blow your mind. He has an account just for A/B testing lighting on window fans. He's one of the top window fan affiliates on TikTok Shop. He prints $1 million a year in commissions. He's crazy. And he has one account where he posts multiple times a day. window fan with a different gradient of lighting. And me and him were really hitting it off in SF because I'm just like, I got made fun of because I was like, oh, you need to A/B test lighting. And then Marcus would be like, A/B test lighting? You're an idiot. We A/B test copy. And I was like, nah, we need to A/B test lighting because on TikTok Shop, the lighting matters so much. And this guy was A/B testing 3 different kinds of lighting every single day. Wow. Hundreds of videos posted just to find the right lighting angle that he could post on his main account to get that to sell really well. And then, he scrapes and indexes the top 100 videos from the top affiliates every single month, organizes that into a testing pipeline with Claude Code, and then pairs that with the logic from his lighting testing. And I'm like, this is what it takes to be a top 1% affiliate. And I knew there was— that is definitely level 3. It's level 3, but I'm just like, you know, now that it's on this transcript, someone can take this transcript and go test it out on their account. Yeah, there's nothing stopping someone from doing that.

Matt Edmundson: Yeah, and that's interesting. And I totally agree. I think like my son, he, um, he has got a really large Instagram account and he— that's where he makes his money is, is, is on Instagram. He is a nutritionist. He does a lot of stuff for nutrition. And he does so much testing, it's quite remarkable. Just the angle of the camera, the distance between him and the camera, and everything is tested and everything has got an exact thought behind it. To the point where he figured out that if he had a tea towel, just like a cloth on his shoulder, when he was shooting the video, the engagement went up.

Sohun Sanka: Wow.

Matt Edmundson: And it's just all these little things. All these little micro adjustments, which took his account from being a great account to being a pretty big account. And you kind of go, that is dedication. That is stuff that I would never have thought about. I'm really good at split testing headlines. Because that's e-com. And it's just the same principle, isn't it? It's like, well, let's split test, you know, if I stand like this, or if I, if I have this on my shoulder or not have this on my shoulder. And I think you have to do that with video content.

Sohun Sanka: Agreed. Fully agreed. And when you use AI, it's you're like, hey, out of this 100 angles, which one performed well and why? And then it can start suggesting your next 5 angles. That's where you plug AI in because we have to make it AI-oriented, right? Yeah. Yeah, yeah, yeah.

Matt Edmundson: Absolutely. Listen, but I'm aware of time and I'm thoroughly engrossed in the conversation as I normally am. If people want to reach out to you, if they want to connect with you, find out more about what you do, what's the best way to do that?

Sohun Sanka: Yeah, just type in Sohun Sanka on LinkedIn. My name's Sohun. There's not that many Sohuns in ecommerce, so I'm sure you'll be able to find me if you type that in. And then clankersapp.com is my website, but I'm very responsive on LinkedIn. I use it more than Instagram, so you can find me there.

Matt Edmundson: Fantastic. We will, of course, link to the LinkedIn profile. I'm just trying to say this in my head. We will link to the LinkedIn profile in the show notes. And the description, which you can just scroll down on your app and click and just click that link. It will take you straight to the LinkedIn profile. But listen, bud, I genuinely love the conversation. Thoroughly, thoroughly fascinating. It's always good to talk with fellow AI eggheads. I don't get to do it much. And so it's always nice hearing what other people have to say. Before I go, 2 questions. Number one, what's your question for me? This is where I take The question, I'll just go out and answer it on social media. But what's your question for me? It can be about anything, by the way. I should have prepped you. The question can literally be about anything. I've had people ask me about, um, whether I think aliens exist, what's my favorite book, to anything to do with ecommerce. I don't care. Uh, what's your question for me?

Sohun Sanka: I would say if you could ask your audience what is the thing that scares them most about getting started with AI, I'm very curious to see what they say to that.

Matt Edmundson: Okay. I will, I'm actually going to go and ask a bunch of them. Uh, if you want to know what the answer is, come follow me on social media. Um, last question. Uh, so much value given today, bud, but, um, I always like to do this saving the best till last thing. What's your top piece of advice for someone who just wants to get started in TikTok Shop using AI right now? Where do they begin?

Sohun Sanka: Connect your Claude connector or your ChatGPT connector into your TikTok Shop outreach tool. This is the repository of every single affiliate activation motion you're doing on TikTok Shop, and just connecting it allows it to learn everything you're doing. These tools have like 110 MCP tools that go and tell you how to read, write, ingest the data. You, you You basically get an expert on all these tools by just making one connection. And what I really like to do is just ask it to pull out everything I'm doing and put it on a nice FigJam board so I can see all my message copy logic oriented in a nice funnel. Ask Claude to add notes about like, hey, what should I be turning on that I shouldn't? Because it'll go and enrich itself and its learning of these tools with the help center documentation. And everything into there. So that could probably save you like a couple thousand in consulting fees, actually.

Matt Edmundson: Yeah.

Sohun Sanka: That's one of the things that I would do if you're coming to me and saying, hey, audit my TikTok Shop, is I'd pull everything out, I'd put it onto a nice, easy-to-view interface, and I'd show you exactly where your leaks are. But AI can help you at least get started with that. You're going to need some TikTok Shop domain expertise to understand what it's saying and correct it. But it's a great place to be like, well, what's going on? What are the things I should do? How should I get started? And it can help you answer those questions.

Matt Edmundson: Fantastic. Fantastic. Well, thank you so much for coming on, man. Genuinely really appreciate it.

Sohun Sanka: Appreciate you having me on. It was a pleasure.

Matt Edmundson: Now, we got there in the end. We had a few technical issues at the start, but we got there in the end. It was definitely worth it. Thank you so much for joining us this week. Uh, it's been a great show. Make sure you like, subscribe, and do all of that good stuff like you do with your podcasts. Um, because we have lots of conversations like this with some awesome people just like today. So make sure you do that. But that's it from me this week. Thank you so much for joining us, uh, wherever you are in the world. I will see you next time.

Sohun Sanka: Bye for now.