Ecom Growth Insider · Podcast episode

The Claude System That Decides Which Meta Ads To Make | Will Sartorius

With Will Sartorius · Published 2026-08-04 · 42 minutes · Hosted by Andrej Tumachowitsch

The Claude System That Decides Which Meta Ads To Make | Will Sartorius

Will Sartorius runs Selfmade, a New York agency that raised $19 million in venture capital, nearly collapsed, and was rebuilt into an AI creative engine. In this episode he lays out the system that decides which Meta ads to make: social listening on your reviews and surveys, a gap analysis of the personas, angles, and emotions your competitors run that you do not, and a time series analysis of your own ad account that resurfaces learnings from ads you turned off months ago.

He also explains his 80-20 rule: 80 percent AI-generated creative for volume and bottom of funnel, 20 percent fully human creative that builds real connection at the top. Both halves speak to the same personas, and each winning angle gets its own landing page and email flow. That is the playbook he credits to Gruns and its 45 angle-specific landing pages.

For founders starting from zero, Will walks through a one-day setup in Claude: one folder of reviews, surveys, and founder interviews, a persona-angle-emotion tree, and every past ad tagged and correlated to CPA or ROAS. The result is a clear list of the ads you should make next and the ones you keep wasting money on.

What you will take away

• Three analyses decide which ads to make Will's system combines social listening on reviews and surveys, a gap analysis of the asset types, personas, angles, and emotions competitors run that you do not, and a time series on your own Meta account.
• The 80-20 rule for AI creative Will's hypothesis: 80 percent AI-generated ads for volume and bottom of funnel, 20 percent entirely human ads for top-of-funnel connection, both attuned to the same personas.
• Old ads hold buried learnings A time series analysis resurfaces personas, angles, and formats that worked before you turned them off. Will's team uses the Meta API for this and says it still beats the newer MCP.
• Build a persona tree in about a day Will's starting move: put reviews, post-purchase surveys, and founder interviews in one folder, then ask Claude for eight core personas, two to three angles each, and two to three emotions per angle. He says the whole setup takes about a day.
• Tag every ad, then correlate to your North Star Will tags each past ad with persona, angle, and emotion, then correlates them to CPA, ROAS, or CPL to see which ads to make next and which keep wasting money.
• Match the message through the whole funnel Ad, landing page, and email flow should carry the same angle so it feels like one conversation. Will points to Gruns, which he says ran 45 angle-specific landing pages.
• Taste means anticipating the client AI steers, humans decide. Will keeps creative strategists and editors in the loop because brands expect a certain look, and saying so on sales calls eases the stigma around AI.
• Iterating winners is overrated Will's pick for the most overrated habit: making 20 iterations of an existing winner. It sometimes works, but he says brands over index on it instead of testing new personas and angles.

Jump to a chapter

• 00:00 - Cold Open and Episode Intro
• 00:47 - From Deutsche Bank to DTC
• 07:27 - ChatGPT Reviews and the Three Analyses
• 10:37 - Rebuilding Selfmade and the Open-Source Mindset
• 17:12 - Can AI Learn Taste?
• 20:37 - The 80-20 Rule and Human Connection
• 27:37 - Full-Funnel Message Match and 45 Landing Pages
• 31:52 - Starting From Zero: The One-Day Playbook
• 37:06 - Team Structure in an AI Creative Org
• 39:44 - Lightning Round and Where to Find Will

Full transcript

Lightly edited from the episode captions for readability. The episode is the record - where they differ, the audio wins.

00:00 - Cold Open and Episode Intro

Gruns got acquired for 1.2 billion and they had 45 landing pages. Like one was talking about kids in pediatrics and, you know, the pediatric doctor saying like why you need this. Another was talking about GLP-1. One was talking about pooping.

My guest today is Will Sartorius, the man behind Selfmade, a New York agency that raised 19 million in venture capital, nearly died and then got rebuilt into an AI creative engine. Will took it over at 28 and rebuilt it. In this episode, the three analysis that tell you exactly which ads to make. Social listening, gap analysis and a time series on your own ad account. His 80-20 rule for AI creative and the full playbook behind those 45 landing pages, step by step, starting from zero. Fair warning, after this one, you'll feel behind on AI. I did. Let's get into it.

00:47 - From Deutsche Bank to DTC

Hey, Will.

Hey, Andrej. How's it going, man?

Going great. What about you?

Good. You know, Friday, I think it's supposed to be like 80 degrees Fahrenheit, which you know, for you Celsius folks, let's see, is 26 degrees this weekend. So like New York City, popping off.

That's nice. That's nice. Here it's way colder than that. It's like 17, 18 degrees Celsius, which is still not too bad, but not 26.

Fair enough.

But cool. You left Deutsche Bank to become the CMO of a D2C shoe brand. What happened there? I don't think that's a normal move.

Yeah, no. It's pretty serendipitous and just like very unconventional. So I studied economics at a liberal arts college up in Maine. That's in the northeast of the U.S. I was very much interested in macro trends, namely also like behavioral economics. It was like still pretty early in the behavioral economics space. But either way, everything I'd sort of been positioning myself put me towards the track, like going to work in finance. And so I did an internship at a boutique investment bank and then I interned at Deutsche Bank. And the internship led to an offer. And I did, you know, consumer investment banking there for two and a half, three years.

And while I thought like the intellectual stimulation was definitely there, what I found was challenging was its cutthroat meritocracy aspect. And I'm all for meritocracy in the sense that like if you do great work, you will get promoted. But at the same time, it created this really unhealthy work environment where you couldn't really trust anyone. And like when you're perpetually getting pitted against your peers in bonus season, it just like it never felt natural to me. And I'm like very much like a team player and a collaborator. And so that was just like always really tough.

And so in the evenings, I knew I wanted to sort of leave at some point, but I wanted to sort of have a bit of a plan. And so the behavioral economics component was always interesting to me, namely the psychology of it all. Like what is driving sort of folks to make, you know, quote unquote, irrational decisions. And so what I started to do is I started to take evening classes at NYU. And what I would do is I would have to like rush out of the office. Normally I'd have to come back afterwards. But, you know, occasionally I'd like stop at Goodwill to like grab a t-shirt. So I like wasn't showing up in a full suit to these like advertising classes. And I just like became like ingrained and hooked. Right. Like we were all making like print advertisements, but it was really, really great. And it was just this nice marriage of analytics and creativity they hadn't necessarily found.

And so I sort of decided to exit stage left and said, hey, you know what, this isn't for me. I had a friend that was building a startup that was like part of the Snapchat incubation. It was sort of like an Infatuation, but for workout classes. The idea there was you'd go to a workout class, you'd rank it, etc. This is right before COVID. COVID sort of tanked that idea. No one was really going to workout classes then.

And then I sort of joined a D2C men's shoe company based out of New York. They have a few stores in New York, one in Nantucket. And the biggest hole in their company was they weren't really doing anything with paid or really any email or really anything. It was largely just based on like word of mouth, organic and foot traffic. And so I started doing that for the company and actually started growing the online business, which was cool because we knew who our buyers were, but we never necessarily started actually running paid ads for them.

And then one thing led to another. And I found myself as an employee of Selfmade and Selfmade has a long story. You know, at some point we can grab a beer and I'll give you the full rundown. But sort of the TLDR is started in 2016, raised a bunch of venture capital money around like $19 million. The initial app was for posting on Instagram. You could sign up. Selfmade would post photos for you. You would upload your assets. The editors would generate or, you know, edit photos. And the whole idea there was that Instagram was going to become the center hub for e-commerce. And this is before everyone sort of had their own Shopify website and everyone sort of wanted like a unique experience. And so that hypothesis ended up sort of being true, but a little too late.

And so they pivoted to become like a conventional agency in around 2018, 2019. I joined the company. I think it was maybe 2020, 2021. I don't remember exactly. And one thing, again, led to another. And there was an acquirer who ended up being a bit of a snake oil salesman. And things just sort of hit the fan. We were probably doing at the time between like $4 and $5 million in revenue. And yeah, the revenue sort of tanked when he sort of took over. And the company started losing a lot of money. And after about a year of that, there were a lot of other red flags. I was sort of tapped to be like, hey, do you want to run this thing?

And I, you know, at the time I was probably 28. And I was like, I don't really want to go look for a job. Like I have all these client relationships at this point. I could like probably give this a run for a few months and like sort of see how it goes. And so I said, sure, you know, why not? And it was just like a few months of really, really intense grinding of, you know, sort of getting the balance sheet positive again.

07:27 - ChatGPT Reviews and the Three Analyses

And this was at the same time that ChatGPT had just come out. And what we had been doing quite effectively at Selfmade at the time was collecting reviews, manually analyzing the reviews and saying like this persona, this angle, this emotion. And what we found very quickly was that using ChatGPT at the time to analyze reviews was a lot faster than having a human do it. There was a lot of hallucinations early on. And I think you had to say many, many times, do not make something up. Do not make something up. Do not make something up.

But having those consumer insights, it started very quickly cascading or word of mouth just sort of piling up. And like those analytical decks that we were generating started selling like hotcakes. And so for many of the very large agencies that you're probably very familiar with, we were selling these for all of their clients. It's like analytical decks about like who their consumers are, who they should be targeting, et cetera. And then from there, it was like, okay, so we should probably, you know, start bringing brands in-house and not just like serving agencies and like actually generating the creatives.

And then, I mean, the AI continued getting better. And so like we started tying the reviews to actually generating creatives. And now we're in a place where, you know, the AI is so good that you pretty much can create like a bit of a feedback loop where your social listening, and that's, you know, the collecting of the reviews, can become like one of the three components of generating great ads.

I always say this, like sort of three things that you really need to make great ads. It's social listening, like understanding what your consumers are saying. It's performing like a gap analysis. And what that is, it's like what asset types, personas, angles, emotions are your competitors running that you're not? And then third, it's like doing a time series analysis on your meta data, not your meta data, but, you know, your Meta ads.

And I think what so many folks miss out on is they probably maybe turned an ad off in January. You know, there were a lot of great learnings in January on that ad, but they turned it off because it maybe fatigued out, just stopped working. And then that learning was sort of lost forever. So again, AI is really good going back to the ChatGPT era of analyzing data. So Meta API is what we use internally. The MCP is relatively new. We've tested it. The Meta API is better. But like running a time series analysis on your data can show you like this persona, this angle, this format has worked for you many times in the past. You just haven't been running it recently. So it sort of resurfaces old learnings.

And so if you sort of combine that time series analysis with, you know, the gap analysis and the social listening, you're in a place where you can create a pretty good feedback loop and have a very good sense of the types of ads you need to create. That's a long-winded answer. But, you know, I think we probably touched on a few things there we can probably dive in a little bit deeper on.

10:37 - Rebuilding Selfmade and the Open-Source Mindset

Yeah, yeah, definitely. A few things where I would definitely love to dive deeper into. So first, like when you took over Selfmade, did you have an idea where that would go and how the agency would transition?

Yeah, great question. I would say no. I had no idea. I was sort of just like faking it till I made it. We were doing a lot of different things. We were doing post-production work for Suna, right? We were just like editing their photos because we had this big editing team from Legacy Selfmade. We were doing these GPT decks for some of these larger agencies. And then we were also doing some just like one-off like monthly, here's a batch of 12 creatives for these like small brands that are maybe spending, you know, 10 to 20K. And it was just like very much just like keep trying and seeing what sticks.

And I ultimately found that when I would sort of explain what I was using with AI, people were like, oh, like this makes a lot of sense. And like you can sort of get from zero to one quite quickly with this. And you could actually see the value add in using AI rather than just like sort of the hype around it. And like what I can't stand in this era that we live in is like people posting like generate 100 ads with this like Claude Code plus, you know, Nano Banana or GPT-2 or Seedance thing instantly. It's like that's not how this works. Like you have to put in sort of the critical thinking and you have to build a system around what you're trying to do. You can't just like assume things are just going to naturally happen for you if you prompt. And I've said this before, it's like I can't remember the last time I actually read a prompt. The prompt is the very last part of the equation for me. Like there are 100 things I want to do first before I actually see a prompt.

And so what I think ended up happening was I just like became hyper fixated on just like in being sort of stubborn. And like how can I generate ads without having a sort of a human in the loop? And then I sort of realized that you probably do need to have a bit of a human in the loop because at least where it stands now, AI doesn't have great taste. But it does have really good ways of steering you in a good direction. And then if you combine that with someone with really good taste, that's like that's a winning formula right there.

So to answer your question, I really had no idea what I was doing. I had no idea how to run a company. I had never in my lifetime sort of wanted to or expected to run a company. Like I don't think I'm necessarily like an entrepreneurial guy. You know, I just like sort of like to hang and, you know, be part of a team. But, you know, that's just what sort of happened. And, you know, what I love the most about doing what I do is, you know, it's like writing my newsletter. It's like it forces me every week to do a deep dive on something where I can actually sort of, you know, get tunnel vision and build. And like building is what is exciting to me. And I've surrounded myself with such a great team at this point that like, yes, I am involved in the agency on a day to day basis. But like I don't have to necessarily be on every call or like be in every Slack channel. Right. And so that has allowed me to sort of push our company forward in ways that would otherwise be impossible if I was, you know, the point person on everything.

Yeah. Yeah. And I've been really enjoying the newsletter. It always has a lot of valuable insights and yeah, just everything that you, the latest tools or processes that you share in terms of how you use AI to come up with new creative ideas or review and go through quality control on certain creatives and landing pages. That has been super valuable to see and also just very easy to then test and implement, like try it out and see what it spits out.

Yeah, totally. And I write for an audience that like has really no knowledge. Right. Because I think, and this is true going back to what we were discussing initially, like this was very true in my finance days. I think nothing frustrates me more than people that overcomplicate things to make themselves seem smarter. And I think folks love to do that. In this AI world like this, these tools are not as complicated as people make them out to be. The barrier to entry is a lot lower than you think it is. You just have to pull up Claude Code and start talking to it the same way you would Claude chat. Right. And so it seems intimidating, but the zero to one is actually not that scary.

And I have made it my mission to take these people down that really have tried to just like gatekeep a lot of this stuff. And, you know, I give credit to Dara Denney, you know, a good friend of mine. And, you know, she's in New York as well. And at first I was like, I feel like I'm giving away too much. Like, I don't know. It's just crazy that I'm telling everyone this. And she's like, no, like this is how we should be working. And I've totally subscribed to that mindset. Like if I have an idea, I'll ping, you know, the Foxwell group or I'll ping like my Hampton group and I'll say like, oh, you know, I've got something here.

And so it's sort of like the open source mindset. And then the other great part about that is that people come back to you and say, like, I took what you built and I added this, this and this. And like, I'm like, oh, shit, this is amazing. Like, I'm going to take that now. So like if you have that mentality, right, it just creates a really nice cascading effect. And that was something that I've always sort of craved. It's like a community where we're all sort of building together. And it's like no one is trying to cut each other down or gatekeep information.

Yeah. Yeah. Those are the best type of partnerships, communities, everything, because you just progress way faster.

17:12 - Can AI Learn Taste?

And you mentioned right now there is still like a human in the loop required to have good, good creative because AI doesn't have taste. Do you think like on the one side, whether that will change and whether you can train AI to develop taste? Because, I mean, at the very high level, taste is usually just pattern recognition and past experience. And in theory, AI should also be able to learn that.

Yeah. It's a really good point. And I think the answer is yes and no. So creative is innately a very emotional thing. Creative dies in bureaucracy, right? Creativity dies in bureaucracy. So when I say you need to have taste, what that really means is that, especially if you're running an agency, like you need to know what your client is sort of like anticipating and expecting. Because I agree, like when I do like my organic or I, you know, am making videos for whatever, like I don't really give a shit, right? Like the AI can help me all day long. And I trust it because like these systems I've built, you know, have the pattern recognition mindset and have like a good sense of like this is a great open rate. This is a great click through rate, whatever, right?

When I say taste, what it really comes down to is brands typically want to have things a certain way. And so like without having that human in the loop for that sort of quality control, you run the risk of alienating the folks that are, you know, on retainer, I will say. And then sort of even more so than that, there feels to be, and I experience this on sales calls all the time. They're like, what is your creative process? Like how much AI are you using? Because we advertise ourselves as an AI creative engine. And I always feel like I have to say like we are using humans in the loop. We still have creative strategists. We still have editors. And that alleviates concern because there is, I mean, there is such a stigma around AI. And like we live in a bubble where everyone's using it every day, all day for marketing. But a lot of these folks are not and they don't trust it. And the previous hallucinations that were happening in 2022 are not happening in 2026, but perhaps they believe they are. And so there's just like a lack of trust with AI and especially with AI and marketing.

And so when I say you need to have taste, you need to sort of be able to anticipate what your client wants and anticipates. And like, yes, eventually you will get clients that are like, I don't give a shit, like just give me performance. And like those are the great clients. But like a lot of these enterprise brands don't feel like that. They say like, I want everything to like look this way and everything to sound this way. And like, yes, AI can absolutely do that. But like having quality control or just like having a human in the loop is still really important, I would say.

20:37 - The 80-20 Rule and Human Connection

Yeah. Where do you think is that with, like Meta, like where is Meta ad creative developing over the next few months and years? Because obviously there are now more and more brands that want to just launch a lot of creative volume with the AI possibilities. And with like basically, obviously, if the brand does it properly, does it right, they add a lot of context, they add the reviews, they add their customer persona and then like previous winning ads. And then they use that to create a lot of new creatives and a lot of new ads. And obviously, like oftentimes the brands that can create the most, or historically, the brands that were able to create the most creatives, they were getting better results. Do you think that will still be the case with AI?

Yeah, I think, and my hypothesis on this has been 80-20, 80 percent AI generated creative, 20 percent entirely human generated. You need to have creative volume and you need to have creative diversity. AI can do a really good job of telling you what sort of ad types you aren't running and what personas, angles, emotions you aren't focusing on. So that in and of itself will allow you to fill the holes that you currently aren't testing for. And, you know, there's a tool I built. I showed this in Lisbon, which is like adlib.getskipper.ai. And it's totally free, you know, because like nothing like this existed. It'll tell you like where the holes are in your strategy.

And the whole point of sort of saying all this is that you ultimately need to be able to sort of synthesize the information in a way that is like sort of like digestible to not only your team, but like sort of to your clients as well. So I lost my train of thought there for a second. But, yeah, what was like sort of the core of the question?

I mean, it was talking about the 80 percent.

Yeah. Sorry. 80, 80, 20. Yeah. Perfect. OK, sorry. My bad. I like got up so early this morning.

All good. All good. Yeah.

So it's 80, 20. So like you need to be able to fill those holes with AI. Right. And they do it. There's a great job producing volume. People are also starved for human connection. Like we are more isolated as a society than ever before. We are seeking dopamine at like an increased rate that is just like so unbelievably unhealthy. And like people are just like working in dopamine deficits perpetually. And so give people that, I hate to say it, but like that facade of connection. And I was chatting with Harry Damaj the other day, and I think he's doing like a really good job on this, where you just like have long form content, you know, maybe two to three minutes. It's just a single clip. And it feels like you're just talking to a friend. And I think that is really needed.

I don't know if you remember, you know, Steve, this is semi-related, but like Steve from Blue's Clues. And if you grew up on Blue's Clues or not, you know, a little too young for that. But, you know, he was like a pivotal part of my childhood. And he came back a few years later and he did this like Instagram live. And it was like one of these things like, hey, like you're doing OK. Like I know the world is really tough right now, but like everything is going to be all right. And like that totally went viral. And people like felt like they could sort of relate to that experience. And even though it was virtual, and obviously it helped that this was a nostalgic character in their childhood. But I think that's what people are craving. They just like want to be seen. They want to be heard. Like my hypothesis about, and I don't want to get like too heady here, but like, you know, human species, like all anyone really wants is to be seen and to be heard in life. And like have your thoughts, you know, internalized by others. And that is where the human connection is totally necessary.

And that's why I never use AI UGC. I think it feels, and even to say AI UGC seems crazy to me. It just like feels like so inauthentic and just like, you know, it's the antithesis of like what we're trying to do here. And, you know, I'll add that like in terms of AI, I'm a bit of a pessimist in terms of like how our future looks with this technology. Like I don't necessarily see this utopian future. And again, like not to get too heady here, but like we are, especially in the U.S., you know, we have a country where, you know, the wealth gap is like getting increasingly, increasingly, you know, more wide. And like you have these tools in the hands of a very few people, that like doesn't necessarily bode well.

So, you know, this is all to say that like where you can make human connection in your ads, like you will win. And so like think, like to go back to sort of fundamentals, think top of funnel. Like who am I talking to? How can I speak to them as if I were talking to a friend, right? Like there's sort of three ways, you know, to convey information. It's like are you talking to them from a position of authority? Are you talking to them, you know, from a position, you know, as a friend, as an equal? Or are you speaking to them from a position of like an aspiration, like who they want to be? And like those are sort of the three. If you look at any ad, they fit into one of those three buckets.

And so, you know, pick one of those three buckets. And when you're doing that 20% of the human ads, you will start capturing people. And then you have this other 80% that's generated with AI that is attuned to the same personas that you're talking to at the top of funnel. That is sweeping up. And, you know, maybe they don't have great click-through rates, but they have really good click-to-purchase rates. And like that's how the full funnel works. So keep humanity in what you're doing. But also like you need to play the game. You need to have volume. And like that is what Meta even says. Like they, you know, at the Meta Summit this week, they were saying like you need a lot of creative. You need a lot of diverse and creative. Like that is how you're going to win. And so brands that aren't doing that are going to fall behind. There's just no doubt about that.

27:37 - Full-Funnel Message Match and 45 Landing Pages

Interesting. So you mentioned to basically use the 20% real, like authentic, human-generated creatives as primarily top of funnel. And then 80% AI-generated creatives for bottom of funnel, right?

Yeah, exactly. Like you take aspects of what is being said in the top of the funnel and you turn that into ads. And then subsequently you take those angles that you learn are working, you turn those into landing pages. And so like this all goes back to the humanity component of it. When you're having a conversation, people don't like to sort of have whiplash in conversations. Like if I started talking about penguins right now, you'd be like, what the fuck? And so like that is to say, like when you see an ad at the top of the funnel, you know, you're selling AirPods. And you're talking about how great, you know, these AirPods are. And then you have a bottom of funnel ad that talks about like the sound quality. And then you go into a landing page that like has like information overload about like all these great features of the AirPods. But I only really care about sound quality. I'm going to be like, ah, fuck this. This is like information overload. Like I came here because I wanted to learn about sound quality. And that information overload is a heuristic. It's a mental heuristic. If you're presented with too much information, you're going to shut down.

And so like this goes back to, you know, what we were chatting about before, where it's like you need to have your ads now go to a landing page that's conveying the exact same information that your ads were conveying. And that way it feels like you're having a conversation with the customer the entire time. Like, listen, I'm not an email guy, but like the next step of that process is like if we're talking about sound quality and AirPods, you know, you should be having an email flow talking about sound quality and AirPods that is only going to the folks that are going to that specific landing page. And it's a pain in the ass to do it. But I will say, you know, Gruns got acquired for $1.2 billion and they had 45 landing pages. Like one was talking about kids and their kids in pediatrics and, you know, the pediatric doctor saying like why you need this. Another was talking about GLP-1. And one was talking about pooping, right? Like they had 45 angles. They had 45 landing pages. Like it works. And like now with AI, it just like feels like stupid not to be doing that.

Yeah, exactly. I was talking about exactly the same thing on, I think, the last podcast or one of the last podcast episodes. How I think with where AI is heading right now, how the e-commerce brands that will win the most will have the most custom tailored funnel for every single avatar, for every single person. Because previously, like you had a lot of different avatars that you were targeting with your ads. You would send them all to the product page, maybe to a custom landing page. But the fastest growing brands right now, like Gruns that you just mentioned, they have for every single avatar, for every messaging angle, for every approach, they have an individual ad. They have an individual landing page. I think the next step will be to also have like custom email flows and email automations for those audiences so that the people still feel, they basically feel that it's personalized to them, that it's custom tailored to exactly what they're looking for. Because that has always been how you increase the conversion rate. Like if you had the same headline in the ad as the headline on the landing page, you would always get a higher conversion rate. But now with AI, it's way easier to create tons of variants, tons of versions, and really show the people what they want to see.

I completely agree. And you just like, if you create, you know, three template, like Markdown files and Claude Code, then you can just like swap them out. You know, add different images, add different copy. The rinse and repeat is almost too easy.

31:52 - Starting From Zero: The One-Day Playbook

Yeah. And you mentioned that the brands that don't adapt to that will get left behind over the next few months, maybe years. If a brand or a founder is listening to that, doing maybe 3 million, 5 million per year in revenue, Meta is their main channel. And potentially they felt the Andromeda update, or the return ad spend since then has declined, but they still don't really know what to do about it. They don't know how to approach like incorporating AI into their process. From your perspective, like how should they start? What should be the first three things that they should do?

Yeah. So like effectively starting from zero and like assuming they haven't even really used the tools at all.

Exactly. Maybe have used ChatGPT a little bit, Claude a little bit, but just like asking questions and just the most basic things.

Yeah. So the first thing I would probably do, and again, like goes back to the sort of the three elements of the equation I was talking about earlier. It's like the social listening, time series analysis, and gap analysis. So the first thing I would probably do is go into Claude Code, or even Claude chat is fine. And I'm sure you probably have all this information, but put all of your reviews in a folder, put a post-purchase survey in a folder. If you have founder interviews, put that in a folder. Put all of these in like a singular folder on your desktop, right? The idea is to like try to not overwhelm with information, but provide information that will ultimately lead to insights. So you have reviews, post-purchase surveys, founder interviews, whatever. Connect that folder to Claude Code or, you know, create a project in Claude chat and put all that information in that project and then say something like, I need to figure out like who my core buyers are, right? Let's just start from zero. Eliminate human bias from this as much as humanly possible. Give me, you know, eight sort of core buyers, personas. For each persona, give me two to three angles that they care about the most. And then for each of those angles, give me two or three emotions. So you have like a bit of a tree, right? You have like the trunk is the personas, you know, the branches are the angles and the leaves are the emotions. And so from there, you can sort of say like, okay, these are all of my, you know, core buyers, like who I should be focusing on, etc.

The next step that I would take is, you know, use Motion or, you know, Motion is a pretty good MCP now. But again, like I said, I built a totally free tool, you know, adlib.getskipper.ai. And like that will show you where the holes are in the ads that you're running right now, where you're effectively seeing like, oh, you know, we should be running more of, you know, this format or this emotion or this persona. And that's sort of like step two. You know, take that information that you just collected, also put that in your Claude folder.

And then sort of finally, if you want to get like a little heady, you know, ask Claude, how do I connect Meta to this chat? And it's actually quite easy. You just hit connectors and you can just add Meta now because the MCP exists. And then you can pull in all of your information from Meta. And in the same way that we just tagged everything with persona, angle and emotion in our first step, we can do that again with the ads that we're already running. So we can say, pull in all of these ads, tag every ad that I've run with persona, angle and emotion, and correlate them to like CPA or ROAS or, like, CPL, like whatever your sort of North Star is. And then you're going to have a ton of information about like what ads I need to be creating, what ads I keep creating that I'm wasting money on. And ultimately, like how much more volume do I need to sort of test?

And then I think that's a really good starting point because then at least at that point, you have really good analytical insights as to what your next step should be. Don't worry so much about like the creative generation and like don't get shiny object syndrome. A lot of these folks are over-promising and under-delivering. There are a hundred percent ways to go from zero to one in generating AI creative, but like the creative strategy is by far the most important part and like what AI can do really quite well. And I don't even want to call it strategy. It's like AI, you know, analysis or just like data synthesis. So that's all to say that this can probably be done in, you know, a day. And that will give you enough information to sort of at least determine like what should my next steps be.

37:06 - Team Structure in an AI Creative Org

Yeah. And with that kind of process, and obviously with developing that process further, how do you think does the team structure for a brand in-house or also for an agency change?

Yeah, I think there are probably a few different components there. I think you sort of have builders in the sense that you have folks that are taking all of those insights and saying, okay, like these are the ads we're going to create. And like, these are like how we're going to create the ads, and like using these tools to like effectively be builders and build the ads for you. At the same time, like AI video editing is sort of mid. And I will say like, we still, you know, have plenty of video editors. But in terms of like static creation, I will say that, you know, at least internally what I have is I hired one person to just like be taking these insights that I'm sort of outlining now, you know, internalizing them and then generating static ads from that information. And of course we have, you know, based on that sort of pile of information that Claude is initially analyzing, of course we have steps in between where, you know, the AI will help generate a brief and the human can review the brief. And if they approve it, then it moves to the next step, et cetera.

But yeah, I think especially at least right now, as it pertains to static ads and gifs and animations, like you just really need probably two or three people who just like know these tools really, really, really well, like know how to develop skills, markdown files, whatever you want to call it, and have the ability to sort of synthesize what the AI is doing and generate, you know, 20, 30, 40 ads a day based on that. And then that's like on the AI side of things. And then on the human side of things, you have someone that is taking that information saying like, okay, you know, we need to be focusing on a mom in her thirties who just had their second kid. Like, let's go find a creator like this. Let's generate a script that's like two to three minutes long and let's run that as an ad. Right. So it's like there's sort of two paths you can take from that data synthesis.

39:44 - Lightning Round and Where to Find Will

Yeah. To end this episode, I have two, three lightning questions. So just answer them as quickly as possible with like one or two sentences. What is the most overrated thing brands spend their time on in their Meta creative strategy right now?

Iterating, I would say. Too many ads take an existing winner and saying, I want to replicate this winner. Like let's create 20 iterations of it. That sometimes works, but I would say a lot of brands over index on that working.

Yeah. What's one brand outside of your clients that you think has really cracked creative strategy?

I mean, Gruns, you know, it's like, they are my North Star. Like if I didn't listen to the podcast interview with the founder after the Unilever, I would never consider like we should be doing landing pages at scale. I've always sort of had that hypothesis, but now it's like a no brainer.

Yeah. Your tool Skipper is in beta right now. Where do you want it to be in 12 months?

Yeah, I think we'll have, you know, in June we're coming out of beta, you know, we're fully launching it. We're doing, having a launch video, et cetera. We need a launch party. We'll see.

Nice. Yeah.

Yeah. New York gets expensive though. I think in 12 months, like listen, the whole point of having Skipper is for someone that doesn't necessarily want to take the time to like learn how to generate AI creatives, but they still need the strategy. So like what Skipper does is it does the social listening. It analyzes your Meta account, et cetera. Like it does everything that I was just outlining and it will generate the ads for you. You know, you have less control than like having a human in the loop running through Claude Code, but it's like a really good starting point. And like for brands that just like need a lot more creative, it's like a very cost-effective way to just like get creative and diversity.

Yeah. Awesome. Will, really appreciate you hopping on. It was great talking to you. And every time after I talk to you, I feel behind on AI. Where can people follow you and learn more about your stuff?

Yeah, for sure. Mostly LinkedIn is my platform of choice. I'm on Twitter. I don't post that much. You know, I doom scroll at a fair amount. And then my newsletter, which you can access via, you know, my LinkedIn as well. But yeah, newsletter is every Friday. So if you're not part of that, like I give away a lot of these workflows for free.

Yeah, it's super valuable. Cool. Cool. Thank you.

Thanks. Yeah. Thanks for having me.