Go back

How an AI Army Replaced His Ad Agency (Up to $150K/Month in Spend)

from eComFuel Podcast

47m 28s

How an AI Army Replaced His Ad Agency (Up to $150K/Month in Spend)

Mariano Patino, co-founder of Airlock, shares how he leveraged AI to transform his business operations, moving from manual, agency-dependent ad management to a fully automated, scalable system. After leaving Shopify, he built Airlock to solve air quality issues in boats and RVs, and now uses AI agents—especially Grokbot—to generate hundreds of ad creatives per month, from copy and video to campaign deployment. The workflow integrates specialized tools like Stefan Brain for high-conversion copy and Higgs Field for authentic b-roll, with AI bots handling scheduling, quality checks, and performance monitoring. Key guardrails prevent costly errors, such as misconfigured ad spend or incorrect targeting, while human oversight ensures brand alignment. Automation has reduced effort dramatically—video editors can now produce endless batches, and campaign management is fully hands-off. The system is supported by robust APIs and real-time analytics, including buyer journey tracking and on-site CRO insights via tools like Post-Hog and Triple Whale. Despite early concerns about platform restrictions, AI operations are now stable and effective. Mariano emphasizes that the future of marketing lies in AI-powered automation, where humans shift from execution to strategy. Airlock’s mission to improve air quality in all boats and RVs is both a product and a reflection of broader consumer health awareness. This deep dive illustrates how AI is enabling seven-figure brands to operate at scale with minimal overhead, while maintaining control and quality.

Transcription

8144 Words, 42724 Characters

English
Hey, it's Andrew here and welcome to the EconFuel Podcast. These should dedicate to helping seven and eight-figure brand owners, build incredible businesses and amazing lives. And, man, have a super addition of AI Geekery today. This one was fun. I'm joined by Mariano Patino Paul, who is the co-founder of a company called Airlock, where they make air purifying and filtering systems for boats and RVs. His background, he used to work for Shopify and transitioned into running this business. And he's a member of our community of the EconFuel private community, and just had been mentioning a lot of really fascinating things about how he has used AI, and specifically more recently Grockbot used to use some other models Grocksis we've been using now, to almost completely automate six figures and monthly ad spend, pumping out and putting out hundreds and hundreds of ad creative per month. They used to have an agency do it, now they do it in house. And I wanted to go deep with him on how they do that, how it works. So we talk about how he built up the management platform for all this, what the tech stack looks like, the process for creating the scripts for creating the copy, the video, how they manage everything, the guardrails they put in place for this, because a lot of these campaigns are completely managed, what team members he has involved, and a bunch more. It was really fun. And he talks about some really unique tools, one that I thought was fascinating, was a tool that actually makes LLM's right phenomenal direct response copy. If you use a lot of these tools, you know that the dream is that they're going to write great copy, but it's almost always, even if you spend the time to get it to work well, there's tells there. And the copy, it feels like an LLM, right? Like I spent a ton of time training a voice model or training it to write my voice, and it was better, but still not great. And I had to go back to writing it myself. Anyway, here's some cool resources that they're using that's worked phenomenally well for them and a bunch of other things. Anyway, a lot of fun on this one. If you're into paid or if you're into AI or both, I think you're going to enjoy this one a lot. So we'll get into it in just a second. One thing before we get going, the hardest part of running e-con business isn't working hard. It's knowing where the work actually moves the needle. And that's why we built the benchmarking guide for e-con owners. Every year we survey 300 store owners inside the e-con fuel private community. It's our community for seven and eight figure owners representing over $3.5 billion in revenue. And we built a benchmarking table, which breaks their numbers down by revenue cohorts. You can see exactly where you're outperforming and where you're leaving money on the table compared to businesses your exact size. It comes with our full annual trends report. And it's free at e-commerce fuel.com forward slash blueprint. All right, let's get into it. All right, Anna, welcome to the show. So good to have you. And I, man, I'm going to struggle this episode because you've got such a cool story. You just came back from China, which we talked about. So we had a little fun conversation about the US versus Chinese dominance who's going to win the next 20 years of super power contest. I could talk about that for an hour with you. We could talk about you grew up in Peru and, you know, kind of that story became a US citizen recently. We could talk about that for an hour. But so I'm going to have to really, really be disciplined here to keep it in on the AI and the paid traffic side of things because there's so much there to talk about. But that being said, I want to quickly touch on talk about you started. You worked at Shopify for, you know, a year or two before jumping into airlock. And I gave people a quick heads up in the kind of preview of airlock, you know, you sell verification systems for boats and for, for RVs. What made you jump from Shopify to do your own thing? Shopify's a pretty good gig, great company to work for. Why do you transition? First of all, thanks for having me under its grade to it here. Long story short, I'm my business partner and just kind of like me, or you know, story of airlock. My, my business partner had been in the industry for a while. So he identified this need and I was kind of like a technical kid. I used to work with him in open projects while I was studying computer science and we ended up coming up with this idea for an air purifier for boats and had it on the side for a few years as I was studying computer science at the University of Florida. Then I became an intern at Shopify working on the mainly on the point of sale on the mobile app with React Native and very interesting stuff. Then I became like a software engineer, a higher full time at Shopify. I was still studying at the University of Florida computer science and still having airlock on the side as a side project that was very interesting and then there were massive layoffs of like 20, 30% people and I was the chosen one with a bunch of people on my team. So I was laid off the answer and then I had two choices. I would either go back to be an engineer for any other company tech or I would jump in at airlock. So what I did is I decided to do a master's degree on the side on computer science while I would be testing out this idea of working full time at airlock which was showing potential. So it went well, I dropped the master's after a year and now I'm here. Dropping out of school almost guarantees that your business is going to start up as a succeed wildly, right? I think that's a hack for you to look at all the big people, you know, jobs and suck it everybody. I mean, I advice people to stuff on the side and then I think that's good advice. If it works, then jump and I had a plan B like my plan B was going to be was laid off from Shopify. I have a master's degree of computer science and then I would be working right now at any software company running agents, right? So that's the other thing and it's very interesting when I think back of this and this was my layoff was three and a half years ago or so. Probably the stuff I would have learned in those three and a half years has already been fixed by AI. Like the level at which AI coding agents are today are, I don't think I missed out on much learning myself because now AI is writing like 99% of that code which is very interesting because the human is becoming an orchestrator of bots and that's a whole different scale that humans are learning and creating on the run. So in the, it turned out well for me, but yeah, I was I was able to see answer it. Yeah. No, it worked out. Super well. I, before we jump into the actual very needed gritty of the architecture, how you're doing this. I'd love for you to kind of frame up maybe the before and after using whatever metric is most helpful in terms of maybe, I don't know, maybe it's hours spent, maybe it's in terms of creative pumped out per week, but what are one or two of the numbers or metrics where you can say the easiest, quickest snapshot to say here's what our life was like before when we were running meta and Google manually. And here's what our life is now that we're able to accomplish kind of using the AI system that we have. Is there a metric or two that come to mind in terms of that? For sure. Well, basically, I think if a video editor was able to do five video batches per week, now they can make endless. That's a point because it doesn't matter anymore because it's fixed. I think it's like the problem now, if it can be done is if we should do it, would we benefit from making the bar create 100 different batches of video creative that are going to go live on meta, would we be able to put the at spend behind to give a proper test and would we set up the campaign in the correct way? And would we choose the angles we're going to test the different styles in the right way? I think if I would tell my video editor to press a button by tomorrow, I'm sure we would have 100 videos, probably they would not be excellent as if I, if himself had edited them, but they would be like 80% their antirectional would give us data by all the data on what works or something else. So that's the way I'm seeing it right now, like the ceiling has grown by a lot. I was talking to the other way with a friend and I was telling him that I once had an employee told me like, I don't feel as passionate as you are AI and my response is, okay, I give you a point like I'm very passionate about AI. But the fact is the ceiling, like the barrier has rising, like now this can be done. So the bar is up there when it used to be right here. So either it's you or it's going to be someone else, but companies have just realized this is possible, obviously, now is, so that's the new expectation across all areas, right? Yeah, let's talk about the tech stack you have built up, I, you know, this is maybe six months ago or so, but I know a while back you had a custom, I think you were using open claw that was connected by API at a minute ads and Google ads and Amazon Shopify, Clavio, Osana, Triple Whale, and kind of you managed it all from your phone by telephone. hologram slash slack, which is pretty cool, is that, you know, is that still kind of the tech stack and the, how is it orchestrated right now? Give us maybe a 30,000 foot view of what, you know, how everything looks in terms of the text layout. Right. So about four months ago, we switched from OpenClaw to Hermes agent, which is by a French lab, I believe, and just had some memory improvements. But what we've been running for the past two months is Grogpot, because it was definitely a breakthrough in AI agents and how they, how useful they are and how trustworthy they are. OpenClaw and Hermes used to have issues in which the, it would break. It would stop working. It was not very consistent. But they did in Grogpot, which is very interesting is that they, instead of having, like people were wondering, should we have one agent or should we have a thousand agents? So what Grogpot does is there's one agent, very, very specific area. And that's all the agent notes and learns how to improve on. And that agent has a sound dedicated computer. So that's the, the tech stack we're running right now, like playing Grogpot, like the business button. And we have our bot for everything, like just yesterday I created a bot whose only job is to improve our sign up rate in the, in the pop up. Got it. So you've got all of the agents are all run by Grogpot. And then they all integrate with all of the other platforms that you talked about, but Grogpot. So you don't have a custom cloud code app that talks to via API anymore. Everything is Grogpot because the headquarters probably have one Grogpot agent that runs them all. And then, and then everything integrates via integrations with Grog. Okay. That makes sense. Yeah. Exactly. So what's very, what's very nice is that it's been very easy to transition from open cloud to Hermes to, to Grogpot. You, you just tell your bot, give me a secure way of passing over all these API keys to a new bot and instructions on how you use them. So yes, what we're running now is Grogpot through connectors. If they have them and if not through API keys, it's been very nice to see every single software has created very good APIs and MCPs for botsees. Got it. I love to have you walk me through the entire process for putting, maybe we can start with a video ad and then we can do a static ad later. What is the whole process for going from, you know, concepts, script, copy, creating the video, publishing scheduling, all that kind of stuff? Yeah, take us on the journey there. There's actually been a breakthrough for me in the past couple of weeks, which is that I found out about this platform called Stefan Brain, which is a, like a custom LLM trained by these direct response copywriter and marketer, very talented guy, both Stefan, Georgie. I believe when he came up with this platform, basically like a cloud, plain cloud and plain open AI, don't write good copy, then you could get into what makes good copy, is it going to convert? In my opinion, if you just open a cloud and tell it to write an ad for this product, the ad is going to be bad and the AI always starts to close the loop and be smart. So it didn't work. We tried that for a while. Like, like building it ourselves, building the skills ourselves to make it a good copy, have it study copywriters and previous ads. We tried that, we didn't succeed at that. I think these guys succeeded. And if you open Stefan Brain, he's platform and you put your products and brand, I think that copy is at a decent good level. It's so much better in this plain cloud and open AI. So what we're doing now is we're connecting, well, a body is connecting through the MCB to Stefan Brain that's generating the brief, of course, based on angles and formats and whatever one I do. And then for statics, it's easier because we have these, well, actually, this is run by my graphic, my designer, who is now like a bot artist through to himself. So I don't have access to the platform myself, but I understand how it works. So basically, it's trained on all of our product, my yes, brand guidelines, et cetera. And basically the platform in this case, Stefan Brain proposes like a copy of the, whatever should be on the creative, right? Like whatever should be on the static. And then I think that uses chat GPT image generation under the hood. And then for any minor tweak, the graphic designer does it himself with just Photoshop and generally feel. And on the video editing side, I know there's been like a breakthrough in video editing in the past two to three to four weeks in which people found out the latest and throughout big model was good enough at editing. And it seems like it is. We're still cracking it. Actually, look, my, our video, my video, I don't know, it was also very tech savvy. He's goal for next week is to come up with the final project in which we can write an angle and script and then have it edited 90% there. And the cool thing is we know initial tests. The output is not just the video, but it's also the edit double file. So a human can always come in, fix it up and then ship it. But it's, so after the assets are ready, I'll stop you just for a second. So on the video editing side, so using like, fable to actually create the video footage itself, the, I don't be role is something you've had a lot of success. Right. So we have, so we used to use this platform that would categorize the footage and the bureau, right? Like just labeling, if you, this would be a laptop podcast, Maria, no, Andrew, microphone, right? And then this was a very expensive and very good platform, I think we charge. But what happened, there is a like theaters wouldn't use it. Great stuff for it. By the way, amazing stuff. So basically, these, my video editor has bipcoded something very similar to that. So the AI has the individual footage labeled. And it can edit it itself based on that, right? And it's not just for the clip, but it can also see different frames and very accurately choose the right clip. And the beautiful thing here is that this is just the worst is going to be, right? At this point, standing time, this is the, the worst is going to be. So my brother, for example, yeah, forgive me for interrupting our, just for clarifying. Are you saying that you have a lot of b-roll that you've shot of, let's say like boats in the harbor and Fort Lauderdale or whatever. And you have that as kind of the source material that then you can hand off to AI to be able to create more b-roll. Or are you saying that, are you saying that, are you actually generating the b-roll in a lot of your video footage from absolute scratch from from from a AI agent? It can, it uses it from scratch. That's a good question. But again, like meta is supposed to identify when like a certain video is the same over and over again, and you're not meant to use the same, like the same picture video across a lot of hats, because then it groups them into its own entity and then it doesn't test property, whatever. These cloud priorities also connected to the Higgs field MCP, which basically generates. It's excellent at generating both interiors. For example, like three, four years ago, I would need to send someone find a boat in Fort Lauderdale, get permission to get in, put our products on the table and record it, which was just crazy. You know, it was, it was very, very hard operation, especially at our size. Now it can generate beautiful b-roll of our products inside a boat or inside an RV, the exact same RV we want in the exact same conditions. And it is excellent b-roll. For humans, I know sealands, which is a model, a video model by Tik Tok or Python. I know it's very good and I know there's people who use it. I think it gets you 95% there, but for our audience that is very authentic. We sell electronics that cost on average, like $300, $400. So we, at the moment, we're staying away from AI generated humans just because we think our audience doesn't perceive it well, but I'm sure it's going to get there. And I'm sure there's people doing ex--we have exence results right now. So Higgs field can create those. Are you using the b-roll that you have? Let's say you've got a 40-foot long win of b-roll, you want to get some b-roll. Are you just saying, "Hey, Higgs field, go ahead and create this for us." And it's smart enough to know that this is kind of what a 40-foot win of b-roll looks like. Or do you have-- So you feed it with e-mails. So there's, for example, let's say you wanted to get a car. So you go into car max, find a specific car, grab all the e-mails, and then you would give that to this specific model. In this case, I think clean is very good at that. It's cheaper than. see that. So that's why I think we're using Queen more. But that's kind of like a solve problem. And that was probably the first video generation problem that was solved about a year ago. So it's been, it's been, it's been amazing to see. So to answer your question, yes, we're, we're grabbing the, the, you might get the reference email, yes, and giving them to them. Got it. Okay. So you've got an army of rock bots who are kind of tying into the, the Stefan Jordy brain via mcp. Let's say you've got like, hey, we've got, you know, we're going to make ads for these 20 different products today. It manages that, gets the copy, it hands that off to, to another bot, which through Higgs field or chat GPT can create either the video or the images. And then you get all these results back. What is your, at that phase, I'm sure there's like a quality control. How much quality control do you have to do to look through the results that are created and, and be able to say, hey, these are great. These aren't, work and rock even do that. So in theory, you could be able to lunch. I mean, am I, my, my barrier for these, I think I have very high tolerance for what, for putting out stuff that it's not very non branded. I mean, if, if I create an ad that ends up getting a lot of bad comments, I'll just turn it off. You know, I know there's people who would never take that risk. Personally, I don't think we're that important for that to be the case. And technically, if you hope create a CBO campaign in meta today and add 200 ads, meta will find the top three ads out of those 200. And I think that, that, that that seems to be true from my experience. So I do want someone looking at them, but I don't think, I mean, if we miss one or two, which has happened, I'm, I don't, I don't have an issue with that, really. Got it. And then once you have you evolve the, let's say you got 30 ad, you know, variations for, for this product, what's the process for getting those actually uploaded into campaigns and meta in Google? Is there another rock who actually goes through and just uploads those all by the MCP and starts running them based on some guidelines for budget and things? Is that also a hands-off task? How much of the campaign management is automated? We're like most of it. It's, it's amazing. So the editors, well, the graphic designer and video editor, we only have one of each. They upload the files to ad NOAA. And I know there's a bunch of alternative for ad launchers and they send them on a specific Slack channel saying, hey, these are ready to go. This is the name of the, of the batch. They don't even say it's for boats, it's for RB's or it's this specific product. They just leave them there on Slack. And then a bot takes them from there from Slack because Rockbot released these Slack integrations. That's really good. And the, these bottles, whose name is ad launcher. He finds the, the ads on, on Slack goes through a creative, it identifies issues. So for example, for Black Friday, we're extending our guardian D from 90 days to 180. And just yesterday, we were uploading an ad with 90 instead of 180. And the bot was like, hey, I'm stopping this one. I'm not launching this one because it says 90 and it's just a 180. And that's the graphic designer for him to, to edit. So after the, it's on ad NOAA, the, this bot goes through all the video, all the creative. Then it hands it off to bot who's on the job is to write like the headline and the description and all that. Then there's like a third bot for quality control, who I call the boomer bot, which basically reads everything to the boomer bot. Yeah, it represents my, my, my ideal customer. So basically that body is trained on our customer feedback, how they speak, how they talk. And they, and if any of the ads over the headline or description say something that would trigger an ascetic boat or RV owner, which happens a lot, then it stops in and provides recommendation. And so that's kind of like a quality control. And then the bot coordinates with our main media buyer bot whose name is suck. And suck the sites where in meta is every ad going. So which specific campaign based on, of course, rules we've set. But it's pretty good. You know, like from the five of the editor consent, the most black and he doesn't even say this are ads for this specific product and for a boat or for an RV. And I, oh, I say it's in. Okay, go ahead, lunch. That's, that's all I do. Used to be so much more complicated than that. And each individual ad would take pointy clicks six months. Yeah, do you, you mentioned ad nova? Can you, for people who aren't familiar with that, can you just give the quick two, three sentence summary of what that does? Right. So they are these software company who my understanding is they're trying to bring all the relevant aspects of ad creative into one single software. So they have what it would be motion frame.io, then an ad launcher and what would have been for play. So for example, for play is to find creatives by other brands and save them in boards. Motion is to analyze meta performance, specific ads, brand reports on creative frame is to give feedback to the video editor. Sorry, the video editing process. So what ad nova is doing is it's my understanding. I just share a Slack channel with them and I think they're great to work with. They have built their own version of each of these tools. So it's everything on ad nova and it's like more, I mean, it's cheaper than than than than each of those. So pretty good experience with ad nova so far. But I know there's people use a little or so far. Yeah, and at the software, right, that's not an agency enabled by software. That's just straight up software ad nova. Yes. Yeah, got it. On the grok side, I know this is a little more recent, but the last, I don't know, maybe weaker. So I know people have been saying that for a while, grok was kind of open season, very limit, you know, very loose limits on usage. And then that's been cracked down the last week or so. Have you guys run into that in terms of getting your usage, hitting your usage limits? And do you have to pay up for that or how you deal with that? So yeah, a couple of days ago, I think they reduced usage for everyone. So good software lesson or product lesson is never decrease the quality of your service users are already paying for, but they did anyway. I think Elon posted on X that it was expensive to run whatever we did run into that issue. And we had to pay credits. I think we just for yes, three, we paid like $400 or $500, which I mean, I think it's 100% worth it. You know, I didn't think twice because I don't only use it for the ad creative, but the media by ear for meta for Google run pretty much independently. The one I talk the most with in the day to days with the media for meta, which requires more like active management and study. You know, for example, in Google, it's Google ads. It's good advice is to just leave it decide and leave it on for like a week and then decide again. You're not meant to be deciding on a day to day basis, maybe as you would be on the meta side, little bit more. So that's the interesting part because the bar, my Google bar could potentially be pushing buttons the whole day forever, right? But it wouldn't be in the best interest of the area. So you must artificially create rules. In this case, you set the rules, right? Like, okay, we're gonna do our Google decisions on Wednesdays at 9 a.m. you run a deep analysis on this and you provide your recommendations. And then you run safety checks. Like, okay, every 20 minutes, you check every single campaign. And if there's anything critically wrong, turn it off. Don't worry if I'm awake. If I'm sleeping, I trust you. If this happens a bunch of times in meta, in which it starts spending a lot of money in a new ad, that it's just very engaging with a five dollar CBM. That's bringing a lot of clicks. We've had happen that multiple times of meta allocated spend on a new ad that is very engaging, but that won't sell based on the metrics we recognize. And the bar steps in and saves us. And that tends to happen over a night. Like, it tends to happen at 3 a.m. Never happens on a Thursday. Of course. And so when you said $400, was that, are you saying on average, you're spending $400 a day to manage kind of the Groc bot army that's helping out with this? Or is that just a one time? How much would you say you're going to have to spend per month on Groc to manage all this? So they changed their pricing recently, but my personal account, well, my business account, but for Mariano is on the highest plan, which I think it's $200 a month. And we had never bought any additional usage on top of that until a couple of days ago, they did this in which they reduced usage. So I mean, and then my employees have the meat to your version because they don't use it as much as Aseidu, and that one's probably like 80 or 100? Okay, so you're not spending $400 a day, you know, maybe this bumps up to 200, maybe this bumps up to $1,000 a month or something, but you're not spending, you know, you're not spending 15 grand a month on and I can I can pull back on some I mean I was I was going all in you know I was telling a bot to go and scroll to here every day to find whatever so maybe I'll stop doing that like there's bots I had that weren't as useful but I I mean I just have the usage so now I'm just gonna pull back on the on the usage but I think it's all going to zero like it's it's gonna be free eventually or extreme cheap because even even now like there's open source models that can run on a mic mini that can do any admin very easy work and it's basically free just whatever energy costs you to have the mic mini on how are you you touch on us a little bit but when you think about can I talk to about creative the quality control the deploying how you get them out there how are you thinking about campaign management and tracking results I think I mean I know for a while you were using yeah north beam hooked up there are you still using an attribution platform like that do you have a bot that just goes through and tracks how things are doing how you how you do that yeah so the bot suck he's connected to to our data that's great ZUCK I'm guessing it's like okay yeah yeah yeah sorry yeah I often tend to to make that that pronunciation mistake suck anyway he's connected to the meta at say bi he's connected to our Amazon triple well and everything you can imagine we also have like an internal tool and this is very interesting on something that AI has enabled there I mean I could always I always had the skill of coding so I there's a lot of stuff I could have done that I wouldn't do because I didn't have enough time but now coding is solved so literally the limitation is your your energy and your decision making and how you feel in that specific day and we have this specific very interesting tool in which we have a database that tracks if a specific link click ended up in someone that being our pop up saying they were a boat a boat or an RV or and this is important because we've had the it happened a bunch of times in which we have a a boating at that is bringing 70% RV years which tells you meta is incorrectly targeting like up during the more right so these RV years are clicking on the add and then they're landing on a boat landing page therefore not engaging not converting and and that just hurts performance so stuff like that I've built with a prompt to my to my coding bot and it has been extremely useful so that's a very good use case like suck is also connected to clarity which is the software by Microsoft that gives you heat maps scroll maps it's connected to our own software and we track this metric cost per pop up sign up and cost fair pop up sign up or not even the sign up so people who say I'm a boat owner and they they then they don't leave their email or we're still tracking that we know how much it costs me to get a boat owner to tap that button which gives signal to make decisions right because as we sell like a highly product we need this additional signal feeding our decisions so the body's suck is extremely good at deciding based on all these inputs and provider recommendations and as you would do with a human employee as well as you onboard it him or her there's a you said expectations you said the objective you determine what good media buying is for your own specific company and brand so for us it's stuff like this like we see that an ad is bringing a major it's spending a lot and it's bringing a majority of our years and we meant to target boat owners and turn it off for all of those things it's excellent and I trust it I trust right now in this precise moment it's he's making good decisions with our account and then he never does anything that is never does anything that could break the account you know yeah what what are some I mean you talk about the breaking what what are some of what's the worst thing the worst mistake here you're kind of AI Grockbott army has made and what are some of the guardrails that you've put in place to keep that from happening again maybe not Grockboard but my open club and Hermes in the past sending emails or responding to customer support tickets we didn't want to respond with that I mean I remember it's all that lady that we would video call her and guide her through the installation which is something we don't do so but that was Hermes or open club with Grockboard it has not happened but I mean the risk would be in stuff it typically does like for example that risk would be that suck right now decides to for our cost gap at $5,000 right I think it's extremely unlikely it's gonna happen unless per the guardrails suck checks every 15 minutes the entire performance on a account and if anything is off he turns off the at the uncertainty company got it what what's the human team involvement here how many people are needed to to run this machine I mean obviously you're involved you mentioned you have I think a video editor give me how many people are tied into this so it's 12 of us in total but dealing actively with agents for sure I'm the one most involved with it I mean and for example the the email manager has her own boss for email and the affiliate TikTok shop manager has her own bots and the video editor runs his own system for video generation the graphic designer has the same but the question there is like yeah now that now that agents can behave as any human employee which they can and have been able able to do it for the past month or two like three months ago or four I mean at least I I wasn't running them as if they were as good as a human right the question is how are humans gonna work with along agents in the in the future right because it's just so obvious that that's where where the world's going what is you know some of the challenges are things you worry about I know that kind of in researching this looking at sometimes meta sometimes Google don't love having a agents run a bunch of these things and they can potentially throw up some red flags or I know if they're going to cancel your account but that's a concern have you run into issues with that being the case what kind of things do you do to prevent against you know shutdowns or those kind of problems so I found that so far the APIs of each of these platforms are very good and very comprehensive and they do let you allow the about to do specific things and not others I think for example you know Amazon doesn't let AI agents shop on Amazon.com and shop if I it does it's probably a business decision right because they're not taking care of the the customers best interest and experience right it's they're making a business call that totally understand though but on the I'm really not worried I think all software companies have had to improve their APIs so they are very easy to use for bots I would even get to just telling you that my me looking at the meta cyber screen has gone down by 90% you know so eventually people are not gonna even open those platforms anymore you know it's all gonna be done through through the AI. Talk about the buyer journey reports that you have that you used to surface insights I thought this was cool. Sure so the bot has well boomer has access to our triple whale buyer journeys and this is not through the API so because each of these bots has its own computer I mean they have a part of a computer technically they are able to see the screen just like you can see it right and it can click it's being one of the things AI has traveled to do the most for the past few months just clicking on a screen because it's not a natural UI for a text-based LLM it needs like some sort of like vision and understanding of UI I remember hearing that OpenAI is gonna release a model or something like that it's gonna be a fixed problem by the end of the year and I truly felt so basically in these occasions in which as far as I'm concerned the triple whale API also released those buyer journeys the bot goes one by one stores them at an uncertain format in our database and then for example we know that Andrew visited us for the first time through these static meta and then open a claim your email and then add a card but then buy it so we can track the whole buyer journey through our agents which again it's something I I always knew was there and I would check it from time to time but I wasn't doing a very proactive job at understanding what was happening with how I could apply those learnings, right? It's just like a feature in Triple Web. On top of that, the API is connected to the API of new commers, which runs our post purchase service. So every day, we get a report on anything interesting, our customers have done, said, etc. It's extremely useful, right? Because it's data that was always there. I remember opening it and saying, "I should open this more often," right? But then now it just pulls it on your face. Yeah, we're doing something similar, not on the buyer journey from a paid-out perspective, but in terms of on-site CRO. You always have a lot of that data, but I think how many people really dive into their site analytics on a regular basis? No one does, right? It's a treasure trove, no one does. We started using this tool called Post-Hog recently. One of the cool things about it is it will watch, it has all the analytics and session data on your site, but it will go through and watch it automatically and be able to surface a report that says, "Hey, we went through 400 different sessions over the last month. Here's the problem in your checkout. Here's the problem in your application page. Here's what people are looking at or they're clicking on this so it'll give you, which is that's so valuable." Like, who trucks if people drop off under PDP on the second carousel image and not the third, you know? Which is like a very important thing. Like everyone sees that carousel. Everyone scrolls through it, literally 100% of your visitors are probably. No one knows if the second or 30 image is more effective at generating interest, right? And it should be iterated on a regular basis, just because of like now the limitation is just me, Mariano, deciding that's important, that's something we should do, right? But for example, on the Google SEO side, AEO, if someone asks Gemini or opens Google and says, "My boat smells, I want Gemini to recommend." So how do we get there? Because I know a lot of people are using Gemini through just Google.com. And you can get there through good SEO practices, which are extremely boring and they're like making sure that code in your website is indexing properly and then you're clicking this bottom on the search console and creating blog posts that are formatted in the right way for Google to see or, you know? Just like, stuff that I brought up, I mean, I wouldn't have hired an agency because it's not as important to me, but it's something very nice and my boat runs that, because it's very low risk, my boat runs that full time. Like I just received a report every then and it's like, these week out of 100 people who ask Gemini if they're why they're both smell, they, Gemini recommended us 50% of the time, which is awesome. And I think we're connected to that through something called Open SEO. - And that'll actually track the kind of the LLM rankings for different search queries for your site, Open SEO. - That's right. We just, which just recently, my recommendation of a boat because what people would use until today would be RIFS, I don't know how to say it, A-H-R-E. - Oh yeah, I don't think anyone knows how to say it. They've been around for like two decades. - You're still in the bad luck. - But I don't make a name, people know how to pronounce. - A-H-R-F, A-R-Fs, nobody knows. The refs, yeah, everyone knows you're talking about. All right, this is awesome. Thank you for doing the dive on this and how deep you're going. People, I mentioned at the top, of course, but airlock. Can you give a quick shout out for, I know we've got a handful of people listening with RVs, with boats that probably are smelling a little manky. So give quick a shout out again. Who you could fit for, if anyone's got one, working, they find your stuff. - Sure, so our goal is to fix air quality in every single boat and RV in American the world. Some boats, of course, don't have enclosed spaces, therefore no use for us, but our goal is to improve air quality. It's something so important. I think people are starting to pay attention to how important air quality is and how it affects your health and how you feel and how just so important, just like water quality and food, it is an extremely important part of health, really. Boats on RVs tend to suffer from smells and that's what's really annoying for people, right? And smells typically mean there's something wrong and that's what we aim to fix with every single product we release. - Remind people about the website, if you would, where they can check that out. - Sure, our website is airlockUSA. That's a-i-r-l-o-c-k.com. - Go on USA.com. - Perfect, thanks. Well, sir, thank you for coming on doing this and also for being in the forums, you've been such an awesome forum member like the reason I reached out to you initially is because you added a ton of value and a number of threads about talking about this, things you were doing and yeah, you've been adding a ton of the conversation there. So if you are in ECF, John Mariano, a little line say hello, say thank you. And yeah, man, thanks for being part of our community and for sharing all the cool stuff you're doing behind the scenes, it's a pretty exciting time and it's fun to geek out about with people that really know what they're doing. So thank you. - It is, it moves so fast. I'm sure in a few months it's gonna be a whole different story and every single company in 2008 or 2012 version of this because it just makes so much sense. Once you're, you're, you're, you're into it, you know. So it's refining everything and yeah, thank you so much for the invitation Andrew on any time. - How quickly things go out of date, you're right. Like when I record episodes like this that talk about AI, I always push them up to the very front of the publish queue 'cause I'm like, man, if I wait three weeks, I get a chance to third of this is just not gonna be relevant which is wild. So yeah, we'll be pushing this one live very shortly. So anyway, man, thank you so much. Look for it, hopefully getting to hang out in the future and happy rocking. - For sure, thank you so much, Andrew. Thank you for having me. (upbeat music) - Before I let you go, we asked 300 store owners to rate their financial knowledge on a scale of one to five. 80% of them rated themselves a four or below. And here's the crazy thing, the jump from a four to a five is where everything changes. Owners who made that lead from a four to five had nearly double the financial run rate. Not 10% more, double. And by runway, I mean cash in the bank. That last step from pretty good to, I really know this stuff is one of the most valuable jumps you can make as an operator. The financial mastery series that we built out at Econ Fuel is built to get you there. It's a free course built for Econ owners and you can grab it at ecommercefuel.com/mastery. And if you're a seven or eight figure owner who wants to be around other owners at this level, thinking about this day in and day out, you wanna check out our community at ecommercefuel.com. Work hard, adventure often. And I'm looking forward to seeing you again. Next fight.

Podcast Summary

Key Points:

  1. Mariano Patino, co-founder of Airlock, transitioned from Shopify to launch a business focused on air purifying systems for boats and RVs after being laid off and leveraging his technical background.
  2. Airlock now uses AI agents, particularly Grokbot, to automate 90% of ad creative production, including copywriting, video editing, and campaign management, drastically increasing output volume and reducing manual effort.
  3. A custom AI workflow integrates tools like Stefan Brain (for direct-response copy) and Higgs Field (for realistic b-roll generation), with human oversight only for quality control and final approvals.
  4. The automation pipeline includes AI-generated creative, Slack-based handoff, bot-driven campaign deployment, and real-time monitoring via bots like "Suck" and "Boomer" that enforce guardrails to prevent costly missteps.
  5. Quality control and data tracking are enhanced through internal tools that analyze user behavior, buyer journeys, and conversion funnels—enabling data-driven decisions on ad performance and site optimization.
  6. Despite initial concerns, platforms like Meta and Google have not blocked AI-driven operations, and APIs are now robust enough to support full automation with built-in safeguards.
  7. The team maintains a lean operational structure with only one dedicated human in each creative role, relying on AI to handle scale, allowing Mariano to focus on strategy and decision-making.
  8. Airlock’s long-term vision is to improve air quality in all boats and RVs, aligning with growing consumer awareness of health and environmental factors in personal spaces.

Summary:

Mariano Patino, co-founder of Airlock, shares how he leveraged AI to transform his business operations, moving from manual, agency-dependent ad management to a fully automated, scalable system. After leaving Shopify, he built Airlock to solve air quality issues in boats and RVs, and now uses AI agents—especially Grokbot—to generate hundreds of ad creatives per month, from copy and video to campaign deployment. The workflow integrates specialized tools like Stefan Brain for high-conversion copy and Higgs Field for authentic b-roll, with AI bots handling scheduling, quality checks, and performance monitoring.

Key guardrails prevent costly errors, such as misconfigured ad spend or incorrect targeting, while human oversight ensures brand alignment. Automation has reduced effort dramatically—video editors can now produce endless batches, and campaign management is fully hands-off. The system is supported by robust APIs and real-time analytics, including buyer journey tracking and on-site CRO insights via tools like Post-Hog and Triple Whale.

Despite early concerns about platform restrictions, AI operations are now stable and effective. Mariano emphasizes that the future of marketing lies in AI-powered automation, where humans shift from execution to strategy. Airlock’s mission to improve air quality in all boats and RVs is both a product and a reflection of broader consumer health awareness.

This deep dive illustrates how AI is enabling seven-figure brands to operate at scale with minimal overhead, while maintaining control and quality.

FAQs

Mariano uses AI agents like Grokbot and Stefan Brain to generate hundreds of ad variations per month, automating script writing, copy creation, video editing, and campaign management with minimal human intervention.

He uses Grokbot for task automation, Stefan Brain for direct-response copy, Higgs Field for AI-generated b-roll, and adNova for creative management, all integrated through bots and APIs.

He employs a multi-layered quality control system: a 'boomer bot' checks for customer tone mismatches, a bot reviews campaign performance, and human editors manually review and refine content before launch.

Mariano’s business plan on Grokbot costs $200 per month, with a one-time usage fee of around $400–$500 due to a recent reduction in API usage limits, which he considers worth the investment.

A bot automatically pulls creatives from Slack, uploads them to ad platforms, and manages campaign settings based on predefined rules, with real-time monitoring and automatic blocking of errors like incorrect campaign durations.

Yes, his AI system tracks buyer journeys on Triple Whale, analyzes on-site heatmaps via Post-Hog, and monitors Google search performance using Open SEO to surface actionable insights.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.