The $5K/Day Side Hustle Built Entirely on AI Tools

Welcome to episode 5 of the AI Operators, and happy Noah Khan album release week to those who celebrate. Ben, we as always are here because of our friends at Richpanel and AfterSell, and we are here to help everyone make sense of this changing world. In the AI space and apply what matters. It, as always, has been a hell of a week. Ben, how you living, dude? How's life?

I just read this morning about this biohacker that claims to have sequenced their own genome, uh, on at their kitchen table using like an M4 Ultra, Mac Studio, and Claude, which sounds like science fiction to me, but, uh, you know, things are happening so quickly.

It sounds like people need new hobbies. I'm doing good. Um, that sounds crazy. I might need to dive in. It's also a reminder I should probably go to the doctor. It's certainly been a couple of years, but we have a lot to talk about this week. So let's talk about what's coming up first. OpenAI launched their new ads manager inside ChatGPT and They just launched ChatGPT Images 2.0. Anthropic, we'll dive deep on this, just launched Claude Design, and we will get into it. All of that in the AI for design and creative space builds on the launch of Seedance 2.0's video model, which just dropped a couple of weeks ago. So your head is spinning, mine is too. There is so much going on in AI for creative. Anthropic pulled a surprise move and switched enterprise billing to just usage-based pricing. So some are calling this the great AI shrinkflation. We will tell you what it means for you as an operator. Klaviyo shipped Composer, which now lets email marketing teams draft full campaigns in a single prompt. PwC has shared that 20% of companies are capturing 74% of AI's value. Lucky for you, Ben's company Scale is one of that 20%. He will peel the curtain back for us and we will sit down with Bar at Boozcoos, Krishna at Sarus Analytics, and Cody from Jones Road to hear how they are leaning into and applying AI. It is going to be a big show. Let's get into it.

Some brands are cutting their support tickets by almost 50% with AI. I've actually been recommending Richpanel to my community for the last 6 months, so I'm really happy they're the headline sponsor of AI Operators. What they're building is pretty different from most AI support tools. Most tools are basically just AI agents. You train them once and hope they behave. Richpanel built what they call Boss AI. You just tell the Boss AI your values, principles, and desired outcomes. It handles the rest. The Boss AI reviews thousands of conversations every week, scores them, and continuously improves them to get your desired outcomes. Brands like Ridge cut ticket volume by 45%, and Jones Road actually cut their support team size in half, and both of them were live in under 2 weeks. Ridgepanel even guarantees 30% ticket reduction in 60 days or your money back. If you want to see what that could look like for your support team, go to ridgepanel.com/demo.

OpenAI launched their Ads Manager and the price of entry drops from $250,000 to just $50K. So. OpenAI quietly launched a self-serve ads manager for ChatGPT. The minimum spend dropped from $250,000 down to just $50K. Right now it's CPM only, so no cost per click, no cost per acquisition, no conversion pixel. Ads are running in the US, Canada, Australia, and New Zealand, and it's matched to conversation topics rather than user data. Here's why I am excited about this. They are investing aggressively. It is rare to have a full self-serve ads manager this early in a platform's life, and they need it to work. OpenAI burned $8 billion last year. They'll burn $25 billion this year, and projections have them losing close to $60 billion next year. So ads have to work. They are existential. The challenge is whether a company that builds world-class AI can also build a basic ads infrastructure. That is a different skillset. My vote is that yes. They can, and I personally think that OpenAI will become the next Meta and Amazon-style ad platform at scale. Ben, you have looked at the actual performance data. What are you seeing at scale?

This rollout was not great. The awesome thing about being one of the first to buy ads on a new platform is that the platform doesn't have much data, so they're trying to drive market penetration, get advertisers in the door. Usually what that means is cheap clicks and low CACs until they optimize and start charging advertisers more. That's not the case with this rollout. Okay, so let's look at the data for a second. ChatGPT ads are running right now at 0.91% click-through rate compared to Google, Google Search, 6.4% for the same categories. That's 7x lower. CPM on ChatGPT, $60 compared to $20 on Meta and far less on Google Display. And there is no conversion pixel, so you don't get accurate data on your conversion. So yes, it's exciting to test a new platform and you can get in there with $50K in total spend. It's not $50K a month, by the way. But honestly, I think OpenAI were a little too greedy and missed the mark for now. I personally would not allocate ad spend there right now, but when they ship conversion tracking and CPC and CPA models, which they said it's coming soon, this can change.

Ben, I don't mean to put you on the spot, but before we go to the next story, is it possible this is just a play like TikTok where, hey, you know, you're driving so much awareness through the channel and maybe you can't adequately provide it back? I go back to what Sean at Ridge shared, that they're seeing 12% conversion rates on traffic from ChatGPT, $5 per visitor. Is it possible that it's just an unattributable part of your mix?

Definitely. It could be a first-click or impression conversion sort of play where people that, or brands that get visibility on ChatGPT then convert better on Amazon or under Shopify through direct traffic or organic traffic. It's too early to say, but as an ad platform comparable to Meta and Google, they're not quite there yet.

On April 14th, the information broke, the story that Anthropic had quietly shifted enterprise customers to usage-based billing. Enterprise customers, keep in mind, have over 150 seats. So if you're less than that, this doesn't apply to you yet. Enterprise customers used to pay $200 per user per month with a pool of tokens included. Now it's $20 per seat, plus every token your team uses gets billed at Anthropic's standard API rates. Fredrik Flipsen, the guy companies hire to negotiate software licensing, estimates that heavy users will see bills double or triple overnight. That's a really big deal, and it means that flat rate plans, $20 per user per month, could quietly go away. And the timing kind of makes it worse. This landed the same week that users are calling, and I am experiencing AI shrinkflation. Anthropic has degraded its product and raised prices on the heaviest users. That is a rough week. So what does it all mean? It means that the real skill moving forward is going to be efficiency, the ability to manage context properly and not burn through tokens. I see this playing out in 3 phases across the 70 different companies that I work with on AI. So in phase 1, you just teach people the basics of the tools, the 5-step prompting process. What are projects, connectors, and skills? In phase 2, everyone goes off and does their own thing. Having had the foundational training, they start to build more advanced use cases. And yeah, they are gonna burn through some tokens. That is okay. You want people using the tools. And then in phase 3, that's when you have to bring everybody back together and manage everything properly. It probably means giving more advanced API access to engineers who know how to be more efficient with this stuff, and everybody else gets their more basic plan with co-work and automations included. So Ben, scale runs Claude across the entire business. Is the flat rate era ending for everyone?

Not for everyone yet, right? So th— this is the Uber Lyft playbook, and, you know, everyone should have seen this coming, honestly. It's, it's, we all remember rides on Uber that used to cost $8 and are now $22, right? It's the same thing with tokens. They subsidize, they focus on market penetration, they get everyone hooked on the product, and then they raise price to increase their profits. And the reason it's obvious, because you have guys out there using 10 billion tokens over 8 months on $200 a month, Max plan, which the API equivalent is $15,000, right? This specific guy I'm talking about paid $1,600 for that amount of tokens. That's just not sustainable for Anthropic. The good news for operators is one, if you're on Team or Max and not on Enterprise, nothing changed for you, not yet at least. And two, open source models are getting better and cheaper to run. So if you're a heavy user and you're planning for the inevitability of price increases that will eventually hit Team and Mac, start looking into open source or how to use cheaper models, um, more efficiently.

Ben, I, I don't want to put you on the spot again here, but in your Uber and Lyft example, you know, ultimately prices go up because the cost of gasoline is fixed or rising, right? So their costs get more expensive. In this world right now, Anthropic, OpenAI, xAI are compute constrained. They don't have enough data centers. They're not able to generate enough compute to do this efficiently, but won't over time the cost of compute and the efficiency of the models go down so that ultimately this might not be as big of a problem?

I don't have a crystal ball, but I know that for now, these companies are losing a ton of money and they need to show investors that they are They have plans to increase profits. So I think that's starting with Enterprise because they have the deepest pockets and we'll see what happens with Team and Max. But you see the tweets out there. Everyone is complaining about how they're watering down those plans. If you have a $25 account, $100 account, even the $200 account, they're starting to get watered down to push people towards API usage, which is 3 times more expensive. So something to think about.

All right, Claude, you hear it here. Take my money. I don't want my stuff to be watered down. I'll pay you $1,500 a month.

Stop.

Okay. So Klaviyo launched Composer on March 24th, an AI agent that builds an entire email and SMS campaign from a single prompt. You type, build me a spring reactivation for lapsed consumers across email and text. And in just minutes, you've got segments, you've got copy, designs, send times, all ready to review. It integrates Claude and ChatGPT under the hood. It is grounded in 14 years of data across Klaviyo's nearly 200,000 brands. And as all good AI agents, nothing goes live without human approval. It's in private beta right now, and you should try to get your hands on it. This is ultimately where everyone is going. Meta, they just want you to put in a target and their platform will do the rest. Same with Google and PMax advertising. Now Klaviyo is doing the exact same thing. Tell us your goals and we will build it for you. So the question is whether with email this will work because it is a channel that's so much more tailored and personalized. But the core argument to me makes total sense. You give it a goal, you upload context about your brand, it designs the campaign and suggests audience based on what it knows. Plus, It's recursive, so it's learning based on past performance in real time. So Ben, you talked last week about Shopify shipping something similar. How does this fit into Scale's marketing operations?

Yeah, just a great time saver, right? Just like Shopify did last week with the AI toolkit we covered, it's where every platform is going, right? As operators and marketers, we spend a a ton of time on setting stuff up, right? Like clicking through, opening another tab, pulling this list, pulling that piece of data. Now you can do it with AI, right? So you can create segments, sends, copy, design in minutes, and then you spend the rest of the time tweaking, optimizing. So it's just a great time saver. It lets you level up because now we have more time to do it.

I think your final point on tweaking and optimizing, that is how to use these tools. So everything we've talked about, think of a football field, right? You don't have to start on the 1-yard line anymore and go 99 yards. With AI tools, you can start at the 50-yard line or even in the red zone, and then you refine, you iterate, you tweak to get there. So these are production tools that help you get faster and better and start closer to the finish line of best in class.

Okay.

I am excited for this story. PwC ran a global AI performance study and saw that 20% of companies are capturing 74% of AI's economic value. I love this story because it's gonna show up in every board deck over the next 6 months. So PwC surveyed 1,200 executives across 25 different sectors, and the headline is brutal and clear: 74% of AI's economic value is going to just 20% of companies. Everyone else is stuck in pilot mode. I have lived in pilot mode for many years. I can promise you it is not a fun place to be.

Okay.

But this is not about the AI use cases that they're deploying about sort of driving productivity. It is what they are pointing AI at. The leaders are 2.6 times more likely to use AI to reinvent their business model and use it to pursue new growth opportunities and category expansion. That is what is most exciting to me. They are 2 times more likely to redesign their workflows around AI instead of just bolting it onto their existing workflows and processes. This is what Jack Dorsey at Block was saying. They are nearly 3 times more likely to have increased autonomous decision-making with humans there overseeing it all. So, okay, Ben, you are leaning into this aggressively at Scale. What does this all mean for you?

See, this is why I've been so bullish on Phase 3 at Scale, right? So a quick reminder for anyone new that's listening, at my company Scale, we think of AI rollout in 3 phases. Phase 1 is using AI to do what you already do faster with prompts and projects. Phase 2 is cutting costs by automating repetitive work. And these are both fine, right? But they're table stakes. Every operator should be doing these already. Phase 3 is the one that matters and the one I'm most excited about. That's asking, what can we do today that we couldn't do before AI existed? And that's a completely different question, right? It's not about efficiency. It's about building new product. It's entering new categories. It's launching customer experiences that weren't economically viable maybe 18 months ago. So this is what this report was measuring. So if you want to be part— of the 20% that generate 7x more value than everyone else with AI, stop asking, how can we get our team to use ChatGPT more? Instead, you need to be asking things like, what line of business can we launch this quarter using AI as a partner? Or what can we enhance with AI that would generate more incremental revenue?

Anthropic is in the building. They have just shipped Claude Design a few days ago. You describe what you want in plain English, a pitch deck, a landing page, a one-pager, and Claude generates the entire thing. We will hear from Cody at Jones Road about how he's leaning into this. Prototypes, decks, interactive UIs, all of it is powered by Claude's Opus 4.7, which dropped just the day before launch. And all of this is in research preview right now, free on Pro, Max, Teams, and Enterprise plans right now. The story is that executives and individuals are now able to bring their ideas to life and pass them off to designers quickly. Your briefs get clearer because you've started the process on your own. You can say, hey, this is what I want, and pass it off with a real visual instead of a paragraph. Brilliant's senior designer said that pages that took 20+ prompts in other tools took Claude Design just 2. Datadog has compressed a week-long brief-to-mockup cycle into one conversation. For me, the long-term question here is about SaaS. This is Figma inside Claude, right? Like Figma stock dropped 5% on the news and is down 85% since its IPO. Mike Krieger, Anthropic's chief product officer, who by the way co-founded Instagram, resigned from Figma's board just 3 days before launch. At some point you've gotta ask if Claude can do the design, if Claude code, can build the tools and if Shopify's AI toolkit can deploy it, what do all these other tools actually do? Ben, what is your take?

Anthropic is eating everyone's lunch. That's my take. Look, look, look, I'm really, really excited about this one because I think it solves a big bottleneck that many companies have around design, right? Especially for landing pages. And I think what I love most about this and what's super unique is the handoff package that Cloud Design does to Cloud Code, right? So Cloud Design packages the components, design, copy, and notes to Cloud Code to implement accurately. And that's just awesome and it saves a ton of time and it's super high quality. So that's why the Figma stock dropped. You don't really need to use it as much or at all. Now, many of these design and design AI tools start from nothing and they generate generic mockups. Cloud Design connects to your database. It understands your design system and then it outputs something relevant to your brand and then you can hand it over to Cloud Code as, as I mentioned, or to the AI toolbox in Shopify and it creates the actual page so you can have your marketers create these drafts and designs, and designers spend their time tweaking it and polishing it and elevating it rather than creating from scratch. And that transition to code is almost seamless. So that's, it's a game changer. One thing to pay attention to, if you don't want generic mockups, make sure when you set up Claude Design, link it to your actual code base, Claude reads. Your components and applies your real brand. Uh, so it makes everything, uh, on brand.

The pace of change and the updates coming from the team at Anthropic are relentless. This one is worth paying attention to. Hat tip, you might wanna check out Claire Vo's How I AI YouTube. She just did a really nice 30-minute tutorial on this that was eye-opening for me. Okay. From Claude Design to OpenAI for ChatGPT Images 2.0, I will never ever forget the first time that I got early access to, uh, ChatGPT's image generator a few months before it went live while at Crocs. We were kind of creating new Jibbitz designs through the tools, and it was truly mind-blowing. And that was a year ago, and now OpenAI has come out with ChatGPT Images 2.0. So what this new model brings is 3 things that really matter for you. One, text rendering, like, actually works, and that's a big thing. Now text is really clear product labels, pricing, user interface elements, like it's clean. Two, the tool now thinks a lot before it draws. So paid users have access to a thinking mode where the model will plan its layout, it'll count objects, and it can even search the web for mid-generation sort of creation, uh, to get its facts right. And then three, you can generate up to 10 coherent images from one single prompt. So think about full design systems or brand briefs through one prompt alone. So this is a really big deal. And to me, what's most interesting is like, this is really an internal weapon, right? So decks, brand designs, prototypes, these things that take weeks, you can now do in seconds. So Ben, what does this change for operators today?

Uh, before I talk about what it changed for operators, I want to talk about the benchmarks, because to me it was mind-blowing. Uh, I was reading about this morning, so GPT Image 2 took the belt from the reigning champ, which held the top spot on LM Arena since February. So to me, the crazy part, and with all the images that I've seen online, is the text accuracy, right? It's like near 100% text accuracy, which was a weak point for a lot of the image generation models before that, and testers are saying that the gap between this GPT Image 2 and the prior reigning champ is the same as the gap between that champ and the old DALL-E. So that's a huge, huge jump.

Yeah, I would encourage everybody go to, you know, the OpenAI announcements on this. You see it running on the screen now, like these images are crazy. They look so accurate. And I think I want to close with a Sam Altman quote from a few years ago that's worth kind of paying attention to, right? He said that 95% of what marketers use agencies, strategists, and creative professionals for today will easily, nearly instantly, and at almost no cost be handled by the AI. And that the AI will likely be able to test that creative against real or synthetic customer focus groups for predicting results and optimizing. Again, all free, all instant, and all nearly perfect. Images, videos, campaign ideas. No problem. That is a direct quote from Sam Altman. And when people tell you what they're building towards, you gotta believe them, right? And so the foundations for all of that is, is here now, right? Like you can spin up an AI version of your customer with a simple GPT, right? You know, copy is effectively dialed now. Images are rendering at, at full fidelity and automations. Like we talked about the Carpathian Loop last week. Like these things are, are just here and compounding now, right? Every run. Sharpens the campaign performance against your goals. So it is here now and it is genuinely crazy stuff. I am so excited to be joined by my friend Cody, the CEO of Jones Road Beauty. Quick story. I'm very fortunate to be able to have a lot of friends in this space, but my wife Ellen is most impressed by Cody. Uh, she loves Jones Road and more than that, I swear, She looks beautiful every time she wears it. So Cody, welcome to the show, man.

Thanks for having me. Love, love to hear it. Always love every time, every time we talk and I hear it. So I appreciate it.

Absolutely. So tell us a little bit about yourself. Tell us about Jones Road. Give us a little bit of context before we dive in.

Yeah, I'm CEO of Jones Road. We're a clean beauty brand founded by Bobbi Brown, mostly D2C. We've got a bunch of our own stores, probably about 50 people full-time, you know, not counting our store employees. Pretty, pretty lean team. Bootstrap business, family business. All right.

And one of the things I'm so excited to chat with you about, particularly following you on, on Twitter, is the way that you as a CEO have leaned into Claude, specifically Claude Code, really aggressively. So why don't you tell us a little bit about a use case that, that's been most impactful for you?

The biggest one that I've been spending a lot of time on lately, and again, I, I just caveat, 'cause people are always like, oh, you shouldn't be spending time doing this yourself as CEO. Like, I do this at night and on weekends for fun. Like, it's not like I'm doing it like during the day, um, is Shopify dev testing. Like, I just, I'm really into this, you know, the convergence of roles and being able to, you know, not just save money, but do things that weren't possible that, you know, we would normally need a 4-person team to do. And so I think the first use case that, you know, I could really find where it's like really good and usable, and I think we can do a— where I'm excited about it, I think we can do better. I think we do a better strategy than what we'd be able to do, but also much faster is like, web development, CRO, I'll call it our whole CRO process. So I think one person can do that from strategy, insights, research, roadmap, all that stuff, you know, design, dev, copywriting, analysis. I've been doing, you know, I would say 90% of it by myself. I'm going to hire somebody. We're recruiting somebody right now. So if anybody's listening, kind of going to show them the ropes and give them the role. But yeah, I mean, I'm building 2 to 3 new tests per week. Bunch of new landing pages, just cooking. But it's what's exciting is not only the speed, the quality is good. You know, the quality is, I think, better. Our website's just getting better. And our hit rate is a lot better because it's built on the right principles that I was able to build the skills on.

And so you're literally designing the webpages. You mentioned the skills. Why don't you take us one step deeper into that? What does the design look like? What is the tool you're using? And what are the skills that you're building to create it?

Yeah, so first step is strategy, right? And I'm big on like first principles approach. Like let's not just come up with something, but let's figure out what we want to build. And so my favorite thing I have is this customer intelligence markdown file where it essentially has all of the, you know, consolidated insights and data about our customer. So I have like 4 to 5 different APIs that feed that. I have Junup for reviews. We use Typeform for a lot of surveys. So I have surveys that feed it. We use ListenLabs for surveys as well, uh, Outer Signal, um, post-purchase surveys. There are certain skills that I have that are built that pull from this file. For example, I have a CRO skill that I trained on, let's call it everything I know about CRO plus way more than I know about CRO, proper frameworks, ICE prioritization frameworks. Oh, it's got, you know, uh, like GA4 or heatmap data that's in it. So just always trying to start everything from, you know, first principles. Customer insights, consumer insights, as well as data. It pulls in all of our past tests. So that would be number one, is it pulls all that in. It can look at our past tests. It can kind of prioritize. So it's built on prioritization frameworks, right? We should be testing our highest traffic pages are the biggest opportunity. We should be testing above the fold. So it's trained on all of those. It'll, you know, do that. It'll analyze, it'll build a brief. My process from there, once I have the brief strategy, you know, and have some skills that do that, is design. The other thing that's been the most helpful is a design system is I've probably spent who knows how many hours of building essentially a markdown file version of a design system. You know, there's HTML visualizations, it's built on a Figma design system. It's built on, you know, Claude building it and giving access to things like your website and whatever. And then a ton of trial and error. 'Cause it's, you know, design is not easy with code. It's getting better, Or, and with Claude, it's not, it's not easy, but it can be done as long as you train it on the proper tokens, but it's just taking a long time to get it there. It's probably a lot easier now with Claude Design. So what I'll normally do is I'll design in HTML, and then I have a skill that once I am happy with, we'll call it like 80% of the way there, I will either send it to Figma and have my design team touch it up, work on it. That's especially if, you know, again, A, I'm not a designer. Also, that first 80%, it doesn't take that long. That, that last 20% probably takes longer than the first 80%. So often I'll just push it to my design team and they work on it. Um, and then from there, I will get it into Shopify. I have a skill that goes from HTML or Figma to Shopify. It's specific to our theme because there's specific theme architecture criteria. Everything is just about taking the time. Upfront building all of the right components, building all of the right skills and logic, and I'll push it there and often refine the design, set up the test directly in IntelliJ. So I just like to see the whole process, like what would be if I had a team of 5 people, like world-class execution, how can I do that with one person with Claude? What's a roadblock I'm hitting at this process? Oh, I don't know how to write the CSS to set up this test in IntelliJ. Well, I did it once, I fumbled my way through it. Let me build a skill so it does it automatically. And that's how we've been going about it. And it's been a lot of fun. It's, I do it at night on weekends. I was, this will sound terrible, but I was away in Bahamas last week, got remote control working. So I'm Whisper Flow on my phone to remote control and I rebuilt our cart that will go live that we're testing this week.

I am, dude, I want to hug you. This is like so impressive. So you said a couple of things. One. This idea of passing it off to Figma is so consistent with what we've talked about, where it gets you 80% of the way. You as the CEO are able to then pass it off and say, look, this is what I want things to look like. So kudos to you. I also want to touch on, you touched on the convergence of roles, the role as a CEO in leaning in aggressively as well. But before those two final questions for me, Claude Design, I know you've been messing around with it over the last couple of days. Clearly design is an important part of your workflow. What are you thinking about Claude Design? What are you impressed by? How are you using it? Give us the rundown.

It's really cool. I'm excited about it. And it's also new. I think that's what everyone's got to keep in mind is like, just, I just love watching Claude iterate on new products and try to see where it's going. And they're very open to feedback. So I think it's awesome. Um, it eats up tokens, so hopefully they'll fix that, but it's very token hungry. It's pretty good. The first thing it does is ask you for a design system, you know, or you can build a new one. I would say it's pretty good and made, it's not definitely perfect. It's not a Figma killer yet, but it's cool. The biggest limitation of it, it actually does a better job following your design system than Claude Code has. The biggest limitation is there's no skills besides the token thing. There's no skills, so I can't like get our whole zero process in there. So for now it's, you know, it's only been out for 4 days. I'm hacking my way around Claude Code HTML design. When I get to a point where I need a little bit design help, I'll screenshot that, give it a little brief, throw that in Cloud Design. What I really like about it, it kind of acts like, I guess, a PM, and it'll ask you like 4 or 5 different interview questions. How do you want this done? What framework do you want? You know, how many variations do you want? You can then iterate from there. Once you get it, you can either just chat with it, you can click on elements and circle them to call out. You can also, uh, move some pixels around yourself, not the same flexibility as Figma, but you can do a little bit. And then you, once you're happy with that, you can copy that, you know, essentially the code to clipboard, put that back into, into Claude code. So that's a workflow I've been doing. Um, I think it's cool and I'm really excited to see, you know, what more they go with it. Hopefully they can just integrate this all into one Claude. That's my hope is like, it's just all as one, but it's cool. It's cool so far.

Kudos to you for leaning in. My last question before I pass it off to Ben, you touched on this convergence of roles and sort of, we've, we've talked about this a little bit off air before. How are you thinking about that as a CEO with Staffing? What do you mean by that when you talk about convergence of roles and some of the new jobs that you're hiring for?

So I think like the CRO e-com role is one, you know, that's normally it would, it would take 4 people to run a really good CRO. Function, and maybe you need an agency for that to fractionalize the cost. Or, 'cause we don't have the budget for separate copywriter, designer, dev, strategist, data analyst. One person who's highly capable and ambitious can do that. And you know, I was talking to a family, you know, in-law the other day, like, and you know, they're a PM in tech and they're like, yeah, it's our role is just changing so much. Like you're expected, if you're a PM, you're expected to code. And if you're, you know, and, and design as well. So I just think that's where it's going. And I, that was the first role in our business that I identified. And so we're recruiting for it now and we're being very clear in the job description. Like, this is a very new role. This is experimental. You're expected to do, like, here's the expectations and yeah, you'll get some support from design and dev, but like, you're expected to do a large amount yourself and be very proficient. So I think that will happen. I think Senior people will be more ICs than just managers. And I'm having that happen where I'm having, you know, directors and senior directors get their hands dirty a lot more because those are the people that can have the biggest impact at your company and get their hands dirty. And it's now much more leverage. And it's funny because everyone's like, oh, you shouldn't be spending your time, you know, doing cloud. I'm like, I'm not coding. I have agents that are coding. Like, I'm managing things. You're just no longer managing people. You're managing agents. Agents. Yeah.

I, I, before I pass it off to Ben, I'm going to say this in one second. I could not disagree with those people more. I think when I work closely with the team at OpenAI while at Crocs, like AI adoption lives and dies at the executive level with an executive mandate. And the fact that you are leaning in so aggressively on nights and weekends and candidly you should be during the day sets the tone for the rest of the organization. I don't think at all that that's a bad thing. And kudos to the leaders who do lean in. I'm, I'm just, I'm surprised to hear some of the pushback there.

Yeah, whatever. It's, it's fine. I mean, I have other responsibilities, so, right? I have to be doing them and, and be smart about it. But, but as I can, I'm doing it. I'm doing a Claude co-training this Friday and I'm going to just pick, it's the second one I've done, and just pick a use case for my team and build something. Like, I'm, I, my team knows if there is a workflow they're struggling with and they want me to automate it, like, I'll do it for fun. It's like, it's fun for me to do. But yeah, obviously part of it is let me set the culture and expectations and want everybody using the tools to the best of their abilities and set really high expectations because it's where I think the, you know, the needs are.

Cody, super inspiring. A lot of what you're saying resonates as a founder CEO myself. You know, we have the things we have to do every day, but Some people say, "Oh, AI is going to save you a ton of time." I actually been working more since I started—

Oh, absolutely.

—in AI, right? Like, because it's just fun. So I do nights and weekends figuring out stuff on Cloud Code and building cool stuff. So completely agree with you. This is an exciting time to be an executive and an IC at the same time. Because it's easier and more fun now. Okay, great. So I want to move on to talk about creative for a second. So you posted recently that AI-generated creative is getting good, but the future belongs to content that can't be faked. Jones Road is a brand built on Bobby Brown's authenticity. How do you think about, you know, drawing the line between AI for ops and design versus keeping the human in front of the customer?

Yeah, I think it depends because we are using it in some capacities. I think static is good. You know, I think like design is good. I'm not super interested in AI-generated people, like even though it can get pretty good, like I don't really want that for us. And then I think the biggest thing, like Adam Mosseri talked about this at the beginning of the year, is as it's so hard to tell You know, like when production was hard, it was kind of a status signal or like a trust signal over what consumer wanted was like produced content because, you know, that showed that you were a trustworthy brand and everything. Now that it's so easy to actually produce stuff and make it look good and it's hard to tell, the things that people are going to want to resonate with and consume as media are going to be the things that are so hard to fake. And so I think live streaming is going to be a bigger thing because you can't really fake a 3-hour live. It's like, not to get political, but it's like candidates not going on the Rogan podcast because they can't sit through a 3-hour, like you can't script that. You know what I mean? It's the unscripted stuff that's long form. It's a really lo-fi content that's just, you know, those ugly ads that's like so obvious that it's like, real. I think that's what people are going to want more of. And then IRL, like it's not a content thing, but I guess maybe it is like as we do events and stuff like that, like, hey, just with AI, everybody be— I'm a huge AI bull, clearly, but with everybody being on their phones, like they are going to, they're going to starve for in-person community stuff. But then even that's going to be good content online in your ads and everything, because it is more trustworthy that you're a real brand.

Totally, totally could not agree with you more. Last question from me, Cody, what's your advice for operators out there that feel overwhelmed with all the AI tools out there and don't know where to start?

It's definitely overwhelming. I was leaving office yesterday and I went to my director of data AI, and AI, we have, her and I have a channel where we just share stuff. And I was like, hey, what do you think the Claude update's going to be today? Because like usually it's like 5 o'clock we get home, 6 o'clock. And like, so it's, it's very overwhelming. Unfortunately, you have no choice right now. You just gotta, you just gotta lean in. Um, you know, otherwise it's, you're not going to develop these skills. So you just got to lean in, set some time aside, prioritize it. Um, and, and do the same for your team where, yeah, we have our entire team and, you know, AI all hands, you know, pretty often and doing a lot of training because we think it's that important. So. Not, I think if you're a CEO, founder, entrepreneur, you gotta be doing nights and weekends a little bit. But for your team, you gotta block at least a few hours per month, per week to allow them to play around with it. 'Cause it is definitely a lot and people are, it's gonna be, well, who knows what the job market will be? I think it could go either way, but the people that are gonna succeed are the ones that are gonna have really strong abilities with these tools.

Amazing. Well, thank you for being here. Where, where can people learn more about you? Obviously you're on the Marketing Ops. Give us the rundown on how we can hear more from you. Yeah.

Marketing Operators Podcast, wherever you get your podcasts. And then on X @CodyPluff, usually tweeting about some cloud stuff. Later, man. Awesome.

Thank you, Cody.

Most brands have deployed AI across acquisition and creative, but the confirmation page, the highest intent moment in commerce, remains a dead end. Rockt Aftersell is the design system for the full post-purchase flow, unifying cart checkout, post-purchase, and confirmation into one intelligent system. Rockt Thanks brings real-time AI decisioning to the confirmation page, powered by Rockt Brain, which analyzes 1.95 trillion— with a T— data points across 7.5 billion transactions annually to determine the next best action. The result: Instead of generic offers, customers see timely, relevant partner offers from 500+ different brands like Hulu, HelloFresh, and Venmo, generating pure profit per order at scale with zero ad spend and zero customer data shared. Brands like Sephora, Nordstrom, True Classic, and Jones Road have generated over $1.5 billion in pure profit this way. Activate Rock Thanks at aftersell.com/operators. And unlock AfterSell's full optimization suite, cart, checkout, and post-purchase, all completely free.

Okay, we're being joined by Krishna Poudha, the CEO of Saros Analytics. Welcome, Krishna. Please tell us a bit about yourself and your company. Thanks guys.

Thanks for having me on the chat here. Krishna here, co-founder and CEO of the company called Saros Analytics. We are an AI infrastructure company. We work with brands. We help them get their data and AI infrastructure set up so that they can automate their business workflows on top of that and drive business value. Very cool. Been in the business for about 10 years now.

Wow. Okay. Well, thanks for coming on the show. Let's get into it. So you work with e-commerce CEOs every day. What's the biggest gap you see between brands that aggressively implement AI and ones that kind of take their time with it?

Got it. Fundamentally, I believe that AI transformation only happens when the CEO of a company or a brand determines that it's a priority. Priority for me is indicative based on the actions that they take rather than just someone saying that it is a priority, right? So So we say health is my priority. I don't hit the gym, I don't eat clean. My actions indicate that it's not a priority, but in reality, but when you hear me say, I'm saying it's a priority. So the ones that I see being successful not only say that AI is a priority, but they're also making active investments into people, technology, coaching, et cetera, to get their team ready. and using these AI tools and building value on top of that.

Totally agree with that. So you're basically saying a good AI transformation or rollout in companies needs to be a CEO directive, not bottoms-up experiment, right? So why does it fail, you think, when it is bottoms-up, right? When the curious person on the team running AI pilots and the CEO is not super connected to AI rollouts?

In my view, you need both bottoms-up innovation and top-down directives as well. When both of them come together is when real magic happens, in my opinion. So a bottom-up initiative might give benefits for an individual team member or a specific team, but a top-down directive will ensure that the entire company is benefiting from AI provided the roadmap is done well, the execution is done well, etc. So yeah, so a top-down initiative definitely forces priorities in terms of actions, and actions lead to measurable outcomes and results. So yeah, so definitely, in my opinion at least, for me at Zara's Analytics, AI is one of the top 3 priorities for the business, for every single department. Including for HR, right? So yeah, I would assume that even for a brand, the same thing would hold because there is some serious value unlocked here with AI.

Very cool. Okay, let's move on to a real AI use case that you automated, right? What does that system actually look like and what changed once it was live?

I can give you a couple use cases. One of them is a daily, weekly, monthly business summary for the CEO and the chief marketing officer. A report that would typically go once a month now goes daily. And after that, a more summarized version of that goes weekly and subsequently at a monthly level. Just putting that report for a brand that is running in multiple countries, multiple channels, running advertising across a whole gamut of systems. For a single person who has access to a single source of truth, an analyst with access to a single source of truth would take hours to put together. We have reduced that to minutes. So the report goes out 8 AM in the morning, the CEO could look at it, understand what moved in the business analysis on why that, you know, a particular metric moved in the direction And all of that, once we developed it, is automated. So we've rolled it out for the CEO. He liked it so much that they've asked the international— the international leader saw that in one of his meetings. They came back and said, hey, can we get that for my team? And so on and so forth, right? So now we are working with the marketing team for the same thing. Not only are these summaries important for business reviews and meeting with the team, reviewing goals and progress and performance. but is also important for forward planning. And to get that report automated, fully automated and delivered, means that the people who are working on it to put this manually together, working for many hours in a week or a month, get their time back to focus on more strategic initiatives. So that's one example I could give you from an executive summary standpoint. Another rather simple use case, but a high-value use case is a customer came back and asked us, guys, when we are running low on a particular SKU with some thresholds, we want you to go check if there's a PO that is pending and that hasn't shipped yet. Because if the PO is not pending and hasn't shipped yet, then we could just get on a call with the supplier and ask them to expedite shipping rather than putting on a board where we put that delivery on a flight, ship it faster so that you don't get into an out-of-stock situation, right? Now, to be able to do that, you need a system that has access to your PO data. You have access to your inventory, consolidated inventory data across all sales channels, et cetera. And by automating that use case, we were able to uncover some scenarios where it could have led to a potential stockout situation, but that was averted because the AI agent caught it on time, that flagged these issues to the buyer, and the buyer could get on a call with the supplier and change the shipping terms, right? So something like that would've gone unnoticed completely, or worse, we notice it, but after stockout has happened, where there's really no recourse for correction.

I love these use cases, and I want to dive deeper on them. You know, automated reporting, we heard from Dan at Topo Designs. Preventing stockouts. Mike at Simple Modern sort of shared some of that. Obviously, Krishna, you talked about the importance of a strong data infrastructure to then unlock all of this, right? So what does that mean in practice? What does a brand, let's say, doing $25 million a year need to have in place to start to unlock some of this automation that you're talking about?

Yeah, I mean, data infrastructure might sound, uh, technical to, uh, many people, but the nuts and bolts of it is you have your order data sitting with Amazon, sitting with Shopify, sitting with Walmart. You have your advertising spend happening in Facebook, Google, Amazon, across the board. If I have a simple question that says, what were my total sales yesterday? And what were my total returns yesterday across all of these channels? I have to go to every single one of these, you know, marketplace web application or a Shopify admin panel, download the report, get that data into a spreadsheet, and then calculate what was the sum of my sales yesterday, right? So for such a simple use case, I need to download 3 CSVs, massage that data into something that would give me meaningful results. A data infrastructure is just that consolidation on steroids where there is data feeding from all of these systems into a single data warehouse, into a single location where the data is clean, organized, and set up just like how you would take your physical warehouses, right? Where you are getting all of your goods in, shipped to the warehouse, but then you are organizing it by lots, you're organizing it by shelves, you're organizing it by trays so that it's easy for people to you know, go find what you need in a particular aisle at this particular lot in this particular place where your product lies in. A digital data warehouse is sort of a manifestation of that where we are getting data from quote-unquote multiple suppliers of data, right? Instead of suppliers of parts that go into a physical warehouse, we are talking about suppliers of data, which are Shopify, Amazon, Facebook, so on and so forth. And we then go in there, organize that data into lots, bins, et cetera, so that any consumer who is interested in parts of that data from the warehouse can pull it, use it, and take decisions based on it. Based on it.

Krishna, we spoke a little bit about this last week. And so once you've got your data infrastructure clean, we also talked about how these tools are just kind of fancy autocompletes in some ways. And the way Ben might query the data might result in a different, you know, deliverable or outcome than I might query the data, even if we're operating off of the same data infrastructure. What is a semantic layer and what is the role of that in ensuring consistency?

Yeah, that's a great question. There's a very technical answer and there's more business-friendly answer. But at the end of the day, how semantics define English as a language. Where it makes it easily understood by everybody. Data also has a semantic structure to it. At the end of the day, semantics layer is helping consumers of data. It runs behind the scenes. They don't really get exposed to it so much. Maybe in one-off situations, semantic layer is helping you understand where is my data, how is that data organized. If I have to go find net sales, how is that net sales calculated? What data is it pulling in from? Right? Does net sales include my Amazon data? Does net sales include canceled orders? Does net sales include return orders? So there is a logic behind how a metric like net sales is calculated. And semantic layer is, to put it in very simplistic terms, is is a mechanism where you actually store these definitions so that anybody who is pointing to net sales as a metric, the underlying logic remains consistent for how to pull it, right? Why is this important? If I'm a marketer, I generally won't be caring so much about return product because my job as a marketer primarily is to drive demand. And if the product is not good, and the product gets returned, that is something a marketer should know but cannot really control, right? So the revenue metric for a marketer may be different. They might be interested more in demand revenue that they're driving, which doesn't include returns. But for a finance person, returns and cancellations, all of these become important because finance needs that true how much money is hitting my bank account sort of proper number, right? So when a marketer says revenue, they might mean one thing. When a finance person says revenue, it might mean another thing. And what Semanticlayer does is it ensures that the right person is being fed with the right definition. But when they say net revenue, they actually get the same definition that is aligned at the business level. Does that make sense? It does.

So if I'm understanding everything correctly, your data sits in a structured warehouse potentially you've set up a connector through Claude, and in that connector, there's a middle ground skill or semantic layer that defines for the entire organization what key terms mean when it's being queried. So I, I guess my, my closing question for you then is what is the difference, uh, for a brand who is doing all of that, is connecting through Claude versus another one who's just uploading spreadsheets all the time. In sort of one-offs. What are the teams who are uploading spreadsheets in these one-off instances potentially getting wrong that the others who are more deeply embedded into Claude are getting from it?

So AI speeds up a lot of workflows. For instance, right? I use AI all the time. Have we, and let me ask you this question also, right? When you use an AI tool, do you ever get an answer where you just copy paste what you get and you continue on with you. You never get that answer, right? Why is that the case? The reason that is the case is because AI is a generalized system where it is giving you an answer that upon further prompting, you refine and you get to the final state, right? Unfortunately, with data, that is not an option. If I get the CEO to use Claude and they go and ask a question, what were my sales yesterday? It gives a number and then the CEO is going, oh, remove canceled orders from this number or remove test orders from this number or don't apply this conversion rate. They're not going to sit and do that, right? They expect from an AI system when they are asking a data-backed question, they want an answer that is 100% accurate qualitatively and quantitatively, right? So that way the AI data universe is slightly different from the other AI universe, like generating a creative or generating a narrative or generating a campaign brief where AI can get it 70-80% right in a matter of a few minutes. And then I'm spending a few more minutes to tidy it up and get it pro version that I would be okay with, right? When you are uploading your spreadsheets, what you lack is one, it becomes your own AI spreadsheet. So this is a funny story, right? So customers generally use spreadsheets to understand how their business is doing now. And that has been the case for many years. You get a number, you ask, you pick 3 different spreadsheets, the number might be very different. The number of customers in one spreadsheet might be different from another spreadsheet to another spreadsheet. When each of these spreadsheets gets uploaded to Claude, the same thing is just getting propagated to Claude, right? Where Claude is now summarizing confidently what is in the spreadsheet, but it is not necessarily fixing the gap that exists in a spreadsheet. Whereas if you replace these spreadsheets with a single source of truth, with a context or a semantic layer on top of it, and every single user in the business is connecting to this underlying infrastructure through an MCP, the same question will give the same answer to every single user, regardless of who's asking that question. That's one benefit. The second benefit is Spreadsheets generally have a drift in definitions because the spreadsheets are generally department level. So the marketing team might make a slight tweak to their spreadsheet that doesn't get communicated to maybe the finance team or the chief of staff who's helping put something together for the CEO, right? Then these, again, the benefit of having the single source of truth and the context layer is these drifts can can be caught quickly. And if you have an approval layer on top of the context, then if you are making a change, that change doesn't get reflected everywhere. That change goes through an approval process. And once the approval process is done, then the context is changed and everybody sees the same thing again. Right? So to put it in simpler terms, spreadsheets created these silos. Spreadsheets uploaded to Claude will accelerate the workflow for that individual user, but they're also perpetuating the silos, in my opinion. And having that unified layer is the right way, has been the right way, continues to be the right way. And with AI coming into play, having that single source of truth will just unlock some serious efficiency for businesses because Earlier, the bottleneck was creating a dashboard to take a month. Right now, creating a dashboard on a data warehouse is a few minutes of working plot, right? Like, that's real power for the users, in my opinion.

You said the keywords for me, which is a single source of truth. We get that all the time, right? We have one dashboard gives you one number, another dashboard gives you another number. They pull from the same data source, but the definitions are different. That semantic layer you spoke about is very, very important. Having a very organized data warehouse is very, very important. So a lot of what you're saying is resonating. And then when you add AI to the equation, you got to make sure that you ask it a question in a different way, different users ask in a different way, it'll give you the same data and the same answer every time. And that's a bit of a challenge. That was great, Krishna. Thank you for joining on the show. Tell operators, where can they find out more about you and SarasAnalytics?

So I'm on LinkedIn under Krishna Poda. Last name is Poda. Website is sarosanalytics.com. Yeah, hit us on the contact us page. Let us know you've heard about me from the Operators Podcast, and I'm happy to jump on a call, talk AI, share my own experiences. Whatever makes sense.

Thanks for joining, Krishna. Thanks, Krishna.

Thanks for having me, guys.

All right. I am so excited to bring my friend Bar. Uh, you know, you know him as the maestro behind D2Ski, the primo, uh, e-com operator event of the year. You know him as my mom's favorite couscous company, the founder of Boost Couscous. Unfortunately, she tells me every day how much she loves it. But she keeps calling it Boostalicious. Bar, welcome to the show, man.

How are you? Yeah, doing great. Yeah, it was great to meet your mom in New York. I'm so stoked she's, she's a fan.

The reason that I'm so excited to have you here is because I think that you are a one-man wrecking crew when it comes to AI adoption for a young emerging company from a non-technical founder. I was blown away by the ways that you were leaning into the tools. Can you just One, I guess, give us a little bit of overview on Booscoos, and then two, talk us through the ways that you are kind of applying AI in your, your day-to-day at the company.

Yeah, so, um, yeah, I think you, you hit it, uh, right on the head. Like, I think right now is kind of like the biggest opportunity for D2C brands to be building. Um, I think I come from a pretty unique background. There's a lot of brand founders that end up in SaaS. Um, I can think of probably like 3 or 4 SaaS businesses that were formed from a brand that like had a pain point and moved over to SaaS. Like I actually can't think of any like SaaS founders that moved into brand and like if there is anyone out there, please reach out. I like wanna connect with you guys. But yeah, I've been building like SaaS tools in D2C for like 10 years and came up with Boost Goose about 2 years ago. It's the world's first protein couscous. You know, I'm Israeli, grew up eating couscous, always loved it. But, you know, it's versatile. It takes 5 minutes to cook, but like realized as I got older that it's like actually not that good for you. It's kind of a carb bomb. And similar to like, you know, there's Banza for pasta, there's Goodles for pasta, RightRice for rice, which they went out of business and a bunch of people are actually moving over to Tabooskoos from Right Rice. There wasn't a healthy version of couscous out there, so I was like, hey, I wonder if I can do it. Took 2 years to like actually make the formulation and, and be able to like scale up the, the koeman to like the demand that we have today. And yeah, we launched about 4 months ago, sold 40,000 boxes. Yeah, it's, you know, we're, we're keeping it really small. It's me and my co-founder. Um, everything is run through AI. So, um, you know, there's a million different use cases that I'm sure we'll, we'll, uh, dive into. But like, I think we, uh, have this massive opportunity that like we're building from the ground up, like thinking through every single function, how can this be AI-led? And of course eventually we'll have humans in there as well. But, um, but yeah, I think there's, a massive opportunity for companies that are just starting now to not have AI tech debt versus like, you know, some of the bigger companies that are trying to bring AI into their systems and might have a little pushback from humans or just tech debt in general.

You mentioned kind of the conference in New York, which you spoke at. One of the things that I was so impressed by, so obviously you come from a SaaS and analytics background at Know. And one of the things you showed off was an OpenClaw instance in which you were combining disparate datasets, meta, post-purchase, customer reviews, and then you were using voice to ask your OpenClaw what was going on. Not only did you combine it, but then you were just talking to it almost as an analyst, uh, you know, to surface insights. Can you share a little bit about that use case with the listeners?

Yes. So, um, again, like I think coming from SaaS, um, a lot of the feedback that we'd always like in a lot of the product requests we always got was like, hey, can you can you update this in the dashboard? Can you update that in the dashboard? And like what it really comes back down to is I think people want to have a dashboard that's specific to them, right? Like Boost Goose as a business is very, very different than a lot of other CPG brands. We're similar in a lot of ways, but like, you know, our subscription business is going to be different than like a Grooms or, you know, a lot of these like supplement brands. And so Um, it kind of started with that frustration of like, hey, can I just like build a dashboard for myself? Um, and coming from the SaaS world, especially I think pre-AI, it was, uh, very much like a slow moving, um, like truck, which is fine, right? Like I think, uh, now like SaaS is just gonna have to like, uh, adjust. Um, but I, I essentially took all of our data sources, Recharge, Shopify, Klaviyo, Nocommerce. Um, really like every single data source, put it into a BigQuery database, um, and then just started like building the dashboards that I wanted to see, right? So, um, I obviously like, uh, you know, I come from NoCommerce. I understand post-purchase surveys quite a bit. Um, and so I started there and I was like, hey, can I actually like visualize some of this open text data? Like we're asking people things like, um, you know, what almost prevented you from buying today? Um, or why did you buy? Um, these are like really, really interesting insights that are like open text responses that, um, you know, if we're able to get like really rich data out of that, that can feed into our top of funnel for our ads, our copy, our website, our email, so many things. Um, and so I started there and like really quickly the like AI, uh, was able to, to come up with like a lot of insights that like I hadn't thought of or hadn't, you know, like even seen. Like as an example, like, you know, it would pull out like things like gluten-free was like, I'm not personally gluten-free, but like the gluten-free market is massive and we're like one of the first in gluten-free couscous. And so it was pulling out like testimonials of like, you know, there was a mom that had a 9-year-old kid that had always asked for couscous, was never able to eat couscous, because couscous is, uh, in, in its typical form has gluten in it. And now like that kid is having couscous for the first time, right? So like these little insights that it's able to like pull out. Um, and then yeah, I just kept building it from there. Like, okay, we're running a bunch of Meta ads. Can I actually like take a look at, um, like the Meta ads dashboard just kind of sucks. I've, you know, before this company, I'd never logged into the Meta ads dashboard. I think there was like those, there were these like tweets of like people saying like, I'd love to see Zuckerberg like upload his upload or like, you know, like go into Meta Ads Manager. And it's like, you know, I'd always seen these tweets of like people talking bad about the Meta Ads Manager. Now I'm like living in it. And I was like, okay, well, can I actually just surface the most important stats from Ads Manager and actually tie that in with like our Shopify data to essentially see like, Hey, are we profitable or not on these ads? Um, so I was able to do that and then I just kept going down the line, like Recharge, you know, we're, we, we have like a 35% take rate for first orders on subscription. I wanted to understand like, um, you know, what's our LTV on subscription orders? Are people canceling? Why are they canceling? A lot of that is again, open text data that I can like start to visualize. Um, so yeah, just to kind of like put a bow on it, it started with this like frustration coming from the SaaS world that you couldn't customize the dashboard to yourself and Then I was like, okay, I can do this myself. And I did it in, you know, very quickly. I just said, hey, Open Cloud, go connect to this API. Here are the API docs. It came back, said, you know, give me this and that, gave it to them. It connected, put all that data into a BigQuery database, and then I was able to kind of build whatever I wanted on top of that.

And is it just a visualization tool or is it an insights tool as well? Are you querying it to then ask it to find certain things?

Yeah, certainly. Yeah, it's, it, it, I would say it's both, you know, it's something that I'm like logging into every day to just take a look at like the health of the business. But certainly like, you know, we, you know, like 2 or 3 weeks ago had like a massive sales weekend. We were hitting like $10,000 per day, which like for some brands is not that much, but like for us that was like a record-breaking day, right? And so I would just like, you know, go talk to OpenClaw and say, hey, you know, we just had like a really big sales spike, uh, over the weekend. Um, can you go take a look at my post-purchase survey data and, um, like bring up any insights that you can like think of? Um, and so it started to like look through like, you know, how did you first hear about us, things like that. Um, and you know, like some of that, like we know, like, you know, Meta, is ripping for us. Like we're a pretty small brand and we're, you know, our media mix is not that big, but like there were some insights where it was like the marketing operators, Connor Rowland, shout out the Other Operators podcast. He loves Booskooz. He's like the only person that eats Booskooz for breakfast that I know of, which I, you know, Connor, I've told you it's weird. I'm going to say it on the air as well. But whatever, anyone can eat Booskooz whenever they want. Um, but anyway, he, he shouted us out on, on there and we got like 3 sales from, um, that segment. And like that came up in our post-purchase survey. I would never have seen that 'cause I'm like not logging in every single day and looking at every single, uh, like report or every single, um, entry. But, um, yeah, I was able to see that through Open Claw.

All right, listeners, you, you heard it here first. We gotta get Barr his first $20,000 day coming outta this. I've got one last question for you, Barr, before I pass it off. To Ben, which is as a solo founder or a founding team of two, you want to scale your output. You want to use AI to 10x you. You shared a little bit about a way that you were kind of delivering one-to-one personalized outreach to customers at scale with the support of some of these tools. Can you talk a little bit about that use case?

Yeah. So, um, yeah, so we, you know, we're, um, we're getting more and more reviews. I think we crossed like 250 reviews and at first it was, you know, like we'd get like 1 or 2 reviews. I would just like reply back to them. The first 50 reviews I replied back to every single one. All of them are 5-star reviews. And so what I ended up doing was I created a skill around that. And I just, you know, really all the LLMs now can like connect to your, to your email. And so I just had it like go through, hey, I said, um, we use Judge.me for reviews, but I just said, hey, any email that comes from support@judge.me that has the review title, um, go look at it, go look at all my replies and build a skill for, uh, like review replies. Make it sound like me and do whatever I would do. Right. And so, um, what it would do is it would just, uh, you know, look at a, a review that would come in. It would, uh, I, I have it so that it still has to like create a draft of an email before it actually sends the email. I think eventually I, I I will like let it just run, but right now I have some like guards on it. But I essentially said, hey, listen, like, go do this. And so yeah, once we got that massive sales spike, like, you know, 3 weeks later we got a massive, like, review spike as well because it gets delivered and reviews go out like 10 days after that. And so it was able to actually like reply back to like most people and it actually looks at the, the review and makes it personalized. So like, you know, if someone said that they made it like a shrimp dish with, uh, you know, I don't know, like pesto, um, you know, it would like, if I haven't tried that before, it would say like, hey, um, thanks so much. That's an awesome rec. I'm actually going to try that this week. Um, and also this means so much to us. We're a small business. A 5-star review goes a long way. Every single person would reply back and be like, oh my God, this is so awesome that the founders reaching out. That's like a cool use case. The more interesting use case was actually once we started getting our first like, um, 1 and 2-star reviews, which we don't have that many. Um, but what it was able to do was actually like take a look at that review and, um, like, uh, look at the feedback, um, from that person, validate that feedback. And we're actually working on a new formulation for Booscoos. Um, Right now you have to rinse it after you cook. This new formulation, you still have to rinse it, but like not as much. And people like, even if you forget, it'll be, it'll come out like just as good of a product. And it offered to just send that person 2 free boxes. Correct. And like asked like once it's delivered to like if they liked it to update their review. I ended up like it flagged it for me too, and I ended up getting on a call with some of these people. And we had like 2 of the 3 1-star reviews go and update their review to 5 stars. And now they're like loyalists. Like we literally, like one of these, you know, this is like the bridge between AI and humans, right? Like I, like it, it was flagged from AI. I ended up talking to, to, uh, this woman from Massachusetts. She is now like a loyalist and was like, hey, as soon as you're like in retail stores in Mass, like let me know. I want to be your sampling lady. And I was like, yep, like, I— you have my phone number now. Like, text me whenever.

So yeah, Ben, I'll pass it off to you. But very quickly before I do, Bar, just that note too, right? Reviews as being such a power to the flywheel for insights like those phone calls, those transcripts from those conversations are so powerful too, right? All of this is context that clearly goes into the LLM. So Ben, I'll let you drive from here.

Yeah, Bar, really inspiring. Two people at the company and a bunch of AI agents. That's really, really cool. And that's kind of a new way to grow new DTC companies. I'm wondering what's the ceiling that you've encountered thus far? Where does AI still fall short for you and what are you waiting for it to be able to do?

Yeah, so, you know, I think we sit in an interesting space where like on D2C, I don't like there, there probably is a ceiling, but we just haven't really like met it yet. Right? Like we, are doing like a lot of our static ads are all like AI generated, our websites, like all our images are AI generated. You know, we're starting to play around with video. I don't think that there's too much constraint there. Eventually, like we, like the first hire I'll make is, is probably like a chief of staff that will like just kind of come in and take everything off my plate and like use the AI agents that I've currently built and just make it even better. Um, the other side of the business is retail, right? Like we want to be in Costco, we want to be in Sam's Club, Walmart, Whole Foods, all those places. And they still operate, um, very much on like a more of a, like a human level. And like, we need to hire, like right now I do the sales there. Eventually, like we're going to raise money and I'll like hire a salesperson for that, like conversation, like you need to have humans in stores, giving samples, walking around, meet, like shaking hands with like the buyers. Um, it's a, you know, it's a pretty like old, old school industry. Like, you know, I literally just got like a, a, like a purchase order from a big retailer in, in California and it's, it's like literally on the bottom of the like form it says last updated in 2017. Right. So that's like literally 10, like 10 years ago they updated their form. Like they're probably, maybe some people in their industry are using AI, but they, they're still pretty, old school. And so that part of the business I do think will need to, um, just like have humans for, but I do think like AI can definitely help there. Like, you know, I, I have an AI sales agent that's, um, like reached out to someone on my LinkedIn that was a Costco buyer, ended up getting every single Costco buyer's email for every single region, um, drafted an email for those people, uh, sent an email to, to those people. They asked for like, a heat map of like our sales, it created the heat map. Like, you know, like I do think we will be able to automate a lot of that and like one human will have the power of like 5 to 10. But right now, like, yeah, that side of the business is very much constrained and needs to have humans.

Very cool. Very cool, Bart. My last question is as follows. So, you know, outside of salespeople, that's still old school at retail where you're going to need real humans. that don't have an AI proficiency per se, for the digital side of the business, if you were to go and hire tomorrow, what kind of AI proficiency are you going to be looking for in this new hire?

Yeah, great question. Um, I think, uh, yeah, like Cody Palfker, um, like I think he's, he's great when it comes to this and he's like, I know he's like hiring someone like for this specific role. Like, hey, listen, I've taken it. From zero to one. I've built an entire cloud code database on like how to like, you know, build a, you know, CRO optimized landing page. I want to bring someone in, train them, and just have 'em keep going. I do think, I do think that like, so like this past week we had D2Ski, right? I talked to like some operators of some of the biggest brands like in, in, in D2C specifically, I'm thinking about like the Groons and the David Protein, like first hires. Both of those hires, interestingly enough, never had marketing experience or D2C experience. They both came from like consulting, you know, like call it like 25, 26-year-old, like 4 or 5 years of consulting, has a really high drive and has a really high, um, like, uh, need and desire to like figure things out. Um, and so of course I think that person is playing with AI. I think I can train most people on the way, like the way that I'm thinking about it right now is like, it's essentially like a video game and similar to like the top influencers who know how to like talk to the algorithm for Meta and Instagram. Like I now feel like I can understand how to talk to an LLM. And make it better. And so, and work for me. And so I can train people on that. I just, I think the, the first hire is going to be someone that has a really high drive and eagerness to learn. And, and then they'll probably take the tools that I've already built and just make them 10 times better.

Bar, thank you so much for being here. To close, one quick sales use case. We'll have to connect on it after. It's amazing for cold outreach, but it's even better if you have a skill that you run your transcripts by, effectively a personalized sales coach after each call that emails you a recap of performance based on what you're specifically working on. You and I'll connect after this. I'll build one for you. It's one of my favorites. Bar, thank you so much. Last question. Aside from going to BoostKoos.com and getting the most delicious and protein-filled and gluten-free BoostKoos on the market, how can people, uh, continue to hear from you?

Yeah. And Jewish mom approved. Um, we got website. Um, uh, yeah, Bar Brujas. I'm on Twitter and, and LinkedIn. Um, probably the only Bar you'll find on there. And so, uh, can't be too hard to find. Thanks, man. All right.

Thank you, Bar.

AI 101 skills. You heard Cody, you heard Bar talking about them earlier. Ben, Take it away. Tell us what skills are and why they are so important.

Here's the simplest way to think about it. Say you want Claude to write product descriptions, right? So a project is where you drop your brand guide and some example products, the stuff that Claude needs to know about your brand and, and your customer. The skill is where you spell out exactly how to write the description, tone, format, what data to pull. What structure to write. You write it once, you save it, and now everyone on your team just says, write a product description for this SKU and gets the same quality every time, the same structure, et cetera. Projects are context. Skill are instructions for a repeatable task. Now, skills work everywhere, regular Claude Chat, Cowork, Claude Code, even the Excel and PowerPoint add-ins of Claude. You just need code execution turned on in settings, go to Customize, then Skills, and then you can browse the built-in ones, install what's useful, or create some of your own skills, would be my recommendation. When to use what, right? So you use a project when you need Claude to understand an ongoing body of work, right? A client, a campaign, a product line. You use a skill when you have a repeatable task that needs to come out the same way every time. Most operators should be running both, honestly, uh, projects for context, skills for the tasks that live inside that context.

Yeah, it's not really an either or, right? It's using the two together. And Ben, you are a brilliant engineer. I unfortunately am just a simple guy. And so what I like to do is dumb things down and I'm going to do that through, uh, everybody's favorite movie, Finding Nemo. And so you think of a a project, almost like the aquarium, right? So when Nemo is caught from the ocean and he's in the dentist's office, he's in the aquarium, that's a project. Everyone is sort of living off of the same context in that shared environment. A skill is when that psychotic fish Gill hatches his escape plan. He says, hey Nemo, we are going to first put the rock in the filter, then it's going to get really green and messy. Then they're going to take us out of the tank and then we're going to roll out into the ocean. So the skill, just like you said, is the step-by-step instructions tied to that environment. So let's say, and I work with a number of customers whom this is the case, you've got a project where you're analyzing sales data from a particular retailer. Your whole merchandising team is analyzing that information. The skill is the specific set of instructions about how you analyze that data, about how the brand looks at performance. And so ultimately it's certainly not either or. All of these are just orchestrating different pieces of context. Think of a project like the aquarium and the skill like Gill the fish. And that is it. Thank you, the listener. If you like the show, please like, rate, review, subscribe, follow. We are here to be your resource to cut through the noise and apply what matters. We will see you next week.