AI Is Moving Marketing Value From Retainers to Your Own Data

Welcome to the Marketing MadMen, and boy, do we have a special episode for you. I cannot go two feet without hearing the words AI, disruption, marketing, and honestly, I'm tired of talking to marketers about it. So I have brought in two actual AI web developers. We have Phil here from several different companies, from Paydax and from Pilotship, correct?

Pilotship mostly.

Mostly. And then we got Mark here from PitchGhost. And the beauty of this, guys, is these are two tools which I use and I've built into my tech stack, into my AI implementation, which I think makes this as practical as it is futuristic. Gentlemen, first of all, how are we doing?

Doing great.

Fantastic. Good to have you both. Let's hit it. I am going to ask both of you a question. I don't think you're expecting this, but this is what makes it fun. Phil, I'm going to have you go first. It seems to me like there are a lot of industries that are being disrupted by AI in any direction, but it seems like marketing is a special one that has been disrupted. I will say, seeing it from the other side, one of the reasons, I think a lot of this industry has been fluff and smoke and mirrors for a long time. Everyone's been afraid of data because the data shows things that they're not capable of doing. I think this was a lot of that old adage of 50% of your marketing is working, you just don't know which 50%. Or the old Henry Ford adage about stopping your marketing. Now I'm going to forget it, but pretty much it's like stopping a clock and not being able to tell time. You cannot stop your marketing. Let's go into this real fast. Phil, why do you believe, from what you see as a developer, marketing was disrupted so much, and how much opportunity do you see in that field for AI implementation?

Fundamentally, in my opinion, the big thing about AI and kind of the common thread is that the goal of AI and the thing that it's doing to every industry and marketing specifically is it's allowing everybody to provide a better service for their customers at a much cheaper price. Marketing, in my opinion, has always been kind of the opposite, where they're just trying to get an automated process out there, whether it's your team doing SEO auditing over time or doing ads, going in there, clicking around, making a few small edits, and then charging you on a monthly retainer but kind of doing work less and less over time.

Welcome to the Marketing MadMen. This is part two of our episode with the boys from Atlanta Tech Village talking about how marketing is especially disrupted by AI. We're going to pick it right up where we left off. We are talking about the 12-month horizon after we had just established that unfortunately this is the time where the bad actors come in. Let's hit it. Go ahead, Mark. You have one thing to say and then, Phil, you're going to call your horizon.

I think the moral of this conversation has been the horizons and the looking back. It varies by person.

For sure.

For you, Claude Code, it's practically a now. For Phil, that was six months ago. For me, that was 12 months. I think it's the same thing. What the future looks like is how aggressive into that curve you want to look.

For sure. And how big your appetite is for it. If your appetite is that's where you want to be. Look, if you're running, and this is not, a window company that is worth 20 million and your goal is to get it to 23 million and give it to your kid, you don't need Fable 5.5 running with 16 things over it. You need a quick audit of your tools. Once a month, give me a report, get a little bit better every month, done. I think it's also your appetite. And I think it's one of the biggest things in business: if you know what you want to accomplish, finding the tools to help you do it is very easy. If you are a company whose only job is you want to sell, AI implementation is the single most important thing you know, because you can increase your enterprise valuation by hundreds of millions of dollars by just building scalability tools into it. Because those smart people from private equity come in and you say the word scalable, they're going to be like, oh, look at this, this guy actually knows what he's talking about. When I do my work, Phil, a lot of it is not the now. It is, if I've gotten this far now, what does a year look like? If these tools can self-learn, what does it look like on the 64 million touchpoints that I've already built into it? Now give me the 12-month horizon.

Yeah, 12-month horizon. I'm going to piggyback off of one quick thing for you guys. I think one of the big things for me is we used to be talking about how the model layer and the infrastructure layer was the most important thing. And that's why the SaaSpocalypse was happening. Now it's the application layer that I think is going to have its moment. The reason for that is I almost don't care about the frontier models anymore. They're at the point where they're solving cancer.

They're marketing each other.

They're not marketing each other. They're trying to jockey for position.

They've gotten so smart that they're PhD level, they're solving cancer. They're doing mathematical equations that nobody's ever solved. The businesses that I work with don't need any of that.

It is good to know that it's still on the horizon. It is still good to know that you're growing towards something, because this is every business. What happens is it goes early adopter, corporate, enterprise, top 500, then it goes to smaller corporations, then it goes to mid-size businesses, then it goes to small businesses. That is the horizon.

I think Phil's point though is almost that it's now back to being the data in the application is the limiter.

My point being is that the models could not get any better for the next 10 years, and we would find ways to apply them in businesses in a million different applications. Because the applications just haven't caught up. You have a million things where you could take this infinite intelligence and apply it into your business. I think for the next 12 months, it's all going to be application layer. It's going to be people taking this level of intelligence and baking it into their software, their business, automating an amazing amount of work. I'll give you guys an example. I'm not going to share the name of the customer that we're working with on this, but they're a steel distributor. We are automating their entire business. Everything from the moment that somebody brings in a quote, it goes through, understands the margins from all of their historical data, understands how they should price that based on their margins. It's in their HubSpot updating everything, going into their QuickBooks, sending out the invoice, routing things to the right vendors. We are fully automating this entire business. It's insane. And we're starting in more of a space where we're just working with a copilot, where somebody is working with it and they are using this tool. The agent is doing the work and they're saying, all right, now do the invoice, now do this, and slowly it's training it. But very soon, and it already has this capability, it can literally run their business for them. It's just getting to the point where you trust it enough to do that level of work. But I think we're going to see some businesses that are truly just operating agentically in a lot of different ways. And I think that's the 12 months down the road, where you see full businesses. And we're talking about a steel distributor. They deal with hard goods. I see exactly how we automate their entire business, from freight, from logistics, all of it, because of MCPs, because of the application layer and just tying everything together. So I think there are businesses like that that are doing this. And I think it's tough, because as a small business, how do you keep up in a world where your competitor is doing that? If your competitor is doing that, they're going to be providing so much more for so much less. They start with keeping the margin for themselves, but how do you actually win? You're the business that charges less than anybody possibly can. You take the whole market.

By building that into a system and then going and buying every steel distributor in the country that is using the old school way, and building a model that scales and plugs them into it, and then it takes over the world. Concern of mine. We'll get to that at the end here.

I think the way that companies have done this in the past, more like an Amazon, is you make it cheaper than everybody else. You put everybody else out of business and then you raise your prices. Amazon takes people out of business all the time.

That's this country. Take a loss, run up $100 trillion in debt, stock price somehow keeps booming, run everybody out of business and then raise your prices because you have your monopoly, which is fine. It's capitalism 101. They employ a lot of people.

We will see some monopolies because of this.

For sure. Here's what I'm going to liken it to, because I think tech works in similar cycles. We are at the point now where cell phone chips have evolved so fast and so quickly that they're still using a game, and I'm going to forget the name, Genshin Impact, from six years ago, as the model of what graphics on a phone can push to. The phones have been the same for four years. It has stagnated. AI is still on its upward boom, but what you're talking about is it doesn't matter what the most advanced models are, because it still hasn't even caught up with what the implementation is. But that moment's going to happen and we're going to start to level off. So here's the question for you. So you said 12 months, you said six months. When do you think the curve will start to level out? In my head, it's got to align with the RAM crisis and RAM prices, and something's dropping. Those things have to be aligned with each other, but when do we get to the point where this starts to level out? Because that's when, in theory, the democratization can come. Because it's not how fast you know how to adopt it, it starts leveling. When do you think this parabolic curve of the development cycles for AI will start leveling out? Mark, you go first.

I'll take probably the more contrarian take here. I really don't think that's any time on the horizon. I think there's a lot, and the real thing everyone likes to cite, even though it's always misquoted, is Moore's law on this. The entire time chips were shrinking, people kept saying no, no, no. And you even bring up the phone chips, I would argue the chips on the phone are the one thing that's really improving. The iPhone 18 Pro is like a two nanometer chip.

No, no, no, they're improving, but the software that you can execute on it has not kept up with it.

I'll give this one example. I've told some people, I compare this to your teacher, and maybe Phil and I are dating ourselves as young people here, but our teachers would tell us don't use Wikipedia as a primary source. Or you're not going to have a calculator in your pocket growing up. That classic one. And to me, early on, everyone disliked AI because it did what? It hallucinated like crazy. It would make stuff up. I think the model providers, they didn't really go public with this, but they realized something huge. They're like, okay, nobody wants us to hallucinate. Let's actually get back a bunch of space in the training data by not feeding it facts. Focus instead on call and response, being really good for agentic work, and instead just train the models at base to always look things up. That is now, if you actually look now.

And that's what Perplexity was out of all of them, because it is the deep research model.

If you ask Claude anything now as opposed to two years ago, two years ago it would have just told you the answer. Claude now will search for the answer. But I think Moore's law didn't just come from magic each time. It came from every year, one new efficiency they pick up that compounds. And I think that's a good example of how these models have basically been able to more than double within a year: they got rid of all the training data that's like, who was president in 1933. I'll just Google it, is how they work. And they are much better now at structure, follow-up prompting, and being able to fan out and just brute force them. That's how they're solving the math problems. And all those ones, it's swarms of hundreds of thousands of agents and you're just throwing bodies at it. It's not that you've created a million Albert Einsteins. It's what could 100,000 people do to cure cancer?

And I completely agree with you, but I don't think that any of these industries level out because of the actual industry itself. It's the economics that go into it. Moore's law, actually, my understanding is, especially in the silicon, the chips for computers, the CPUs, went out of date five years ago. They stopped doubling in speed every year. They stagnated. The new thing is power. So they're shrinking down the dies. I get all that. But here's the point. When we get to a point where the capitalism stops and these companies stop growing. Because you can say right now, I don't think that GPT or Claude are going to be the first $10 trillion company. I just don't think it's going to happen. I think Apple, Google, they're taking so long to get themselves to 2, 3, 4 trillion. I think the economies of scale are going to stop the growth, because stagnation comes from a downturn in the economy. All of a sudden, what if the productivity gains that we were supposed to gain from AI don't show up in the economy and the economy crashes? The big worry now is not that the AI models aren't going to be worth it, but that we baked this cost into the economy, saying that the output gains from people are going to help build this thing. What if that crashes?

I think we have a good five years. And I think in the same way that the phone, you said, hasn't improved in four years, people still buy the iPhone 18 Pro. Apple's still going to probably have a record stock day from that.

They did. They just had their number one ever.

Yeah, we've been talking nonstop about how AI gives you an edge. And so long as there is a Fable 5.2 over a Fable 5.1, there'll be people who gain an edge by using it.

But the curve will not be parabolic. The curve is going to have to level off a little bit.

It will, but I think there will be a solid amount of time. I think there were people who predicted a very early doom and gloom, that AI training back on itself would be kind of this collapse.

By the way, mark this on the record, please. This goes on all things. So you say five years.

Five years. I'll stand by that. I think for five years it's going to continue to get better.

But the curve will start to level out. So five years will be when it flattens, or when it starts leveling? Let's be hyper clear.

And then we're going to put this on Kalshi.

I'm talking about when the parabolic curve starts leveling out.

Is this an economic impact or a…

Growth of the growth. Exponential growth. It has nothing to do with the stock price, but it has to do with the output. Will the output of these tools start to level out when there's not as much financial incentive for them to grow?

The output being like the level of intelligence?

Yeah, I guess I'm confused.

Are we saying…

Okay, so let's put it this way. The level of intelligence that is capably used by its customers to accomplish tasks. Yes, it has to do with how good the model is, but do people stop caring about 5.6 because they're comfortable with 5.5, and they don't adopt it and they don't use it as much? I'm trying to say, if you multiply the number of users times the data they use, when will that parabolic curve start to level?

I will say two contradictory things. One, I think AI is a tool to change the world. If you stopped right now at Fable 5.1, the world would be unrecognizable.

It self-learns though. Let's be clear. One of the reasons PlayStation 5 is so good is because it's only built for video games, and it outpaces graphics cards that cost 10 times more money, because the rendering and the layering and all the software gets better and better and better. Even if you just stop at 5.5, tools will be made to get more out of 5.5. Aren't they shrinking tokens now? You run RAGs and all these systems, then you can shrink the token usage. There's a lot of stuff where you're not changing the model, you're just changing the tech stack built around it. So I agree with that completely.

I think we're talking about two separate curves. In my opinion, I think we're already leveling out on the intelligence, like how much smarter it can get. And I think people will stop caring as much, because there are just fewer use cases for this ultra level of intelligence, which is curing cancer, really important stuff. AlphaFold.

But you know what, though? I will take some speed. The tasks still take a lot longer, and I'm running a pretty robust setup here. They still take a lot longer than I think or want them to. I get caught in a loop to do a task. It's a 12-minute execution.

My point is two things can be true. Fable 5.1 can be good enough. And you can also be first in line when Fable 5.2 comes out. I think there's still plenty, even though I'm totally fine with the current models, there's still plenty to get better. Em dashes, writing style, taste.

But I don't think that has much to do with model intelligence so much as people getting better at training models for specific tasks that are cheaper and run faster. And here's my take on the whole RAM thing. Right now, people use a ton of compute in order to train the smartest models out there. I think what we're going to see is that people are going to get more efficient with power. They're going to get more efficient with training data. They're going to create small language models.

But we're going to argue those are because of economies of scale, because the industry will be built more robustly around it.

There's going to be more tokens, more usage. And I think with the catching up of people actually adopting this technology, we could literally not get any smarter on the models and it would take 10 years for everybody to figure out all the ways to apply this technology. And that's where you see GDP increase. I think that's where you see GDP increase. I think on the compute side of things, to your point, Nick, I think Anthropic and OpenAI stop being the most important companies, because frontier intelligence isn't what people are still using. I think people are going to get better at using stuff like Muse, for example, which is a super cheap, easy model to use.

It's like they're putting advertising behind it. You see it everywhere, the advertising.

The point being is it's not the smartest model, but it is just great at one thing, which is tool calling on the internet. It's the only one right now that does a good job of actually just reserving something for you on OpenTable or something like that. They trained it to be really good at doing everyday personal tasks. It doesn't need to get that much better.

The specialization model, I think. I wonder if the specialization model could be a layer on top. You can very easily have a specialization layer that layers over Anthropic to help with one field. That's going to be a thing.

I think right now we've got intelligence that has just scaled up like this. Usage is here. I think you're going to see we're leveling out on intelligence, because it's not that much more important to spend a bunch of money training these models to make them any better. What we should be spending money on is the application of everything.

I think I disagree strongly. I think if you look at where all these guys are even trying to do CapEx, this is OpenAI, Anthropic, they're not saying it's time to slow down.

Yeah, but they're being unlimitedly funded right now.

Who would? There's a huge reason to, though. If Anthropic were to stop and say 5.1 is enough, they would instantly become profitable, because their one loss center right now is the model training. They'd instantly be cash flow positive. They'd be growing.

But they don't have to declare any of that stuff yet because they're not public. So it doesn't matter.

But my point is, you even brought up power. I would actually argue Microsoft, all these guys, they've been restarting power plants. They've wanted wind and solar.

They're trying to launch data centers into space. They're trying to launch them to orbit the earth, because the sun is an unlimited source of energy. They said in the next eight years, they bet you that it'll be possible.

But all of the things you're talking about there, that's why I don't think we're anywhere near the end of the curve, because there's still so much of this stuff that's literally just limited by how fast humans can move.

So you don't see a dot-com bubble crash.

I also think the SaaSpocalypse is overplayed, because while the stock market hates those companies, Monday, Figma, HubSpot, all of them beat earnings.

I'll tell you right now, the SaaSpocalypse just needs to be adapted to real life. You cannot lock people into three-year contracts anymore. If you're doing a month-to-month cancelable contract, there's nothing wrong with SaaS. The problem is that someone goes, builds this great tool, gets bought out, they lock you in for a year, and they tell you to [expletive] when you challenge what is going on.

They all weaponized switching costs, which are now gone. But it doesn't mean that HubSpot itself did not provide value.

Well, HubSpot's problem is they're trying to lock you into HubSpot and they're trying to make…

You can now, in one prompt, switch yourself between HubSpot and Salesforce.

Open source is the key to everything. You and I have talked about this, about how with the scraping, everybody has to start being open source, because it's too easy to switch now. It's too easy to have a tool do the switching for you. All right, we've got eight minutes left because, believe it or not, I have my call with Google that I forgot about, my favorite people. Four minutes each. Talk about PitchGhost. I'm most curious. For anybody who's thinking about the space, tell them the real behind-the-scenes story of what it's like to get funding, come up with a business, what the end goal is. Talk more about the business side of it than the actual execution side of it, and then give a plug for how somebody would use it.

To start it off, I'm the founder of PitchGhost. PitchGhost is a tool that helps you find and leverage whatever you want on social media. We scan and monitor LinkedIn, Reddit, Facebook, Twitter, platforms like that, helping you source literally the conversations and posts in real time so that you can engage with them for whatever reason. Typically, that's sales and leads. So that's somebody saying they have your exact problem, so you can jump in on it.

I'm using it to have people find this podcast. Hey, that's a question, this episode, blah, blah, blah, answers it. Check this podcast out.

Yeah. It can also be used for brand monitoring, for competitor analysis, monitoring your own customers. For instance, I'll give a plug. We've been talking a lot about the ATV, the Atlanta Tech Village. They are a customer of ours. They follow all the founders in the building on LinkedIn, and anytime that we say anything that they should cheer on, that's what they get alerted to. So if I announce I'm hiring, I'm raising, all those things, that's what they really engage with. So we really help, I would say, mid-market tech-enabled businesses, so 10 to 200 person companies who are really trying to level up. Especially, like I mentioned, if you're doing this manually to any degree, if you have anybody who's a social media manager, an SDR or BDR on your team who is manually going around on LinkedIn sourcing deal flow, we can help make that a lot better. We connect with your CRM. Anyone who you have in your CRM, we can just tell you when they're posting.

The tech stack is unbelievable. The fact that you can take the emails of everybody that you already have, put them in, follow them around, put data in front of them. The way you use Reddit, the sky's the limit. And imagine that the more user data you get, the smarter it gets too. Because what I've noticed is I get recommendations like, here's two tweaks you should make to this Ghost based on XYZ. It becomes self-learning, which is the end goal. What's the process been like? You said you had a company beforehand. This was a tool that you used to start that company. What's been the process with this one? What have you done differently? What do you expect the end result to be?

PitchGhost started as an internal tool for my previous company, just to help me do more marketing. Like I mentioned, my background is as a coding developer, so I am not necessarily the best marketer. I've had to fall into that a little bit, [expletive]. But I actually think our story of starting bootstrapped and then becoming a VC-backed startup actually relates to…

It's what people want. They don't want some pie in the sky telling them how to do it. They want someone who built it themselves, has a use case and can be practical.

It relates to AI. In the early days, when this stuff was moving slower, I was totally comfortable taking the profits, eat what you kill kind of style, every customer we brought in, reinvest right back into the business, growing from there. But AI is just moving at such a pace now that we looked around and went, this almost religious belief that we shouldn't raise money is hindering us. It's time to change that. I saw this tweet the other day that basically said, to keep up with AI, as we've been talking about today, you practically have to be unemployed. You have to have so much free time. You have to make it your full-time job.

Yeah, pretty much.

That is one of the main reasons why we decided to raise, because money was becoming our limiting factor. I feel like for people out there, and you have to be a raisable business, but if you are a business that in theory could raise and you feel like money is your limiting factor, that is the time.

I would agree, because if you don't get the raise, then the business isn't ready to be raised in the first place. This is not the free money days of 10 years ago where they're just throwing checks around to everything out there. It has gotten a lot harder. VC firms have gotten harder. Especially here in Atlanta.

My number one advice to any founder who happens to be here, and I guess this is true advice for pretty much anywhere if you're not New York or SF: you must, must, must have revenue. Do not go to people with an idea, especially in an AI era where we're past ideas. You need to be able to prove it. When we raised our most recent round, we had already passed $150,000 annually in revenue we were generating. And that is a requirement, I think, to even get in the door, and we did not raise a big round. That was the bare minimum.

To just get in the door.

If you haven't tried to build it yourself, a VC won't even look at you. Unless you're working on some really difficult problem and you're from OpenAI.

Yeah, and they have you sign an NDA, otherwise they're going to steal your deal. Okay, so Phil, how about you? As much as I want you to brag on your own company, let's use a use case of Solomon Brothers, developing the website, some of the stuff you've done for us, because we are big enough of a business to have the money to do these things, but small enough that we were operating with FoxPro database systems from 2006. Talk about what this project has been like and how you guys at Pilotship take on these kinds of projects especially.

Yeah. Pilotship.io is my company. I'm a co-founder of it. Really, what we're focused on is being the tech team or the AI team for companies that never really felt like they were technology companies. We work with traditionally non-technical businesses in that $5 to $100 million range, small to mid-sized businesses. And we're the team that comes in and helps them understand what is possible. We're also the team that comes in when they're getting quoted 40, 50K for a website. We come in and we say, guys, with AI nowadays, you can do this for much cheaper, and we can build a lot more compared to what some of these old agencies were doing.

Yeah, that happened, by the way. That's why marketing is struggling if you're a website developer.

But anyways, it's been an incredible project, mostly because we get to see what it looks like to take a business that is operating on old infrastructure, old systems, but growing incredibly fast and trying to catch up with the market. Solomon Brothers is pushing really heavily on e-commerce and becoming more of a nationwide name, where you guys are selling your jewelry not just in the Atlanta market, where you guys are the big dog. Being able to put yourself out there and build that infrastructure has been incredible. From our perspective, a lot of what we've been able to do is go from just working on a website and building the marketing side of things, which is the easy side, to pushing into, okay, how are we connecting to y'all's infrastructure, solving more difficult problems, like how do you manage inventory and handle syncing inventory from an old system like FoxPro into the Shopify era. A lot of what we love to do with you guys is also just show you, hey, you can connect your AI agent to Shopify and do a lot of this incredible work. And that, I think, is…

And that's the next step, right? You talk about these MCPs, you talk about all these connectors. Well, when you've got a million connectors, now you're going to have an output tool to get all of the information you've built. Because look, all I care about is a website that can get customers and close customers. When people search, we land, this is cool, and then a mechanism to convert. The rest of the stuff, the pretty pictures, that's very important, don't get me wrong. But without the technical stuff, the pretty pictures are worthless. Yeah, no, it's great, man. Honestly, I think this is something that we have to do more often. I think we can get more people in this room, believe it or not, and I think we can keep these going. Especially the marketing world, they're not hearing these stories from the right side. They're hearing this from some of the bad actors and how they're taking some of this tech and charging you $6,000 more than they were previously because of this tech implementation. The biggest piece of advice I'm going to give you, and I think these guys would agree with me, before you go to somebody and start asking: you have years of incredibly robust, powerful data. Anyone who tells you otherwise is just trying to steal your money. Your data sets plugged into the right inputs, pulling the right outputs from a Claude, Shopify, OpenClaw, I don't care, is such powerful stuff. Anyone telling you otherwise is part of the problem. You've been listening to the Marketing MadMen. Mark, Phil, thank you guys. I appreciate it. We'll catch you next time.