The AI Models That Are 100x Cheaper

I actually spend my time on two things. As an organization. I don't actually spend too much time on recruiting anymore. Being quite frank, when we were much smaller, I did spend the time on recruiting and building up the team. Now my team spends the time recruiting other great people. But if you have great people, they don't want to manage them. So they'll work to find really good people and they deal with the headaches of recruiting because it is a long process. It's not simple. I don't care if you use AI, you still have to build the relationships with the people and talk to them to figure out who's the right fit or not. I'm not saying AI can't help you speed up some of the process, but it just takes quite a bit of time. The second thing is I spend a lot of time hitting up companies to buy and building the relationships there. I spend an arm, like I would say, that is probably on a daily basis at this point. 60% of my time, it's maybe 70% is just reaching out to businesses myself and trying to see which ones I want to buy.

0:00

And how are you thinking about them now? Because some people have the angle where it's like, okay, I'll just buy these logos and then get rid of the people. Or are you thinking about, hey, I'm going to buy these logos. But also maybe I'm internationalizing. I'm internationalizing, um, because the clients that we want, they require us to be in these areas.

1:06

Yeah.

1:23

Uh,

1:25

so the way I look at it is we've done enough deals where I did one deal where the founders did not stay. Actually I did two deals where the founders did not stay. One of them was a software company called Answer the Public. So we already knew the founders weren't staying and we felt they were doing things wrong and we didn't want the founders to stay. So I didn't mind that one. We did another deal in Asia Pacific called Search Guru where the founders were checked out, they didn't want to stay, they had a guaranteed earn out. I wouldn't ever structure a deal like that anymore. But the upside was very capped and I didn't understand how relationship heavy Asia Pacific was. So for a service based business, whether the founders are checked out or not, I'll never buy a business again where A, the founders are checked out and B, they don't want to continue. That's two different things. You could be checked out and wanting to continue, but if you're, if you're checked out, that means you're not putting enough effort into there. So why would I want a business where you're not willing to spend your own time, especially if I'm compensating you. And if you don't want to work there, it just creates more risk on me. So I'm looking for businesses where the founders want to stay in the game for at least three years. And I'm ideally looking for founders who are like, yeah, you know, it would be great if we can work out a deal where we continue longer than that because it shows they really care and they believe in the future. And my fear is when you buy businesses and people aren't in it long enough, there's knowledge transfer, client relationships, and the list goes on and on of stuff that may not be able to be ported over, which just creates risk. So I try to find companies and I build relationships with the right people and then go from there. And the way I look at M and A is what are the regions? So if you look at ad agencies specifically, and I'll ask you this question, do you know where most of the global RFPs come from?

1:28

Like what countries, I would imagine.

3:33

Let's go top five. Let's rank them.

3:37

Top five.

3:39

Us for sure is number one. Huh. What's number two?

3:39

You probably have. I would say I'll give you somewhere in Europe.

3:42

That's English speaking. What's number two? Uk.

3:46

Yeah, yeah.

3:51

Number three. Close to your homeland?

3:52

Japan.

3:57

Your homeland.

3:58

That's close to my homeland.

4:00

I'm saying your homeland because you were born in Taiwan, right?

4:02

I'm not.

4:05

Oh, you were born here, but your parents are from Taiwan.

4:05

Taiwanese, Cantonese. So China.

4:07

China is number three, huh?

4:09

Yeah.

4:10

Number four and five is kind of like a split Middle Eastern. Number four and five is a split usually between France and Germany. Number six is usually some other areas in apac.

4:12

Yeah. And so where were you going with this?

4:24

So I'm trying to infill in the regions where we get the RFPs, so that way I don't have to travel as much to those regions.

4:27

That's a lie.

4:34

You would still travel. I'm not saying I won't travel. I would still travel, but I wouldn't need to travel as much because when we look at the ROI of US me traveling, what we found is building relationships has helped building the pipe for the long haul. It would just be more effective if I traveled less, bought the companies, and then paid to speak at the biggest events within those regions, I guess.

4:35

Let me ask you this question. So you had said before doing this Podcast, you're reaching out to at least what, 10 people per day on the. For sales and then maybe like how many people are for M and A per day?

4:58

I've slowed down on the sales side, so I'm probably reaching out to three, four people max a day for sales. M and a, maybe 15, 16 a day.

5:11

Yep. And so my point of asking this question is like, this stuff changes. Right. Cause what I'm doing right now is probably I reach out to 15, 20 people a day that I know because it like my thing has all my. It knows my connections. Right. For single brain, these single brain pilots. Because we're right now we're just trying to figure out like, how do we just have more of these customer conversations? And inevitably like most of them lead to. They want to engage with us.

5:20

Right.

5:44

And then from there we do customer development and then it's like, okay, how do we scale it? But then in a year or two that might change where it's like, oh, damn, this stuff is working really well now. The scaffolding is working really well. Maybe it shifts over to something more strategic calls, or maybe it shifts more into, you know, executive recruiting, or maybe it shifts over to M and A. Right. So this stuff ebbs and flows. It's not to say that sale, you should always focus on sales every day. It's what is the highest leverage thing that you can do every single day, that if you did that one thing, the day will be good. And that's the way to think about it.

5:44

Yeah. And I think that should change every single day or every single week for most entrepreneurs. And I think a lot of them don't keep adjusting. It's like just because something's working and you're focusing on it doesn't mean it's the thing that you should continually focus on over the next 30 days or 60 days or even year.

6:15

Yep. By the way, so I want to talk about why open weights matter for marketing. So open weights are basically these open source models. Right. So you can use like a GLM 5.2 or a Kimi K3 as of this recording. These are strong open weights. And, and so when you use these, when you think about the cost savings, you're talking about 25 to even a hundred X savings on these open weights. Right. So a lot of people are talking about, oh, like why should I pay the frontier models like anthropic and OpenAI if we can get these open weights? And so this matters for marketing because if you think about when you're generating a ton of creative volume. Okay, you think about you're generating maybe AEO SEO pages. You think you're generating all these, all the strategy, right? Like how do you think about mixing this up from, from a token optimization standpoint? So my stance on this is that 5% of your the strongest strategic thoughts that you have should go to the frontier models, right? That's where the most expensive stuff is. 15% goes to, maybe you're paying subscriptions, for example, and then maybe 80% goes to these open weights. Right. And there's plenty of things that you can use right now, like open router to, to kind of handle this. But everything is downstream because marketing costs a lot of money. Like you think about all, again, all the things, all the emails you need to send, all the sequences that you need to do, maybe like all the stuff that you need to do, right? And so again, keep in mind you're talking about a 25-100x savings. But that's not to say that you should use your open weights for, for everything because sometimes you do need, you know, the more powerful weights.

6:32

Well, we work with quite a few organizations. I wouldn't say it's the majority, but it's a decent enough number because they're publicly traded. Some have a lot of restrictions on what they're allowed to use when it comes to open source and what they're not allowed to use. Some companies are restricted on using technology, let's say from China and places like that that we work with. And what I always tell people in marketing is there's a big problem in which I see a lot of marketers using AI and specifically frontier models for really basic stuff. If you want to end up using the frontier models, use them for stuff that's more complex and what they're really needed for. You can use some of the older models, which are much cheaper, or you can just use your subscriptions to get a lot of the basic stuff done. And you don't need to pay the latest, you know, model released by Claude for something basic like keyword research. Like it's crazy on how much more expensive that's going to be versus paying an older model that can do just as good of a job.

8:00

Yeah. So let me show you the meme here that I like looking at. So the meme here is, let's see,

9:01

it's this one over here.

9:08

So, Neil, do you see my screen?

9:09

Right, A saxophone. Oh, no, it's a flamethrower.

9:12

It's a blowtorch. So this guy is using a blowtorch to light his big cigar. Right. And so when you use frontier models to just do like a simple search, you are wasting tokens. Right. So that's what Neil's ultimately talking about. So I think this meme goes a long way to explain that. Yeah.

9:16

And almost every single company we work with, especially the larger ones, they're very cost sensitive now to how much they're spending on AI because most of them have not seen the revenue growth compared to the cost that they're spending.

9:32

You know what's interesting? So Neil, I was on a chat last Monday and so I'm in these like AI executive groups and again these are people that operate at, you know, we're talking about nine figure, ten figure companies. And some of these people are like the AI transformation people, for example. So the topic of the conversation was what's the ROI of AI and how are people managing it? Right. Nobody really had good answers is what I'll say. It's like, you know, you shouldn't message, you shouldn't manage like token usage or measuring people on token usage is stupid. Which I do agree, like primary line token usage is stupid. But you know, nobody has a clear model right now. And then one of the people even said like why are we even trying to measure this? Because this is akin to the Internet, which I think is wrong because this is not the Internet by the way, costs. You like it's a flat cost, right, that you're paying, but when you're, when you're paying API usage costs, it's not the same cost. Like you do have to measure in some way, but also when someone else like, oh, you know, measuring on time, time, estimated time saved is also not a smart thing because it's estimated time saved. Right. So how do you measure it? Ultimately at the end of the day, I think is still a conversation that we're, we're, we're having. But again, on a call with like 10, 12 people that think about this stuff all day, nobody had a good answer. Isn't that interesting?

9:46

It is. And even people who are really smart and sophisticated, I don't think there's exact solution because I honestly don't think most companies have figured out how to use AI to create more revenue other than the AI companies or some really tech forward companies. I believe the majority of organizations who have used AI and see financial ROI have seen it from cost savings, not necessarily revenue growth. And that's the harder one to tackle in which yes, we all want cost savings, but what we really want is more revenue growth. Revenue Fixes everything. Cost savings is definitely a plus, but how do you get more revenue from using AI? And there's some obvious ways like, hey, how about rank on ChatGPT? But when it comes to actually token costs and can you using it for work, what we're finding is companies are using it to get more done, but their customers are expecting more. So it's not like they can do the same amount as before or less and still make the same amount of revenue. And that's the conundrum. It's, in theory, you should be able to get more growth, but because everyone's doing it, the customer's expecting more.

11:00

So what I would say is this. I find that measuring revenue per employee is a healthy way to look at it as of today. So one of the people on the call, their revenue per employee last year was 600,000. Today it's 1.2 million. And so I'm like, okay, there's some efficiencies there, right? And they're not necessarily looking to cut. I will say one more thing. The narratives that a lot of the CEOs that they, that they work for, previously all the CEOs were like, how do we cut more people? How do we cut more people? How do we cut more people? And they're all saying now that that doesn't work anymore. And they've like backtracked on that. And so I find that interesting. Like, by the way, you didn't see the founders in the room reacting to that much, right? But the people that were working at companies, like, yeah, you know, we had to backtrack on that. So I'm just reporting on kind of what happened in there. But I think revenue per employee is one thing. I will say one more thing too. All right, So I wanted to take a moment to tell you about my podcast co host, Neil's agency called MP Digital. And they work with a whole host of global companies or a global organization. Also, Neil has SEO tools such as ubersuggest. And as to public, all you have to do is go to npdigital.com to learn more and we'll see you on the other side. With these single brain implementations that we've been doing right now where we set up these, these, these, these, these agents, these managed agents, we've noticed that a lot of these calls Neil, it's more like tech support now. So we've realized that a lot of these companies, even though we set these, these agents up in Slack, that these, these marketing agents, they need a lot of support and, and At a certain point, it's just like, whoa, like, why don't we just charge them on outcomes, right? Because it's like, you just want these outcomes and like, your team's not gonna learn how to use this stuff, even though they live inside of Slack. And so what I kind of landed on, you know, talking to my CTO yesterday was like, you know, at the end of the day with this, this AI stuff right now, maybe the edge is being on the cutting edge, right? And that is very much something that is. It's an advantage. Because I remember reading a tweet, it's like, oh, if you're on Twitter right now and you're into AI, maybe you're six months behind a Frontier Labs now the rest of the world is probably like way behind you, right? And so I. That's why I think that if you're in services right now, I don't care what service you're in, if you have the AI augmentation piece, if you're in services, there's going to be more demand for your stuff. Because to Neil's point, the expectations keep getting higher and higher. So that actually means more people need to be hired.

12:12

I don't know if I said this on a previous podcast, but we're not really seeing too many customers saying, hey, we expect more because of AI now do a lot more for less money. We're seeing customers now focus on how do we grow more in this competitive market? What do we need to do to win? I'm not saying that doesn't mean they want more for less or whatnot, but the way they're describing their problems now is very different in which they're more focused on, we need growth. What do we need to do to grow? Because if you're publicly traded or even if you're privately owned, you can increase your profitability, but if you can't grow your top line, people don't like that business as well. And you can measure things like revenue per employee and even profit per employee, but those metrics, depending on the company type you are, only take you so far. The reason I say that is a lot of the people listening to this marketing school podcast are global audiences. There's in many situations where you may decide to fire people in certain countries and hire two people in some other countries to get a little bit more done, but there may be a fourth of the price each. So your revenue per employee may decrease, but your overall and your revenue or your profitability per employee may decrease, but your overall profitability may increase. In your overall revenue may start growing because you're able to get more done due to the fact that you have total greater headcount, but you're paying them less. Less per person.

14:28

So I think with any metric, you need to have a pairing metric. What I mean by that is you can say revenue per employee. That's one. But every metric by itself can be gained. Right. So if you do revenue per employee plus, to Neil's point, your growth rate, that's a good mix. Right. And. And so I would just encourage you to look at that. And I think the key takeaway here is that nobody's quite figured it out yet. We're all kind of figuring it out as we go, so don't feel like you're.

15:56

You're.

16:17

You're that behind. So, yeah, I think that's my take on. Anything else you want to add to this?

16:17

No. And I think you nailed it with the very last point. You guys don't have to solve all these problems right away. We're still really early into this whole AI revolution. If it's a baseball analogy, I think we're only in the first or second inning, and I think we have a long way to go, and time will tell what happens. And I think for people to try to solve all their problems right now, especially when it comes to marketing and how you implement this technology, I wouldn't worry too much because it's changing so fast, and there's gonna be a lot of solutions that affect you for your whole corporation and not just marketing. So just be flexible. Don't be tied to any one solution, and just be patient.

16:23

Let me talk about something with open weights, because, you know Jensen, you know Jensen, my Jensen Huang, my Taiwanese brother signed up for X. This is his first post. So before I call out the post at all, I just want to show you this. His First Post got 64 million views. Okay, so you want to talk about all this personal brand stuff. I'm sure Neil and I, we create content. You know, whether we like it or not is.

17:01

Is.

17:23

Is. Is one thing, but I do enjoy hanging out. Like, it forces us to hang out. Like, that's a cool thing, right? But here's the thing. Does Jensen have to create content? No, not necessarily. Right.

17:23

He just.

17:33

I'm oversimplifying it, but he creates a product that people want, and then when. When he talks, people pay attention.

17:34

Right?

17:39

And it just so happens that it's one of the biggest revolutions right now, and he's at the forefront of it. So of course he's going to get more views. But he put out an open letter on open weights. So he's just saying, look man, like open models strengthen safety and cybersecurity, accelerate innovation diffusion and enable sovereignty. Because everyone's talking about AI sovereignty right now. You don't want to be just tied to one model. I just think this is great because he put this out there, he put this letter out there, but then immediately it got co signed by Satya Nadella and it got co signed by all these people. Even Sam Altman co signed on it. The only company that didn't co sign on it is who Anthropic. Right. Which I think is hilarious. So it's like a nice marketing play because he's here to support everyone. Like the open weight ecosystem helps Nvidia, but he also wants to get everyone on site as well. And it also shows who's not on site. Yeah.

17:40

And for them it benefits because he's selling the chips. Right. So he's making money. No matter what models you want to end up running.

18:25

He's good with whatever. So he needs both ecosystems. But I agree, you shouldn't just have like a duopoly that control all the power. So like by the way, if you talked to Neil and I a year ago, I don't know about you, I don't want to speak for you Neil, but you talked to me a year ago, like you asked would I ever use open source models or whatever. I wouldn't even know what you're talking about. Right? Not that I don't know what you're talking about. More so like I wouldn't want to put the time into it, but now that it's such a big thing as it ties into our businesses, it doesn't make sense for me to not understand it. And so I would encourage you all to play with open source technologies. Right? Like Buzz that just came out from Jack Dorsey, like that's open source. Hermes is open source. Open clause, open source. Right. Using these models like GLM 5.2 or Kimi K3, these are all very powerful things and you combine them with what else you're using. And then the cool thing is I actually believe, Neil, that your company or my company, we're going to need to build our own language models.

18:34

Okay.

19:28

So we might fork like a Kimmy K3 that's close to frontier model. Right. We might take our opinion, the data that we have on helping our clients grow their revenues and we might make our own models. Right. I think that's very much going to be A thing just like other companies like Cursor has done that they forked a version of Kimi and they made their composer and they have a lot of data on people engineering and coding and they made their own. Right? So I think that's going to become more of a thing.

19:29

My opinion from a year ago, I still hold the same on it, in which I believe Microsoft and Google are eventually just going to undercut the market and give away a lot for free and let companies do a ton for free and you're just going to be on their ecosystem and it's powered by all the other revenue that they make and eventually they'll charge you some here and there. And yes, you can use different open source models or whatnot, but the Google versions and the Microsoft versions will be so cheap or affordable or free in some cases you won't care and you won't end up experimenting with a lot of this stuff. That's what I still believe is going to happen in the future. I just think it's a question of time because unlike a lot of these other AI companies, Google and Microsoft just make so much from their core business, they can just keep undercutting everyone on pricing.

19:57

So when I talked to the Google Ads team, so we had this conversation, they're like, so what do we need to do to get you guys to use Google more or whatever? I'm like, dude, everyone likes to use Google, but you guys need an MCP that's available to the public first and foremost so we can run these things, right? They're like, oh, okay. They're like, you know, we've been offering Gemini for free to these people. We're offering more tokens so they can, they can use Gemini more. But a lot of these companies just laugh at us, right? And I was like, the reason they laugh at you right now, just right now, is because your models aren't good enough, right? And I, I said, look, you know, people might be using Nano Banana, which is our image gen model, but chat GPT images too is, is really good, right? So I'm just saying, like, it's just a matter of. You got Daniel's point. It's a matter of time. I believe you guys will catch up. But if you bundle poo poo, like, I don't care if you bundle anything, it's just poo poo, right? So not saying Gemini is complete poo poo right now, but it's poo poo compared to the Frontier models. And so I think they'll get there, but it's just interesting that they're just, they are trying to do that, they're trying to bundle and they're trying to undercut, but they can't quite do it yet because the product is not quite good enough. So.

20:41

But on some of these things I would have to say, okay, William, go potty then use the bathroom in the theater.

21:44

Potty on the ground.

21:52

So use the bathroom in the theater. So the, with the, with some of these models I have to say, like when you're having it do analysis, like I was having it review a cost seg model. It's for real estate and bonus depreciation. It's like some accounting thing.

21:54

Of course you were a neo. Of course.

22:10

Uh, Gemini was the most accurate LLM out there for analysis by far.

22:13

Now whether I think Gemini sucks or not is one thing, but I used it to plan a lot of my travel and like that was last minute trip to Italy last week and it, it did a really good job. I'll just say that.

22:22

Yeah. And when you do a cost seg, you pay a professional to also do it. So I, I had Gemini do it, I had Claude do it, I had Chat G B T. Right. And I'm telling you, Gemini produced the best results.

22:31

And were the other ones by a factor of like what?

22:45

Okay, so let's just take a dollar amount. Let's say it was a hundred dollars, right?

22:50

Yeah.

22:54

Gemini was off by $3, so 3%. Okay, chat GPT. And Claude, one of them, I don't know which one was what. One of them was off by 30 something percent. The other one was off by 50 something percent.

22:56

And we're not talking about a hundred dollars here, guys, we're talking about a lot of money.

23:08

So we're talking millions of dollars. With millions of dollars in something being off, it's like, no, this is not good.

23:11

Yeah, yeah, yeah, yeah. So I think, look, I think every model has, I will still, I will still say for today, Gemini is not quite there yet and they've kind of delayed their new flagship model. I do think they're going to catch up. They're going to be fine. They have all the infrastructure in the world. And to Neil's point, I do think they're going to undercut. But yeah, so I do want to call out something else. So if you look at anthropic. Okay, so a couple, like last year actually when we were speaking at the HubSpot conference, I called out how important MCPs were. So MCPs are model context protocols, right. And you know, it's basically like an enhanced version of an API. It was very expensive back then. It was very janky. Right? And so it's like at a certain point I was like, okay, maybe we don't need to use MCPS anymore. We need to use, just use the APIs for that application programming interface. Let me tell you where I'm going with this. Anthropic just launched a new version of their mcps and I didn't understand all the technical jargon. So I'm going to explain why Anthropic's new MCPs matter for you, especially if you're in marketing. Because I actually had a table made for me, Neil, and this is pretty cool. So I had a conversation with my friend Grok. Okay, so Grok, I was just like, hey, like simplify this for me. I have clickflow. Okay, and I have Carrot. And if we use the mcp, what are the practical implications there? Like what are all these capabilities allow us to do? So keep in mind new MCP came out, there's a lot more capabilities now. I actually think every software needs an MCP now and they need to upgrade to this new version. Okay, so first and foremost a new capability with Anthropics MCPS is mid process human approval clarification. Okay, so now if within clickflow, which is where we have a content creator. So agent might start rewriting a declining page. So let's say the traffic's coming down, it can pause and ask hey, do you prefer a more technical tone or more benefit focused? Right. And then it can continue and save the draft or Carrot, which is where we make LinkedIn account based marketing ads. Agent might generate personalized LinkedIn ads for 20 accounts. Pause and it asks hey, approved is created for company X or regenerate with stronger social proof. That's one. Okay, I'm not going to go that entire list here, but you can have it run bulk jobs which it couldn't before. Like these things would just break. It would just break with like very fragile. Right. Reliable, high volume use case. You can have team members, eight multiple agents running this. So you can use this inside of like Buzz for example interactive previews. That's cool, right? So you can preview the live before and after page or you can like preview side by side ad creatives. Right. So for ubersuggest or answer to public, which is Neil's tools or these tools that I have over here, you can just do a lot more here. Oh, it's cleaner for enterprise or team authentication, so easier SSO managed authentication. So my point of calling this out is that as a marketer it's important for you to understand the enhanced capabilities of your software or whatever it is that you're building. And then these MCPs, by the way, other agents are going to find these capabilities and they're going to use them. And if they use them more and more, maybe there's going to be more, you know, maybe whatever tool that you have is going to become more popular. But I think as a marketer it's helpful for you. Whenever someone tweet anthropic, just release this new thing, I just hit the grok button. What does this mean for my businesses over here? What are the practical implications? Can you make a table here? And then I understand it and then I can disseminate this to my team. So don't get caught up just because you're quote unquote not technical. So that is it for today and we will see you tomorrow.

23:17