Very True by Verissimo

Alex (00:00)
Every AI company on Earth is making the same bet over and over again every few weeks, whether they know it or not. And their gross margin is keeping score. I spent the last few months trying to figure out how you'd actually measure who's winning that bet, not with a vibe, with a number. And I think I found one. As far as I can tell, nobody has written it down yet. It didn't come from a spreadsheet. It came from a mechanical engineering class I took 17 years ago about what happens when a car hits a speed bump.

That is not a joke. And by the end of this, you'll see exactly why it's the same problem. So today I'm going to tell you why SaaS worked, why AI breaks it, and the number I think that tells you whether an AI company is built to last or built to get eaten. This is the first pass at an idea I'll keep digging into on future episodes and over on my substack. So subscribe and stay tuned. There's a lot more here than one episode can hold. Let's get into it.

Alex (01:46)
Open Claude or ChatGPT or Gemini, it doesn't matter. Click the model selector. Count what's in there. Four or five model families, several versions of each, an effort setting with levels, a thinking toggle that doubles all of it, and you're looking at something like 80 permutations in a product that was supposed to replace a single empty search bar and be able to do everything. Most people look at that drop-down and think it's a product design failure. It isn't. It's deeper than that.

That drop down is the entire strategic problem of every AI company in the world, rendered as a UI element. Every one of these companies is making the same choice, except they're making it across their whole product surface every few weeks and with their gross margin as the scoreboard. I spent the last few months trying to work out how you'd actually score it. I think there's a number. I don't think anyone has written it down yet, and that's what this episode is about.

To kick this off, I want to start with SaaS. SaaS seems to have become the victim, the punching bag, whatever you want as it relates to AI. SaaS is dead, long live AI, but there's something that's simply incongruent about talking about SaaS and AI in the same sentence. Anyone who actually uses these products, understands that They're not mutually exclusive. So what what's that coming to? And the answer is actually pretty simple. SaaS is a business model.

Software as a service. It's software sold as a service, software delivered as a service. That is a business model. Multi-tenant cloud-hosted software, that's the technology. The underlying technology that supports the product and the business model that drove the SaaS gold mine over a 30-year period.

AI, on the other hand, is a technology. And there's not yet a business model that sits on top of it. Certainly nothing as beautiful as the SaaS business model. It's not a criticism. It's just where we're at right now. And so before we jump into all the AI stuff, I want to drill down on what actually made this gold mine of SaaS work so well for so long.

And then ultimately where the wheels came off.

The first thing I'm going to mention here, and hopefully I don't lose people in the details, is the DCF, the discounted cash flow analysis. Basically, the DCF is how we define value in a business. It is the net present value of the sum of all future cash flows. And that is how we value a business. And the idea is that because it's all future cash flows, it takes into consideration everything from IP to brand, not just pure.

numbers on a page, finance accounting, boring stuff. Because over all time, if those things, brand, IP, whatever, doesn't make itself known on paper, then it doesn't really matter. So I learned about DCFs when I was in college. I did a bunch of them when I was in investment banking. When I was in investment banking training, one of the people got up and said, DCF, yeah, that's just fun with Excel. You can make it say whatever you want. And it's true.

Because there's a lot of variables and they're all very sensitive. Now, if you play with those variables, you start to understand their sensitivities and how they impact the ultimate output. And the more mature a company is, the more liquid its capital markets are, the more accurate this can be. But fundamentally and on a conceptual level, it is, I believe, absolutely true.

To describe the value of a business. So if we just look at the value of a business and we isolate it into four simple variables, which can map to all the qualitative things that we love to say about companies, those four variables, and I've covered this before, and I'll link to a more detailed explanation, but they are short-term growth, long-term growth, free cash flow margin, and discount rate.

And effectively I'll explain what those are real quick. So big G, we'll call it, which is short-term growth, which is in your growth period of your DCF, for how long you want that to be, it could be five years, could be 10 years, it could be 15 years, that you're actually specifically modeling out the company, how fast is it growing? And you can do that as a CAGR a compounded annual growth growth rate across that entire period. But that's that first variable. The second variable, long-term growth, which is little g.

Which effectively represents the market size and how much faster we expect this to grow than GDP, for example. The third variable is free cash flow margin. this is generally gonna be steady state free cash flow margin, but it's really gonna be averaged across the growth period and then whatever that ends at and takes you into that terminal value is what we call it, that last term where that little g comes in.

And free cash flow margin obviously matters a lot. You can earn a dollar of revenue, but if it only brings in, five cents of free cash flow, then that's not great. Whereas if it brings in, 80 cents of free cash flow, that's pretty amazing, just in terms of the power of the business. And then the fourth variable is the discount rate, which is how we quantify the risk of a business.

And the way that we do that without getting too nerdy into finance is using the weighted average cost of capital, which relies on the capital asset pricing model. This is all nerdy finance stuff. That is to say, for a mature company with robust access to the capital markets, both debt and equity, we actually can understand the quantified risk of that business based on how much.

they are able to charge for their equity and charge for their debt to the active markets. And that is the combination of those two things and their weighting and how the company's capitalized represents those things. So everything we do in a business.

You can speak to one or multiple of these variables. And SaaS, especially Enterprise SaaS, just knocked the cover off the ball when it came to all four of these variables. So I'll run through it quickly just so that it's clear on why it was really just so special. and why it's not anymore, pretty quickly.

so the the first one, big G is how fast can it grow? So if you take a company like Service Now or a Workday or Salesforce or a Spotify and you think about how fast can it grow? And the answer is as fast as you can sell it. Right? The demand is there, it's multi-tenant cloud software. You just literally click a button, maybe it's not that simple, but you spin up an AWS instance, whatever it may be, and it can just scale as fast as you can sell it. And that's amazing. And then the demand for it off the charts. So great. Check, check.

The second variable is little g, that long-term growth rate. How much faster can it grow than GDP? And if you look at some of these companies that are at billions and billions of dollars of revenue scale, still outpacing GDP growth by four to five times, you see that the reason this works so well is people never thought these things would grow as fast as they did. And I attribute that to this idea that technology became a layer of the economy rather than a sector inside the economy. The third variable is free cash flow margin.

Now, free cash flow margin incorporates a bunch of different stuff. The first is the gross margin. So that's kind of the first thing you take out, which is variable expenses once you come off revenue top line. And that gets you to your gross profit. and that's kind of first base in a lot of ways. Like that's where that's that's where you start from. And then how efficiently you run the business on a fixed basis is the R&D.

The sales and marketing, the G&A, and all four of those costs relative to the revenue are kind of how we measure a business. Now, where things got a little bit weird with SaaS is that the modern accounting rules do not really allow you to describe what goes on with SaaS that well. And so a lot of the categorizations of what is R&D versus what should actually be COGS, for example, like customer service or solutions engineering, whatever you want to call it.

get really muddy in there. And so frankly, SaaS doesn't have the gross margins that people think it does, because of how you have to service it. and then the fourth variable is that discount rate that I mentioned, which is the risk associated with the business quantified. Now it's very difficult to quantify the actual risk of a business. I'm not gonna get into how we do that with beta and

you know, market risk premiums and everything for public companies because it's not really relevant for what we're working on here. but it does matter a lot. And so everything from, hey, this is a repeat founding team that's been successful together before, like that, that reduces the risk. Hey, this is a known big market. Like, okay, that reduces the risk. But other things that are much more technical can also reduce the risk of a business. And one is like IP, like if you're patent protected, like if you're a

drug company like a Pfizer, like yeah, that massively reduces the risk. but another thing is just the the contracts. What what kind of contracts can you get with your customers? If you're a SaaS company that is selling digital goods in the form of software to companies and getting them to pay you a year or more up front, regardless of what happens to them or what happens to you even,

Most cases, like that's wild, right? And it's actually on an accounting basis from a cash flow perspective, it's your customers giving you a loan. So not only is the revenue locked in, but the cash gets paid up front. And again, this is all about the sooner you get cash, the more valuable it is. That's the underlying idea here. So all four of these variables are critical. and they just they clicked. Now, where did the wheels fall off the bus?

the answer is competition. So if we run now competition through everything I just said, which is what happens when you have a gold mine, it gets overmined, everyone rushes in, and then when there's too much of it, the commodity that you're mining, the price goes down. the answer is quite simple. When competition rushes in, you lose pricing leverage. And when you lose pricing leverage, you automatically have a lower gross profit because it costs you the same amount to deliver. you're just starting from a worse spot.

Effectively. The the second thing that you get hit on is sales marketing expense, right? When there's more competition, it costs more to sell it, More dollars on marketing, more cost to buy the ad words, more expensive salespeople, longer processes. the sales and marketing goes up. and then the other thing happens in R&D, which is that you gotta keep pushing new products, you gotta keep your customers happy, you gotta keep innovating. That's not free.

if you compare that to perpetual license software, you know, the Oracles, the old school names, they there was just like, hey, you're gonna pay us a million bucks and then you're gonna pay us a hundred thousand dollars a year just to keep it working. And if you want new products, good luck. Like you're stuck with us. It's gonna cost you more money to get our stuff out than it is for you to just keep paying us. And that's like evil genius level business model stuff, but that's implicit here and and

Matters lot. The other thing that I always say about SaaS is that you are pre-negotiating your breakup with your customer. That is that was the original idea behind it is hey, we think we can reduce the customer acquisition cost to get a customer if we imply that if they don't like it, they can leave. So people talk about churn as the enemy. Churn is not the enemy. Churn is actually like a prenup with your customers. It's what you tell them so that.

they come is that they can leave, that they're not trapped, that they that that kind of like PTSD that they might have from the Oracle days is not is not there anymore. And then you hope that they end up just staying forever. But in order to do that you have to deliver new products and customer service. So that's why SaaS worked so well. It knocked the cover off the ball on all four of those variables in a really, really elegant way. So you have the high gross margin, you have the low discount rate, you have the high free cash flow margins. And because most importantly

When you have subscriptions, you actually accumulate revenue. So one of the realizations I had back in 2014 when my old boss Harry Weller asked me to nerd out on SaaS, which I've been doing for the the last 13 years or so, is figure out the math behind it. Now people would always say, the what's the calculus?

In this case, it actually is calculus. And this realization I had back in 2014 was that.

The integral of the ARR function is recognized revenue. Revenue is an accumulation function over a period of time. ARR is actually a point in time. So it's the same thing as velocity versus distance covered. And that's cute math, but why it matters is that ARR accumulates into revenue. So once you get that stream going of ARR, it's basically gonna stay forever. And that's incredible.

I'll jump in now to this idea of these like SaaS metrics, LTV to CAC, SaaS magic number, and burn multiple and all these different things that people tried to cook up. Now, I was always kind of not interested in these things because once you understand the math and the actual theory behind it, you don't need benchmarks. I always say if if a if a company needs to know what other companies are doing to figure out if it's doing a good job or not, it's probably not doing a good job. And by the way, even if it is, it's because it's getting lucky and they don't really understand what's working and what's not.

So let's pick on the SaaS magic number thing, which by the way, silliest name for a financial business metric ever. But credit to the concept behind it The idea is just net new ARR divided by associated sales and marketing spend. you have this engine, this growth engine, where you put sales and marketing spend in and you get net new ARR out. So that incorporates churn, upsell, downsell, all that stuff, and obviously new customers.

For a given spend on sales and marketing. And that basically says, hey, if I'm an investor looking at a company, or I'm a CEO running a company, looking at it more from like, let's say, a project finance perspective, I can put a dollar in and get two dollars of ARR out. Now, one of the things that always got kind of glazed over with ARR was that you actually needed to gross margin adjust the ARR to get a proper read on that magic number. A lot of people forgot to do it and

all the benchmarks obviously change and all the health changes. But the truth is that even with that, it makes a huge difference if you're a company at five million dollars of revenue, a hundred million dollars of revenue, or billion dollars of revenue, how all this works and what good actually is. If you're growing five to ten, let's say, then you better have a SaaS Magic number like above two, or it's just gonna be a long road ahead. If you have a billion dollars of ARR,

the overall efficiency when you already have a fat recurring revenue stream coming in, and you look at like again the cash flow that you can throw off, when you have that big chunk that's guaranteed, it it works amazingly. So I'll explain that just to make it a little bit more visceral. Let's say you spend a hundred million dollars to get a hundred million dollars of net new ARR, and that's starting from zero.

Assume that it's really gross margin adjusted. I'm just gonna use the term ARR here for a second, which is exactly what I'm saying the problem is, but it's just a lot simpler to explain the math. So again, put $100 million of sales and marketing spend in, get $100 million of net new ARR. So you go from zero to a hundred. So your margin was basically zero. Again, you have other expenses in the business, so it's gonna be negative, but we'll leave that out.

Now let's say you do the exact same thing next year, but this time you're starting from 100. So now you went from 100 to 200 and spent that same hundred million dollars. So now your your margin on that is 50%.

Now let's say you do it a third year, same exact operation. You go from 200 to 300, you spent 100. Now you're now you've only burned a third of what you earned. And that continues on and on and on. And that again, that accumulation function is what's so powerful. And it also relies on this idea that gross margin is going to be high. So you've got the underlying technology, which supports the business model dynamics, and then the billing mechanism, which also supports the business model dynamics. And those ingredients are absolutely critical.

to the success here.

So now let's jump into an AI powered product and what it loses. So what it loses is two key things. The first is that switching cost.

That lock-in. Yes, you can bill Claude as a subscription, but is it really a subscription when I can just take all my memory and everything and in less than five minutes switch all of my workloads to ChatGPT? That's I that that's why they they pushed out co-work and all these things. They want to make it stickier. They want to get in your face. They'd love to have a device, something that makes it stickier, because in a lot of ways, the main value of AI is that it actually made APIs like automatic.

So moving data between these two things when the work methodology is actually the same. Work methodology is one of the key drivers of value in SaaS, like Adobe, ServiceNow, Workday, like these companies at Alassian all drove how people do their jobs. But if it's all prompt and you interact with generally all the AIs in the same way, then the stickiness and the the learnings of it is not there. I always say that the more valuable a software company is actually the worse the product is probably.

To work with. Like I said, the ultimate level is to have it be a skill on someone's resume. If it's a pain to use, that means it's sticky. So now it's like, let's make this all really easy to use. Again, now the switching costs go to zero. So calling something ARR annualized recurring revenue, implying that there's some contractual recurring nature, it really rocks the boat quite a bit. But that's actually

I would argue it's it's elemental and it's important, but it's second fiddle to the gross margin question. losing money on your variable costs. It's like breaking Newton laws of physics. You just don't do it. Now I heard on a podcast, well, actually, if you're an AI-powered business and you're losing money,

on gross margin, which again breaks every law of finance that anyone's ever learned, then that's good because it means people are using your product a lot. Now, that might be the case, but that also underlies the assumption that there's real meaningful baked-in stickiness, which I don't believe there is because every

Two to four months, somebody is the new hot, great model, and a lot of workloads switch. So the recurring nature, not so much there. and then again, people are making excuses one for the other. it's okay that we have a negative gross margin or a super low gross margin because people are using it and that's good and it'll be sticky. Well, it's not that sticky because we made it so easy to use that everyone just switches all the time. Now, I'm sure there will be but you know, Claude code is just so much better. And like,

In my anec- data of talking to people of who uses Claude Code versus who uses Codex versus who uses cognition, like it's a lot of flavors. It's just a lot of flavors. It's like Hubspot versus Salesforce versus pipe drive, except these things change fast. So there's huge questions about this. But what one of the things that has become a little bit more clear already is that

Value accrues to the coding harness So the ability to actually take a prompt and try to figure out what it means and make that useful for a computer, that actually seems to be the hard part. And that's why people are switching between these things. It actually has less to do with the underlying model. And this is something that Cal Newport talked about as well, which I've seen, which is like at a certain point, these things get so good that for 99% of use cases, it just doesn't matter.

And what matters instead is that usability and that kind of post-model training. So that's all kind of saying the same thing. I'm not a super expert here. I might have used some of these terms wrong. Apologies. Everything's changing very, very quickly. I'm just coming from my perspective. So people also might argue too that, well, an enterprise, if you can really get baked in, then you're gonna have that same stickiness that you had with ServiceNow. To which I say.

I don't think enterprise AI is like working yet. Again, I'm not in an enterprise or buying it, but that's kind of what I've gathered. And we'll see. We'll see how sticky it is. you know, enterprise SaaS, very, very sticky. Enterprise AI, which again, if what's gonna make it sticky, it's actually gonna be the SaaS portion of that AI. It's not gonna be the underlying AI itself. And you might say, well, they're also using AI and AI for implementation. Okay, congratulations. That's great. but there's another

big important thing here, which is that the fact that these model releases are happening so quickly. This isn't like such a novel idea.

All the cloud computing platforms dealt with the same exact thing. Like every time there's an upgrade to EC2 from AWS, engineers have to get used to it, learn how to use it, implement it, figure out how to maximize the products around. and then they push that to the product and the sales team, and they gotta figure that out. So I'll leave that there for now. And then let's push this into kind of the chunk here, which is

I figured out this calculus thing like 12 years ago, and it was like a massive light bulb moment. And every person that understands math and physics and finance and accounting and SaaS is like, wow, that actually is exactly how it works. And I found myself scratching my head after a conversation I had with a friend in June, and just being like, What? There's gotta be math that underlies this. Like, what's the answer? And like I said, we still don't know exactly the business model, but we can understand the business dynamic. And

It took me to one of my favorite classes at Stanford. And I'll before I talk about that class, I'll take talk about one of my we'll say least memorable classes at Stanford, which was math 53 differential equations. basically the way differential equations are taught abstractly is like there's like five different

Ways of solving them, and then you kind of memorize them, and then you just figure out which one to use, and like that's it. And then you just take the class. fall quarter junior year, I took a class called ME 161, Dynamic Systems Vibration Control Design. And this is where the rubber meets the road. All of a sudden, we had a use case for differential equations. So one of the things you learned in that class is about a sprung damped mass encountering an external force.

That sounds really like nerdy science-y complicated, but it's as simple as a car going over a speed bump. A car is a sprung damped mass that has springs and dampers. and then that external force is just running into a speed bump. So we've all experienced this. We all intuitively know that there's a correct speed at which you can hit that speed bump. And that's basically that's the whole thing. We'll end there? No, I'm just kidding. I'll get into it. So

This might seem a little bit less intuitive and the math is certainly more complicated than using like the power rule like you can in calculus. but it's a similar idea here. And so the way that I map this to modern AI powered businesses is that

Effectively, every company is that sprung damped function. And it's figuring out how to do that as dynamically and intelligently as possible in the optimal way. So what does that mean? Effectively the input force on the system is every time there's a new model release. Every time, you know, a new model comes out, everyone's scrambling to figure out how to use it.

It's this new technology. And the question is twofold. The first question is: how can we make our existing workflows and workloads run as efficiently as possible? So push them to the cheapest models, improve our prompting, our harnesses, the memory, everything to make that as efficient and as cheap as possible. then the other question is.

Same thing as SaaS, we got to push new products. If we don't push new products, if we don't take advantage of the latest, hottest, most exciting new advances and understand them and then develop products around them, our competition will eat us alive. So, in my view, this is actually what a company is designed to do. Take amazing technology, put it inside a fantastic product.

deliver that product, sell it, market it, whatever you gotta do, and solve people's problems and do that in the most efficient and timely way possible. And that is actually the power of a business. Now what is a business? It's just a bunch of people, right? It's a a bunch of people that have the same, you know, email domain on their email address.

And that's that's their jobs every day collectively is to take that new powerful technology and as efficiently and as quickly as possible develop the most exciting products to deliver to the market. And that's why we say like companies fundamentally do two things. They make stuff and they sell stuff, and and that's what they gotta do. The iteration cycle just became a lot faster. Now, let's run through a little bit of where that gets complicated. And

The answer is actually in this idea of oscillation. So again, new EC2 came out however often it came out, not as often as every new frontier model. And engineers had to figure out how to use it. And they developed new products and great, wonderful. Now that's happening like every quarter, every month. And every time there's a new shock to the system, something happens. The first is old stuff, can we make it cheaper? New stuff, how creative can we get?

And how quickly can we do both of those things? So, going back to this sprung damped system that encounters an external force, the question is how effectively can you damp? And there's this variable called zeta, which represents the damping coefficient. And you can take a system and figure out how effectively it damps. So an underdamped system would oscillate, right? It hits a new thing and it oscillates up and down, up and down, up and down. An over-damped system.

Basically isn't so affected by the thing and it takes a long time for it to settle. And then you have the critically damped system, which is exactly where you want to be. So if we map that to AI powered businesses and gross margins. So again, the AI powered business, that is a business that's an operation, that's that's hitting a speed bump in a BMW seven series, right? It just feels great versus hitting a speed bump in a dump truck where like the thing like gets air,

And then the gross margin is actually the output here. It's it's what the business experiences and how we measure its efficacy. So overdamped in our case would be a company that's not even using frontier models. They say, hey, a new model came out. We don't even care. Like in in the in the world of Claude, like we're not even using Opus. We don't care about Fable. All of our stuff's still running on haiku. And they're not gonna be so affected by this kind of

you know, speed bump in theory, which brings about a bunch of other issues, mostly competition product. If everyone else is using these great things, like how are you going to compete? The second issue, or the second way you can kind of be wrong is to be underdamped So would be new model comes out, company pushes a hundred percent of its workloads to the new model. so now if we map that back to gross margin, if you push everything to the newest model, your gross margin is going to crater.

Your cost of compute of inference is just gonna go through the roof. Whereas if you are overdamped, meaning you're not even using the newest models, basically, your cost is gonna stay relatively the same, but you're gonna get eaten alive on the competition perspective and on your pricing leverage. So the answer is somewhere in between. And that's what companies need to do. You've got to take this stuff, figure out what you can push to the old models, what can you make as cheap as efficient as possible, and what can you

push to the new models and deliver new exciting experiences and products to your customers. So this variable, I don't think we need to rename it. I call it Zeta because that's what we called it in engineering class. It is the efficiency of a business as it relates to building on AI.

Alex (27:39)
So there's a lot of math, there's a lot of finance. One of the things we haven't talked about is the actual product. And that's obviously not what this podcast or my podcast in general is about, But I think it needs to be at least mentioned here because AI has changed the game when it comes to product. It's reduced the barriers to custom products and adoption to near zero. Now you've still gotta have the idea and think of it and all the details in between.

but making that happen just got really, really fast. And then, like I said, switching between things, porting data, AI is so good at that stuff. But if we just think about the strength of a business, that's what we really care about as operators, as investors.

As employees, you want to have a company, that group of people, that organization that can effectively understand, adopt, utilize, and deploy the newest and best technologies in the most valuable, creative, and accretive ways possible. And if you look at a lot of the greatest businesses that have been created over the last 30 years, everything from Uber to Shopify to Salesforce, Tesla.

These companies did not invent technology. I don't think these companies own any core IP. They figured out how to take a core technology or a couple of core technologies, see the moment in the market, and actually construct the business model and the business operation that enabled them to effectively take those technologies to scale in those markets and continue to do that really, really well. And this is where, the value of a business.

is derived from is the consistency with which you display your ability to do that. So getting back to this concept of Zeta, the idea with Zeta is it can behave like this SaaS magic number. But in a lot of ways it's actually even better. Because like I mentioned, SaaS is an accumulation function, it actually hides a lot.

And that's, I would say, the other biggest distinction here between the SaaS world, the SaaS business model, and this AI-powered business model that we're still looking for. In SaaS, companies could go public with negative 150% operating margins, and people didn't bat an eye. And this is exactly why. Because it's an accumulation function, there was an understanding that if you developed a reputation and an ability that was demonstrated to

Deliver products to sell and to consistently grow, then the investment world said, Great, I want to deploy dollars into this because I know that those dollars are going to be converted into revenue growth that is consistent, and that is what value is, is a company's ability to take capital from the markets and turn it into enterprise value by delivering value to your customers in a consistent and predictable way. So

The one thing that we have to mention with SaaS, because gross margins were high, and because we were running an accumulation function with our subscription, really predictable, consistent, contractual annual subscriptions, growth covered up all problems and ultimately, in a lot of cases, cured all problems. And I was wrong here a lot. I looked at companies at certain scales and saw inefficiencies when I sliced the data and looked at the cohorts.

And you know what? It didn't matter because they grew out of it. You can literally grow out of problems if you are A running an accumulation function and B have high gross margins. Now, if we shift back to the AI powered business models, they look like they're running an accumulation function right now. And I think a lot of finance people, investors, CEOs, operators, whatever are still looking at them that way. They're talking about net dollar retention, they're talking about

magic number, they're they're using all the same terminology that worked and made sense and was developed under a lot of the core fundamental assumptions that underpin SaaS. When those assumptions change those little hacky metrics and cute ways to say that my business is great, they fall apart. And the simplest way to explain that is like I said in a in a high gross margin accumulation function, growth cures all.

When you have a business that has variable, variable costs, meaning, like we're talking about, your ability to adapt and damp and absorb the new models, which are expensive but very valuable. If you mess that up, you actually can go gross margin negative. And if you consistently mess that up, then instead of growth solving your problems, growth actually

reveals your problems. And it's put very, very simply. If you have a chart that goes up and to the right and then you multiply it by a negative number, all of a sudden it starts going down to the right. And the more of that it does, the worse off you are. This is a lot of what happened with the misunderstandings around We work, right? It was a negative gross margin business. They were renting things for less than it cost them to rent and manage it.

And at a small scale, people said, that's okay because they're building a brand. And they did build a brand and we're sitting in one right now. but when there's not a fundamental accumulation function and a high gross margin point that you can know and understand in the future, then that just doesn't work. So that is the fundamental difference here, is that growth is going to expose these things. So the question now is what do we do with this?

Okay, so we've got okay, Alex just spent half an hour trying to explain some really complicated engineering math that I have no relation to because I'm a product computer scientist, an investor, a finance guy, whatever. I'm just not a mechanical engineer. so why does this all matter? Now, I'll bring it back to the SaaS magic number idea.

Alex (33:07)
So the reason that the SaaS Magic number was so kind of beloved and used almost universally in one way or another. is that it was really simple to calculate and really portable or seemingly portable across companies at different stages and different markets and everything. And

Question is now you're giving me all this differential equations, dynamic systems stuff. Like that doesn't sound portable, that doesn't sound easy to calculate. But I'll draw a very important distinction between the world we live in now and when we were calculating SaaS magic number, you know, in 2013. And it is very simple. And it's actually AI. It comes back to it. surprise. we used to have to look at monthly or quarterly net new ARR, and then look at the

Quarterly or monthly S&M spend. And then people realized, well, actually, we can do much better attribution than just looking at that broader S&M spend. And then people realized, well, we can actually do that on a more of a micro basis. And then Stripe started managing subscription billing. And then sure enough, Stripe just bought OpenRouter. So actually now it understands the cost side in addition to the revenue side. So

When we look at these absorption functions and this ability to damp these new models and these new technologies, it's really important to understand that we can actually access this data with a simple prompt. All we have to do is come up with our function for how we understand gross profit, which is not that complicated. It's just what revenue did we earn on each part of our product.

And then how much did delivering that part of our product cost us? Now I say part of our product because like I mentioned earlier, there's two fundamental problems at play here that then need to be collapsed into one. But if we can't do the effective, cohorting, then we're not gonna be able to pull it back together in a way that's meaningful. So you've got to look at, okay, for our existing workloads, like what are we doing?

And how efficient can we make those? And that's like one part of the team's job. And then the other part of the team's job is: okay, we need to figure out, hey, these new capabilities just came out with this new frontier model. How do we make the most of that? How do we make that absolutely most valuable version of whatever our company can manifest across its existing and potentially new customer base? so what you look at is what is the revenue being driven by a certain

model and what is that model costing and you can do that in every portion of your product and you can slice and dice that a million different ways with a single prompt and all of that now sits inside Stripe because Stripe bought OpenRouter. Now Ramp, other companies are all releasing similar things that help you manage the cost here. I would also point to the fact that Stripe buying a company for billions of dollars, Ramp rolling out a new product and dozens of other companies focusing on this should tell us something.

Know everyone's focused on well, if this costs so much, that means it must mean it's so valuable. And just using that as justification for like crazy AI valuations. But if it's costing so much, why don't we just take that and face value? It's costing so much and it's representing a real risk to the business. Again, when when cloud cost optimization companies started becoming almost ubiquitous.

When sales and marketing optimization companies started becoming ubiquitous, that meant this stuff was getting really expensive at scale and that it starts really impacting the overall unit economics of the business. So again, to kind of bring it all home, like if you're building a company or investing in companies that are powered by AI, which most companies are, then you've got to figure out a way to understand your own efficiency.

And your efficiency is comprised of two main things. One is existing products getting cheaper and more efficient, and new products using the latest technologies that are out there and getting more exciting and interesting and competitive. And then you've got to figure out the rhythm of doing that. And this is where the oscillation function comes in: you'll see every time a new model release comes out that it's

Going to go through a process of inefficiency to efficiency. And that is the name of the game. That actually takes me back to a really interesting analog, which was when I was working at Monday.com back in 2018 and I started understanding what they were doing operationally. I was working on strategic finance and their metrics and things like that.

Being in their office, you start to understand like what's the strategy behind this. And the genius of it was the first two weeks of every month, they would spend blowing out their marketing budgets across every channel.

And then they would spend the second half of that same month making each of those channels as as efficient as possible, studying every little detail.

What worked, what didn't work, and then optimizing and optimizing and optimizing. So that their base case for the next month was starting at a place that was both more scaled and more efficient than when they're where they started the previous month. And that ability, again, it's an oscillation, it's it's an expanding and contracting. It's like it's like breathing. And that is effectively what every great company needs to do. Now that's just more on the product side than the sales and marketing side. Now, again, that still needs to happen on the sales and marketing side, but

You could argue that everyone using AI powered tools for also sales and marketing, it's really part of that same skill. How can you use the best and brightest tools and people to deliver your product? That means designing it, building it, marketing it, selling it, servicing it, the whole full stack as efficiently as possible. And that will always go up and down. The question is how high does it go when it goes up and how low does it go when it goes down? And what we want to do.

Across every business is figure out how to make the lows less deep and the highs both higher and longer. And that's the name of the game. And so I'll leave you with that. I'll leave you with Zeta, this concept of being critically damped as a company that's powered by AI and the consistent demonstrated ability to both make the company more efficient on existing technologies and develop creative, new, innovative products on the leading newest technologies.

If you are a founder and you're running your sales, marketing, engineering, product, HR, finance full stack operation, what you constantly want to be asking yourself is: Am I critically damped? And if I'm not critically damped, then how do I get there? Are we moving fast enough? Are we moving too fast? Are we efficient? Are we inefficient? And by the way, this is not independent of the ability to raise capital.

There is an argument that if you can raise borderline infinite capital, then you might as well be as inefficient as possible and try to swallow up the whole market. But also have to be careful there because, like I said, switching costs are very, very low. And the ability to spin up new things is not that hard. So it's still unclear in this AI-powered world that we are living in and rapidly moving through.

Who exactly is gonna be the ones who maintain control? who's gonna have who's who are gonna be the stalwarts, who's gonna have that low discount rate, that high growth, that big market, that high free cash flow margin that develops really, really valuable businesses. And I think in my experience, the answer is the continuation of a trend that we saw in enterprise software over the last 10 to 15 years already, which was what I call the decreasing shelf life of enterprise software.

If you look at Oracle, SAP, some of these old tools, they are still active. Whereas if you look at the hottest enterprise dev tool from 10 years ago, you may not have even heard about it now. Because its ability to stay relevant is just harder. Things are just moving and developing faster. And AI just put jet fuel on that across not just dev tools and infrastructure, but all areas of product. So the ability to adapt.

I would argue is the most important demonstrated ability right now. It's not actually, owning markets and throwing as much money at things as you can. And just I I think that the strategy has been sharpened in that way. And the agility to move and be dynamic and adapt and push new products and make the business more efficient at scale is going to be where value accrues going forward.