810 | From 20% to 80% Margins (The NEW AI Agency Model explained) | Frans Riemersma

Agencies are running on 80% to 20% margins today, but they could shift to a size like 60% to 80% margin in a market of services that is 53 times bigger. That’s Frans Riemersma, martech analyst, consultant, and founder of Martech Tribe. He’s benchmarked over 1500 Martech stacks across industries using the world’s largest independent Martech data warehouse to help brands like HP, Adidas, and Audi cut waste and maximize ROI. Now, he’s making a bigger claim. AI won’t just automate workflows, it will automate expertise, turning agencies into scalable, SaaS-like businesses in what he calls the golden age of agencies. But how does he do that? AI is now executing it on behalf of the human.

The human is still in the loop. You need a conductor. In this episode, you will learn what service as a software actually means and why it changes everything. Digital transformation 1.0 was agencies using software. 2.0 is agencies becoming software. What’s that? 50 people? I think they, you know, did some 6 billion. Why AI might finally bring the golden age of agencies? The more you have automated or standardized your stuff and your processes, the more you can automate it with a generic AI. The biggest traps guardrails agencies need to know before making the shift. If you focus on the software right from the start, I would say that’s almost like a mortal sin. You’ve said that it feels like that we’re in the golden age for agencies at one point. And I just wanted to ask, are you talking with respect to margins?

Are you talking about scale or are you talking about something else? So where SaaS in the old days was really about, you know, helping humans to execute a task in a repetitive way. It’s almost like AI is now executing it on behalf of the human, on behalf of, because it doesn’t mean we’re out of the loop. The human is still in the loop. You need a conductor. And that’s the shift we see happening. So agencies excel in knowledge. And what you see happening now with genitive AI is that basically there’s a new breed introduced next to SaaS software as a service, which is service as a software. And those services can only be automated and make repetitive if they are in some kind of protocol, you know, standardized. So many agents who have standardized agencies who have standardized their knowledge in procedures and protocols are in a good shape.

So, you know, think of, you know, pilots in airplanes or even dentists or doctors, but also lawyers. That is all professional services, but also agencies, you know, the more you have automated or standardized your stuff and your processes, the more you can automate it with genitive AI. The margins are also, of course, in SaaS, software as a service, very different from service as a software or circuit from services, service as a software, unlocks, you know, the margins maybe of a software as a service. So you go from not, you know, what is it, five to 10% maybe to 60, 80% depending on how well you have standardized your knowledge. If you haven’t, then that is something where you can start, you know, but a lot of the stuff is very repetitive. And I’m not talking so much about the processes to get to an outcome, but the outcome itself, you know, think about, you know, Harvey AI or for lawyers, you know, that’s a tool. Yeah, so that is a nice example of how you have standardized knowledge. And yeah, that is, that is the playground of genitive AI, the more structure in your unstructured relative unstructured data, like language, like copy, the better it is.

If you’re a 10% agency right now, and you want to start making this shift to and take advantage of the trend, this service as a software trend, where would you recommend that they start to unpack these things? I think the best place to start is where have you standardized most of your outcome, your services outcome? And that’s one. But also, does it bring money? So follow the money, make sure, you know, it brings in revenue. Otherwise, you’re just having a great product, and it’s not bringing any value to the customer that they are willing to pay for. But that is, that is a really important thing to do. So what you see is that in the old days, there were agencies that help you, you know, find leads and nurture them.

Right now, this is done by AI tooling. And some people, you know, in the background, one or two persons helping with the output and cleaning it up. But actually, it’s a lot less. So it’s not an entire AC anymore. If that is something you can automate, and this is a process we see a lot that’s being automated in lead intelligence, lead mining, lead sourcing. That’s a typical example that’s absolutely booming right now. And look at marketing agencies, and you’re talking about agencies now becoming or moving toward this trend of service as a software. Also seen a trend where it seems like more software companies, or SaaS companies, are getting into services.

Who’s moving faster in which direction here? And with AI commoditizing rapidly, I think the advantage of being tech savvy from software vendors is decreasing rapidly. I’m not saying that they’re in a losing game, because on the other hand, you know, many software companies do a lot of services as well. So they understand what is the pain and what. But because of the SaaS model, if you want to do an exit with your company, it’s not so good if you have services. So you want to get rid of that. This is often why you have implementation partners. But maybe this will change, and they’ll just go like, no, but we have the templates, we know exactly in your industry, you need to switch on these buttons, and there you go.

And that’s the best practice. This is our standardized knowledge, and they become more and more from software as a service to service as a software. So I think the lines will be blurring, if that makes sense. For someone that’s retooling their agency, and they’re retooling their stack. Like, are there pitfalls or mistakes that you’ve seen people making? What is a big mistake, and this is also something we’ve seen in SaaS, is that you build an entire tool or environment without testing it. This is where I always think about the, I call it, started calling it the hackback stack approach. So you need to growth hack.

And once you have traction, and only then, you start cleaning up your prototype. Well, how often do we do this? Normally, we just bring the prototype to the market, and then, you know, make money right away, but you have to clean it up to keep it lean and mean, before you put it in your stack. We want to have concrete outcomes, like a tool, something like a service or a product. We don’t have the courage yet, found yet, to say, you know what, we live with a bit of uncertainty. And we’ll see what, you know, it’s not like wait and see, but you experiment, you experiment until you have the product market fit.

Now, if you focus on the software right from the start, I would say that’s almost like a mortal sin, because your back is facing the audience and the clients. And this we see a lot. So I think now with AI doing a lot of the heavy lifting, there’s almost no escape, but to listen to the customer and talk to them and say, Do you like this? What do you like? What are you using? And maybe not even asking, but seeing what they really do, because that’s sometimes completely different. So there’s no escape. We have to show a little bit more empathy with our customers and try to understand what they’re trying to solve and what answers they need. So that avoids you from creating a kind of a Ferrari and spending a lot of money in there without making sure that it’s really bringing any value. So I think that’s also preventing you from creating bloat or what some people call Frankenstech. And that’s a big risk. So you implement too many tools around it and make sure that you evolve a big stack before even launching your product. So I would say start small and don’t try to boil the ocean with the data and with

a big team. And this is an interesting point. How many big companies started with 10, 20 people? Not many. I know only one. And I don’t know if you happen to know that the name is company. It’s Intel. So they started with, I believe, some 10 people. Why? Because they were students and they didn’t like a professor. So they kicked him out and started the company themselves. This is the rough version. But listen, any other company was one or two founders. So why then we expect a team of 10, 20 people to come up with a new product? This is the atomization of labor or maybe ideas or innovation. Make it small and there are always bright minds in your company.

Maybe I’m not saying isolate them, but just give them the mandate and the budget to do some stuff. And if there’s good outcomes hacking, then you’re in a good space and then hand it over to another team who cleans it up and then bring it to market. You talk a lot about composability.

I wanted to ask you what role that plays in the shift that we’re seeing here. That’s a very interesting one. And also having an impact on agencies because you started with SAS is really about the tooling, the task and agencies about people.

You see that in technology and the landscape that Scott Brink and myself are studying for now over a decade, you see that there are constantly more tools available, but those are smaller tools. They’re not the enterprise big companies. And these smaller tools are very well integrated into their ecosystem. They do one job really well, one job to be done. And I started to call this atomization like a Lego block. And composability is creating and crafting journeys with those Lego blocks. Now that is what you see in technology. And what I see happening also in labor, the labor market is also kind of an atomization. I think in the 1930s, you needed like 10,000 people to have a billion dollar company. Well, what about what’s that? 50 people? I think they know did some 6 billion Instagram 12 people. There you go.

We’re busy. How do you balance the day to day running a business successfully with this need to experiment and evolve and test?

Yeah, it’s almost like a kind of a split personality type of thing. But that would not be a very good qualification for the really successful CEOs who do this. And what is it that sets them apart? Well, they have the knowledge or the experience that you need actually a stack that performs two tasks at the same time. And it almost feels like one is you know, dancing salsa and the other is doing techno. That doesn’t work together. And it’s the dynamics are very different. And I would say so one is really your factory. And the other one is really your laboratory. This is where marketing sits. And this is where it sits. This is where your current revenue comes from. And this is where your future revenue comes from. So this shows up on the balance sheet like license fees to the cost of ownership.

This one doesn’t. So missed opportunities. So here’s the secret, I would say CEOs that are the founder of the company are probably better equipped to do both. Think about Richard Branch and type, you know, or Elon Musk, they understand, yes, there is a certain amount that needs innovation. Otherwise, we become irrelevant. If you look at more hired CEOs, they’re often forced by venture capital or invest whoever I’m not blaming. But what you see is that they go like, yeah, but the balance sheet is more important. And there goes your innovation. So you need to these two, you know, legs to stand on as a company. So your technology stack has two different roles. And that’s why you need to go from hacking, you know, to stacking with a team in the middle, which is doing the packing, which is cleaning up, making sure it can go into the stack as a performant clean piece of software or service, if you like.

If you’re looking at today, and where everybody’s experimenting with AI and adopting it at different rates, what have you seen? Who’s overperforming and who’s underperforming? And what are they doing differently? I manage a database with stacks of real life companies, and we track them, we follow them, and we speak to them and their clients of ours. And we always, you know, see that the top 30% is behaving differently. And what the real shocker is, especially for the ones that are not outperforming in a certain industry, they go like, your data must be wrong. Because the outperformers don’t do all the headless composite fill in the blanks with the buzzword, you know, the latest greatest. And that’s exactly right. They keep their stack lean and mean, because that’s current revenue, you know, and on the other hand, it doesn’t mean they’re not playing with AI agents. Actually, they have teams like laboratories, labs working on that. So you need, again, they reserve that balance and keep the balance in place. And that is how they outperform, because if you don’t, you know, innovate today, you will be outdated in five to 10 years.

And they really understand this. Some tools that might have been applicable, you know, a couple of years before, there’s better things out there, and there’s better things that fit together. And it’s really easy to get bloat. Isn’t there? Then the CEO talk comes in, it goes like, we have too many tools, let’s cut down. And by the way, research shows that 33% of the tools are used. So most of it isn’t used. I think 33% is still too high. If you think of it from a composability perspective, I worked for a market company and we did some research and we found that we don’t, our customers don’t even need all those buttons and features. They just need the right ones. So with the research recently from another company that said 12% of all our features are used most, not 12% for every single customer is the same, but it is not more than 12%. So you have to find as a customer, your 12%, so to because 12% brings probably, you know, the rest of the value or it’s kind of a 28 year old.

And to find that value, you have to go back like, what if we switch it off? Will we miss revenue? Will this most important customer of ours not get the experience they need now or in a couple of weeks or months? And if you don’t do that, then yeah, you just believe that software will actually solve a thing and software doesn’t solve a thing. I think you know, there are two laws of software. One, it makes an efficient process more efficient. The second law is it makes an inefficient process more inefficient. So you have to clean up, come up with brilliant concepts. They go like, let’s ask ourselves the magic question. And the magic question is this, if we switch this off, will we lose revenue? Simple as that. And to just go through this tech and say, no, not really, not really, you know, sometimes it’s indirect value, but direct value, you will immediately spot.

There was another client of mine who said, you know what, we will implement now a new rule, which is we only implement software with decommissioning in mind. So the vendor has to tell us how do we get out our data and our users? Otherwise, we will not implement. I mean, those are really composable principles that you can play by. And then you keep your stack very simple. You know, when AI is changing things so fast, what types of skills do you think people should be doubling down on? I did this post on service as a software, you know, being 53 types, you know, bigger market than maybe SaaS potentially, the product, you know, the software is commoditized.

So I’m commoditizing. So the only thing you really need to look at is distribution. You need mode, you need a community to sell to. Or actually, I know some growth hackers that had only a community to try things out. And the people know they’re trying things out, just to get that, you know, growth hacking, hacking type of thing. And then they move into packing and stacking. So I think it’s will become a skew number, if that makes sense. Makes perfect sense. I like that a lot. So and it’s not like with Harvey AI, of course, it’s a it’s a big book of law, you know, that’s a big, but I think even on smaller levels, I think atomization is not only there for technology and labor, but maybe also for outcomes. We don’t need always a big, big outcome. So you can deliver expertise as a skew number. This is what you need to know.

There you go. And if you think about it, I think at breakneck speed, chat, GPT and Claude and all those companies and tools are educating the market to get instant insights and abundant advice. So this is the new new. This is what we’re used to. So in the old days, it was like, here is the tool that solves everything, replaces everything and everybody. So I think we’re now clear that life is more like Lego or maybe Minecraft. My son would love that. Mine too.

So we’re building our own stuff. And I think that makes life also maybe even more pleasant because you can wrap those services around your convenience, around your comfort zone, what you really need. So I think that is where the imagination of, you know, the suppliers should go and what is it we should solve? And I heard a very nice quote the other day, if you’re obsessed with the customer problem, you will never run out of business. Because there’s always something to add to refine, to improve. If you’re only obsessed with your own solution, like we said earlier, one of the pitfalls is, you know, you’re done pretty pretty.