Marcel Santilli
For us, content is really language. It’s the foundation of how you communicate information, ideas, knowledge… how you build credibility, how you build trust. If brands don’t show up and are not positioned well there, they either don’t exist.
Saket Saurabh
What is that approach about content and how do you really build that value? You’re just trying new things and you’re validating. It’s a combination of moments, right?
Marcel Santilli
And hopefully those moments turn into signals that then you can capture and figure out… okay, this is working, let’s double down on it.
Saket Saurabh
But then AI comes in and it seems like content generation is a little bit too easy.
Marcel Santilli
People are just forgetting that the more AI automates things or makes things easier, the more important it is that humans with good judgment are asking the right questions and figuring out what the right input should be.
Saket Saurabh
Hi everyone. Thank you for listening to another episode of Data Innovators and Builders. Today I’m speaking with Marcel Santilli, CEO of Growthex. Marcel, thank you for being here and chatting with us.
Marcel Santilli
Yeah, thanks for having me.
Saket Saurabh
So Marcel, you have an incredible story getting into starting Growthex and I want to dive right into that and your journey becoming an entrepreneur.
Marcel Santilli
Yeah, I think there’s not that many CMOs that have turned into founders. I could be wrong here. For me it’s a slightly unique story… I started my career early on in large enterprises and was constantly building these growth engines for different companies. At IBM and then HP. I was at HashiCorp leading growth… we went from six to 100 million in two years. They eventually went public. ServiceTitan eventually went public as well. Skilli had a huge inflection curve on growth. Eventually I was CMO of a company called Deepgram. Amazing company… they’re actually a customer of ours now and doing amazing in the voice AI space.
I take a lot of what I learned along the years, and the way I think about it is very simple: content is the atomic unit for marketing. If you can get content right and approach it that way… and the outcome you care about is growth… you want to build an engine that figures out what to prioritize to build a growth engine. That’s what I kept doing throughout my career.
At Deepgram, I started using AI workflows to take my expert processes and add humans in the loop in a unique way. That ended up helping us publish over 3,000 pages. We continuously improved those pages and eventually 24x’d our traffic in three months. It helped the company 4x revenue… a huge inflection curve. From there I started teaching workshops and people came to those workshops. We had about 170 people pay a little over $500 to attend, and that’s kind of how GrowthX got started… a lot of people asking us to help them end to end, not just offer one slice of the equation.
Now we do that for a lot of different companies. We’re working with a little over 60 companies… from Ramp and Brex to Lovable, Webflow, Surgei, Neural Security, Sentinel One, and more. We have a unique model where we get to be selective about the companies we work with. We just hired our first AE about three months ago and we’ve gotten to a little over $13 million in ARR with very little to no sales… all inbound. It’s been a fun journey, but hard as hell.
Saket Saurabh
It’s such an incredible story. As I see your journey, I feel like you were destined to start this company. You started this concept pretty early on… focusing on content and earning the attention of the market. A lot of attention is all you need is kind of the foundation for all this AI stuff. When I talk to AI builders, everybody’s trying to figure out how to get that attention from the end user. We’re all building cool stuff. So talk to us a little bit about what that approach to content looks like and how do you really build that value?
Marcel Santilli
There’s no one-size-fits-all for how to grow and build a company, but a lot of the companies that have been successful… like some of our customers like Lovable… take a different approach to everything. They’re really delivering value first, and there are multiple ways to deliver value.
If you have a product that delivers a lot of value up front… like Lovable, Ramp, and Brex… you can sign up and get a lot of value pretty quickly without signing a big contract. But you can also do it without being product-led. For us, we started doing workshops, creating value, teaching people, educating the market on what they were trying to do, and understanding where the pain points were beyond education.
Then it’s like iterating fast and trying new things. Even today, there are companies thinking through every little word… they think their homepage is going to be the end-all-be-all forever, and it takes them three months to update it. We updated ours last week in one week. You’re just trying new things, validating, not thinking one-and-done. It’s a combination of moments… and hopefully those moments turn into signals that you can capture and figure out: okay, this is working, let’s double down on it. Then you build systems to scale those things.
Building in public is so critical. As you find the signals that work, you’re trying to systematize them with both AI and humans. A lot of that plays together in this new approach where people are building more in public, being value-first, and building systems around it. At the end of the day, speed and distribution is what’s going to win.
For us at GrowthX, we’re really focused on thinking of content overall as the substance… the atomic unit for marketing. Think of an ad, an email, the substance on a page… that’s all content. And content is really hard. We’re thinking about it as a growth engine that can be a moat for companies.
One of my favorite examples… they’re not a customer, but I think they’re one of the most incredible examples of this… is Canva. They’ve been profitable for 10-plus years. They get about 280 million organic visitors per month. That traffic is worth $100 million a month… that’s what they’d pay if they were buying those clicks. So if you’re trying to build a slightly better version of Canva today, it’s going to be really, really hard. That gives you incredible defensibility and moat.
For companies just starting their journey, it can be really daunting. But if you look at Canva almost 10 years ago, they had less than 200 pages on their website. They started putting those bets down 10 years ago, and now they have over 45,000 pages delivering over 280 million visitors. Sure, they do a bunch of other things, but if you took that traffic away, I bet it would really hurt the company.
And so speed learning… how do you build that engine that goes from 200 pages to 45,000 pages? Think of a page as an output that’s really an asset you have to nurture, enrich, watch the signals on, and constantly iterate… ideally with AI and humans in the loop.
Saket Saurabh
The point you make about AI and humans in the loop is super critical. We’ve always thought about content as someone with domain knowledge and expertise who translates that into articles and puts it out there. You proved that model way before AI was there. But then AI comes in and it seems like content generation is almost too easy. You still have to bring that expertise out, still have to cut through the noise. How does GrowthX do that from a product perspective? How do you bring together human and technology?
Marcel Santilli
A lot of the lessons and how we approach scaling that… it’s not just about content. For us, content is really language. It’s the foundation of how you communicate information, ideas, knowledge… how you build credibility, how you build trust. Everything is content. And technically, content drives the interface. It’s language, so it drives the interface with AI as well.
If you get good at content, you can get good at pretty much anything these days. So we break it down to a few different layers.
The first is context. The context engineering piece goes beyond just context engineering… it goes into seeking a deeper understanding of your company, your product, what you’re building, the market, and most importantly your audience and who you’re for. If you don’t have clarity on that and can’t spell it out, you can’t really scale.
The way I think about it: if you hire an intern off the street with very little context, can you delegate successfully to that person? If you use the same context and inputs with AI… if a smart intern couldn’t be successful with that context without talking to someone and spending the next month figuring things out… then AI is not going to be successful either. So we always take this approach: if you can’t delegate to a capable human, you can’t delegate to AI.
Context is the foundation. Then there’s strategy… how do you map out different opportunities, the kinds of questions and topics your audience cares about that will eventually lead to you and your product? How do you systematically become by far the best answer? “Best answer” doesn’t necessarily mean the best-written or longest answer. It’s best for that audience. If you have clarity on who you’re for, you can figure out what is best for that audience and optimize for it.
You want to map it out, but it’s an infinite game because there’s an infinite long tail now. You go into ChatGPT or AI answer engines and you’re not just putting in a few words… those answers are also pulling your own context or preferences. You’re creating content that normally happened post-sales, and today it’s happening inside an AI answer. If brands don’t show up and are not positioned well there, they either don’t exist. People are ingesting those answers as if they were expert word-of-mouth advice. So you want to influence that.
Then there’s execution… working backwards based on the output. If your output is a how-to guide, a newsletter, a social post… within the context of your company and the audience you’re going after… you can work backwards on what it takes to do a really good job.
The key primitives are: defining the brief, pulling in the right context, doing deep research on the topic, planning how to execute, drafting and executing, and then post-processing and polishing.
You want humans in that process, especially at the beginning, because you’re calibrating to figure out what great looks like and your definition of done. That calibration requires a lot of human judgment… do I need to update my context? Do I need to ingest more or better data? Is the deep researcher using bad sources? You’re really decomposing everything into first principles, building a benchmark of what a great output should look like, and then having a human as an output bar raiser to figure out where you can improve the system… not just this output… in order to consistently achieve the level of quality and results you need.
There’s a lot of human judgment in that loop. That judgment isn’t there to stay there forever… it’s there to build leverage for the next cycle and the next iteration.
And then there are all the signals out there in the market… is your brand ranking, is it getting cited, are you showing up in the right keywords, is it getting traffic, is that traffic engaging, is it converting? Those are all signals. You take those signals and start that loop again… what do you prioritize next? Do you refresh a piece of content or go do something else? That’s kind of how we think, end to end.
Saket Saurabh
This is incredible… it’s a whole engine you’ve built, going from the goal of earning the user’s attention to understanding who that person is, to taking signals back and tuning the whole thing. I really like that part where you have a human raising the bar on what’s coming out. You’ve mentioned speed… these cycles are running in very tight loops. Is that right?
Marcel Santilli
Yeah. It’s tight loops but also has to be contextualized based on the space, domain, brand, credibility, and goals. If you’re in the medical space… we have a customer that is a supplement brand for women in different phases of life, from fertility to premenopausal to menopausal… they have two medical reviewers on every piece of content. That content is really about establishing their credibility as a brand and communicating how they’ve done years of studies picking the right ingredients. That can make or break their brand, so that lane is really focused on quality over volume.
But then there might be other things you can scale a lot more. For instance, for Ramp we built a vendor catalog that pulls their proprietary information and data on how much people are spending on different vendors, and we create a catalog of alternatives, reviews, and comparisons. With that you can be more programmatic, but it still comes from a question of how do I serve the user… it just becomes different. You spend more time on the template, the data sources, building quality checks across different pipelines.
Think of the pipeline as a spectrum. On one end is thought leadership. If you’re writing an essay about the future of your company, you’re obviously not going to delegate and automate that. You might delegate some of the deep research. On the other end of the spectrum, there are brands scaling at a massive scope… companies like Hims & Hers that need to do drug comparisons, talk about how different drugs interact, build content around that. There are so many drugs and combinations that it would be impossible for a human to manage a catalog of a thousand pages. You do need to use technology and AI for that.
Saket Saurabh
You mentioned you ran about 170 workshops, people were willing to pay for them, and that led you to formulating GrowthX as a company. For other AI founders out there… when should they think of workshops as an effective mechanism for validating and building toward an idea.
Marcel Santilli
A good anecdotal example here. I was interviewing someone for a job recently and he was at a company that ran out of money. I said, “Imagine you’re the CEO of that company six months ago… what would you do differently?”
The main thing was they were doing something in AI around context engineering, and people just wouldn’t pay for the pilots. We went through this whole thought exercise and the conclusion was: okay, just refine your ICP. But the chances you’re still going to run out of money are still pretty high. So I said you’re solving the wrong problem… take a step back. Why didn’t you just offer a consulting service? The people at that company… I would have hired them and paid them more than all their lifetime revenue. I’m a Series A company, and I would have paid for that specific help. If I can buy acceleration to solving a problem that matters to me, I will pay money for it. I want the outcome… either solving the problem or achieving something. I don’t want to buy a tool.
A lot of founders have been indoctrinated in the VC industrial complex… burn, burn, raise. Series A investors just care that you get to Series B, and so on. It does help generate amazing companies and technologies, so we do need it… but as a founder, you’re there to figure out what you’re building, solve a problem, and find conviction in that.
If you can use mechanisms… forward deploy, services, workshops, educating… just figure out how to meet the market where they are. For us, the workshop was just an easy way to say: I genuinely want to teach people what I did. And people were like, “This is awesome.” In that process I figured out what people actually wanted to do themselves and what they wanted someone else to handle.
People do not want to drag and drop workflows, edit prompts, learn prompt engineering, figure out how to scale a runtime layer, and run eval. I’m a marketer… I do not want to do that at all.
Saket Saurabh
From the outside, AI often looks like it can do everything easily. But as you get in, you realize it takes a lot of effort and tuning to get it right. The model you’re talking about… bringing AI as an accelerator or augment to human effort… makes a lot of sense. You have to fit it into a certain process, system, and outcome to really find where it fits well.
Marcel Santilli
Yeah, it’s been a fun journey and it’s way messier than it probably looks from the outside.
Saket Saurabh
You work with a very broad set of AI-first companies today. Do you see this concept of the GTM engineer… where marketing itself becomes engineering-driven and programmatic… as a central way marketing will continue to build and grow? Or is this a phase in the evolution and adoption of AI? How should we think about that function itself?
Marcel Santilli
I think sometimes as technology evolves, roles come up that are more a reflection of the tools than a reflection of what people actually wish they were doing. Most people who call themselves go-to-market engineers… and I know many of them… are revops people mostly, with a few exceptions. They’re tinkerers, hackers. Great to validate zero to one, but they are not engineers first and foremost.
There are a lot of first-principle things you can apply to think like an engineer and approach problems like an engineer. More people should do that… and I think that’s a trend that’s going to happen.
But the way I think about it is: you have process architects, input calibrators, and output bar raisers. Those are three roles that matter regardless of title.
A process architect… a go-to-market engineer at times… may not necessarily be the best salesperson on your team or a go-to-market strategist. At HashiCorp, the founders had this principle: workflows before technology. Figure out how work gets done… the expert processes, the logic you follow to research your customer, figure out what to say, how to say it. That’s how I think about process architects. You don’t need a bunch of them. Once you have the blueprint, you can have a real engineer build it properly in code and go back to primitives that work.
Then you have input calibrators. People are forgetting that the more AI automates things, the more important it is that humans with good judgment are asking the right questions, figuring out the right inputs, calibrating expectations, learning from the whole loop, and connecting the dots. You need people with great taste who are there to ask questions. A lot of what we do is get on a kickoff call, do a deep dive with an expert, and learn from them… then use AI to process it. You still need someone there for that judgment.
Then there are the output bar raisers… people who hold the benchmark for quality and are improving the system, not just the individual output.
Most go-to-market engineers from companies are just being asked to set up HubSpot or build some Clay automations for basic enrichment. That’s great… but it’s very different from engineers we have in our company.
Saket Saurabh
That’s an incredible framework… understanding what workflow you want is the most important thing. Coding it up into tools and automating it comes as a secondary part, and then you can always iterate. And what you said about the input calibrator is really important… it gets underestimated. Building that context, the prompts, and tuning them as you iterate… all of that sort of input wires the whole system.
Talking about input… what role does data play? Marketing has always been data-driven, but is that role increasing and what challenges do you see from that perspective?
Marcel Santilli
Here’s a good example. We’re working with an AI-native company in the vibe coding space. They have a lot of data… millions of users signing up… and a really rich data set. Not just session data, but the prompts people put in, what came back, how long during the session, when they upgraded.
We were trying to figure out the right strategy for programmatic pages to create for them. We connected with their data team… a couple of data scientists… and said, “Do you have any insights on use cases and ICP?” They had built dashboards, but we asked if we could get a data dump… a slice of the sessions. That included sessions where users did a bunch of things, got stuck, never came back; got stuck, then came back; did more things, and eventually signed up for the pro plan.
They couldn’t surface a lot of insights. So what we did was build a workflow that went through that data and described a narrative of each session in plain English. Think of it as: this user was an agency, they were a solopreneur, they tried to do this, got stuck, pivoted and built this instead.
After that, it was much easier to run a few other processes to categorize them with tags… taxonomy we created… and do a cohort analysis. Simple tables, color-coded, showing the strongest correlations: types of use cases more likely to be successful with the tool. Within those were actual examples… internal tools, landing pages, specific website types. Those were the ones we should go create templates for. Done. Go.
That whole process took less than two days. The data science team… they’re great and smart, but the data was the same. The difference is: are you going to use AI to make a prettier chart, or are you going to take the context you have and use the data just to answer the thing you really need to answer? Once you have confidence in that answer, go execute. Later you can watch for signals to figure out if it’s working.
We use AI to process things and enrich the data in a way that the data tells a clearer story and makes it easier to analyze patterns.
Saket Saurabh
Looking at data at an individual level and figuring out the story behind it… versus just dashboarding or visualizing it… brings more context into the data.
One thing I want to ask because you’ve mentioned velocity and speed multiple times: it’s much easier for startups to build that culture. What about AI teams in bigger companies trying to be AI-first? Are there things you’ve seen or practiced that they should learn from?
Marcel Santilli
Maybe I have a hot take here. If you can clarify who the people are in those three different roles… process architect, input calibrator, output bar raiser… who are absolutely incredible, keep those around. Everybody else? If they’re not building leverage for themselves, not being clear, not great input calibrators, not connecting dots, not true output bar raisers… they’re slowing everybody else down and making it slower and harder to get to a decision, get data, and build velocity and momentum.
You’ve got to build momentum. There are teams with 50 or 60 people in marketing and it’s so much harder. We see customers where we can ship the same thing with the same inputs and the same process in one day, and another company takes two to three weeks. You’re waiting around, trying to get consensus, trying to get a hold of someone… “Oh, they’re busy, they have five other things.” You have 5,000 employees and 100 people in marketing and it’s way harder to get an answer and get the same output out. That’s hard to justify in my opinion.
This can sound a little bit mean, but I truly believe you’re giving people the opportunity to retool themselves and reinvent themselves… and it’s up to them whether they do it or not. Some of the best people we have today, at our company and out there, are people who forced themselves to take that break… or had it forced on them. They took that time to go super deep, build intellectual curiosity, and reinvent themselves. I’m not talking years… I’m talking one or two months to go super deep.
My co-founder did the same thing. He left his last company he founded and took a year off just studying and going super deep before we partnered. You’ve got to build that room. If you still have a job, figure out a way to create that room to retool yourself.
If you’re running a bigger team, look at the output people are delivering. If it’s not absolutely incredible, out-of-this-world output… what are you doing here? I think large companies are doing that, cut by cut, unfortunately or fortunately.
Saket Saurabh
AI is definitely here to augment humans and help us be more productive, but it does require a change in work style… and that curiosity to go deep and get creative with it, because everybody else is moving fast. The world’s not going to stop. Better to get on that model and become native to it.
Marcel Santilli
Ask every single person in your company: what is the outcome you’re optimizing for? Then ask: what is the output you can generate that will influence that outcome? And three: what are the controllable inputs that will drive the volume, quality, or velocity of those outputs that influence the outcome you care about?
If you can answer those three things with clarity… and you can hold people accountable to those three things, starting with the inputs that drive the right outputs that deliver the right outcomes… then you have clarity for everybody.
Saket Saurabh
Well said. So as we come to a close, Marcel… tell me one tool or technology you can’t live without.
Marcel Santilli
Wispr Flow
Saket Saurabh
And tell us something that most people may not have heard of but you’re already seeing as the next up-and-coming thing.
Marcel Santilli
I don’t know if no one has heard of it, but context engineering is just so critical. And be really good at writing. If you don’t have good first-principles thinking and you’re not good at writing, it’s a muscle… just build it. Get good at writing, good at organizing your thoughts, and you’re going to be incredibly valuable for the rest of your existence.
Saket Saurabh
Context engineering is definitely coming up and very critical to getting the best outcomes from AI. Would you define context engineering for our audience and how you see that?
Marcel Santilli
To me it’s big picture. I always like the human analogy. Humans have quite large context windows in their brains. But if you go to a brand new employee on their first day… I just did our vision meeting for new hires this week, this very morning… and I spend the next 10 hours dumping everything I know about the company at once, they might absorb some of it, but their brains are going to be flooded. They’re not going to know what to prioritize, what great looks like, or anything like that.
Just because you have a million tokens in a context window doesn’t mean you should flood it. One input, one output… if you flood one step in a context window, that deteriorates performance. Now multiply that between humans, AI, and agentic systems with evals across very complex workflows… it just gets way harder.
How do you distill it down to the essence? How do you manage that context so you know when to share it, how, and how to measure if it’s being effective? That’s what a CEO or a leader should do. You are the context. You’re driving it. You’re setting the goals, defining what great looks like, what done looks like, how much someone should know as an employee, what this department should do, how to measure that. It’s a pretty hard challenge for humans… and for AI, it’s not going to be any easier.
Saket Saurabh
Amazing, amazing stuff we covered today, Marcel. We talked about process design, input calibration, output bar raisers… thinking about work in this framework is really valuable. The speed and velocity approach, how to think clearly about the outcomes we’re all creating, and of course the approach you’ve taken with building workshops, understanding what you’re solving, and being very focused on building the solution. Lot of great topics today. Thank you so much Marcel. It’s been incredible.
Marcel Santilli
Thanks for having me. This is fun.
Saket Saurabh
Thank you.
Express — 
