20Growth：如何打造1亿美元增长引擎：Wispr Flow 与 Superhuman 的经验｜为什么今天就该投放付费广告，以及如何做好｜如何搭建最佳推荐计划、把 UGC 做到极致——Matt Swulinski

Probably fire most of your marketing team. My philosophy is that the e-commerce playbook is the right playbook for SaaS. Every single cent needs to equal a purchase or an add to cart.

Paid is the easiest way to validate that you have PLG. I say there are the core 3 of any acquisition engine: Meta, Google, and lifecycle. You probably need at least 400 to 500 new creatives a month. We have kids who are 17, 18, 19 and making $20,000 to $30,000 a month just making a couple of ads for us.

Matt, it is so good to have you on the show. Thank you so much for joining me, Matt.

Yeah, thanks for being here.

Now, you've worked with some of the best companies, from Superhuman to Whisper Flow, and now with Victor. We've seen these companies and a huge number of SaaS companies rise with a product-led growth motion, and now we're entering a world of agents. What changes in the growth world when we shift from humans to agents with product-led growth?

1. Agents Change The PLG Playbook

That's a great question. All of the startups I've worked with have been PLG. In a world of agents, there's definitely a layer where PLG still matters because it assumes you have a perfect self-serve funnel to the human. In the same way, you're definitely going to want to optimize it for how an agent is doing research and how it's picking the tools and APIs that it's actually using.

I think it just takes the same framing and diligence to understand why the product already worked from the human end and how agents are making decisions. So the best companies focused on PLG right now will also be best positioned, in my opinion, for the agentic layer—essentially, decisions being made without a human in the loop when tools are being selected for certain tasks.

When you look back at, say, Superhuman, if we go back in time chronologically, it grew very craftsman-like with founder referrals. What are your reflections or lessons from that experience on Superhuman's growth that you've taken with you?

2. Superhuman Finds A New Growth Playbook

The taste-making of what it means to create a strong product was born for me while I was at Superhuman. At the end of the day, if you've used Superhuman, you've likely stayed using Superhuman just because every little element of that product is hyper-refined to make you want to continue using it.

But I think the difference in Superhuman's early scale, where it was super founder-led and word-of-mouth referral, was that there was essentially an asymptote that the product hit where they had to follow a different playbook. This was roughly 3 years ago. Back then, with SaaS—what we now call AI SaaS—everyone was running more of the PLG motion: referral and word of mouth without doing paid. And they were laughing, like, e-commerce has been doing UGC programs and hundreds of thousands of ads; paid ad spend is the way to drive performance. The resistance was definitely there while I was at Superhuman.

Seeing what I've been able to do alongside a great team at Whisper Flow and now Victor, my philosophy is that the e-commerce playbook is the right playbook for SaaS. If you look at e-commerce, every single cent needs to equal a purchase or an add to cart. You have hundreds of UGC creators, a variety of creative, and a ton of channels that you're essentially balancing to showcase the entirety of the brand.

That's the model that I took and applied at the end, before the Grammarly acquisition at Superhuman, to scale paid, and then followed that exact same motion for Wispr Flow. That's really what put it on the map and got it to where it is. As we were talking about before, distribution to me is the only moat, and you have to have that strong of a playbook when it comes to marketing. In today's world, that's the only way to succeed.

Dude, I'm going to love this show because most people say the same thing. What everyone tells me is, “No, no, you should wait to do paid. You should wait to do paid. It's a dangerous drug that you can get hooked on.” When is the right time, do you think, then?

3. Paid Starts From Day One

Right away. At the end of the day, paid is the easiest way to validate that you have PLG and that you have a product that can scale in any way, shape, or form. There's a ton of narrative that says, “Focus on brand, go to organic content, build all that base.” That just takes too much time.

Yes, do that motion while also doing all of paid, because you can refine messaging, test creative, test your funnels, and test your positioning all within a week on paid. How long is that going to take you if you do some content writing, talk about it on a podcast, see if it sticks, and do some user interviews? Great. Do all of that, but then you have a much faster engine of validation on the paid amplification end.

When we actually break that down, if I'm an early-stage founder, what should I actually do? How much money should I spend on paid? Should I do it on 1 channel versus 10 channels? If I'm an early-stage company, I don't have a mega-budget. What do I actually do?

Yeah, I would definitely start with what I call the core 2. I say there are the core 3 of any acquisition engine: Meta, Google, and lifecycle. Obviously, your site and everything else are in the middle, because you have your strong video-intent platforms in both Meta and YouTube Ads. You have the search intent of people not knowing your product exists but searching for something via keywords.

And then I add in lifecycle as well, because if you don't have a net to nudge people and show up in the right place—email, SMS, whatever it is, push if you have an app—you want those 3 things spun up. You can scale to your first $1 million or $10 million in ARR just off those 3 things.

Then you're investing in the site. You're investing in founder-led content, and there are the other things in there. But in paid, just focus on Meta and Google, and you'll totally be fine. Everyone's like, “Oh, we have to do TikTok and Reddit.” There comes a time and place to add in a million things, because in the same way that a lot of people get lost with agents, they start doing a million things with agents and do them all poorly. It's the same thing here, right? If you do a million channels and do them all poorly, it's not going to help you out.

Okay. So we have Meta, Google, and lifecycle. We start spending—I don't know, let's just put $100K on it. Say we've raised a $3 million seed round. Is $100K fair?

Yeah, I think it depends on your average revenue per user and the unit economics, right? Because it depends on how long your funnel is and what that $100K is going to give you. But usually I'd say, yeah, if you raise, let's say, $3 million to $5 million, that's a good start budget to essentially validate that if we put money in, people are actually going to want my product.

What should we be looking for, then? We have $100K, and we put it across Meta and Google. Let's just say those 2. What is a sign that it is working versus not working?

4. SaaS Needs Better Measurement

I'll also frame what I think is a massive gap in the SaaS space, where I'm waiting for startups to be made that fill this gap. If you look at e-commerce, there are companies like Triple Whale and Elevar that have existed for many years. What those tools do is—Triple Whale is essentially a mutual exclusivity platform that tells you, “Okay, you add a pixel to your website. It's out of the box. You plug in all your ad spend, and then it tells you”—because there's usually overlap. If you run Meta ads and Google, they're double-counting, right? And then lifecycle is also triple-counting at that point.

What a tool like that does for e-commerce is say, “Great. I plugged in all my variables. Tell me exactly what is driving revenue from a new-customer basis.” You have that out of the box. You don't have to set anything up. The other side is conversion tracking, which is add to cart, purchase, server-side. There's a ton of technical stuff around tracking that's just a massive headache, but there's a solution in e-commerce that's out of the box. It takes 15 minutes to set up, and it's done. You don't have to think about it.

For SaaS, it's a completely different world because you have to have an engineering team that builds this from scratch, right? When it comes to analytics, there's a DB and a BI layer, so maybe you're using ClickHouse and Hex. Last-click attribution, first-click attribution—all of this is homegrown at every single startup. There is no out-of-the-box SaaS that fills that gap.

That's why all of the companies that struggle don't do that first: they don't have the right martech stack to understand, if we're spending, what is actually moving the needle? The platforms won't tell you because you have the wrong setup for conversion tracking. You don't have measurement, and you're spending into the air. I'd say 90% of companies don't do that as a first step. Before you spend your first cent, have everything set up.

How do I set that up? If we don't have it in SaaS, how do I set it up as a founder?

This is what I essentially did both at Superhuman and Whisper Flow, and now at Victor: coming in and defining what that looks like, right? It comes down to having a strong analytics leader, as well as a strong analytics developer who can essentially set up all of that on your website and inside your product to make sure that you are—most people stop at, “We have analytics to track our product.” Are all the marketing platforms getting the same signals?

There's a thing called match rate on Meta, and then there's an enrichment score on Google. If you're not maxing that out, you're missing 50%. It's just ghost people. Meta doesn't see them. It's not attributing conversions.

The way the platforms work in today's environment is that each platform's algorithm does what you tell it. Let's say we have poor conversion tracking that doesn't track when someone subscribes to our product, and it's like a 50/50 match, right? If I optimize for that subscription, the higher volume of events I have, the more data it has on what a subscriber looks like. When I optimize for that, Meta will give me more of those people.

But if you have poor conversion tracking, Meta doesn't know who those people are, and it just randomly targets people. You'll have a super-high CAC and say, “Paid doesn't work for me.” I'd say most of the time, people haven't done the actual setup correctly before they can say, “Paid doesn't work for me.”

So we get that analytics data and get that set up in place. What are we looking for, then? What are good metrics, and what metrics should we be optimizing for? Is it acquisition? Is it retention? Is it spend?

When you're first getting started, the main thing that you optimize for is the most important event to you as a business. In SaaS, it's the subscription or, let's say, starting a trial. In Whisper Flow's example, it was the download—a super-upper-funnel event—because we had a desktop client and an iOS app, which made it super hard to track.

Pick what that event is and essentially validate: We spend $100K; what is traffic cost? What is our acquisition cost per user? Just focus on acquisition cost, because PLG and the product team are going to focus on retention. As long as you bring in a quality user, that's what matters. That's why, over time, you'll probably change the conversion action to whatever the most important event is.

For example, at Vanta, it's a pretty hard funnel to go through. You either have to be an admin in Slack, or, let's say, you're a smaller team and anyone can add apps, or you need to be a tenant in Microsoft Teams. You go through onboarding, and then you add Vanta into Slack or Microsoft Teams. That's our conversion action. It's super down-funnel, but because we're super-high ARPU, we know that if you optimize for that, we can have a couple thousand dollars in CAC and the unit economics still work.

But you need to take the time to find that out, right? What does success look like at the channel level? It's just acquisition in the beginning.

Okay, so in the beginning it's like cost per download.

Exactly.

Okay, cost per download. If we're thinking about good ratios, we hear 5:1 CAC to LTV, or LTV to CAC. Obviously, in the early days, is 1:1 okay? What is okay? What is good? What is great? What is worrying?

I think it depends on how much you've raised, to be honest.

If you've raised a lot, you just burn a shitload doing 1:1, and that's okay.

Yeah, because, as we were saying, distribution is everything in today's market. You open up X every day, and there are 100 new products: 5 that are in your category and 2 that have absolutely just cloned your website. This happens every single day, and the only way you outcompete them is distribution.

You want to run 1:1 to get as many users using your product and believing in you as soon as possible. You don't want to do it below that, because then you're burning money, and you have to analyze the website and the funnel and try to get it to that 1:1.

Then I think there comes a point of scale where you have enough other channels—sponsorships, you're pumping on AE, and all this other stuff starts to lift your LTV-to-CAC ratio. In terms of economics, everyone is trying to shoot for around 3:1.

It also changes depending on what kind of SaaS you are, right? LTV to CAC is usually not enough. You probably want to look at LTV gross profit, specifically if you have token costs, because your biggest cost item is usage at the end of the day. The economics are a little different depending on whether you're Whisper Flow or Victor.

How does that change, then, with Wispr Flow and Vanta, where you do have a token cost being much more significant than you had in a SaaS world where there were no token costs?

The biggest difference is, first, that Meta and Google are not optimized for B2B SaaS usage-based products. They go absolutely nuts with, “So you're telling me there's predictive revenue where one person could be a $50,000 annual contract while another person will be at $50?” There's just huge variance. A single person can be a high user, and the algorithm has no idea what to do with that.

The other side of it is that we obviously have way higher average revenue per user, but the average cost per user is also higher, right? You want to make sure that you actually have a good grasp on what cost is, and a lot of people are like, “Oh, well, that's my Anthropic bill.” But what about Modal? What about inference costs? What about all the other things that are allowing you to service your products?

Have a good finance leader who's going to be able to essentially gut-check you: “Yeah, you've been missing these cost items. This is what you should be optimizing for so we don't burn money.” I think a mistake a lot of people make is not doing that analysis at first.

How much should I spend on creative for these paid channels? If we're going to do Instagram marketing, we need a video. We need creative. How much should I spend on that? Because that can't be an afterthought.

5. Creative Becomes The Targeting

Yeah. I would say you probably want to be spending, not at scale, but in the beginning, as much on creative as you do on ad spend, because creative is everything in today's environment. People have heard this buzzword thrown around. There's an update in Meta called Andromeda that essentially changed the targeting algorithm, where the creative is the targeting.

What Meta did is say, "Stop telling me in the audience settings in the campaigns who people are. We're going to analyze the creative, and based on who we know you're trying to target, we'll find those people for you." So essentially, that removed all the media buyers who were tinkering with campaigns and had strategies for how to do it. That went out the door, and their whole job had to become creative strategy.

In the beginning, you'll see this for Victor as well as Whisper Flow: We have hundreds of creators doing different demos and different use cases for different audiences. That's the only way you can scale spend. Let's say you have a $100K Meta budget. You probably need at least 400 to 500 new creatives a month; otherwise, you're going to plateau and get outcompeted.

How do you do 400 to 500 new creatives a month?

You have a creator program. At Victor, we have our own creator program where we give a percentage of ad spend to people to essentially create ads for us. We obviously have some storytelling, and they use the product, so it takes a little bit to acquire them. But once they do that, they create 3 to 4 videos a week, and we have a couple hundred of them.

We work with 5 agencies, and we also have our in-house creative team. Again, it's the e-commerce model. E-commerce has been doing this for over a decade. That is how the entirety of e-commerce has functioned: UGC, creator programs, and hundreds of thousands of variations of creative. They've been doing that, and it's in their DNA. SaaS had to grow up to this, and because I come from that world, that was the instilling factor that I always brought in.

Again, I'm a startup founder. You're my advisor and angel investor. Well done. This is going to be a very unprofitable investment for you. How do I create a UGC network? I'm a startup. I need 300 to 400 creatives.

There are a million agencies now, right? In all of these gaps and opportunities, there's a sheer amount of aggregation and service-based businesses. There are hundreds of thousands of UGC agencies.

But most of them are shared. We get a lot of people reaching out. Every single day, I get 100-plus messages saying, "I can create clips for you. I can create..." What should I do as a startup founder to know which agency or marketplace to work with versus which one not to?

Yeah, I think it comes from asking peers, "Who do you use?" That's the nice thing about creating UGC. It doesn't actually go on their page, so they can represent every brand, right? They can do ads for every competitor.

Ah, okay. So in this case, I'm creating it for you, and then you use it across different channels.

Yeah. I'd pump $1 million into that ad.

Ah.

It never shows up on your channel, right? It's just pumping creative videos. You can also run partnership ads. Those work really well. You should have roughly 30% of your spend going to someone's posts in collaboration with you. Put spend behind that.

It could be someone who has no experience. Kids are making bank nowadays doing UGC for brands that will pay them a percentage of ad spend. We have kids who are 17, 18, and 19 making $20K to $30K a month just making a couple of ads for us. The nice thing is, if we keep spending on it and it's a great ad, they don't have to make another one for a while.

Have you found that 80% of the ad revenue comes from 20% of the creators?

Yeah, 100%, right? That's also just how the algorithm works: It finds the best ad and pumps all the spend into it.

Have you found there are commonalities among those top 20% performers? Is it a girl versus a boy, direct-to-camera versus not? I don't know if you've seen this, but there's a trend of cutting things in a kitchen while talking about something, or making coffee while talking about something. It's a hook.

Yeah. I think pattern disruption, as well as getting you to stop, is important because of the sheer amount of content that we digest in a day. If it all looks the same, you're going to get drowned out.

The best creative is essentially a rough shake of the camera that seems like it's a mistake, but it gets your attention, right? Stuff like that works really well, where it's just a messy start: "Oh, hey. Yeah, what's up?" Then you talk about the product.

At the end of the day, the key thing about creative is not the one creative. It's the package of all your assets together. They have to be different. To the algorithm, they have to look and feel different: different ages, different genders, different settings, and different hooks.

If you only focus on what works and pump just that, performance will crater. You need to constantly create net-new, completely wild, weird ideas. Otherwise, your acquisition costs will start to rise.

To what extent will that be replaced by AI creative? We had Cliff Weitzman, the founder of Speechly, on the show, and he spoke about how he did it once, and then they used AI to change the background and his clothing. One turned into thousands of variations.

Yeah. I think that's the superpower of AI in creative: spinning up variations. But if you look at full videos that are AI-generated, they're slop. You can tell immediately that it's an AI video.

I think there's a place for it. Maybe 5% of your account can be weird AI videos that are just diverse. But I think that's the secret sauce: being able to create variety. In no way, shape, or form is AI in creative at the point where it can replace real people.

I also don't buy that the algos don't deprioritize AI content versus human content.

I agree.

I just don't.

No, you're 100% right.

Okay. We've got this $100K, and we've got this UGC network. It's not really working. What does "not really working" look like? It's not doing the one-to-one. How long should I give it? Should we do this for 3 months straight? Should we do it for 6 months?

You'll usually be able to tell. If you just get started, the campaigns have to warm up. There are certain things you have to have already in place. As we said, you have your analytics and your conversion tracking. Let it run.

Of course, you already have to have customers coming in the door. Ideally, that's why product-led growth and product excellence are still the beginning steps, right? Have a product that people genuinely love and want to tell their friends about.

If that's already happening, then your conversion tracking is working on your website, your app, or whatever, and you get to 50 conversions. Then you can start paid. That's the minimum line for whatever your conversion action is, where the algorithm is going to say, "Okay, I have enough data to understand who your best target customer is."

I'd say that within 2 to 3 weeks, if you launch a campaign and go through a couple of iteration cycles—you launch some creative, see what worked and what didn't, ask why, throw in some diverse stuff, and do some variations—you probably want to give it a full 3 months.

But you'll be able to say, "Yes, I can start printing money today in 2 to 3 weeks," or, "I have to go back to the drawing board on all of these things," because there are a lot of moving pieces. It's the ad creative, the ad copy, the landing experience, the messaging on that landing experience, your page speed, and whether your screenshots in the App Store actually make sense.

All of those things are microlevers that compound. You need to be refining everything along the way as you spend. Otherwise, things will plateau.

You said a couple of things there that I need a bit of help on. The landing experience—what do you mean by that, and what should I know as a founder?

A lot of people have really poor UI, UX, website design, or branding. Absolutely awful, right?

I don't want to throw shade, but have you seen PostHog? I go on their site and I'm like, "I have no idea what's going on. This is Beijing's version of UI."

The thing is, that's a pattern-disruption thing for them. They know who their target audience is, and the target audience appreciates the weirdness.

Oh, yeah. James is amazing, and it works. It's a brilliant company—phenomenal. But I look at it and go, "Ooh."

No. On the flip side, it's just: Do you know what your traffic mix is? Is your site mobile-optimized? The sheer number of sites I see where the main CTA is a scroll and a half away—come on.

That’s the first thing that you start with, and a basic audit will show you, right? That’s what I mean by the landing experience. If I read nothing else but your headline, do I know what you do? It’s hard sometimes to explain what this crazy AI SaaS niche product does. Focus on your messaging.

Actually have something that tells a user, if they didn’t move their thumb, they didn’t do anything, “I want to try that out,” right? Go through that proxy. Same thing with the creative. Same thing with the copy. That’s a frame that you want to take throughout the entire funnel.

Page speed really matters.

Page speed matters because it depends, obviously, on who and where your target customers are, right? For the US and UK, we have good data and it’ll be fine, but where page speed really starts to matter is both AEO and SEO. It is one of the core elements in what gets your site to rank in, let’s say, a top 10 position on Google.

The same thing applies if, for example, a bot crawler goes to your website. It’s an easy call to see how long your page takes to load. If you’re on a low rating from A to F, the average person in the world will probably not load your site immediately. Any second of improvement is a drastic jump in conversion, and the numbers show that, right? Even though it seems like, “Oh, yeah, it loads in half a second versus 200 milliseconds,” it does make a difference.

Okay, so we have this $100K. We’ve done 3 months. I’m really enjoying this because, as I said, for me, it’s nice to follow the chronology. You have $100K. It’s going well, I think. How do I know whether to pour fuel on the fire, keep the budget as it is and just keep testing in a more measured fashion, or, “Uh-uh, this is not fucking working”?

6. Scaling Spend Tests Incrementality

You want to chart your total spend against your acquisition and how elastic it is. There’s always going to be a lag where I spend today and then convert to paid in 14 days, right? We’re looking at 2 main things. Revenue is the most important thing from an ARR standpoint because we’re in SaaS. As I spend up, does that ARR also spike? How elastic that relationship is will immediately tell you: oil on the fire, or let’s pump the brakes.

As we spend up, this is a very common thing that you’ll hear: How incremental is your spend? It’s a question you should ask relatively early because once things are working, it’s really easy just to dial it up to 13 on a scale of 10.

Sure.

What ends up happening if you’re not very diligent is that all that additive spend could have not been there, and you would have gotten the same results. Then you’re just giving money to Meta and Google, right? There’s that fine line, and that’s why I’d say people who truly understand growth are on the rarer side. Usually, someone is an expert in Meta or an expert in some other channel.

Being able to sit above the whole thing and say, “Okay, here’s the interrelationship of this much spend here, this many sponsorships, this many podcast features, and this is how much PR we’re pushing”—all of that interrelationship is how you essentially ensure the most elastic spend-to-revenue impact. That’s further down the $100K, but that’s the end state. You are everywhere all at once at the right moments.

Why did Superhuman’s growth asymptote, then?

At the end of the day, it was very much a premium product with an immediate paywall, and the core ICP was founders, right? We also experienced this at Whisper, right? The early-adopter tech founder is not an infinite audience.

I think today it’s an ever-growing audience because everyone can be a founder, right? But if you look back even 2 or 3 years ago, the TAM of that was finite. That was the first asymptote.

Essentially, going after additional ICPs like sales: How do we turn Superhuman into more of a sales engine for people? Looking at the recent opens on the right, that’s my favorite feature, right? I can see who’s creeping on my emails and send them a reply. Adapting the product to more ICPs is where you get those additional step changes.

The asymptote is because it was hyper-focused on 1 ICP, and if you don’t proactively open up the layers of the onion, you’re going to hit some kind of level.

But if you look at revenue scaling and you see Fixed Scale faster than Superhuman in the top-line revenue number, do you not just go, “We fucked up paid”?

Well, yes, right? Fixer followed the e-commerce playbook: UGC, tons of ads, strong on Meta. The thing is, it’s a slightly different value proposition—

Different market.

Different market, different consumer, right? This is someone who wants a simple tool that adds labels to their Gmail.

Would you say Fixer should continue to just spend the hell out of this market, then?

From what I know, their scale was both on the B2C side through Meta, but they also made big revenue jumps through the hyper-frustrated Outlook consumer and big enterprise deals, right? That’s also where Superhuman Mail did the same thing: going after big logos and ensuring that you can essentially have sales-led growth alongside PLG.

I think Fixer did that also super well. They pushed hard on the sales side because that was also their background—the relationships they had from the email space. That’s the same thing for Whisper, right? It was really interesting to see that we were B2C PLG, but a very large component of our ARR was enterprise before we had a single AE, because it was self-service enterprise adoption, team licenses, and that kind of stuff.

You also want your product to be easily adoptable by enterprise. I think that’s a nice unlock as well.

We mentioned the 3 channels: Meta, Google, and lifecycle. What are your lessons on how to do YouTube well? We hear about YouTube being the acquisition engine and machine that it is. Any lessons that I should know as a young founder?

It’s a totally different type of creative. If you look at Meta and YouTube, they have similar but slightly different hook curves, where the hook happens slightly later on Instagram and Facebook. You want that immediate validation, while on YouTube you’re there to spend the time and digest the video. You have a little bit more of a longer attention span.

A lot of people are terrified of YouTube because the best-performing YouTube ads are 16:9, so they’re landscape. Most people are like, “Well, I have a UGC program. I just have Story placements. What am I supposed to do?”

We did a really cool thing at Victor. We had Victor build an app for us that takes any Story video and adds it into a static template that has logos that use us, a G2 rating, and a CTA. The video is the replaceable asset. We turned all of our Story ads into YouTube ads.

Every single one ended up on YouTube. Not all of them worked the same. We had to edit them a little bit to fit the format. That’s usually why people are terrified to get started with YouTube, but just go into Figma, make a template that’s a static asset, throw your video in that, and export it. Welcome to YouTube, right? You can run ads that way.

There’s also a different kind of storytelling where you want it to feel like an organic YouTube video. Someone is talking about a hack or a use case, like, “Yo, I just started using this and it absolutely changed my life. I’m going to show you how,” right? You’re not going to say that in a Meta ad, right? It’s a very different script, framing, and filming.

We have a separate team that does the scripts and films the YouTube videos. It’s an entirely separate program from our UGC. We use some things from UGC in YouTube, but it deserves its own focus.

We’re going to go into the team composition later. I want to stay on channels. We now have a working program.

Mm-hmm.

How do we determine how much fuel to pour on the fire?

I think it’s all dependent on the elasticity that we defined. At Whisper, we did what I think everyone should do: go as hard as you absolutely can, see it blow up, understand where and how it blew up, pull back, and use that as information. Now we know what is incremental. We know where our ceilings are. We know that we hit audience saturation or need more creative.

We did that last year at Whisper. We 5X’d the budget from 1 month to another. Where do the pieces fall through? The reason why we did that instead of trying to figure it out over time is that you don’t have 6 months to say, “Okay, we need to be growing 30%, 40% month over month over month,” and the only way that we can do that is, “Okay, let’s see where our ceiling is.”

Maybe we would’ve been surprised that we would’ve been scaling infinitely, which didn’t happen, right? We then knew, “Okay, Meta needs to be second to Google. Here’s why. Here’s how all these things worked together.” These newsletters were shit. This podcast doesn’t work.

How do the pieces fit together? In my opinion, go hard to understand what doesn’t work and where your mix needs to be adjusted, then bring things down and scale back up with those learnings. Then you can get a repeatable rep.

If something doesn’t work today, do you just turn it off, or do you wait to give it time?

I would say it depends on how poor the metrics are. If there’s no world in which this is ever going to make sense, turn it off immediately. But let’s say some leading indicators are saying, “Eh, maybe.” It’s pretty crap, but let’s give it a little bit of time to see if there are other things impacting this.

Like I said, there are a million dials. From the ad to the conversion, there are a million things that you can tinker with along the way, and that’s why you want to give it time. This asset is going to the homepage. What if we had a hyper-personalized landing page that says the exact same thing that’s being said in the ad and even includes the creator’s face on the page? It’s a random idea, right? Does that move the needle? Test some things, give it some time, and then scratch it off.

What single channel worked best at Whisper, and what did you learn from that?

Definitely Google, in terms of—

That is in traditional ICO?

—Google Ads.

Ah.

Yeah, Google Ads. Google Ads was, and still is, the main driver. It was surprising across literally everything: non-branded search, PMax, YouTube ads—everything performs there. The reason why is that it’s the middle of the funnel, but you essentially have everything from top to bottom right in 1 platform. If you have all the right conversion tracking, Google is a big marketplace with a lot of surfaces.

Just optimizing for app download, because it’s a freemium product, gives us a lot more fine-tuned control than we have in Meta when it comes to CAC and how much we want to spend in India versus the US because of the different prices in those markets. We don’t have as much fine-tuned control in Meta because, like I said, they’ve removed all the manual settings. Creative is your targeting: either it works in a geo or it doesn’t. You just iterate on creative.

It’s a completely different job. You have a pretty structured beast to essentially understand how to scale there. So it was Google, and it’s also because of longer-form video for a pretty difficult product to explain. “Okay, I talk, it turns to text, but where? In text boxes?” Explain that to the average Joe walking down the street. They’re going to look at you weird. “Yeah, voice notes.” “What do you mean? I send them in WhatsApp, right?”

There’s education that needs to happen. The 30-second to 1-minute videos on YouTube that educated the base ran for a while and got millions—hundreds of millions—of impressions. They then fed into non-branded search, PMax, different placements, and display, and eventually those people converted. All those pieces worked together.

Is performance always best at the start of a channel, or does it get better over time?

It gets better over time, but again, it has its own scaling laws. There’s a point where you have to diversify into other platforms before you see an additional unlock—when you hit some kind of audience fatigue with whatever you’re targeting. But in the beginning, it’s always shit because you’re waiting for the conversions.

When you hit audience fatigue, do you keep spending, waiting for the next unlock, or do you just go, “I’ve tapped out this channel”?

Well, you want to unlock the next audience, right? For example, when we were looking at Victor, when we launched, we purposely had “the AI employee for everyone.” We did this to see who came in the door. We had agencies, e-commerce brands, and SMBs that first showed up.

Essentially, what we did was go through in order. We said, “Okay, we have evergreen stuff. We’ll continue running that, but now let’s go after agencies. Can we acquire them? What does the funnel look like? What does the creative look like?” Once we unlocked 1 audience and it seemed like the economics worked, there was a repeatable process to refine it. Then we could go on to the next audience.

You’ll hit audience fatigue in what you’re currently targeting, and marketing now is very ICP-specific. We have all the latest and greatest tools that can tell you, in a prompt, whether you’re using Victor or whatever tool, “Simulate this person. What do they read? What do they think about decision-making?” Run that prompt, and then you have the richest ICP research you possibly can have.

Create it for that person. Create an experience from the landing page through the onboarding and product experience for that person. You can continue scaling by just going down to the niche, audience by audience.

Because I don’t think, as an investor in the company, that was a good tagline, and you may hate me for that. But “the AI employee for everyone”—I don’t know what the fuck that means. If you were to say “the AI video editor,” great, amazing. If you were to say “the AI copywriter,” amazing. “The AI employee”—are you doing my tax? Are you doing my creative?

But my point is, are you able to do horizontal taglines and then verticalize, and that’s the right approach?

I think because the category is so wide—and I wouldn’t say we’re the first, but we’re the first to say “AI employee” and actually deliver on that promise, where work is being given and the usage is actually ROI-positive, actually valuable—we started with that because, again, how else are you going to frame it as being different from Claude or ChatGPT and these other players?

For us, it’s the positioning around knowledge work. There was nothing other than “AI coworker” and “AI employee.” There’s no other positioning that really makes sense, and because it hasn’t been used a lot, I think it’s confusing.

But now that there are more use cases and stories to back it up, we can shift the framing to, “I hired Victor, and here’s what he did.” Then it’s, “Oh, it’s a guy’s name. It’s an employee. Oh, he’s an AI tool?” You can invert the process now, but we had to start with something to essentially build those use cases.

One of the things that’s supposed to be a growth engine as well is referrals and referral programs. Any lessons from Whisper or Superhuman that are really important for me and founders to know about how to build great referral engines?

7. Referrals Follow The Aha Moment

I think it’s one of the core things. We talked about Meta, Google, and lifecycle. When you think about PLG, have that referral program set up, and have it be very front and center and easy to discover in the product, to make it effortless for someone to share the product and get something for doing that.

At Superhuman, it was “Give 1 month, get 1 month,” and we had people who had hundreds of referrals—hundreds of months they were never going to pay for for the rest of their lives—because it was delightful and easy to discover. It was very tactile: I’m getting something for doing this.

It was a similar thing with Whisper. When you hit your, let’s say, 2,000-word limit—we’re changing a little bit of how that trial works now—but essentially, you want to find the right moment to show the referral program and make it tangible. You’re about to hit your word limit: “Hey, did you know that if you share this with your friend, you can get the next month free?”

Okay, I’m going to take 5 seconds and go do that. I’ll tell 1 friend, “Hey, sign up for this,” and if they sign up, I get that 1 month. Now I have another person who can spread the word, and that K-factor is what makes or breaks that.

So align the referral program to usage limits?

If that is your product, right? What is the thing that, if you have more of it, won’t frustrate you? For Victor, the referral program that we have is all token-based. You get credits to essentially use in Victor. Whenever you’re near the limit, we give you a decision tree, depending on who you are, of what you could do to earn free credits.

We have a creator program, so if you just post on LinkedIn about Victor, we give you a payout via CPM directly to your account in credits. That gives us social and virality that we can essentially push forward on. Or you can be a part of the actual referral program, where you can invite companies and get a percentage revenue share of the referred company’s paying plan in credits every month.

If I’m a company and I know another cool company that should be using Victor, and I know that Victor’s expensive—we have 8-person teams that are spending $15,000 to $20,000 a month to use Clay, and it replaced all their hiring. We can go into that later—but, for example, if that’s a cost item and I have another friend who has an agency, they’ll probably spend $5,000, and I’ll get 20% of that as revenue share, which lowers my cost.

To us, that makes sense from a CAC basis. We’re willing to spend $1,000 to $2,000, depending on the cohort, to acquire a customer. We’d rather have our customers finding those best customers for us.

Yeah. Fuck, if I’m a startup founder and I’ve got a load of founder friends in noncompetitive markets, great.

Game on.

Yeah.

I totally get that. What are the ways founders fuck up referral programs that they should avoid?

I mean, it’s either not making them tangible, where it’s just, “Hey, refer your friends,” and it’s, “But what am I getting for it?” Or it’s just some “You get swag,” or things that don’t really matter. Or they create these hyper-complicated 20 tiers—do these actions, gamifying the referral program—and then no one wants to use it because it’s a pain.

So I think it’s somewhere in that spectrum: you’re either doing that or that. You need to hit the sweet spot of what that looks like.

Unless it’s Palantir swag, in which case—

Yeah.

That’s really quite cool, and I would love that. We mentioned how the timing of when you put that referral in front of people matters. The paywall is a difficult one, and it’s also about where to put it—whether it needs to be hard and immediate, or whether you need to show value. What lessons and advice would you give founders on the immediacy of the paywall and where it sits best?

I think even before the paywall analysis, the main thing you want to adjust for is when that magical aha moment happens. You want that magical moment to happen as soon as possible, but you also want the paywall near that moment. That moment is either your first discovery of, “Wow, this is amazing,” but you want to continue that over time.

You have that magical aha moment, and then you stay in that kind of honeymoon phase of, “This is just amazing.” And then within that phase, you put up the paywall, right? That’s whether it’s words per week or how many credits you get in Victor.

This is a big game that we’re trying to hyper-refine now: how many free credits should we give you, and how do we nudge Victor to show you really useful workflows? Because the moment you have one that goes, “This is game-changing,” that’s when you put up the paywall. You want to open up your pocketbook because you’ve felt something magical that you haven’t felt before.

Because there are so many products, those true, unique, magical “This is in this product? This is insane” moments matter. This happened at Whisper Flow when we had that first rocket-ship moment, when LinkedIn was going absolutely nuts, saying, “There’s this thing, Wispr Flow, and I’m not typing anymore.” A lot of that was organic, right?

But that’s because people hit that magical moment. It’s similar to the amount of posts that are happening about Victor. I’d say half of them are organic, and half of them are from the creator program that I mentioned: you post and you can get credits. But it’s all genuine, right? It’s that magic moment that matters.

Should credits go in marketing budgets?

We call this fully loaded CAC. Our free-trial credits get summed with marketing spend. If you’re not doing that, then you’re not really calculating your acquisition cost.

No.

Trial credits, or anything free that you’re giving away, are a cost, and they go under a marketing cost item.

We mentioned another one that was very interesting. You said AEO, which is obviously answer engine optimization. We invest in Peak, which—

Yeah.

—essentially optimizes this. How do you think about that as a new channel? How do we embrace it? What should we know?

8. AEO Extends Organic Distribution

AEO in general followed the core tenets of SEO, where what used to be old-school SEO—pumping out thousands of pages—came back again because Google, because they own the SEO algorithm, essentially penalized people who were just pumping out pages for SEO. So, over the last couple of years, people stopped doing that.

But the sheer amount of pages that you have in AEO is also changing a little bit. You can ask any agent to analyze the sitemap and essentially do, like, cron. We do this with Victor. We look at every competitor and their sitemap. We see how many pages they’re making. All these companies are doing 100 to 200 pages a week.

A lot of it is just generated AI slop, so that’s also what you don’t want to do at scale. You actually want valuable content that says valuable things because the AI crawler essentially ingests that. Then what you’re seeing is what gets written about you when someone asks, “What is the best tool for this?” or “I’m trying to do this,” and it’s generating these citations.

The things I’ll say that are most important to AEO now are: yes, have an engine that sends content through your website on specific things, but the most important things are YouTube, Reddit, and the social narrative. One of the early levers that every founder should use when they’re first starting to scale is investing in good YouTube reviews.

YouTube reviews, because they’re long-form, rank for the long tail. They’re one of the main things that get picked up by the answer engine. You’ll have your website, PR, and external sources talking about you, and one of the biggest and best external sources is how many reviews and videos you have on YouTube.

They’re a really high citation on ChatGPT, and that’s one of the early, strong things that every founder should focus on.

Dude, is traditional PR and news dead? What I mean by that is, being in TechCrunch used to matter.

I still think it matters in the sense that I have the opinion that, when you launch, you should do Product Hunt and get featured in TechCrunch. That’s just founder initiation.

Almost table stakes.

Yeah. I’d say it builds a level of credibility and trust that you’re on the map, in the same way that everyone went through the process. But the reason why it’s important is that PR is less about PR blowing up and creating a traffic wave.

Really, that PR is mainly for citations around AEO. How many external, credible sources are writing about you? You’ll see your AEO traffic significantly increase because the more that a non-owned narrative is being talked about positively about you, the better off the product is.

Should I bother doing TikTok?

I think there’s a time and place for it, and it depends on whether you’re more B2C than anything else.

I’ve never met a CMO or a head of growth who’s been able to crack it. I’ve never.

I think there’s a time and place to do some audience diversification. But, to be honest, in any of the 3 core companies that I’ve been at so far, it hasn’t worked yet. Maybe I’ll get a chance to change that at Victor.

Totally get that. Final one, again, before we move to team composition: what happens if we turn off paid as much? We’re spending really aggressively, and it’s working. It’s working, and we get to 30, 40, 50 million in revenue, and we’re like, “We probably have a bit of a paid engine now that we’re a bit hooked on.” Turn off paid, dial it really far back, and growth goes down?

That means you have other problems, right? I think the sweet spot for the organic mix should be around 35% to 45% of acquisition coming from word of mouth, organic, and people truly discovering it based on everything else that you’ve built.

In the beginning, because you just started and there’s not a lot of stuff out there about you, it is entirely tied to paid. But in the later stage, if you turn everything off, that means you’ve forgotten about the other half of the job, which is the SEO, the AEO, the reviews, and everything being written.

Sometimes you want to dial down and see how elastic it is, going back to the previous point we were making. You definitely want a strong MMM model for incrementality. The moment you start spending north of $1 million a month, definitely have a model that says, “This channel is incremental. This is not.”

The model will always be wrong in the beginning. Make some changes, dial things down, do holdouts, whatever, and then see if the model proves true. The model gets better over time, and then you’re less in the dark. You actually understand what levers go up and down and where you can spend a lot.

So I think spending down to understand how much drop-off you have is a very strong learning moment.

How many good growth leaders do you actually think there are?

Not a lot.

Yeah.

If I’m going to be honest.

Really not a lot. There are a lot of founders who’ll be listening to this and going, “Wow. I’m thinking through 3 companies in particular,” where I’m like, “Shit, I wish I could put Matt in there.”

Yeah, I mean, I think I’ve had an interesting growth through marketing because I started in corporate consulting, which was an absolute life suck. I hated it. That got me to found my first agency, where I had to learn how to do everything myself from zero, right?

Because I had to do that, and I did that in e-commerce, SaaS, and enterprise B2B. My agency’s still around today, and I’m not involved in the day-to-day. But because I had to do that, I understand how to launch a Meta ad.

I’m not a head of growth or a CMO who has no idea how to do execution. This is why, as head of growth at Whisper, up until December of last year, I was the only person doing execution on a $3 million, $4 million, or $5 million budget. I was doing everything myself.

But this is what I say to all university students and young people, which is: “It’s never been more important to be full-stack.”

Yep.

You need to be writing the creative, shooting the video, editing the video, putting the video on Meta—whatever it is, Instagram, YouTube, you name it—full-stack. The whole “I just do creative and copywriting” thing? Fuck that.

I think a lot of people have been saying this, and I definitely echo that: hyper-specialists are dying a slow death. You need to be a specialized generalist. You have to pick something that you’re better than the rest at, but get your hands dirty and learn how to do a little bit of everything.

A lot of growth leaders get lost in the marketing analytics and the conversion tracking. That stuff is usually where they don’t really understand how it works. If you don’t understand the fundamentals, you actually don’t know what makes the system tick.

We also live in a world where ChatGPT, Claude, Victor—they are the best learning vehicles ever in the history of humanity. If you don’t know how to do something, ask it. Have it tell you, “Give me everything step by step that I need to do,” and go and do it. Then see where things break. See what you don’t understand.

There’s agency if you’re hungry and a learner. The best people who win are the nerds, the curious people. That’s the line for me.

If we go to the team itself, the requirements for talent change. How has what you look for in talent changed in the last year or two?

9. AI Changes Growth Team Design

Yeah, I think it’s a complete 180. I’d say that a year ago, I would have been looking for someone who had 10 years of experience in, let’s say, Meta ads. They’d have scaled to hundreds of millions of dollars and been the best in the world as a Meta ads media buyer. That would have been the main thing I’d be looking for. Let’s say that’s the person I’m hiring.

Now I’m looking for that, but they don’t necessarily have to be the best in the world historically. If they’re AI-native or a systems thinker and can deconstruct what makes their job hum, that person plus experience will outcompete someone who just has experience and isn’t AI-native.

I’m seeing that happen, and I’m going to have a hot take here: I think a lot of teams are trying to make their teams AI-native, and they’re failing.

Why are they failing?

The people who are in those roles are not systems thinkers. If you’re in a marketing or, let’s say, ops role and you’re not a systems thinker, you don’t actually know—

How do you think about it? What does “systems thinker” mean?

To me, that means being able to step back from a task that’s a part of my job, take one degree of separation from the task itself, and say, “In my role, what are all the moving pieces that I need to do on a day-to-day basis? What are the interrelationships? Where’s the boring admin? Where’s the reporting? What’s all of that?”

It means being able to map that out and understand how your job as a system actually works. Then you can say, “Okay, I can apply an agent or AI to this part. Now I can focus on this higher-leverage part of my job.”

A lot of people go to ChatGPT and ask it a question, and then they keep doing their job the manual way. If you take a step back and say, “What would it take for me to fully automate all of my job?”—of course, it’s not going to work, and things are going to break apart—but run yourself through that thought experiment.

If you ask a lot of people that, it turns out they don’t really understand their job. Because they’re not systems thinkers, they don’t actually know what all the inputs and outputs are. That’s the line for me. When I interview people to fill out the team, it’s partially like an engineering interview, where systems thinking is the bar.

So what do you ask them to do? Well, I want a job with you, okay? You’re clearly fucking brilliant, and I’m a podcaster. What are you going to put me through? How are you going to test me?

A core question is: How do you use AI as a workflow at work? But I’m actually more curious: How do you use AI workflows in your personal life? Do you have any systems that you feed through AI that make you a better person outside of work? Maybe the line blurs, because what you do is also work and personal, whatever.

And so a good answer versus a bad answer would be?

A bad answer would be, “I have a skill or a chat thread that’s a project in ChatGPT that I talk to, and it has some context.” Yes, that’s factually correct. That’s how you should use a memory layer and a context layer, and you have one proactive chat conversation, so it works.

Chatting with the thing is not a workflow, right? If there’s no feedback loop, that’s a problem. One of the core things that I always test for is: Let’s say an AI gives you slop. What do you do with it?

A lot of people will take the slop and say, “This is slop.” Maybe it changes, and they get frustrated. Then they take the document and start writing it themselves. If you don’t go through the pain of giving it feedback, that’s the first kind of feedback loop that I test for.

But the next thing is, let’s say you have a loop. I used this example before: I built a sponsorship system at Whisper, but essentially the workflow is a self-improving loop.

If a newsletter sponsor emails me, my agent knows that this is a newsletter request. It asks them for their rates, researches what their audience is, and does that first part of the negotiation for me. I step in to approve, “Yes, we want to work with them.” We get a contract, and that’s where my manual step in the newsletter ends.

I throw the contract into the agentic system. It ingests all the cost data into the file system, does the copywriting and all of the email sending, creates all the links, creates all the conversion tracking, and sends everything out to the partner. A newsletter is a boring, antiquated email: I will email you, and you will tell me the performance.

Based on the copywriting and the cost, it will also know what the performance looked like and then say, “Do we continue on with the partner?” That’s also an agentic decision, because my benchmark is: CPM should be this, conversion should be this. Let the run fizzle out if it didn’t work out.

Again, that’s a workflow. I have a couple of human touchpoints, but if it analyzes the data—or if it generates copy—it knows every link’s performance and can say, “This copy works; this doesn’t.” So the next time it generates copy, it’s based on all historical data.

What do you use for this?

So this was first the marketing OS I built at Whisper. It was Claude Code on a computer, with just the folder system, skill, and .md files. That was the beginning of that system.

I had Claude plugged into my email, with cron set up to check at 9:00, at noon, and at 6:00. It would check for any new newsletter requests, run a workflow to respond to those things, and then bubble up a message to me in Slack saying, “Hey, these need your decisions. Here are some contracts that got sent over.”

Then that system evolved. That was my own OS, and I had to get control of that folder to share it with the team so other people could use the workflow.

For essentially a year and a half, I was the one person at Whisper doing all the execution. Imagine $1 million a month on Google, hundreds of thousands of keywords, ads, experiments, and landing pages. I owned every KPI from eyeball to download, and then product took over onboarding.

Then add Meta and newsletters. At a given point, I was working with between 70 and 120 unique newsletter providers that would send me emails and ask for copy approval. I needed to create unique images.

So I said, “Why are newsletters still in the stone ages? There’s no platform. It’s all through email, and you’re negotiating and haggling with these people, trying to get you on a CPM, right?” That was one of the first things that I created a workflow around.

I don’t code. I learned how to do this by doing it: “I want to automate this. Claude Code, how would you do this?” I started by building the OS by asking how to build an OS.

Every day, 90% to 95% of my work throughout the entire day was fed through the AI. I’d say, “I want to optimize the ads. Can you pull the reporting? What are the keywords?” I would make all the changes and see all the copy through Claude Code in the terminal, because that allowed me to give feedback to the feedback loop. It got a little better every day.

There came a point where the models also got good enough. I was an early adopter of Claude Code, so December 2025 was when the Claude Code moment happened. I adopted it in September, before it was good.

Immediately, you saw that insane ramp.

It was wrong a lot. The copy was poor, and things broke. It was early days, but I was curious. I’m a tinkerer, and I used that to essentially learn how to do the whole thing.

I know we deviated from the team question, but that was me having to learn how to do the workflow, essentially, because I had to do it without knowing how to make that work from A to Z. That’s what I test for: can you take a hard problem and build a system around it that you have solved?

How many candidates do you actually meet who have that level of depth?

Less than 1%.

Exactly.

Yeah.

Can you even build a team on that?

But that’s the thing, right? I think the team composition of today and tomorrow is way leaner than it was before. So, at Grammarly, let’s say—a bigger startup in SaaS; “startup” is in air quotes there, big corporate SaaS—they have an influencer marketing team, right? They have 3, 4, or 5 people.

Today and tomorrow, that is 1 really good person with an amazing suite of agents. Imagine you scale to 100,000 influencers that are posting about you in a month. If you have an agentic system and you know how to weed through the data and understand where you need to be plugged in, you could do that.

That’s why, even though it’s 1% of candidates, that’s fine. If I find that 1 person who is going to be able to fully own that, build the systems, and, right now—and this is my bet for the next 3 to 5 years—80% of our work is still manual and 20% is done by agents, hence why my bet with Victor is essentially to solve all of knowledge work.

I think in 3 years, companies will essentially be like a board of directors, where 20% of the work is the strategy and the thinking, and then agents just do 80% of the execution. Right now, editing is still manual; sending an email is still manual. Even though we have the tools, there’s still a human layer in that.

I think that will change as quality gets better, and as usage and actual adoption toward an ROI basis get there. That’s my bet, and that’s also why this stuff is so important, right? Because if you can’t understand how the systems work, you’re never going to be able to steer agents.

So, less than 1% of B2B CMOs today—are they fucked? I don’t mean that too bluntly, but I don’t know any B2B CMO in a scaled company who has any fucking clue what you’re talking about.

They hire those people around themselves.

Do they really?

They try to. I mean, I—

I’m not going to name companies because I don’t want to shit on people, but if you look at some of the big providers, are you seriously saying, in your—

No.

Anaplans and your Coupas, that anyone has a fucking clue what you’re talking about?

Yeah, I mean, that’s also why I think the companies that do will quickly gain market share against those that don’t, right? I think we live in very interesting times in terms of the competitive environment, where the incumbents that don’t adopt this at the senior level will essentially see their growth rates stand still, while the hyper-growers that are doing exactly this framing will essentially gain that market share and outcompete them.

I think you’re right. Leadership needs to change. Like I said, I think a lot of teams—and my hot take was probably: fire most of your marketing team that is not a systems thinker. Stop brute-forcing people in. Hire the right people into the roles, because the role has changed.

The JD is no longer the same as it was a year ago, right? People are still brute-forcing people into these new JDs. I think there needs to be a bit of a shakeup.

Okay, I’m a founder. I’m looking at my marketing team today. There are 10 people. I’m a Series A or B firm, whatever. Who should I fire? How do I test it?

When you’re not looking and you come back, there are suddenly systems, and that 1 person is 10X-ing themselves, you’re not going to fire that person. But then you will very clearly see who hasn’t done that.

That difference is being felt: the A players are becoming S-tier players, and the people who were B players are becoming D players. That rift is created.

The other thing I just think is that consumer marketing is not consumer marketing and B2B marketing is not B2B marketing. It’s just marketing.

Yeah.

At the end of the day, there’s a consumer who has a buying decision at a big enterprise or as a person.

100%. That is also why people were questioning our initial marketing strategy at Victor, because, like I said, super-high-CAC B2B SaaS, right? You’re hiring an employee into your team and you’re paying thousands of dollars for this thing. That’s a B2B SaaS tool, but we approach the marketing like a prosumer app, right?

But there is a 19- or 20-year-old at Polymarket. He’s called Tobin. I hope he doesn’t mind me saying this.

I mean, he was at Whisper, right? He was originally at Whisper.

He was at Whisper?

Yeah. He built the original UGC viral program at Whisper before he went to Polymarket, so I know Tobin.

Oh, wow. Well, this is perfect then. Yes. I’m like, you would pay through the nose for talent like Tobin, who just gets it, who is 10X what anyone else would be in that position. Do you not agree?

Yeah. I think in each lane there are those people, and that’s also why, from an ML researcher standpoint, you’ll pay millions of dollars for a person who is going to move a model forward.

I think we’re coming to the place where people in these marketing roles—they’re unicorns. They’re rare, and you want them in your company. I think that same ML crazy moment is maybe going to happen over the next year in the marketing acquisition side.

Dude, are you ready for a quick-fire? I’ve loved this. What is the most underappreciated growth channel today?

There’s a channel we haven’t mentioned, and I think a lot of people do this poorly. It actually compounds if you do it well, and that’s affiliate, where you’re paying, let’s say, a percentage of revenue or some kind of CAC to people who are not your customers, and they go out of their way to create content and reviews and all of that.

At Whisper and Victor, it is the highest-ROI channel. If you set up a strong affiliate program and bring people through the door, and they feel like they can earn money even before they’re your customers, they will become your customers.

It’s one that’s a little bit outside of the referral program because it’s not through the product. It’s about making this sexy external program that someone who is just an affiliate wants to use. They have a surface area, a website, or some audience, and they want to sell your thing to that audience. Acquire them that way, and that’s one of the highest early-ROI channels that, if you start doing it in the beginning, ramps.

For Victor, a good 10% to 15% of our acquisition on a monthly basis comes through our affiliate program—people who want to earn money—because we give them between 10% and 15% revenue share. If that affiliate gets a company that pays 10K a month, they’re making 1,500 bucks a month from 1 sale as an affiliate.

Create a program that helps people earn money. This is the other thing that we said.

Can I do this creator program? I swear I’ve sent, like, 10 companies to Fred [?]. Mom, I’m retiring.

Yeah. We have some affiliates who are making 20K or 30K a month.

What’s the most polluted channel? What’s the, “Ugh, this is just a shit pit”?

I’m going to say X.

Just the launch videos are boring. They’re boring.

That’s exactly what I mean, right? Everyone is trying to do exactly the same viral-launch thing. It’s also polluted in the sense that X ads don’t work.

They don’t work?

I’ve yet to meet a SaaS head of growth or performance marketer who says that X ads print. So if someone has gotten X ads to print, please tell me, because I would love to be challenged on that.

That’s fascinating. Okay, so X, I’m with you. I’m just so bored of this. I don’t know if you saw it, but there was 1 video yesterday—and we’re talking about it, so it obviously worked—but it’s this company, Iconic Weave [?], I think it’s called, where they jumped out of a plane and did the copy while jumping out of a plane.

It’s like, have we got to that stage? Is it that mature a market that you have to jump out of a plane and read it while doing that?

The formula of shock-and-awe launch videos is, to be honest, specifically for those that are like, “Well, we don’t really have a product.” I’m not saying that about their example, but the sheer amount of “We launched this for that,” where we spent way more time on our launch video than on our product—I see so much of that, and that’s why I say it’s polluted.

What would you most like to change about the world of growth?

Have more good growth leaders. But, in honesty, I think—

My request for startups is: I don't know why we don't have a school that basically says, “Hey, completely understand you need to be AI-pilled. Great. What the fuck does that mean? Come to our school, and for 6 weeks we will teach you how to get jacked up on every system, process, and model. For 6 weeks, 9:00 to 5:00, this is your job, and you pay 10,000 bucks.” I would say that's more valuable than university, say, for teenagers.

I'll go start another company.

Do you—

No, I mean, you're 100% right, Nick—

I would fund that company today with millions of dollars. I'm being serious. If you're J.P. Morgan or Goldman Sachs and you put that at the bottom of Canary Wharf, would you not send every single employee there?

Because then you built the hyperscalers of tomorrow. Super interesting. I went to NYU for college, and what's interesting is that I did not take a single—I mean, zero—marketing classes, and I learned everything by doing.

I think what's interesting is that the education around marketing is also, like, “Sit still.” And I'm like, again, I haven't been into a college today. Maybe that has changed, but are they teaching you how to install a Facebook Pixel? No, right? The framing of how it works comes after you start doing the execution, right? You can do and then understand, versus understand and sit in a room and learn, and then go and try to do the thing. You're going to learn so much faster just by doing.

So I agree with you. I've yet to see a true A-to-Z boot camp of what it means to be a growth leader. Reforge and Ryan Balfour have tried to do this, but it's still—

I'm actually not even saying a growth leader. I'm just saying how to get knowledgeable to the extent where you can build processes—

Yep.

—systems, start replacing yourself with Claude—with, you name the provider—where you can start doing expenses, copy, and design.

Yeah, you're 100% right. In that frame, I agree as well. I'm applying it to doing that in marketing, but the grounding is that people don't know how to do the beginning step.

The—

You're 100% right.

The very beginning. So how do I simply set up my Claude to be efficient with integrations so it gets access to all my files in the right way with the right permissions?

Yeah.

So you said 9:00, 12:00, and 6:00, where it does the check-in. Dude, I think you forget where people are.

Yeah, no. What I think is table stakes—and I always forget this—is that people don't set things up this way. For example, this is when I was using Claude Code. Now, I'd say 80% of my work is Victor. But for the people who aren't using Cursor, if you're using Claude Code, do one thing: set up what I call the session-end skill.

If you set up one skill that's going to let your work compound, it's what I call session end. I have a terminal or a Claude Code session, and I'm doing some realm of work. It's almost done, or I want to end the session.

What's inside that skill is essentially this: I have connected to my Claude Code an Obsidian vault that is a daily log of everything that I put through the terminal. All decisions, all open tasks, all learnings, essential notes and atoms, and everything that I'm doing in every session is being stored in Claude Memory. But then you don't have access to that, right?

What session end does is analyze the session, distill what we worked on, what the frame was, whether we accomplished this thing, and what the outstanding items are. Then it moves that over into Obsidian, which is a node-based system that lets me go back to any day and understand what I worked on that day.

You can see this web grow over time, and that's one massive hack to essentially turn every session into a compounding one. The node-based system of Obsidian says, “A week ago you worked on this, and then you hit a wall here. You worked on it again today. You solved it.”

Because the agent can also analyze everything, it creates those interrelationships for you: you worked on Meta Ads with this creative type here in 2 places, and normally you'll forget that, right? And so will the system.

If you have a place for your work and your memory, that, to me, is probably the most valuable resource for people. If you can document every little thing that you're building, tweaking, and refining, that's like how social capital was the big thing. I think what you have on your computer that's saved in your AI output, as we walk into the next couple of years, is the treasure trove for a lot of people.

Do you want to hear something wild? Every single week, we enter the IC—the Investment Committee—for the fund, and we now have all of our calls graded by an AI call grader that measures every single call that we've ever done. We have this, obviously, feedback loop.

It collects every single call, and it simply stack-ranks them out of 10 and ranks them across 5 different variables, like founder-market fit, product-market fit, the deal, and a load of other things—like quality of product, blah, blah, blah. Then we just go down the AI rankings in terms of priority, and we actually fully delegate trust in prioritization to the AI ranking system.

I mean, but if you put the thought into the decision tree of prioritization, then you can trust the system, right?

Mm.

Because it's based on your framing.

It's never been wrong in its ranking. We've only done it for 12 weeks, but 12 out of—

Well, I mean—

—pretty good. Yeah, it's over—

Yeah.

—1,000 companies.

And same thing: I do this. I use a Botlist meeting recorder that records all of my meetings, and at the end of the day, it ingests all of them and says, “What did I promise to do on calls that didn't end up in a to-do software somewhere?” Then it drops all of that. It says, “Did you do any of these?” And then I have my to-do list for tomorrow.

Oh, God, no. I have a girlfriend for that, and that's challenging enough. Final one for you: Which company growth strategy that you're not tied to do you most respect, and what do you learn from it?

I think the craft of product, the hypotheses, the deep experimentation, and building true user-experience excellence—that, to me, is just sexy. If you look at the sheer volume of products that are coming out every single day, very few of them have product excellence, and the people who steer that craft are true product leaders.

I learn so much from them and apply those things to my world. Seeing the way their brains work, the way that they're analyzing the data, and how they understand the customer is really inspiring. A lot of people can learn how to do that.

I'd say this is less across the company and more about the people in companies. But, to be honest, in terms of companies, I'm more impressed by the net-new things that are starting from zero and, even at step 1, are doing pretty well. We thought through the branding, we launched into this marketing thing—that's what I appreciate.

Give me a company.

Again, this is more tooting my horn a bit, but just seeing what the original launch was for Whisper. Seeing that product correlation with marketing—even though I had a part in that, it truly entered the map because of that.

The same thing: Yes, I was a part of the ride there, but the adoption and the fact that a Polish startup that, of course, raised a big Series A became massively seen on the map and was very quickly loved—those are the kinds of things that make me want to get up in the morning.

We did a thing. We put a lot of thought into it, and there was pickup around it. What fascinates me is what all the moving pieces are that go into finally launching the thing—and then it goddamn works.

I think that's the beauty of why I do what I do, because those are some special moments.

Matt, I've so enjoyed this. I can't thank you enough for agreeing to share all of your secrets and for letting me drill down into all the strategies.

No, this was an absolute blast. To the next one. More nuggets next time.