70. Why is Freemium Back in Fashion?

0:00Welcome to The Difference Engine, the show for tech founders, investors and innovators.

0:09Right, Jono, what have we got coming up today? Well, we'll ask what OpenAI's latest purchase tells us about the new reality of category power. And we're building a blueprint for overthrowing an entrenched tech incumbent. Could you really replace PowerPoint? But first, we all know that tech trends move in cycles. But who saw this one coming back?

0:35okay Jonathan the French say plus ça change what do you say I think a rough translation of that lovely French phrase is the more things change the more they stay the same and with decades of experience in the tech industry we've seen many fads come go come back again sometimes turning up as the new tech messiah, even though they were in the bin 10 years before. One of the examples of this is that freemium, remember freemium, is now the route to AI category domination. And it has become obvious that for freemium and product-led growth to work at scale, you clearly need millions of users. If it's just thousands, you just have to go and sell it because the numbers don't actually work.

1:25And as you know, if you've been paying attention, the AI era has validated that because of the rapid scaling of AI users, despite the fact that the market appeared to lose faith in freemium for a while, probably just a decline in numbers of the network effects enabling applications. But that was before AI kicked in. Yeah, I think AI has really kicked it back off. there is a bit of a difference here between B2C, where freemium is the go-to, and B2B, where it's been tried with various amounts of success. But as AI goes across both, freemium isn't just back, it's the default growth and category domination weapon for AI.

2:08And the entire current age of AI, B2B and B2C is built on that. ChatGPT was the fastest growing product ever seen in tech is the clearest proof point we've ever had for freemium category leadership building. Now, is it being caught? Separate conversation, but certainly it was freemium. Yeah, sure. And people used to argue that the freemium model was broken. And now we have massive undeniable evidence that it is not. ChatGPT alone has secured around 800 million weekly active users, converts roughly two to five percent and so enters 2026 at about 20 billion in arr now that is twice the arr it was quoting at mid-year yeah uh it might need all of that money though because that conversion version rate does translate to the bottom line it's not it's not magical it's exactly what's been obvious for years you don't win freemium with better conversion rates that's just too hard you win with orders of magnitudes more users uh size matters a lot and the category presence that that size creates goes with it yeah right and uh i'm just gonna do some maths here so say uh 20 dollars a month 12 to 15 million paying users now that gets you the sliced bread of a few billion in consumer arr the rest api enterprise say 200 a month pro version is the very tasty gem on top But none of that matters without a small cut of hundreds of millions of free users at the top of the funnel.

3:50That's your massive sales pipeline there to convert. But even if they don't convert, inertia keeps them loyal. So they ain't going to anybody else's funnel. And your category crown probably stays firmly stuck on. Yeah, and that is one of the very obvious reasons that AI companies are dominating categories. so far. So they're getting the investment that they need to match it so far. Gemini crossed 650 million active users, or MAUs, mores as they're called, with three plus paid tiers, and there's more coming. Claude hit 30 million MAUs and the 5 billion ARR, albeit mostly enterprise and API customers.

4:33And Perplexity, one of our favorites, hit 30 million monthly average users and over 100 million in annual recurring revenues, still using a freemium model. Mid-journey, early doors, smashed it to half a billion in revenue with around 20 million users, no VC money, and even at a time when Sandhill Roadmob, that's the US VCs, were busting to invest. And they did it with just 100 people. Yeah, that's the part people miss, isn't it? You know, freemium at AI scale doesn't just go revenue, you, it floods the market and locks the category. Once you're the free default, the tens or hundreds of millions, competitors aren't competing with your product, they're competing with your distribution, you know, just like how Microsoft secured the pre-cloud desktop.

5:23And controversially, maybe the reason that TikTok isn't being banned in the States, it's just too big, right? So freemium Freemium actually becomes a bit of a moat and the numbers haven't really changed that much. RevenueCats data shows 1.7 % on average free to paid conversion across subscription apps. So you do need the big numbers and AI didn't change those basic economics. It just changed the sample size. Freemium simply needs mega scale to work. Right. And that also explains why so many founders still get Freemium tragically wrong. They plan for a completely unrealistic 10 % to 15 % conversion and wonder why it fails.

6:05AI has shown us, in reality, freemium needs 10x to 100x more users than people typically expect. And it always takes longer to compound. And it's never going to go more than a few percent conversion. But if so many of your costs are marginal, who cares? Yeah, I do think this is a big contrast. And this is why SaaS is in so much trouble. Well, the classic SaaS examples are Dropbox, which needed 700 million free users to get around 2.5 billion ARR. We previously worked with one of their rivals, Ignite. Same thing. Wix, 134 million users for just 1.8 billion ARR and probably going down. Calendly hit a$3 billion valuation, largely on freemium, although I'd question that valuation.

6:52And Atlassian, who really are the dons here, added freemium and tripled signups. New Relic, controversial company, went freemium again and saw 10x signups. There's a lesson for SaaS companies who've done all the investment to create their products, but maybe need to readjust their pricing. Yeah, and it's interesting what happens if the companies survive. I mean, think about Zoom and Slack. Their entire empire is built on free first, but they, like many others, eventually lent into the enterprise. Box is over$1 billion in revenue, but freemium is now just under 10 % of it. Now, the lesson is if you want to do freemium, you need big backing to build the numbers.

7:32And because, you know, as a category leader that you can then start creaming off the paid for subscriptions, it's all about the width of your funnel. And that's what gives you the choice. Yeah, I think there's a there is a sub theme here, which we need to get back to, which is just the speed at which everything's changing. And the real lesson from what we've seen from AI so far and for B2B founders, not B2C founders, is freemium helps dominate categories first. The enterprise revenue will consolidate and then thereafter maximizes the opportunity. You've got to invest up front. Yeah, and you can move fast in a hot market.

8:10I mean, look at Claude. 70 % to 75 % of revenue is already enterprise and API. That's how to get to billions with only tens of millions of users. and your contract revenue value, that's ACV for acronym geeks, when that changes everything. And of course, investors love it. Yeah. And I think Claude's a very interesting example. They were very deliberate in their choice of ICPs, ideal customers, and very deliberate in trying to create the category to be the dons of code AI. So if you want to do that, having a no sales team myth, the product-led growth nonsense of the sashiers seems to us to be mostly dead.

8:51ChatGPT already has millions of paid business users via enterprise, and Midjourney is doing enterprise partnerships as well these days. So at scale, it's PLG plus sales, not PLG alone. Right. So the takeaway here seems pretty clear. Freemium works. ARPU matters. Scale matters more. And in AI, freemium isn't just a growth tactic. It's how you capture and dominate the category. before anyone else even has a chance. Exactly. It reminds me a little bit of our conversation with Rich Waterworth, who, remember, brought both TikTok and YouTube into Europe on a freemium model. And AI has not invented freemium.

9:30It just proved at historical scale that the premise, basically maths, was right all along. It's just that the numbers put into these business models were often wrong. Indeed they were.

9:48So I wanted to talk about a sort of new reality that's coming in in terms of category power. I mean, did you see this one? OpenAI just paid about$100 million to acquire a four-person startup. The company's called Torch. That's roughly, by my maths, that's$100 million divided by four. That's$25 million per employee. Now, on the surface, that sounds insane, but it actually tells us a lot about where tech category is heading, and particularly about category concentration, which we discussed, I believe, in episode 68. Yeah, well, like 100 million, I'd missed it because it's not a big number for OpenAI.

10:27It's not a big number for Google. It's chump change. The 25 million per employee, pretty impressive. But it only looks crazy if you think that what they're buying was people. like many of the aqua hires that we've seen recently, including, one could argue, OpenAI to Microsoft. But they weren't. What was being bought by OpenAI here was strategic control in an emerging category, healthcare AI infrastructure. Hot, hot, hot. Yeah, and it's much more than strategic. It's a control. It's a point. It's a point in an emerging category. And as you say, that's healthcare AI infrastructure. So they certainly weren't a flashy chat bot or something similar.

11:07They were building something much more fundamental. And that is a system that lets AI view a patient's full medical history in one place. That's not a healthcare records app. That's a data layer. Well, yeah. And data layers are where categories consolidate. I have been speaking to a bunch of friends who are medics, GPs, about this for a while. and they seem to overestimate the moat that they've got. And especially in regulated industries like healthcare, that's easy to understand. Now, OpenAI could have built all of this tech themselves. It would have taken years, not just technically, but to my point about entrenched players, politically and operationally.

11:49Healthcare data is messy, fragmented, deeply contextual, very, very private. And that's where the healthcare regulation and compliance shapes business models. Yeah, yeah, which is why this acquisition lines up perfectly with OpenAI's move into ChatGPT Health. I mean, if you look at OpenAI's own report, one in four users submits a health-related prompt every week. That's 40 million people are doing that daily. Yeah, that's statistics critical. And, you know, I think 40 million is conservative. We've all done it. I've done it. You've done it. and Elon Musk did it and recommends you put your blood work in there.

12:30He says Grok almost never hallucinates. It shows how user behavior is already there. It's not moving there. It's already there. People trust AI to interpret symptoms and lab results. The natural next step isn't more chat, it's agency. AI can reason over your medical history, just as you would debate it with a, you know, the gatekeeper, who's normally the medical professional, and not just answer the isolated questions and hedge it. Again, this is all about context and context over content. Yeah, exactly. And this is also where category concentration starts to kick in. Healthcare AI is moving from, you know, possibly helpful chatbot, let's debate that later, to an effective trusted agent.

13:16Whoever owns the agent needs to own the history. And that means owning the data substrate. Yeah, of course. Which brings us back to why on earth OpenAI paid what they paid. Well, probably three things, right? They bought some tech, they bought some technology, a medical data engine that would have taken years to replicate, well, months maybe. But still, you know, secondly, timing, a head start on healthcare AI as the world moves from conversational, well, it moves from having a chat with your doctor to conversational online interfaces to now agent-based systems who can take action and make recommendations.

13:55And third, they bought, and this is the acqui-hire bit, real-world healthcare experience and, you know, scars and all for that. Oh, see what you did there. But I think it also fits into a broader acquisition pattern. If you look at their recent moves they paid, this is where you get nice big money, Paul. Yeah, this is real money, this. Real proper money, right. So from memory, about 1.1 billion for Statsig. That's really about experimentation and telemetry. An even bigger number, 6.5 billion for I.O., which is essentially hardware and infrastructure. And then this paltry, 100 million for Torch, which is all about the healthcare data.

14:38Could they not have negotiated harder? They might have done. But the beauty of this is for OpenAI, ties all the healthcare data together. Yeah, and lighting all of those up, the strategy punches you in the face. I'm not sure that's very healthy for health professionals, but OpenAI isn't just building models anymore. They've moved right on. Foundational models are becoming table stakes. The real competitive advantage is vertical control. Infrastructure plus experience plus workflow ownership, the entire category value stack. Indeed. And this is classic category concentration behavior. Early on, potential categories are fragmented.

15:19You have lots of startups, lots of surface level differentiation. Then once usage explodes, and it has done very, very quickly with our lens, power moves towards whoever controls the deepest levels of the stack. Yeah, you see this a lot. So we're talking about healthcare in this instance, but you can see it very much in agentic commerce. and in this phase of the categories as it's forming the key questions are not who has the best model anymore that's all that's all gone they're much more structural and strategic who owns the agent who owns the data history who owns the workflow and who do people actually trust yeah and why are we talking about this well well essentially healthcare just happens to make this visible faster because the stakes are so high because we all care about our own health Trust, compliance, and continuity matter more than novelty.

16:10You can't swap vendors easily once the AI becomes embedded in clinical decision-making. You know, what is that? That's vendor nirvana. And that's why this deal really is not about a four-person startup or a paltry amount of money. It's about open AI planting a flag even in the ground and saying, we've got intentions to own the healthcare AI category, end to end. Yeah. And what can we learn from this? There's a broader lesson for any tech company watching this behavior. Categories currently are concentrating around control points. And by control points, we mean data, agents, workflows, not features.

16:49Not features. We go on and on and on about this. Category domination is not about adding new features. It's about strategic control. If you don't own one of these data agents workflows, you're likely to be a plugin, not a platform, and probably pretty soon dead in the water. Or an acquisition or aqua hire target. Yeah, yeah, if you've been schmoozing the big people in tech. And we did say this in the last episode, if you have been flirting with category owners from the beginning, you might have a chance. but probably not at 25 million per employee, unless you're sitting on a category defining choke point, which Torch were.

17:33That's the new reality of category power. It's moving fast. And some people are making the moves that are going to define their categories.

17:48So Paul, what are we going to look at next? We want to ask ourselves, how do you beat a 40-year-old tech incumbent at its own game? Well, that is indeed an interesting question, a category conundrum. So in order to explain this, let's go a little bit business school here. Let's do a case study. How would you displace good old PowerPoint? Where do you start? There's a whole graveyard of presentation tools that tried. I don't want to list them, there are too many. I just wish the upper one would stop coming up on my screen. Same slide model, slightly nicer templates, essentially the same product.

18:25Yeah, absolutely no category into there whatsoever. Exactly. Everything that's come along, it seems, in the last 40 years seems to be based at better, and we know better is not what you do in category, slide-based presentations. um but you know let's be real that really doesn't matter when the incumbent is pre-installed on basically every desk in every organization and frankly these days every home on earth yeah um and as our friends in the ai world are finding out distribution matters a lot and tech distribution often beats out product polish uh in fact pretty much every time that's um that's been microsoft's business model from the beginning and it's worked pretty well for them.

19:14Okay so the real question in all of this is how do you actually compete with something like PowerPoint or any other entrenched mega app? Let me guess you stop trying to beat it out on features and forget all of that better nonsense. Absolutely you think category so what you do is you start by looking at the business model, not the tech. So not tactical feature gaps, strategic structural advantages that you could create. Like what? Would you care to elucidate? Well, think about PowerPoint. Clearly the biggest real world advantage is distribution through the office bundle. So you get PowerPoint whether you want it or not.

19:56That's how an originally very second rate teams took users off the massively better zoom post-covid and why people are using copilot now in preference to other ai assistant products it's just there but in the case of powerpoint it also has the.pptx file format that everybody has to support but if you think about it the biggest advantage has been built over time it's an invisible moat it's what we call the mental model especially in large corporates the slide has become the basic unit of work of structuring ideas and plans particularly when people equate meetings with actual productive graft and creating a deck seems to be the response to any challenge i think that's really interesting um certainly mckinsey has built a business of making slides.

20:54And it certainly is true that people regarded making slides as somewhat of a hard task, whereas AI can just knock them out for fun. But Microsoft's hegemony can then be also a weakness, because it means that should anyone cannibalize this advantage, which of course is the innovator's dilemma, so they wouldn't want to move off it necessarily themselves until they needed to probably the best a competitor could do uh is take this product to a new probably a ai powered new level uh but what they should do really is change the game destroy the culture of slide presentations which i believe is happening and move things on after all powerpoint is simply a better version of slide carousels uh kodak that went well overhead projector foils oh my god remember 3m at least they're still in business making sticky tape and quite a few other things to be fair.

21:48But decades ago, Microsoft created this new category by delivering new utility and linking it with ever-growing distribution via all their global partners on a new PC platform and Microsoft's bridgehead operating system. Exactly. These are just three of the things that made them strong and sticky. So once you absorb that information, what next well you have a really hard look at them as a business and you look for their untouchable you know the inertia ridden constraints in the business so the way to think about that is is to ask yourself if they try to copy my imaginary product what would they have to give up to do it that they're the really really really big pants decisions yeah and i think they're not alone in in some of the issues they face here they probably have to abandon their file format format um uh that's that's a large thing to do um ask adobe or the really big banger change the actual basic premise of productivity and what a key productivity tool would comprise uh in or outside of the office um template right so the big big pants decision would be switching away from creating slides as a core productivity approach so let's let that sink in for a moment completely move away from creating slides as a way to communicate information in corporates and everywhere else, right?

23:17That's a paradigm shift that means facing the prospect of retraining hundreds of millions of users and fundamentally changing global corporate culture. Is that big enough? Well, that's massive. But if you think they could do all of that, what next? Well, if If you genuinely think, oh, they could do all that, clearly your idea is too close. You're still playing their game following their paradigm. Right. So you're saying the move has to change the game entirely. Of course. And that's where tech shifts come in. Cloud, mobile, AI. Each shift gave you the opportunity to change the basic way things are done, which is the essence of category design.

24:03And it works a lot of the time, but often that tech shift alone isn't enough to build the category, is it? Yeah, exactly. You've got to change customer expectations. I think one of my examples in this, a close example, could be Figma. Now, Figma Design, according to Figma, is now the primary tool for people to create, share and test designs for websites, mobile apps, and other digital products and experience. the thing about figma it was designed from the beginning to be browser-based and multiplayer by default and there's sort of other examples of this do you think paul well yeah but on figma though if i recall correctly they um it was the brits that turned down adobe's um merger with them and they got a nice billion dollar payoff because they put a penalty into the into the deal so um Figma actually benefited from other people not being able to move the category.

25:03You could say, why the hell didn't Adobe build its own Figma? And they're not the only player coming at Microsoft and Adobe in this space. Now you've also got Canva. So there's a lot going on. Speaking of Adobe, Photoshop, which was probably still is massively the incumbent, couldn't follow without blowing up its file format. and the business model that Figma was looking to usurp. And you can bet there's lots of people out there thinking about how to use AI in the same way on Figma to change the game again. I've seen it. And that's the inevitable pattern. And the reality of making such a move is you choose constraints that might make you perhaps a little bit worse at the old tasks, which are going to go away anyway because you're going to redefine them, but dramatically.

25:52and this is the only time I'd ever use better, better at the new ones, which of course are things you have defined. Yeah, so they're new, so they're different. And so how might this play out if there really is that PowerPoint killer out there? Okay, so you won't be surprised that in this day and age it would require using AI, of course, but not to mindlessly auto-generate slides. That's just PowerPoint slop edition. Instead, use AI to help structure what you might think of as text-based narrative, prompts for natural presentation direct to the audience. Yeah, I like that. So the idea would be that slides, where people generally talk to them, are no longer the basic route for presentations.

26:42Makes me think video could be. And then what you need is free-flowing, structured, prompted narratives. Notice I said free-flowing, structured, prompted narratives. didn't talk about the delivery mechanism. It's more like that old trick of using cards with the ability to flash up AI-generated illustrations from within the app. That's a different approach. But remember, this is not an AI play. This is about restructuring behavior. AI should only accelerate the basics, not replace them. Ground-up thinking is what gets you to true differentiation. Right, right. A much better format for a mobile-led world, I think.

27:19you know if you think about it this is story building that works whether for reading listening or presenting rather than word for reading and powerpoint for presenting and and some yet to be standardized podcast format for listening is much more suited to the formats of what is now an inherently mobile world yeah a totally different mental model not a platform model and after all PowerPoint is still firmly rooted in the PC era and an 80s way of thinking. Yeah, yeah. And the beauty of this, if you have the backing and if you have the bravery, is that PowerPoint couldn't copy this model without disrupting itself.

28:01That, I think, is the very essence of new category potential. Wonder how long that thing will take to die. I'm sure we'll be talking about it 50 years from now. So if you're competing with an entrenched incumbent, what have we learned here? What's the one question that people who want to design new categories should ask themselves? Right. So in what way could your product be so different that the incumbent can't follow without disrupting themselves? Oh, I like it. And if your answer right now is better design or we use AI. Yeah, then you're still focused on being better, not different. And the real different is being the only one.

28:38The person that got there first with their thinking, the one that really forces others to follow.