Canva cofounder and COO Cliff Obrecht in conversation with John Collison

Transcript

Canva is an amazing story. I think everyone here is familiar with us. Do you want to give us just the quick update on where you guys are today? And in particular, we're going to spend a lot of time talking about AI, but what you guys are in the middle of at the moment. Yeah, so we were born, I guess, as a company in 2013. Started the company with my wife, who's my boss. It's always fun. That's always the case, whether you realize it or not. You're in a similar situation. Exactly. Building a business with someone you've got a close relationship with. But no, we're very exciting, 13-year run. We're in the middle of the biggest transition of our whole lives as a company, taking Canva from a creative platform in the SaaS world to really being the combination of a creative platform, productivity platform, and intelligence platform in the AI world. So that comes with a lot of challenges and excitement, and it's super fun. How did Canva get to scale? What's the backstory of how you guys got massive? I think all companies get massive by having the best product in the world that's ruthlessly focused on solving customer problems. And then part of the special source was really our freemium model. We didn't try to over-monetize too early. When we first launched Canva, there's a lot of people over there. Hey, everyone. They snuck in. Yeah, you could only buy images for a dollar. So it was very much a free platform with $1 purchases. And then after a while, we realized that we should add a subscription model to this. And that's when it really started to take off. But we really always respected that free user base. And we really saw that as our word-of-mouth marketing channel. And it's that channel that's really always growing the business. Did you do a lot to engineer the virality, or did it come organically? We did all the basic things. Obviously, really building in seamless sharing. We never wanted to watermark the product and deprecate the experience of our free users in order to get that brand awareness more externally. So we actually thought the whole strategy was being very free, being very generous, giving people what they wanted. And we saw a pattern over time and over the many years customers use Canva that more and more of them would upgrade, particularly as we added more premium features. The more value they got, the more acceptance they had to pay. How do you think about where you should... So most of your users are free users. And then the business bills are paid by the paid users. And so you're always drawing a graduation point, right? Where it's like, it's time to leave the nest, free user. Yeah, exactly. Flap your rings and become a paying Canva subscriber. Yeah. So when are they ready to leave the nest? We were very happy with our free-to-paid monetization rates forever. And they've always been very, very healthy. We also have a lot of education users that are essentially perpetually free, other than when they upgrade coming into college. But AI has totally changed that because the cost to serve a user has gone from cents per month to now many, many more cents per month. And so that's really changed the dynamics of the cost to serve and also the dynamics of the business. It's been an evolution. And so concretely, that means just for anyone, we know this is Stripe. I'm sure a lot of you in the audience are feeling this. Just building any kind of non-deterministic AI product, that's a totally different kind of product than the free products they're replacing. And so they come with a different monetization path, and you can't just naturally import whatever the graduation path was from free to paid in the old product. That has to be different in the AI-powered product. Correct. Correct, yeah. So the cost to create a design historically, it cost us essentially nothing. But now if you're to use a full agentic loop and you're to do web research, you're to call a brand, you're to create a whole campaign being many different designs that include images, graphs, all types of things, and then you're to deploy it, then you want to understand how that content is actually performing once it's been deployed. Intelligence needs to be run behind all of these things, and many different systems need to be connected, and all those tokens cost money. And so the cost to create a design before, which was essentially nothing, has now turned into cents. And that, at hundreds of millions of users scale, really has a fundamental effect on the business that we've had to adapt to. And I feel like in the discussion about AI, we get asked about the effect of AI on payments, which we're happy to talk at length about. But I feel like as people talk about the effects of AI in the economy generally, design is often one of the examples that people use. Where it's like the robots-eating-the-jobs example. Obviously, it's much more complex than that because also you're empowering people to do their own design where they don't need a designer anymore. You can just have much more design happening. What is your grand unified theory for design in the age of AI? I think it's very similar to when we launched Canva. Before Canva, you only had the Adobe products that are very difficult to use. They were only accessible by cost and by the actual need to learn these tools to less than one percent of the internet-using population. We introduced Canva, and there was a lot of sentiment. Oh, Canva is destroying design jobs. Ultimately, we created hundreds of millions of more designers and lots and lots of opportunities. And then we empowered those professional designers to build templates, submit them on our platform. And we are now one of the only platforms, if not the only platform, where you can generate with AI and reference templates created by real humans that we then pay those real humans for those input images rather than just training on the internet. So we really believe human in the loop is critical, particularly when it comes to creativity. And we really believe in paying out creators for the beautiful content and the design trends that they're setting that people can leverage with AI. So just maybe to go deeper on this, because I think, again, it's an interesting micro story for what's going on more broadly. I agree with you that it's clearly the effect that Canva is making much more design possible. And so for a given job, previously you needed to go get a designer to do it. If you're a restaurant looking for a menu design or something, previously you needed to get a designer to do it. Now you do not. It's also the case that the number of design jobs has just gone up over time. And so for those designers, how is it, if you were previously designing restaurant menus, what are you doing now? How has the design job changed? Are you now making Canva templates, or just how has, for people whose full-time job as a designer, that work changed? Yeah, especially in large organizations, designers are really becoming the brand architects and the design architects. And so similar to how designers in organizations used to create templates, now they can create mood boards, styles, brands for all their different sub-assets within an organization. Every campaign has its own brand. And then they can empower the rest of the organization to either create with those templates or leverage those base bits of creativity, feed that into AI to create things like that. And then they can create things for whatever they want, but within that same brand framework. Despite this obviously being true and the numbers being what they are, you do still get this persistent robots-are-reaching-all-the-jobs, design-jobs-under-threat-from-AI. The headlines still are what they are. Does that bother you? Do you go out trying to correct the record? How do you feel about the fact that people want to write a certain story? And again, I presume in particular over Canva's history, a lot of people have been writing the story of Canva replacing designers. It actually hasn't been too heavy recently. It was more so when we launched, but yeah, there certainly is that general sentiment. But just like when we launched Canva, we really think it does empower professional creatives to do more. And actually with AI, having on-brand content is actually more important than ever. There is so much AI slop out there that you really need custodians of the brand within organizations. And so I still think the role of designers, brand designers, and anyone in that space really has a lot of value to add and are increasingly becoming more and more valuable. So I feel like a lot of AI tools have a certain aesthetic. People talk about with the LLMs, they have this very distinctive writing style, right? You've hit on a great point. Here's the quietly note-bearing part. Just like, it's so distinctive. And similarly with visual styles, mid-journey, when the early image models had this particular aesthetic that it went for, But just like when we launched Canva, we really think it does empower professional creatives to do more. And actually, with AI, having on-brand content is actually more important than ever. There is so much AI slop out there that you really need custodians of the brand within organizations. And so I still think the role of designers, brand designers, and anyone in that space really has a lot of value to add and are increasingly becoming more and more valuable. So I feel like a lot of AI tools have a certain aesthetic. People talk about the LLMs. They have this very distinctive writing style, right? You've hit on a great point. Here's the quietly note-bearing part. It's so distinctive. And similarly with visual styles, Midjourney, when the early image models had this particular aesthetic that it went for, I noticed that whenever AI builds a website, you've probably noticed there's a default CSS that the AI-built websites really like going for. What do you think of as the visual style or the design values that you're trying to inculcate in Canva's products? Well, really, what we want is to give optionality, and these models can do amazing things, but they can do even better things if you give them guidance. So if you're a brand, how do you set up the brand architecture to give the model that guidance so it adheres to what you want, but also expresses creativity? And the human can also express their creativity by adding more prompts, and they can be visual prompts or written prompts. And the combination of that brand architecture with the human interaction can lead to better outcomes that you can guide, or you can let it be massively creative. And it's about the inputs to what is generated that ultimately guides it to being the generic brand slop or things that are broader scale. But can't it not always be input? If you're buying a house and you're touring different houses, at a certain point you maybe say, oh, this is the one, I really like this. But if you'd been asked to specify the house that you want, you wouldn't necessarily have come up with all those design cues. And so people may be, to some extent, picking and won't necessarily be able to prompt exactly what it is they want. That's exactly what we do. We’ve got over 165 million templates, images, and graphic assets in the library. And we really don't think the prompt box is the right vector for creating visual content. People don't know they want a psychedelic retro '80s-themed birthday party invitation. But when they see it, they're like, oh geez, I really like that. And they can use that as the input to then add their own content to and make it their own. I see. Yeah. So you are. Just template first. Exposing them to design space. Yeah, yeah, yeah. I really like that. And I guess we jump into. And we pay the creators. So if you spot that psychedelic template, et cetera, et cetera, then if you input that into your creation, then we also give the attribution and payment to that creator. Are those creators people who do this for fun? Is this people where it's their full-time job? Do they have some other business? And they also put the stuff on Canva. How does the creator side of things work? It's all of the above. We have full-time creators. We have people doing it in their spare time. We have graphic designers that are adding their content that they don't use for a customer to it. So yeah, all of the above. We really help them as well, guide them. Hey, there's a gap in this market for this particular doc type, and put a call out. And then they fill those call-outs based on our search queries. Yes. Do youngsters do it? I remember when I was growing up, if you're an enterprising youngster designing websites and building websites for the olds who didn't really know who the editor, but you could work with a local business owner and help them build a website, and they would pay you a thousand euros for it. To a teenager, that is an infinite amount of money. And so do you also see this as how young people are getting into design and getting into creative? Well, yeah, Canva is a critical skill on Upwork, Fiverr, LinkedIn. And so there's multiple things. You can be a youngster and contribute to our library and get paid. Or you can just do jobs on Canva. And most people, a lot of people now, require their content to be delivered in Canva format because it's editable. They don't want an Illustrator file. They don't want just the JPEG. They want a Canva format because they want to then reuse that design within their small business. Do you find businesses wanting a deliverable, essentially a design language that the business can then work with? Sometimes. That's what we call a brand kit and brand templates. Or it could just be, hey, I have a campaign. I want 20 different Facebook ads or Instagram ads advertising my new shoes. And we jumped into all the AI stuff, but we have to back up a bit because you are broadly known as Canva, the design tool. You are now becoming Canva, the AI lab developing image models. Do you want to just talk about that transition? Yeah, I mean, when we launched, or tried to launch, our product, Canva Create, in April, we launched it with a lot of frontier intelligence. So third-party image models, third-party LLMs. And they were amazing at what they did, but they're also very expensive. And if you're to deploy AI to hundreds of millions of people, you need to really insource a lot of that model development, particularly core competencies around image and design generation. And you also need to own a lot of the genetic loops and also manage the inference as well to get costs down. And so since we launched in April, we actually had to really slow down that rollout whilst we get those costs in check, but whilst also hitting quality and latency goals as well. And it's the combination of quality, cost, and latency that is the holy trinity. And it's only when you get those three things right, you can actually afford to give a really great, generous free product to our free users, as well as a generous product to our paying customers. Okay, so you think it's just not possible to build the kind of Canva product that you want to build when you are shelling out to someone else's image model? You can build it if we had 10,000 customers that were all paying us the economic. But at Canva scale. At our scale, when we've got 200-plus million free users and, say, 30-plus million paying users, the paying users need to subsidize the free users as well as get a generous product themselves. And so to serve 200-plus million users a very generous free AI product comes with a huge cost. And so we needed to get that cost down, and we've managed to do that by about 90%, which means we can really scale it up now. There are lots of people who do lots of AI who choose not to train their own frontier models. And if you rewind to the Canva leadership team offsite in, I don't know, whatever year it was you guys were making that decision, was it a decision where we could go the Amazon route, where Amazon does lots of AI with all their customers, but they choose to help customers use others' AI models as opposed to train their own? Was that a path you guys seriously considered, or was it just preordained by the business logic that you're describing? We actually got into AI over six years ago. One of the most loved features in Canva was the background remover. So you could essentially remove a background from anything. And so we partnered with a great company called Kaleido in Austria, of all places, where my grandfather was from. And that product upgraded so many users, we decided to acquire that company, and that company had deep AI. It is how I ended up paying for Canva now that you see it. Yeah. I need to remove the background from an image. Templates and background remover used to be our number one drivers of upgrades. And so yes, that's exactly right. So that was a visual AI lab that essentially had to do the deep research in order to figure out how to see an image and then decompose the different layers. And so that kind of inadvertently got us into the AI space. We had a research team that constantly worked on that. And then once we realized that image generation and design generation was such a core competency, we acquired a local company called Leonardo, who had also built their foundational model. And really, you can take the scattergun approach of running evals, what's the best cost, quality, latency. But if you really want to own those things and optimize those two, three things, you need to own those models yourselves. It is how I ended up paying for Canva now that you see it. Yeah. I need to remove the background from an image. Templates and background remover used to be our number one drivers of upgrades. And so, yes, that's exactly right. So that was a visual AI lab that essentially had to do the deep research in order to figure out how to see an image and then decompose the different layers. And so that inadvertently got us into the AI space. We had a research team that constantly worked on that. And then once we realized that image generation and design generation was such a core competency, we acquired a local company called Leonardo, who had also built their foundational model. And it's really, you can take the scattergun approach of running evals, what's the best cost, quality, latency. But if you really want to own those two things and optimize those two, three things, you need to own those models yourselves. Because you can probably get the quality, but the cost and latency you're going to lose out on. And only by building your own research team and training your own models, particularly for image and design. LLMs is a bit more fungible. But for image and design, we really needed to be in control of our own destiny there. I want to go back to the background image removal because I think there's a deep point there about iterating on product market fit as a scaling up company, where in the beginning you have to make something that works for some set of customers. But I feel like the 1x to 10x to 100x to 1000x often comes from getting the loop right of seeing behaviors that work well on the platform and then extending the product based on them. My favorite example of this is Twitter, where all of the Twitter features you can think of were invented by the community and then regularized by Twitter. So the notion of hashtags, people started doing it in their tweets, and they're like, I guess we should make them clickable and searchable. Or replies or tweet threads or putting images in tweets or all this stuff. People just started doing it, and previously they would just have an ugly-looking URL in the image or in the tweet. I guess we should actually put the image in the tweet. Anyway, so I think there's this mechanic for scaling up companies where you want to see how people are using the product and then really go whole hog on supporting it and extending it. And so I think most growth loops in companies, companies often don't have infinite growth loops, and the majority of their growth comes from three or four key things that work really well for them. So for example, at Canva, we have a huge template library, and we thought, well, let's expose these templates to SEO. Let's do it for, you always start small, right? Holy shit, this is number one on Google for business card templates. Then it's like, okay, let's do the SEO research. Let's figure out everything everyone's looking for, which then informs the strategy to actually create that content, which people ultimately want in the platform. And then you expose that to the public and then you, well, we ranked number one in all the categories for the longest period of time. And then that becomes essentially, but that was 60% of our growth for a while. And then it's, okay, people are designing. So that being business card creation? Oh, no, no, no, the broad spectrum of all templates. Okay, templates. Yeah, yeah. And then you go down to subcategories. You can go like three-year-old birthday something in, I don't know what I mean? Psychedelic birthday. Yeah, yeah, yeah. But then you realize people are using your product, they're collaborating and sharing. So how do you make that sharing collaboration feature really seamless so you can connect people, connect teams, and then that becomes another great flywheel, et cetera, et cetera. And I feel like one of the challenges to, yeah, I totally agree in these growth loops, and we have had similar growth loops at Stripe over time, where an early version of this at Stripe was the sharing economy, where we noticed that people cared about scaled payout logic more than pay-ins because they were building apps like Lyft and Instacart and DoorDash and all these things. And so that very quickly led us to building for marketplace functionality, and that through line has continued to this day. Okay, I feel like one of the challenges with these growth loops is companies are organized, sorry, this will be a bit complex, but I'm actually very curious. Companies are organized functionally. And so when you say people are doing templates or people want to remove image backgrounds, there's part of that that's product, where you need the image background removal to be good. There's part of that that's SEO, where you want, how do I remove the background to an image? You want to rank first on that, and that could be a huge source of traffic. There's content to drive the SEO. There's a load of other things that go into it. And so how do you actually make these growth loops work when they live in seven different areas? There's different parts of the organization. So, as you know, this is the complexity of running a large organization. Yes, I want to know. Asking for a friend. It's a trade-off between deep alignment and self-contained teams with everything they need in order to deliver, which is always the fastest way to deliver a great product to your customers, compared to a matrixed product. And so now we have a matrixed organization. And so content is, so you have the verticals, which for us are the product surface areas. And we want as few product surface areas as possible because having too many product surfaces creates sprawl, and it creates even more alignment challenges between the horizontals and the verticals. And then you have the horizontals, which may be content and personalization. And so content needs to show up in landing pages. It needs to show up in our marketplace. It needs to show up in the homepage. It needs to also be deeply guided by personalization. Once it hits the homepage, you are a small business in Mexico. You need to see vastly different content than a large organization in Japan. And so it's how you have the matrix work together, how you define priorities, because that content team has a million different teams asking for its services. And you need to be really clear at defining the goals at the top of the organizational structure in order to define how those goals flow down and what to prioritize within the different teams and all parts of the org. Does that answer your question? It does. Okay, so you've got this matrix, functions on one axis, and then how granular do the product areas, or I'm not sure what you called them, how granular do you get on those other sides of the matrix? It sounds like pretty broad. On the product surfaces? Product surfaces. Well, you need to architect your product, and so it all is essentially an integrated platform that's serviceable. So that's why I was talking about the intersection of personalization and content, and then if you look at the homepage, the homepage isn't custom configured for every user. It's configured just like Netflix with a bunch of configurable rows, and based on the attributes we have on the customer, we feed content from the content service. So if you're that small business in Mexico, we go call the Mexican small business content and then deliver it within the platform. So you need to architect it, and so it's not just whack-a-mole in trying to fix these things per market, per user intent. It's really a system that's led by machine learning and feed them and all that stuff. What products did Canva displace as we talk about these use cases? You mentioned the Adobe suite is one, but in general, what were people using before that Canva was simplificating? No one really had the ability to create designs like PowerPoint, Paint. The amount of people that were designing in PowerPoint and Paint, that was the status quo. Or they would figure out Photoshop, they'd go pirate Photoshop, and they'd figure out how to get their job done, for example. Microsoft Word was also a big thing, pushing that to its absolute limits. I mean, we started Canva with a school yearbook company. I used to be a school teacher, and Mel was at university teaching design, teaching the Adobe suite, and realized this is really tough, but we're in Perth, Western Australia. We had no money, we were, how do we make money out of this thing? I was like, well, school yearbooks are something where some school teacher gets allocated this job to do, they hate it, and they have to create a 200-page, essentially, professional magazine with no design skills. And so we took those two ideas, we merged them together, created a company called Fusion Books, which solved the yearbook problem. Or they would figure out Photoshop, they'd go pirate Photoshop, and they'd figure out how to get their job done, for example. Microsoft Word was also a big thing, pushing that to its absolute limits. We started Canva with a school yearbook company. I used to be a school teacher, and Mel was at university teaching design, teaching the Adobe suite, and realized this is really tough, but we're in Perth, Western Australia. We had no money. We were, how do we make money out of this thing? I was like, well, school yearbooks are something where some school teacher gets allocated this job to do, they hate it, and they have to create a 200-page, essentially, professional magazine with no design skills. And so we took those two ideas, we merged them together, created a company called Fusion Books, which solved the yearbook problem. But then all these schools started saying, can we use it for newsletters, can we use it for email design, all these different things. And we were like, okay, there's something beyond school yearbooks here. Which was the intention from the start. People wanted to design things more than school yearbooks. Yeah, yeah, believe it or not. And so that's when, so we self-funded that first business. And we always had the idea, well, Mel always had the idea for Canva, but we needed to deploy it to a market where we could make money and be profitable immediately. But five years into that journey, we were like, holy shit, no one's done this yet. Let's go take on the big league. That's when we went to Silicon Valley for the first time, got rejected over 100 times because we were two people that didn't know what a startup was from Perth, Western Australia, going to pitch these fancy investors in Silicon Valley. And we got rejected all these times until there was one pivotal turning point where we said we've got this other business. And this one curious investor, and this was on us, right? We should have led with this. The curious investor said, tell me more about your other business. And we're like, well, we'll show you. We've built this software. We're used by, I think it was a thousand schools at the time. They've created over a million pages per year. It's making over a million dollars. We're profitable. We've got a team of 20. We run our own printers and stuff like that as well. And he stops in his tracks and he's like, holy shit, you're doing better than 90% of my portfolio. And then he's like, you've got to jazz it up a bit. You've got to be more American. And so then the next meeting we went into, we're like, yeah, we got this startup. It's doing a million bucks. We build a product, got a team. And they were like, yeah, dog. So did you do the pitch in that accent? Yeah. That probably helped a lot. I don't even know what accent that is. American. Yeah. And you waved your cowboy hat around. But, okay, very good. I forgot what that question was. But yeah. As do I. So many places we could go from that. And sorry, this was, what stage were you when you first raised venture money? Well, we launched in 2013. So that must mean 2012. Okay. Yeah. And did you ever consider moving to US? Yeah, we did deeply. All our investors wanted to. That's why I asked. Yeah. One of the many reasons we got rejected, that was the days where investors used to literally say we need to ride our bike. And we heard this. I know it's been publicized, but we only invest in companies we can ride our bike to. Fuck, you hope you get a good bike. We're in Perth, Western Australia. You're going to have to have a bike with wings. And so, yeah, so we thought deeply about it. But then we were literally on a knife's edge. But the Australian government had the R&D tax concession and they had a Commercialisation Australia grant. So we raised $1.5 million on an $8 million valuation, which seemed like tons at the time. Now, I don't think a startup raises under 50 mil value just off a paper pitch deck, which is like the glory days. I know we're both old. This is how you get into it. Uphill both ways to Sand Hill Road. Yeah. But yeah, I forget what I was saying. Okay, but so you're saying it was very contingent. It was a real knife edge. Oh yeah, knife edge. Yeah, yeah. But this is, so we raised $1.5 million, but the Australian government gave us another million dollars. And so that essentially doubled our runway. And we're like, well, that's the clincher to stay in Australia. So we moved from Perth to Sydney, the big smoke. And yeah. And then we just love Sydney. We think there's great talent here. We're a bigger fish in a small pond, which is a double-edged sword. We get written about in the press a lot. Where we wouldn't in the US. But no, the talent here is incredible. And we really saw it as a strategic advantage. Because we're a products company. We don't like being distracted by all the hubbub in the US, the Twittersphere and stuff. We like to stay out of that and just keep building for our customers, which we think is an advantage. No investor is trying to ride their bikes to your office. We're just getting folks in building. How is... Sorry, what's the big smoke reference? There's some Australian rivalry going on. Sydney. But it's lovely. It's not smoky at all. You've never heard the big smoke? Like moving to the big smoke? It's like moving to the city? Okay. Is it just me? That's a common reference. I need to... Must be an Australianism. I can imagine London being the big smoke, but Sydney's... Well, from Perth. I'm from the most isolated city in Australia. Yeah, yeah, yeah. No, you can't ride your bike there. Yeah. So, in what way is Canva now an Australian company? Like if you think about how the organization is and works and just, yeah, how it functions and how that's different from when you... Had you built in the US? Or you can reject the premise of the question and say people get too much into these... No, no, no. We're a global team now. So our largest market's the US. Our second largest market's Brazil. We're very much a global company with a global team. For example, the majority of our enterprise sales comes from the US, followed by Europe. And so to serve the customers, we've had to get close to the customers, put boots on the ground. And so we've just evolved the business as it's needed to be evolved. We've got R&D hubs with acquired companies in different markets. And so, yeah, they've all added to the fabric of Canva. And we really try and centralize all around a few core hubs. And I'm very much coming back to the belief that the best work is done in person, which we're not enforcing in-person work. But it's just so much better. I'm working in the shed. We've got this tiny little office, no, it's literally a shed with a roller door, with the AI team. And just locking in with the team, being able to yell across desks again and not be on Zoom all day is definitely the best way to work. Yeah. And there's also a thing of... So you want to put teams together. Exactly. Yeah. But, okay, but it sounds like you think you would describe Canva more as a global company than a culturally Australian company. Global company with an Australian heart. Hmm. As you think about how you build products, I find it right that sometimes there's deprogramming of people that you need to do when they come in the door. That they've picked up bad habits from other places. My classic one is I think people really overcomplicate and overengineer the product development loop. It's a bit like the, all the early British explorations of the Arctic and Greenland and the North Pole and everything. They first tried loading up the great galleon with enough supplies for three months and the fine china and the library of scientific books they would need, and a hundred men and everything like that. And they'd go off on this expedition. They'd all die. And then they figured out that it should be like two guys and a dog sled and a kayak. And they crossed Greenland and got to the North Pole and everything like that. And I think there's something similar in product development where people want to overengineer it a bit and do too much user UX research and a million documents and things like that. It's a bit like all the early British explorations of the Arctic and Greenland and the North Pole and everything. They first tried loading up the Great Galleon with enough supplies for three months, the fine china, the library of scientific books they would need, and a hundred men and everything like that. And they'd go off on this expedition. They'd all die. And then they figured out that it should be two guys and a dog sled and a kayak. And they crossed Greenland and got to the North Pole and everything like that. And I think there's something similar in product development where people want to overengineer it a bit and do too much user UX research and a million documents and things like that. And you've got to get something in it. You've got to build a yearbook creator. You've got to get something into the hands of users that they can like or not like and then optimize from there. Anyway, that's my example of some of the deprogramming we try to do of people as they come into Stripe, where you're trying to break the bad habits and nip them in the bud and steer them a bit more in the right direction. Do you guys find those ways that Canva works where people come in the door, and you're like, look, this is how we do it here? And you have to steer them into the Canva direction of working. So I've got a, I don't think it's a spicy take, but absolutely, without a doubt, what you said is true. We're also needing to deprogram from our own old ways of working. And that's how we used to do product a year ago. AI has totally transformed the way we do product and our expectations of that product loop. And so we used to come with a customer research, here is a pitch deck that includes the why, how we're going to build it, the chronology of how we're going to build it, how many resources it's going to take, the different milestones, et cetera, et cetera. That's the old way. Now it's, come to me with a prototype and the smallest team possible to get a first spike out and get customer feedback. And then, if it's doing well, we can scale it from there. And that is a lot faster, requires a lot less people, and definitely gets you closer to the customer a lot quicker, getting real interactions. So you're moving from a pitch deck and hypothetical research to actual real implementation and adding real value. And so we're deprogramming from our own old ways of doing things to adapt to this AI world. And that's more important to us than deprogramming elsewhere. Now we filter it in the interview process. AI appeals, how do you think about building products? And if it's not that way, then they're not coming in. So it's like Taylor Swift. Old Canva can't come to the phone because old Canva's dead. I'm not a big enough Taylor Swift fan to get that reference. Is that the Call Me Baby song? No, that's Justin Bieber. Shit. I'm going to give up. I'm really old. Cliff's cultural knowledge is in the visual arts and not the pop musical arts. So, okay, that's all very good. We have a lot of folks at larger companies, at earlier-stage startups, a whole mix of firms in the audience. What advice would you give, based on what you've seen, for, again, you guys are totally changing your product, totally changing your revenue model, totally changing how you work. It's a pretty big transformation for everyone, but especially for you guys. What generalizable advice do you think you have, or as you look at other companies and what they're doing, where do you see people going a bit too slow or messing up? I think AI gives you superpowers as a leader to access insights you were never able to access before. If you've got AI connected to your code repository, all of your data sources, like Snowflake in our case, into Stripe, so you're pulling all the financial data, if you've connected all of your core data sources, you can essentially ask it any question and get really great guidance and get down to the root cause of the data and the issue or the opportunity that you were never able to do so directly before. Which means you can just get into so many different things and understand how it's performing and what the opportunities are. And historically that would have been, I need to go talk to the leader of that space, they need to go pull the data, or I need to go look at dashboards. The dashboards, you can never marry up the financial data with the Snowflake data with this and this and this. And it used to be a real exercise to get deep insights into a particular part of the business. Now I can just natural language query what we've got, this internal system that spans all that data. And I can pull out really, really deep insights, and they're generally 90% correct. And then I can go into the team or the area or the opportunity and I can ask the right questions. And I can also help guide it to a much faster path forward than what I was historically able to do. Because historically every bit of data you got was also clouded with human interpretation, human opinions, biases, their own ambitions, et cetera, et cetera. And now being able to have the data as the root nucleus of everything and being able to get access to that has just totally transformed the way we lead. What do you think about that? Oh, we see that effect as well where it's... That's for all leaders, sorry. That's not just available to me, that's available to... AI kind of effectively makes your organization feel smaller because you can be closer to the actual ground truth or the work or the code or whatever it is. The code, exactly. I can see which... I found out we have over 400 experiments running, and many of them have been running as experiments for six months. You shouldn't have that many... Just things of that nature, and things that are deprecating performance or this or that. And it's like having that omnipresent view is just insane. Yeah, I definitely noticed that. I also think there's this totally hackneyed, cliched line that people in the AI industry use, that the models are the worst they'll ever be. But it is true where I get more excited for what's to come because I think there's all this stuff that is clearly going to be much better in two or three years' time. Where one, in the data context, is really hard, where if you ask a model something where the answer existed on the public internet, it does so much better. And what everyone's trying to do now is get all the stuff that they need, be it the relevant customer data or the business data or whatever, into... A queryable... A queryable form. Yeah, it's the structure of the data that we've spent... And so, I'm sorry to interrupt, but I got super excited about this two years ago. And I designed this system I used to call Maestro that gave me insights into everything. But then we ran into the problem: all our data structures weren't compatible, and it was a bit of a mess, and shit in, shit out, all that kind of stuff. And so we spent a lot of time getting those structures correct that allows you to be able to... Yes. And I think a good example of this is, is there a text box at your company into which you can type, how did our spend on airfares increase last year? Or if you shouldn't necessarily have answer-to-the-entire-company thing, how did my spend on airfares increase last year? And I think that's an interesting example because it gets to wiring up all the permissions and things like that. And basically, I know very few companies that have a good answer to that. Stripe is a pretty good answer. But again, wiring up all the data to be useful is hard. That gets to the second thing, which is... So we've just had, we used to have a pretty good version of it. And only over the last six weeks, we've had the next-level version. Is there a text box you can type that query into? Yes, exactly that.