EP 118: Will Gaybrick (President of Product and Business, Stripe) on Capital Allocation, Org Design, AI, and Global Growth

30:16We have a product internally called Compass. and it is it was originally built just for product development tracking and actually it was only recently that we made it mandatory it had a ton of use already but we use it we use it for like npi new product introduction initially and then we realized was that it could be so much more and so now we run weekly updates out of it you can follow projects it'll send you a digest All of our product development processes are sort of institutionalized and systematized in it. So we have various gates towards launch. One thing that we do is something called implementation review.

31:00It's like a peer review internal to Stripe. So before you ship your thing to public preview, it has to run through another Stripe, a trained implementation reviewer doing an end-to-end integration, writing up a friction log, saying everything they found, then you go through it. You say, what are you going to change? What are you not going to change? Ultimately, there's sort of a rapid decision-making. That's the form of decision-making we do, where the head of the business unit actually says, which changes need to be made? So it's this complex process and the whole thing just runs through this tool.

31:34Super cool. What's the biggest challenge that you're thinking about today for Stripe? You know, there are a lot of them. One that is persistent, but very top of mind right now, is migrating users to new versions of our API. We've made, you know, always made the very user-friendly choice, which we're going to keep making, to sort of have a very, very generous end-of-life policy. So basically, if you're using Stripe, and you're not sort of trying to adopt more of it, it's just going to keep working. right now because we sort of moved not just beyond payments, but to this world where we have this billing suite that we announced publicly.

32:18It's a half a billion dollar business growing like crazy. And we've got all these products in connect land where you can sort of store money in the cloud. We're having to refactor all of our abstractions because a lot of the guarantees that our API sort of looked at. Like if I process a payment, then all the settlement will come through Stripe's Rails are being relaxed. One of the top user asks for billing is, I really want to use Stripe billing, but I have a four-year contract with TSIS or WorldPay or some other payments processor. So can you plug that in? We announced this publicly at Stripe Sessions.

33:00That's actually now in private preview, so you can do that. But some of these changes sort of run through a lot of our abstractions. So we've actually now cut a v2 of our API, which actually makes some updates to the API semantics as well. And we are going to have to maintain backwards compatibility. So if you're a user out there, don't worry. But we also are going to be able to ship so much value so quickly by getting people into new versions. And there are just thousands of lines of Stripe code written into a lot of users. So figuring out how to do that is a big challenge. Yeah. As you think about payments and Stripe, I guess, in the next five years, 10 years, what are some of the biggest changes that you think are inevitable that might be unintuitive to people just going about their day.

33:56I mean, I mentioned it in passing a couple of times already, crypto. In what way? Yeah, I'm so excited about it. I'm so excited about stable coins. So Stripe's history on crypto. So I think it was before I joined, it was 2014. We were the first major payments company or whatever, if we were major at the time, to support crypto pans. And it was Bitcoin pans. And then we sunset that a few years later, maybe it was 2017 or 2018, because it was just all fraud, just all bad actors trying to cheat the system. We lost a bunch of money on it. And it's not surprising. Bitcoin is not a good instrument to make payments with.

34:41It's a much better store of value or sort of long-term asset to hold. Sure. And we took some heat at the time because, you know, crypto then sort of started heating up again. And then we sort of jumped back in a couple of years later, building a crypto on-ramp. And then, you know, we had this next crypto winter. And I think a lot of people sort of abandoned crypto. And the thing we've just seen as a through line that is just compounding and compounding and compounding is stable coins. So we mentioned recently that we're shipping stable coin pans. So you'll be able to, as a straight merchant with no integration changes, accept payments in stable coins.

35:26I think it's just USDC to start, but we'll be able to very quickly expand that. And we're now in very recently in private preview. And I think it took two days for our cumulative volume in stablecoin pans to pass our cumulative volume over a year and a half or two years on Bitcoin. So it is just going crazy. What is the use case for stablecoins? And maybe a quick primer for people that don't know what stablecoins are, if you could just explain those too. So stablecoins are just crypto assets that are pegged to a fiat currency value. And there are now EUR rails. I'm sure there's other rails too.

36:16I'm sure there's GPP rails, but really USD rails are sort of winning the day. So there's USDC, which is a sort of fully collateralized stable coin that was started by a company called Circle. A big partner in creating it was Coinbase. And so for this, it's like dollar for dollar collateralized. So in theory, it can't lose its peg. Fun fact, it actually did briefly lose its peg when SVB went down because they're storing some funds there. But in theory, it's sort of securely backed. And then there's algorithmic stable coins where via trading algorithms, they sort of maintain their peg. So Tether, USDT, is the most popular stablecoin.

36:58Yeah. And so then what is the use case that you guys see over time or how will that change people's experience as you look out 10 years from now? It's kind of funny because we were all fumbling around for the killer app in crypto for a long time. And you had the ETH world computer, which by the way is a beautiful vision. And there's like so much value in Ethereum as a platform. In fact, a lot of stablecoins run on top of Ethereum or L2s on top of them. You had all of the speculation that was happening, Dogecoin and all the other fun coins. Dogecoin, sort of the GameStop of stablecoins, I mean of crypto assets.

37:42And it turns out that just the killer app is just money. It's just crypto works well as money, as long as you sort of stabilize its value. So many parts of the world just don't have currencies that you can trust. When you look at Southeast Asia, you look at, I think in Vietnam, the adoption of stable coins is just booming and the economy is actually taking off there. If you look at the charts, it actually looks pretty correlated. Very young population, very sort of crypto literate. So for us, part of Stripe's mission has always been to increase the pace of globalization in commerce, to break down barriers to cross-border commerce.

38:24And there's a lot of those barriers that exist, that are sort of written deeply into the existing financial system. And the notion that you could have one globally applicable protocol to program money in the cloud, and you can see how it would get us really excited. So what we're seeing now is that companies that ship software, B2B SaaS companies, already major conversion uplifts for them when they accept stablecoins. Because if you're in Rwanda, it's pretty hard to get an electronic payment instrument that Stripe accepts. But if you can get on chain with stablecoins and you want to buy a piece of software that someone built in India or San Francisco or wherever else, USDC works.

39:17And so now you're seeing this in mostly international markets that have volatile domestic currencies. Is that the primary place that this is unlocking? I think that's where it'll gestate. And it is gestating quickly. I think it was something like $4.6 trillion over the past 12 months in stablecoin transactions. Now there's like a Visa report saying that a lot of those weren't actually payments and that's fair. Some of it's probably high interest rates and people just wanting to be in stables because you actually earn pretty good interest rates just from holding stablecoins. But there's a real exponential curve here.

40:01And my sense is that the emerging world will sort of normalize stable coins and they'll sort of bring all of us who have very stable currencies into the... They'll sort of acclimatize us to the notion that it's a perfectly normal and well-accepted way to transact and that over time, moving between stables and fiat just won't be a big deal. And you think that the adoption versus Bitcoin, the reason it's growing so much quicker, is that the confidence in the price versus Bitcoin is going to be far more volatile? Or why is it seeing such faster adoption? I think so. And because the performance is a lot better.

40:50So a lot of stable coins will run on L2 chains. So these are sort of chains on top of chains that will batch together transactions to increase the rate at which they're confirmed and decrease the cost of them. And so Bitcoin, there is a lightning network, but just Bitcoin in general is a pretty slow and expensive protocol. I think even just on Solana, which is a very popular L1 chain, I think just native on Solana, no L2s. I think it's 400 milliseconds now for a USDC transaction, which is pretty amazing. Yeah. Artificial intelligence is obviously one of the topics everyone's talking about. You, in prep for this, I went back and listened to a 2015 talk you did at Yale, and you actually said that that was the technology that you were most excited about.

41:45I think you were even still at Thrive at the time when you were doing that. I'm curious what your perspective has been both as a former investor and sort of looking at the opportunities that are there, but also as an operator within Stripe of all the AI stuff we've seen over the course of the last two years. Well, listen, it's hard to match my enthusiasm about stable coins, but I'd say - I know. I would say that a lot of what's happening in AI is right up there too. Yeah, I mean, I think the thing in 2015 that I was seeing was just deep learning looked like it was really going to work. And so if you believed in sort of continued Moore's law effects, then You would see this sort of exponential improvement in the ability for deep neural networks to just do powerful things for the world.

42:43The video is out there. So this is not some hindsight revisionist history. The video exists of you saying this. I will say that was the extent of my prescience. It was very basic. I had no notion of transformers, even though those were near at hand, or large language models or just the concept of foundation models in general. I think it's all so fascinating. And I can talk a little bit about the applications at Stripe. So at Stripe, we've been using machine learning for a long time. Sure. I'm sure fraud detection or whatever. It's fundamentals of the business. Exactly. Yeah. And we have about something like 100 models in production to augment our products.

43:30There's 43 that are sort of critical, like sort of in the charge path or doing major risk mitigation things. First model was shipped in, I think, 2015. It was like a linear regression model, even how it worked to mitigate fraud. So that's how sophisticated we were at the time. but flash forward to today you know I think we've we've publicly stated that we've been toying with our own foundation models any foundation model basically any large model trained on you know large means many different things you know for us it means you know order of a hundred million parameters you know for the GPT ends of the world it's you know many billions of parameters or hundreds of billions of parameters, I guess at this point.

44:17And it was like a sort of self-supervision. So there's no sort of, here's the input and here's the output we expect. And it's actually just sort of trying to take concepts and embed them in context. And then you can sort of run models on top of that. And this sort of embedding in context gives rise to this word embeddings that everybody uses all over the place and loosely. But just to give you a very simple example of something that's really working at Stripe, card testing has been on the rise on the internet, maybe correlated with the rise of AI and ML because it's just easier to spin up machines to do basic things like this.

45:05So card testing is, you know, steal or buy a bunch of credit card numbers and then go to just random, you know, websites and perform transactions, just small transactions to test, does this card work? And then you could sell it yourself or use it for something else or otherwise. And these used to be pretty easy to detect because they would be sort of one merchant and they would spike. It must be sort of, oh my God, MerchantX is seeing a big card testing attack. We're sort of working with them to mitigate that. And we provide various protections built into Radar, our fraud suite, and so on. We're now seeing some more sophisticated ones where card testers are sort of dribbling attacks across different sites with just two or three transactions here, two or three here, two or three here.

45:57So very hard to detect because no one merchant is really suffering. but Stripe overall is seeing a huge spike and you know that can hurt the sort of you know integrity of the ecosystem violate card network rules things like that and so we found is actually taking just raw payments data so just take like a giant like JSON blob of payment payments data like here's what a payment means to Stripe stringifying it you know just turning it into pure text, tokenizing it, just slicing that up into tokens and feeding it into a transform model has led to the ability to embed payments in a way that is completely unintuitive to a human.

46:46We can't point to any single cluster and say, oh, this is exactly why those payments are clustering there, these are there. But we've been able to identify just in the clustering that happens from these payments, just which ones are more or less likely to lead to card testing. So we just take a new payment coming in. Maybe it's just one of three card testing payments on a site. There's no spike for us to detect. And this model will be able to tell us in a way that no hand curated features ever could have that, hey, just so you know, that one random transaction that doesn't look suspicious to you is actually very likely part of when it's distributed attacks.

47:25So it's been really powerful internally already. Are you doing that on top of one of the big model providers or how is that actually like technically working? So in this case, no, it's actually just ground up our own model. We are toying with leveraging some of the existing models to actually do some multimodal embeddings, which we're excited about. There actually are some really cool, like just almost off the shelf applications that have worked very well for us. So one of the things that is a core part of our strategy is to be sort of conspicuously good at risk mitigation and to be incredible stewards and sort of guardians of the global financial ecosystem.

48:12We invest just a ton in risk management. And one of our leaders once said, risk management to Stripe is like Amazon, logistics is to Amazon. Actually, it's a pretty good analogy. We just need to be best in the world at it. And there's a notion that fintech companies play it fast and loose. And there actually clearly was some bad behavior with FTX and obviously this isn't FinTech, but with SVB going down and FRB going down, there's been some tightening in the regulatory environment. And so we've just been doubling and tripling down on risk mitigation at Stripe. And big banks do a very good job of complying with, literally with what regulators want.

49:02We always want to do that too. And we want to go a step further and just think, how can we be doing this better for the world? And so we built something called Inspector GPT, which lets us do sort of continuous TOS compliance for users, TOS compliance. There's a lot of different very, very idiosyncratic rules about what you can and cannot sell coming in from different financial partners up and down the stack. So something as sort of scary as AR-15s, maybe that's not idiosyncratic, to something as seemingly innocuous as CBD, or actually even more innocuous like tarot cards. So financial partners don't want you to sell tarot cards.

49:51What we've historically done is we've underwritten our users when they're joining Stripe. We use models to do this, like classifier models and so on. And then we'll review them again from time to time at different checkpoints of they are now hit$500 ,000 of volume or 10 million of volume or whatever. But we're actually able to leverage sort of commercial chatbots, ChatGPT or Claude or Gemini, really any of them have worked so far, feed them our terms of service and just point them at our users' websites. And they just will periodically go and say, hey, I took your 25-page super intensive TOS, and I think I found a violation on the 26-page of their product catalog.

50:42And it's because this spa, which sells 308 services, one of them is tarot cards. And that violates provision like 3A, triple I, whatever. of no tarot cards, according to this financial partner. And just, I think the state of the art for this is sort of like sending humans with the printed terms of service docs and maybe screenshots and having them do this in low cost locations. And it's just been really, really impressive and really create a lot of efficiency. I think everyone talks about the human replacement side of this, but the superhuman element of it is super cool when you can find that.

51:26It's not just, oh, you're making engineering productivity, whatever, this much better. We're automating customer support, so we don't need to hire people in Manila or whatever it is to do this. But actually being able to do things that humans couldn't do or maybe would be very hard to do as effectively as that, like going over terms of service and being able to catch things on. distant web pages. I totally agree. And actually, one of the things that I've been thinking about a lot on this front is just the sort of, what are the implications for human knowledge period of the black boxiness of neural nets and transformers in particular?

52:08Just, you know, we don't know how our model was able to tell us that that particular payment was likely card testing and the model isn't able to explain why it was. There are ways in which we're sort of getting better at explainability. And you found that it sounds like the accuracy to be - Very high. Outperforming what we had with hand curated features by a whole lot. And so there's something deep there about how we need to think about developing our products and enforcing risk standards? Like if we are virtually certain that the model is right, but we don't know why it is, what does that mean?

52:50Yeah. I mean, there's so many moral questions with AI in general, but the potential of bias or whatever, like it's great when it catches the positive, but the false negative and the stuff when it can't explain how it came to be. And it It could be making racist assumptions or like you don't know what the primitives are by which it's making those decisions, which is fundamentally interesting. How do you go about your corpus of opportunity to apply artificial intelligence is got to be up there for almost any organization? and you also uniquely have the competency internally to do a lot of these things.

53:41Like maybe Exxon has similar, but I don't think they have the competency to go execute it on it right now, not to disparage anyone from Exxon listening. But how do you guys come up with like what is worth testing and going after from an artificial intelligence standpoint versus what falls below the line and we should just keep doing it the way it is? Yeah, I mean, some of this is cultural. You want to, well, I guess there's a couple of things. So one is from a systems and tooling standpoint, you want to just give people the agency, the technological agency to start using these technologies. And so we've made a bunch of investments on those fronts.

54:19And that just means having access to these tools. Yeah. So one of the first things we did after the whole sort of explosion in chat bots with ChatGPT coming out was we just created a Go link internally to go slash LLM into a Stripe computer and it takes you right to an LLM interface. At that point, it was just GPT-3. Now you've got GPT-4, you've got Claude, a bunch of others sort of in there. And we sort of ragged this, meaning we sort of gave it some additional specific data on Stripe's internal documents. We purged any user specific data because we didn't want anything that's native to users to get into these tools.

55:09But Stripe's documentation, it's trained on things like that. And just sort of gave that to people. It took about 10 days before usage sort of started to level off. And it's actually kind of stayed there. About a third of the company is weekly active. And the vast majority of the company is monthly active. And it's kind of cool too, because we created a community around it where Stripes would sort of share their prompts. So we actually have a prompt now that can take any corpus of text and rewrite it in the Stripe tone. We are, as I mentioned earlier, really motivated internally by creating something beautiful and being, as we say, internally meticulous in your craft.

55:57And so if you're a sales, a sales rep or an SDR, and you just want to make sure you're writing a good email and that it abides our style system and our tone and so on, you can just toss it right in and it'll make some changes and shoot it back to you. Second step was just making sure that from a technical standpoint, we had API endpoints set up to do development with LLMs really easily. And then the third, which just I think goes to product development generally at Stripe, is making sure that you make room for your teams to have better ideas than you have and surface those. So there's definitely ways that we've been sort of tops down prioritizing our investments in AI, but we also have seen some really cool bottoms up applications.

56:44In terms of like empowering autonomy of decisioning within the organization and just letting people go do that stuff, are there any, clearly it's partially how you hire and what you incent people to do, but are there any forums or means of facilitation that you guys have done like a, I don't know, hackathon days or, or it sounds like that prompt sharing is one, but anything that stands out there? Well, so in general, um, you know, I, I mentioned this in passing, but I didn't define it. So we, we actually have introduced a formal decision making framework and to anyone out there who was like worried about adopting one of these sort of very corporate frameworks, I will tell you it has worked so well.

57:27Uh, we use when I think it was developed by Bain called Rapid, where you have a recommender, agreeers, input, performers, I guess inform, I always say input, but it's kind of both, perform and decide. And particularly given that we don't have titles, this has been very, very valuable for us. So basically any material decision is, you define the rapid. I was worried about sort of the recursive meta decision of like who what's the rapid for defining the rapid yeah it turns out no one really cares they just want to know that there's like a formal way this is being made we have a decision log we built a decision isn't real if it's not in the decision log and discoverable and so on and so um you know that's how we've like really worked on decision making it's actually helped a lot um certain decisions go to certain forums you know uh actually you know I make every pricing decision personally just for like pricing consistency across our entire product base.

58:29So we have just sort of various defined rules and systems. For ML, we have maybe two formal ways we've made decisions about how to invest. So one is we actually do run what looks kind of like an internal accelerator. It's a bit of an experiment right now, but we've been doing it for a couple of years or a couple of cycles. So one year it's called our experimental projects program. And this is just a program. It evolved from what used to be called our crazy ideas program. Crazy ideas program was just, we sent around a Google doc to the whole company and you could write stuff you thought we should do in it.

59:07And then we'd read it. Got a little bit big for that. And so now you can apply to just sort of be seconded onto your own team or a team working on the thing you're proposing for a period of starting six months and then you can keep going if things are going well. And we ran one of those in particular for ML. So it's just sort of like, hey, we want an entire batch of teams just saying, I want to do this with MLAI, I want to do this. And some of the ideas or things that would be sort of maybe obvious, Docs AI came out of that. That's just a more AI driven way to interface with our docs. There's some cool ones that I shouldn't talk about yet, but they'll be coming out soon.

59:53And then the other was actually centralizing some of how we operate in ML and AI. We used to have a lot of different nodes in the organization. And that's how we ended up with over 100 models and some of them performing well, but no shared features across them. and we've now created in what we call our information org sort of a center of gravity and center of excellence. We hire researchers onto the team. We're actually going to be announcing soon some pretty cool research areas that we're undertaking at Stripe and this sort of core group is responsible for this sort of foundation model intelligence layer where you have the shared embeddings and shared features across everything we do at Stripe.

1:00:39As an investor at Thrive, you obviously made a very significant and successful investment in Stripe specifically, and then you doubled down on it. What were the things that you looked for in companies or founders or teams? Obviously, the Stripe investment has proven to be a good one for Thrive and for you at a personal level, but was there specific things that you really sought out when making investments? Well, I think you're a B2B SaaS investor, so you can discount all of this, but I can give you my framework at the time. Yeah, I'd like it. You know, when you're an investor, and it's true, I guess, when you're a capital allocator at a company, you're always trying to reduce dimensionality as much as possible.

1:01:25There's just like a million different considerations, like what really matters. And so from a market standpoint for B2B investments, I always just thought about simple two by two of one axis universality and the other axis just sort of core versus non-core. So does everybody need it or do only a few people need it? And then for core versus non-core, sort of the question is like, do people want the best one? Are they okay with just any old one? So I used to be on the board of a company, a great company. We use them at Stripe. it's really awesome, called Greenhouse. And that investment has gone well for us at Thrive or former us, now the Thrive guys.

1:02:07That's sort of a classic universal non-core where everyone needs an applicant tracking system, a way to track the recruiting process. But it's hard to get a lot of pricing power, hard to get really deeply integrated into the company. It's just not sort of a core consideration for senior leadership. Stripe to me was just like the top right that two by two where it's like, what is more core than your money movement? And maybe you needed to have a sense of the vision for Stripe. It wasn't just accept card payments. It was, hey, we're going to be commerce infrastructure for the world. And then everyone needs it.

1:02:42And now it's going to be money around. And the pricing power you have because of the best of breedness being 0.1 % better or could lead to significant pricing power because for an individual merchant, that could be material dollars. Yeah, absolutely. And the sociology of payments procurement is very interesting. I won't bore you with that unless you're keen to understand it. But with larger users, the breadth of what you do for them really matters. And they want to see the pricing tied to the value in each dimension. But yes, the short answer is yes. And then for entrepreneurs, and I guess for leaders in general it's like a good pattern match on When you find someone who just seems to have more hours in the day than you can possibly believe.

1:03:35That tends to be a good entrepreneur. John and Patrick, I just was struck by how prolificing the day. were. Probably reciprocally with you going to law school and PhD and all this, they probably were drawn to that on your side. Failed PhD, but yeah. But yeah, that struck me about them and the judgment point that I mentioned as well. One of your responsibilities has been leading product within Stripe. I guess I'm curious, how is Stripe's product organized? How do you think about what is the platform versus what are the applications that sit on top or what are the primitives versus what people get individual decisioning authority over?

1:04:21Yeah. And I'll give you the marketexture view and then the institutional organizational view, because I really do believe in Conway's law. It's served us very well to follow it. So from a architecture standpoint, Stripe started back in 2011, accepting card payments in the US evolved a lot since then. Phase two was maybe accepting card payments in other countries and then adding in some local payment method support. In Europe, for example, we expanded to relatively quickly. every country has some local scheme that really matters to payments in that country. And if you don't accept that scheme, you lose a lot of transactions.

1:05:06People are just like, oh, it's only cards I wanted to pay with. Ideal in the Netherlands or Swish in Sweden or whatever. So that was sort of phase two. Phase three, we started dabbling in other ways of money movement. It was just pay-ins initially, but then we had Shopify, incidentally, sort of asking to white label our pay-ins API. This is what sort of has led to a really deep partnership with them. And around the same time, we had Lyft. We've been working with them on payments. And as the story goes, I wasn't at Stripe at the time. we were in their office and we saw that their finance team was sort of uploading flat files to make payouts to drivers.

1:05:56And we said, we can create a better way for them to do that. And so that led to an early version of what we call the payouts API, which is sort of load money into Stripe and send it. And so this sort of multi-party money movement, even those two different paradigms, we rolled together into something called Stripe Connect. And Stripe Connect has been just an enormous driver of Stripe. 13 ,000 platforms and marketplaces use it today. And it's a very, very significant driver of vertical SaaS, on-demand economy, things like that. And so this whole thing, like pay-ins, like money storage in the cloud, pay-outs, converting currencies in the cloud, is kind of what I would call the core business.

1:06:41We call it like a global payment suite, or sometimes we refer to the global payments and treasury network. Just basically saying there's all these different ways to move in store money. We have one single API. We can handle KYC for you, mitigate fraud, move money in or out of accounts and send it globally. And we now support pay-ins across 50 some countries and payouts to over a hundred. So that's, you know, so the core business that you've probably heard of, 2018, we created something called Stripe Billing. Are you familiar with billing? Yeah. So billing was originally just this very basic subscriptions API that we shipped way back in 2011, 2012, where you could literally just add a subscription to a customer object in the API, and it would just sort of bill that person on some cadence, some amount, very limited.

1:07:36You could only have one subscription per customer. It was not fit for purpose unless you had the most basic use cases. 2018, we shipped billing, which was just sort of a small team doubling down on this, making it a little bit bigger. And incidentally, I think B2B SaaS was really booming then and got a lot more pickup than we expected. And so this whole area of becoming your billing system and helping you implement your revenue models, track your customers, innovate on your commercial model has become enormous for us and for our users. There's now something like 300 ,000 users of Stripe billing and actually growing very, very quickly.

1:08:25And the sort of adage I have is, and this goes for Stripe as well, everyone has a billing system. Everyone hates their billing system. And it's just one of these things where like, if you've had to work on billing systems, you know, as you hear me talking, you probably feel it viscerally. It's like, you know, your, your, your go-to-market team says, Hey, we want to, uh, rebundle this in this way, or we actually want to pre-bill for this thing or whatever. And you're like, Oh man, okay, well, we'll give you that in three quarters. And so what we're trying to do is, you know, just take that entire team, allow you to repurpose them, or at least give them a lot more leverage to do what they're doing today.

1:09:06And we sort of rolled this all together with some adjacent services like tax, like assessing sales tax globally, analyzing your revenue, revenue recognition, things like that. This sort of like area between the CRM and the ERP into what we call revenue and finance automation. And this is a major, major sort of bet for us today with billing as the core. And it's a really fun product space. We sort of stumbled upon it to some extent and sort of can't believe that it's as greenfield as it is. And digging in, we sort of understand why. It's actually really hard. We've had to rejigger our strategy several times.

1:09:51So there'll be a fun announcement on this in the future, but we've now sort of prototyped our own language that sort of will open source at some point that underwrites all the billing primitives because we need it to have certain aspects of static analysis. We also want it to be pretty simple so that you could effectively test changes to your billing system before actually making them. So that's a huge area for us. And there's a few others as well. But these sort of two foundation, or I guess maybe three foundation orgs, core payments, the connect org, and then revenue and finance automation. And they all sit on top of a sort of product platform org.

1:10:31And that org sort of ties it all together for Stripe. We used to refer to ourselves as AWS for money. And I think that was just like a very poor analogy. And the reason is when you look at sort of, say, S3 and Aurora, like different AWS services, there's no sort of logical tie between them. Like how you model data in one versus the other is kind of like your business and how Aurora and AWS work doesn't really need to be, they don't need to contemplate each other in any way. Whereas the way that straight billing works with Connect and with payments needs to be graceful and subtle. If we model Logan as a customer and payments, then how our invoicing API thinks about a sort of credit balance for you as a customer in a different product line needs to cohere.

1:11:29And so this platform org is very important being sort of the sinews of tying it all together. And so are there discrete leaders for each of those functional areas and then the platform underpins it? Or how does engineering resources get allocated and functionally, how does that work? Yeah. So each of those are unit organized. Payments org, connect, we actually call it money as a service, revenue and finance automation. And then we have the platform also unit organized. So product and engineering, other embedded functions in each of them. Interesting. And is that the way it's been for totality of time or like how has it evolved?

1:12:10So one thing we've done is we have hoisted more sort of technical unification upwards. So it used to be that you had sort of the classic, you know, Amazon phalanx of, you know, just sort of unit orgs all the way down. That hadn't, didn't work very well for us. Because of the need for interoperability exactly exactly yeah spot on uh so we've hoisted that upwards uh so you typically have you know one of these very very we don't have titles but you know some svp evp level person with a head of engineering uh in each of those areas and and uh and so then when you're when you're starting a new project like you referenced billing was kind of i don't know skunk works is a fair term but in the early days it's kind of a small group that got going similar to the the connect product sounds like that was smallish in the early days.

1:12:58How, how do those things manifest themselves? Like, do you just sort of send some people in the wild and say, have at it and sort of see if there's traction behind it or is it more top down and telling people to go try these things? Yeah. You know, we've, we've had the benefit and, you know, I mentioned this before, like this sort of infinite problem, um, problem domain, um, had the benefit of just vehement user demands for us to expand into different areas. A good example is Stripe Tax. Just every year, Patrick or someone will tweet out, what do you want us to build next year? And tax was just rising to the top of the list.

1:13:37And so we decided we're going to shoot off a team to go work on that. You have the classic questions of a capital allocation where it never makes sense from a near-term ROI to spend money over there versus shipping the things your larger users want in the core. So you have to protect them for a while and so on. But usually we have a few gates for each of them, but typically they're non-speculative bets. They're just sort of like, you know, we need to go execute or maybe areas like, you know, when we brought back crypto a couple of years ago, it's a little bit more, at least that time speculative.

1:14:14Now it feels very non-speculative, but yeah, then it was. One of the things I heard you say is that companies are often hesitant to force an ROI calculation for early projects. Can you elaborate on that point and why thinking about ROI is an important thing in the early days of building products? I guess my more specific push is just that, and actually I'll kind of parrot John Collison on this, But we work in an industry where fiscal discipline for the winners is just not that important. Maybe it was very important to Amazon in the early days, but with Meta and Google, just soon enough, ads money is geysering out of the ground.

1:15:03And so they're not always the best run companies from the standpoint of just precision around how many people should be doing this thing. Do we know what they're doing? What's our observability into the growth there? How can we think about the incremental benefit to investing more? And then as you grow up as a company, you end up bringing leaders in from these great companies because of the survivorship bias of that's where the leaders are. And so this skillset of being really, really disciplined with your resources and thinking about that ROI and self-solving, just typically, you know, the sort of leader products don't ship with that feature.

1:15:48And so you want to really, you know, drive it home. You made a comment earlier about titles, not having those specific titles. I guess I'm curious, there's kind of these, there's this Mark Zuckerberg view on titles, which is that they're very expensive. And there's this Mark Andreessen view of titles that they're very inexpensive. How do you think about titling within Stripe and the cost of giving out titles? So I'll caveat this, that I've always had a title with Stripe, so I shouldn't be one to talk, but... Good for me now for the day. Exactly. Yeah. We get various

1:16:31pushes to introduce titles to Stripe. I think that there's a few reasons to do so, and a few reasons not to. And I just think about it very pragmatically in terms of the costs and benefits of it. And to date, just pragmatically, the costs seem to still outweigh, or the prospective costs seem to still outweigh the prospective benefits. Culturally, I think it's actually quite good not to have them because it just speaks to a pretty flat culture. And, you know, they introduce just semantically a notion of hierarchy that, as I hear it, at least from Stripes, we don't have. And I think it's part of that.

1:17:15You also end up, you know, in a situation where, you know, it's actually not just titles. We don't expose levels internally. So, you know, if you were a level six and I were level five at Stripe, officially I wouldn't know that you were a higher level than I am. Now, I might know because people talk, but the sort of second order sociological effects aren't there where the average person doesn't know that you're a six and I'm a five. So maybe I don't care as much. Whereas if you're a senior director and I'm a director, then that's just apparent and everyone knows it. And it just is sort of, I think maybe can lead to bad incentives.

1:17:56A lot of times also the more senior title is correlated with a larger scope and a bigger org. And there's very bad incentives related to that. On the positive side, we've been a big push over the past couple of years on decision-making. And I do think that it gives you a language for decision-making. It's like, this is an SVP level decision. This is a director of this and so on um and actually maybe a subtle benefit of not having them is just people don't know how to mine your employee base people try to mine stripes employee base a lot i can imagine um uh yeah so some of the considerations in terms of like org chart design then is the org chart exposed internally?

1:18:43And so are people able to de facto back into levels of responsibility? I realize headcount is not a perfect proxy for that, but. Yeah, we've made a big effort to sort of create divergence between sort of where you are in the org chart and your level. And so you have some people who are line managers who are some of our most senior engineering leaders. Typically, they report in pretty high. So, you know, there's some ways of getting kind of back into it. But yeah, we do expose the org chart. And it's very useful. I actually really believe a lot in org design, you know, far more than I used to, but it just maybe it's a part and parcel of growing as a company, but actually it's came up at a Q &A for the whole company on Friday around like, what are we doing to make sure we don't ship the org chart?

1:19:41And to me, that question is very little. We're going to ship it. It better be right. There are certain mechanisms we have to sort of tie things together, as I mentioned, but yeah, we're actually, we kind of put the org chart front and center so you know how it's structured. What have you learned about like org design, I guess, now that you've reflected on it. Is there a takeaway or any insights that you wished you had known in the early days? Maybe the thing that comes to mind is not any particular time where it was grossly off, but I feel like I've learned that the incantation that companies reorg too often is probably literally true, but sort of spiritually off.

1:20:32I think there's a

1:20:36Toyota design principle or whatever they're called. There's one on continuous improvement. And I really think if you are operating in this continuous improvement mindset, and if you're growing your ambitions and building new things as quickly as we want to be at Stripe, there's sufficient change that you're probably not, if you're not changing your org chart relatively regularly, you're probably sort of actually destroying productivity. So you're sort of like always weighing the benefits of organizational stability against the benefits of sort of really being organized against your new strategy, your new goals and so on.

1:21:18So I typically think about any org change we make as having a life cycle of 12, maybe 18 months on the outside. And then there's obviously pruning you'll do and gardening you'll do further down the org chart on an ongoing basis. Do you guys still do metrics roulette within product teams? No, not formally. Really? Not formally. Maybe explain for people what it was and why the principle behind it. Yeah. So we used to do, I think this is one of my bad ideas. We used to get together at our leadership team meeting. This was a 30 person meeting. I think it was every other week. And we had a little script that would just choose a team and a leaf node team, like a line managed team.

1:22:13and it would choose them the day before. And it was just using whatever, some Unix rand command or something like that to generate a random number. So every now and then a team would get picked like twice in four weeks, which they did not love. And they'd come in and just present their metrics of like, here are the four metrics that we run our team on. And here's where we measure success. Here's our targets. Here's where we are against them. so on. And so in theory, it was a good mechanism because you would know that it could be you on any given day and sort of encourage you to really be on top of your metrics.

1:22:54Three things that I think were problems with it. So one is this sort of panopticon design of like, you can't tell whether or not we're staring at you, so you better be on your best behavior. just I felt a little culturally off, felt a little infantilizing I think to people at times. Two is that we actually sort of were looking at metrics kind of at the wrong level in terms of the leadership team to leaf node teams. And what you actually really wanted to happen was your reports or their reports sort of ensuring that they have the right WBRs And, you know, those may be cascaded down. There'd be BRs, weekly.

1:23:39Weekly business reviews. Yeah, yeah, yeah. Weekly business reviews. And so, you know, you would end up having this sort of, you know, junior team to central leadership team conversation that just, you know, I think kind of disempowered leaders and didn't really sort of have the right types of conversations. And then the third one was just, you have 10 minutes to do that. And it's a team that you don't have deep expertise working with. You actually just couldn't give very good feedback. And so, yeah, we did that for a year. And actually a different mechanism and one that I think is actually working quite well.

1:24:17So are you familiar with a guy named Alan Mullally? No. So Alan is a leader I find very inspiring. So he was CEO of Boeing, one of their divisions. I think they're sort of like aeronautics division or I can't remember what the exact name was, but longtime Boeing executive. And then he became the CEO of Ford, I think in 2008 or something like that. And at this point, Ford was teetering on the brink of bankruptcy already. huge problems with unions. It was sort of the laggard of the big three at the time. I think the stock price in his early tenure dipped down to like one or two bucks a share. And by the time he left Ford, which was seven or eight years later, stock price was in the high teens.

1:25:16I think it peaked around$24 later. He had something like a 99 % approval rate with the union members, which was crazy. And they were the only one of the three major car companies that took zero federal money. So pretty just incredible guy. And I read this story and I was like, man, I got to figure out how the hell this guy did this. And so he has like a system. He calls it the working together system. And I sort of, just like with Toyota, other great companies, just stolen little parts of it. But one thing we do now, which I'm just raising him because I have to credit him with some of it is every other Thursday, the top 80 or so leaders get together and we run a meeting called the Run the Business Review.

1:26:03And it's a two-hour meeting. So one hour a week that you're dedicating to this. And we go through every single budgetary line item. And everyone has a set of initials. It's the owner. And every single company OKR, We have about 82 of them. And everyone, red, yellow, green. If you are green, you can actually skip it, get commentary if you want to. Yellow means you have a plan to get back on track, but you're not on track. And we talked about that plan. Red means you're off track with no plan. And so we do this together every other Thursday. We do a QBR once a quarter too. And what I found is this meeting is sort of the right level for collaboration among leaders.

1:26:50If you're red for three weeks in a row, we jump out to what we call special attention meeting where the right subgroup gets together to talk about how are we making a plan to get back on track. And I always find like organizational design mechanisms, like if they're good, they'll be fractal. Like you'll find that, oh, if this thing is working, I'll start to see it happen down below in various places. And this has actually turned into sort of an operating model for the whole company where the leaders will have their own RTBRs that sort of happen two or three days before and everything sort of rolls all the way up to this meeting.

1:27:25So I found this has been much more empowering of leaders, maybe a little less scary for line managers. And yeah, it's been working well for two years. How many metrics or things are on that graph? So budgetary items that actually have targets is probably about 30. Then we have 82 company KRs. There's whatever hundreds or thousands of KRs at Teams Track, but there's 82 tracks centrally. We also have a top 10 among those. And 82 seems like a lot, but it actually works out to be about one per 120 employees and if you're 120 person company presumably you'd have more than one so it seems like a good number yeah totally uh if you were still doing uh venture today do you think you would have gravitated to doing ai like would that be where you where you're spending your time or do you think you'd be off doing something totally different I mean, I think you have to be, right?

1:28:25It's just setting aside whether the current crop of companies are going to be wildly profitable or wildly unprofitable. There's all this sort of

1:28:38obvious jibes at inference costs and so on. Just a lot of things are going to change. you can just tell. And so I think whenever the world is on the precipice of changing as much as it is, I think you can't ignore it. Yeah. You've worked now with John and Patrick for nine years and I think people hold them in, well, I guess before you made the investment, what, a year before that or two years before that? A year before, yeah. A year before. So, so, so 10 years now, nine years as an employee. What's something that people don't appreciate or maybe is less appreciated about, um, about those two, either of them, I guess you want to call it out separately, although they tend to get grouped together, uh, um, that, uh, you know, that, that maybe people don't appreciate, uh, that aren't working as close to them.

1:29:29Yeah. I mean, there's so much to appreciate. I mean, obviously they're extremely intelligent. Um, uh, I think, um, uh, maybe it doesn't get talked about that much, but extremely tasteful, like just one of the core, core things that make Stripe, Stripe is a sort of level of taste that's hard to teach. But the two that sort of struck me when I was first getting to know them that have really been borne out over the years is one, they're very kind. They're kind, humble people. And I think that keeps people at Stripe. You're just not working for jerks. And so we've had a lot of our just senior most leaders have been around for a long time.

1:30:08And they're very wise, great judgment. Sometimes you just, at least I do, I think, have you guys seen this before and you're just not telling me? And I think that's probably correlated with them being very curious and just having a lot of conversations with people who have seen it before. But that's what jumps out. Yeah. Interesting. You mentioned studying Toyota and studying some of the stuff related to Boeing and Ford's turnaround. Are there other leaders that you've looked up to or been influenced by outside of Stripe, like just people in particular that you admire or have taken principles from?

1:30:48Well, one I know you talked to was Claire Hughes Johnson. And Stripe would not be Stripe without her. And I sort of consider her in many ways to be perhaps my only personal mentor. And she was COO for... Yeah, she was COO. She started a year before me, 2014. I think she was there until 2020. Yeah. And I think just her understanding of organizations and people and how they work, her intuition around what's really going on and why something's not working. was just really amazing. I encourage anyone to read her book, Scaling People, I guess it's called. So definitely her. Marianne Lake, who now runs Chase, who was at JPMorgan, was a really good sort of mentor in my early days as a CFO.

1:31:49And just seeing the rigor that she used to run JPM. She's sort of bouncing between talking to regulators and doing capital allocation, doing product reviews. So she comes to mind. I'm trying to think about other leaders I've I've spent a bunch of time reading about, oh, a good one, actually. Frank Slootman. Yeah. He impresses me so much. And there's just clearly a formula, right? And it works. One of the things he says that I just, I like to repeat a lot internally, particularly being, you know, someone who has sort of grown up on the product and engineering side is that sales is reality. You know, he proudly calls his companies sales-led companies.

1:32:37And I think in my early days as CFO and CPO, I just spent a lot of my time on all the cool stuff we could do. And then you go sit in the room with the customer and what they want to talk about is completely different. And the reality of whether or not your product is what they want it to be is right there in the room. You're talking to them about, will you pay more for this thing? Will you buy this thing? Will you move off of XYZ to us? And so I think his approach to being very sales-led leads to a customer centricity that I really like. Are there any unpopular opinions that you hold or anything that get espoused as Silicon Valley truisms that you find to be untrue?

1:33:25I don't know if this is an unpopular opinion or not, but I feel like VCs get a hard time. and take all the politics out of it. You know, just, you know, the venture industry sometimes gets, you know, people talk about how the venture is dead and, you know, it's just a reverse auction now in terms of pricing and no one has pricing power and blah, blah, blah, and so on. And that everybody knows the hot deals and there's like romantic notion of, you know, how Oracle was discovered by being in the same office park as Sequoia or something. I can't remember exactly what it was, but Larry Ellison was working nearby and that's how Oracle got funded.

1:34:03You look at a place like China where you had this venture boom in 2018 and you had 50 ,000 startups backed or something like that. I think this year it'll be maybe below 1 ,000. It's literally gone down by 98 plus percent. And so just the culture of being enthusiastic to back entrepreneurs, I would rather have a system where companies are systematically overvalued rather than undervalued. And so I feel like there's sometimes this notion of the persnickety longtime VC saying, oh, valuations these days. And I actually remember saying the same thing in like 2013, right? When valuations were a fraction of what they are today.

1:34:54And so I think the venture industry is just one of the most core engines of growth for the US, for the world. So maybe it's an uncommon view, maybe not unpopular. Yeah. I mean, I think I agree wholehearted with you. I remember there was a very prominent VC that publicly retired on Twitter. This is now three or four years ago. And the person said, like valuations are just too out of hand and I'm stepping away. It's just not enjoyable anymore. And it was funny if you deconstructed the investments they made implicitly, this person probably made$500 million personally off of low valuations. And I was like, of all the times you should be investing.

1:35:48And maybe they were even right with the time. But I was like, in some ways, you could say like took advantage of entrepreneurs at lower valuations back then. It's almost like you owe a service to continue playing it through, not taking your ball and going home uh especially since i think it was the person's personal capital with it yeah i i think there's been this meme uh that i think was largely true 10 plus years ago that probably some high percentage of vcs were a little tone deaf and uh and and the the founders came to them and they were masters of the universe because they had access to the purse strings of capital and I think that whatever that ratio is, is a far smaller percentage than it is today of people just waking up and trying to do their best in founders.

1:36:41And it's certainly a lucrative business if done well. And so no one should feel particularly bad for any venture capitalists out there. You don't need to start a hug your venture capitalists campaign or anything. But yeah, it does feel like it's, the meme has kind of outstripped the reality at this point, which I think is a good thing. I don't know. All the jokes led to self-awareness, which I think are cancellations, in this case, yes, exactly. And that's true in any industry, for sure. And so it's not a sort of, like you say, a blanket hug for every venture capitalist. But I just think that we sometimes deride the industry when, in fact, I really think it is something that makes the world great in many ways.

1:37:25just I remember probably 2010 or 2011 you know one of the VCs I look up to the most Jeremy Levine who I'm sure you know from Bessemer. Yeah did a podcast with him Oh nice I should listen to that one he's probably saying much more interesting stuff than I am but he I think we were at Balthazar just up the street and I remember him talking about Toby from Shopify just talking about how So he had this founder who every board meeting just surprised him in a new way. And this would have been around the time that Toby was getting started with Shopify payments and plugging payment systems into Shopify.

1:38:06And Shopify went public, I don't know, maybe a few years later at a valuation of one point something billion, maybe lower. It was either one one or nine hundred. It was something like that. Something like that. Right. And now it's a, you know, whatever, a hundred billion dollar company. It's amazing. And so, you know, to that end, valuations needed to rise. Yes. Yeah, for sure. For sure. What, I guess, as we wrap, what keeps you motivated today? Well, for one, it's still fun. And that's the most important thing. Just, you know, does it get you out of bed every day? I think a lot of people who I've talked to in my career, you know, predicted I would have founded a company by now, but I love what I'm doing at Stripe.

1:38:49and you know there's just always something new um back in 2020 uh we were about a year and a half into straight billing but it's still a bit of a you know a toy project with a handful of people on the side and uh we were talking to atlassian uh and you know atlassian uh amazing company uh you know coupled together uh in some ways you have a bunch of acquisitions almost like a a holding company for a bunch of different SaaS services, doing an amazing job of cross-sell, cross-subsidy, and so on. As you can imagine, their commerce system was tricky because they're billing for Bitbucket in this way and Jira in this way and Confluence in this way.

1:39:30And we thought to ourselves, what better way to test billing and its future applicability than to take them on as a customer? And so if I just think about the evolution from that moment in 2024 to today, working with RSIs and GSIs on selling the billing integrations to hundreds of thousands of merchants. It's a fun four years. And I just see many more in the future for Stripe. I think people talk a lot about AI as a disrupting technology. They also talk about a lot as a sustaining technology. As you noted a little while ago, we have some of those interesting data in the world. And I think there's a lot of ways we can leverage that data to advance the mission of growing the GDP of the internet.

1:40:22Capital, for example, Stripe Capital, we're finding just that when you can get a small business alone, it actually helps them grow and grow sustainably, which is sort of amazing. So Stripe Connect, as I mentioned, is our product for a complex money movement for marketplaces and platforms. One place we're seeing just really amazing growth is our Connect users providing loans to their users via our programmatic Capital API. And so I'm really excited to see what foundation models, and I guess LLMs is more of a, everyone thinks of chatbots. It actually is really just about autoregressive transformers, but what they can do for underwriting and can we 10X the applicability of credit.

1:41:20So I just find And, you know, there's a lot more fun to be had and looking forward to, you know, the next 10 years. Yeah. Well, good. Well, thanks for doing this. We did a full podcast and no question of when you guys are going public. Appreciate that. Yeah. Thanks, Logan.

1:41:41Thank you for joining this episode of The Logan Bartlett Show with President of Stripe, Will Gabrick. If you enjoyed this conversation, we'd really appreciate it. if you press subscribe on whatever podcast platform you're listening to us on, as well as share with anyone else that you think might find it interesting. We look forward to seeing you back here with another great guest on the Logan Bartlett Show next week. Have a great weekend, everyone.