š§š Inside Stripe: Stablecoins, AI, and (not) Going Public with Will Gaybrick, President of Technology and Business at Stripe
How AI will change commerce, building a payments foundation model, how Stripe builds products internally, lessons from Thrive Fund 1
Will Gaybrick joined Stripe as CFO after investing in the company as an investor at Thrive Capital. Over the past 10 years, heās run teams and business units across the entire organization, spanning product, technology, business, and finance.
This conversation explores how Stripe thinks about building new products, how AI will change commerce and payments, ways Stripe uses AI internally, the interesting data theyāre seeing around AI-native companies, and how Stripe thinks about going public.
Some pieces I thought were interesting:
Stripe deploys to production over 1,400x per day.
Over 50% of code at Stripe is written by LLMās.
The median Top 100 AI company on Stripe sells into 55 countries. Lovable, which was the fastest company ever to $100 million in revenue on Stripe (six months) sells into over 150 markets.
The median Top 100 AI company on Stripe is reaching $10 million in revenue in 10 months. If you look at the top 100 SaaS companies on Stripe, the median reached 10 million in revenue in three years. So theyāre growing 3.5x faster.
Stripe is seeing higher response rates on sales emails that are AI generated.
Half of the Fortune 100 uses Stripe. Stripe processes $1.4 trillion of global payment volume per year, or about 1-2% of global GDP.
Stripe has always been profitable, and has never burned a dollar of investor capital.
Support this Episodeās Sponsors
Numeral: The end-to-end platform for sales tax and compliance. Try it here.
Amplitude: Listen to your users at scale with AI.ā Try Amplitudeās new AI Feedback.
To inquire about sponsoring future episodes, click here.
š Stream on Spotify and Apple
Timestamps to jump in:
2:42 Willās new job
5:29 Build vs Buy in AI
6:39 Inside the Bridge acquisition
8:54 Stripeās stablecoin strategy
11:20 Why building Stripe is so complicated
13:22 How Stripe builds new products
18:29 AI companies growing 3.5x faster than SaaS
22:58 New fraud vectors in AI businesses
25:19 Agentic commerce in ChatGPT
29:34 Building modular products
34:05 How Stripe uses AI internally
42:12 Building the first payments foundation model
48:23 Link, Stripeās 200M MAU consumer product
56:52 Will Stripe ever IPO?
59:14 Blurring of private and public companies
1:03:39 Starting Hack Yale
1:08:23 Joining Thriveās $5 million Fund 1
1:12:30 Low margin businesses are underrated
1:16:30 Joining Stripe as CFO
1:21:59 How Stripeās go-to-market has evolved
1:25:56 Stripeās margins
1:29:52 Why financial services are so hard to use
1:33:04 Lessons from Alan Mulally
Referenced:
Find Will on X / Twitter and LinkedIn
Related Episodes
š Stream on YouTube, Spotify, and Apple
Transcript
Find transcripts of all prior episodes here.
Turner Novak:
Will, welcome to the show.
Will Gaybrick:
Thanks for having me. Excited to be here.
Turner Novak:
So I think a good place to start, you have a very specific title at Stripe. Can you just explain what it is, and what you do at Stripe?
Will Gaybrick:
Yeah, yeah. So president of technology and business, agree, pretty specific. We already have a president, thatās John Collison, and so weāve given me a somewhat more specific title to be more descriptive of what I actually oversee, which is technology and business. Broadly, that means I manage product engineering, so across all of our product development areas and infrastructure security, things like that. And on the business side, teams like partnerships, risk, GTM operations, core company operations, support. JD is deliver the business and deliver the roadmap, and thatās what Iām focused on every day.
Turner Novak:
Make the chart go up.
Will Gaybrick:
Exactly. Charts.
Turner Novak:
Chart charts. Yeah, many charts. But used to be, instead of technology, it was product.
Will Gaybrick:
Yes.
Turner Novak:
That was a change a couple months ago. Well, whatās the significance of the change? Did you do more product related to things before?
Will Gaybrick:
Well, still spending just as much time on product. I guess more hours per day on Stripe total, if thatās even possible, but we decided that it would be more efficient if we brought some of the infrastructure teams closer to product teams. It wasnāt like there was some sort of huge disconnect between how they were executing, but weāre always looking for ways to create new efficiencies in how we run the company.
Turner Novak:
And is that infrastructure related to product moving faster, being able to bring the product closer to the rails of financial services kind of a thing?
Will Gaybrick:
Well, thereās sort of several layers of infrastructure at Stripe. There is cloud infrastructure and the services that product teams use to build products. There is sort of product infrastructure, like you said, rails, integrations with banks, integrations with payment networks and so on. And Stripeās... one of our number one selling points and top focus areas, will always be a top focus area, is reliability and security. And so, infrastructure focuses basically on that. And then, accelerating product development.
Turner Novak:
How do you accelerate product development with better infrastructure?
Will Gaybrick:
So, in so many ways. We spent a lot of time on developer productivity. So one of the principles that we think about a lot at Stripe is, what are the non-speculative, forever compounding investment areas? And I think developer productivity is one of those. You make your developers happier, more productive, thatās just going to make you a better place for engineers to work every year and just make your engineers more effective. And so, we are constantly thinking about new ways to empower developers. Built something cool recently called Minions, which is basically using large language models to automate a lot of the run load thatās coming to our product teams. So you can actually sort of click a button in a Jira ticket and just a minion will go off and fix the thing, or update the thing. Youāll see the PR, click a button to approve, and thatās actually getting a ton of use internally.
Turner Novak:
Interesting. Is this something that you built internally, or is an external product that you use?
Will Gaybrick:
Built internally, yeah.
Turner Novak:
Interesting. So whyād you decide to build that internally versus buy it from somewhere? Was there nothing like that on the market that existed or?
Will Gaybrick:
Thereās probably things like it. I canāt think of any services off the top of my head that do exactly what it does, but in a lot of cases, build versus buy, for us, is dictated by just how strategic something is to the long-term arc of the company, and creating value for users, and for developer productivity, we often find that our systems are so specific that sort of building on top of them is a lot more efficient than trying to bolt things on.
Turner Novak:
Interesting. So yeah, I think you actually kind of broke the internet maybe with the Bridge acquisition. I forget when this was. It was maybe a year ago, maybe eight months ago.
Will Gaybrick:
Yeah, 10 months ago, yeah.
Turner Novak:
10 months ago? So I was in the middle, or right around it. So then, how did you approach deciding that you should acquire versus start to build that internally?
Will Gaybrick:
Yeah. So I think with acquisitions, the adage that you donāt acquire companies, you acquire founders and founding teams has always run true, to me, and to us at Stripe. We had highly convergent roadmaps. They were building things sort of right as far as we could assess, and we just thought Zach and Sean were extraordinary founders, and we could go faster by working with them.
Turner Novak:
How do you convince such capable founders that are probably... they could probably do it without you, all right, in theory? Howād you convince them to join?
Will Gaybrick:
Itās sort of interesting. I think their calculus was very similar to ours. We can go faster with Stripe than going it alone. When we announced the acquisition, it was almost like the gun at a starting line for the stablecoin industry. There was already a lot of TVL, or a lot of stablecoins in circulation at the time, but in terms of the idea of bringing stablecoins mainstream, it was still just sort of an idea. And I think one of the things that Zach and Sean would tell you is, as soon as we announced the acquisition, the inbound was sort of DDoSing their very small sales team and just creating that demand, seeing the new use cases has been really powerful in driving our roadmap, their roadmap. And I think, also, just a lot of the competencies that Stripe has around navigating global regulatory landscape, working with banks, existing partnerships and things like that was attractive to them.
Turner Novak:
Yeah, because an interesting just component of crypto stablecoins as a whole, is you have to interact with the fiat system. As much as you want to say we donāt need it, in its own world, which it is, itās like its own economy, but also, the real world does exist, and youāve got to interact with it.
Will Gaybrick:
Totally.
Turner Novak:
So can you actually just real quick explain what stablecoins are for somebody who doesnāt actually know, and also for my benefit, because I feel like youāre probably... know a hundred times more than I do. How would you explain to someone who doesnāt know, but is really smart?
Will Gaybrick:
Yeah. Theyāre digital assets on chain that are backed by real-world fiat assets. Not real-world, but fiat assets. Obviously, today the most popular stablecoins are backed by USD. Typically, these are custodied at major financial institutions, and theyāre US treasuries or other liquid instruments. What this basically means is that stablecoins allow you to start thinking about running your business on digital assets.
Turner Novak:
Why donāt you just use USD? Why do you need to use a stablecoin in the first place? Is it because the underlying currency in certain markets are just as volatile as Bitcoin?
Will Gaybrick:
Itās actually not really about volatility so much as how money moves globally. So Stripeās mission is to grow the GDP of the internet. What that means tactically is that weāre very often just trying to make money and financial services work the way that we all would expect them to. So you would think, oh, well, itās easy using just electricity and computers to move money all over the world. It really isnāt, because just the global financial system is so fragmented, and so you are thinking about stitching together all these disparate rails. Youāre thinking about correspondent banking, and moving money is often very expensive. Wires cost dozens of dollars, or are very slow, and so we have always been extremely excited about crypto, less because of the opportunity to make a quick buck on trading across coins, and more about the underlying financial platform that allows you to move money globally via just one protocol that all developers can sort of... Itās like a shelling point for developers.
Itās, here, if we all just use Solana, or if we all use ETH and thereās stablecoins sitting on top of both, then money, you can just move seamlessly.
Turner Novak:
How complicated is it under the surface running the sort of financial network that Stripe runs, on a scale of 1 to 100, or 1 to 1000? What would be surprising about it to people that theyāre just like, wow, thatās a lot more complex than I wouldāve realized?
Will Gaybrick:
Well, on Stripe, you can use a single API to get access to... on the order of 150 payment methods. If you were trying to integrate each of those individually, it would take many hundreds of person years, thousands of person years, maybe even more. And so, finding a way to normalize the underlying components such that we can create that unified experience on the front end, thatās pretty hard, and itās been something weāve been investing in now for 14 years. One of the things that Iāve always found interesting is that a lot of times youāll find in the early days that users will say, āHey, we would never do X.ā So we heard from a lot of enterprise users, āWe would never give up the core interface of our checkouts. Thatās too important, thatās our IP. We have to own that experience every pixel.ā
Because weāve been able to, via a single API, allow you to get access to these 150 payment methods, we now find that 72% of enterprise customers going live on Stripe delegate the front end to us. Itās all brandable, they can express their brand, and itās all configurable, but thereās just so much complexity behind these payment integrations that theyāre sort of happy for us to take that load off of them.
Turner Novak:
Interesting. Okay. One maybe slightly different question, but maybe sort of related, deciding that stablecoins was worth going after, how do you at Stripe build conviction around knowing what to do? Maybe this is Will personally, maybe this is Stripe as a team, what goes into those decisions?
Will Gaybrick:
Users. You are on this infinite journey of discovering customer needs. I am sort of in awe of this because it truly has guided every epoch of Stripe. In the early days when Stripe launched, it was a single API to start accepting payments nearly instantly. And the state-of-the-art back in the late-2000s was, you would have to work with banks, or legacy merchant acquirers. It would take days to get going. And so, Patrick and John, years before I joined the company, with the sort of founding team, sort of solve this problem by just saying, āHey, how can we make this instant? Developers expect instantaneous self-serve experiences. How can we enable that?ā We then started to see the on-demand economy, and the platform economy emerge.
Turner Novak:
So this was like Instacart, Uber?
Will Gaybrick:
Exactly. Thatās sort of the on-demand side. And then the platform economy, you had companies like Shopify, or Squarespace, Wix, commerce.
Turner Novak:
Which was enabling mass long tail of new businesses that you probably wouldnāt... Stripe wouldnāt go in and acquire the seller on Etsy thatās knitting some scarves and they make a couple thousand dollars a year?
Will Gaybrick:
Yes, itās certainly not a core competency to go serve that user directly, but we can be the infrastructure backing the platforms that do that. And so, as we started to see that emerge, and it gave rise to what we call Stripe Connect, which is infrastructure for multi-party money movement, and allowing platforms to embed payments, create their own payment solutions. Then in the late 20-teens, you started to see the rise of FinTech. Suddenly everyone to build a consumer or merchant financial services platform. So we built products like Stripe Treasury, Issuing, to give them the core building blocks to build their own financial services. Then, most recently, you see this boom and customers looking for stablecoins and global money movement infrastructure, and then of course all the AI companies, and they have very particular and interesting needs. One of those is just serving customers globally, because most AI companies are providing digital goods that can work across borders.
Turner Novak:
Which probably unique compared to some of these other previous waves.
Will Gaybrick:
Exactly, yeah. In most cases, making money work well in a single market is actually not that hard, but as soon as you cross borders, just everything gets a whole lot more complex.
Turner Novak:
Is it just regulations, laws, uniqueness, everythingās different kind of a thing?
Will Gaybrick:
Exactly. And money tends to move slower, and passes through several financial institutions, and by being a unified platform, money movement, we can speed it up. We can take the complexity off our userās plates and so on. But yeah, the median AI company, the median AI company among the top 100 on Stripe sells into 55 countries. But when you look at companies like Lovable, which is a vibe coding platform, actually the fastest company ever to $100 million in revenue on Stripe. It took them six months. It was pretty amazing. They sell into over 150 markets today, and for them, they just want to create a seamless experience for their customers. Every payment method that customers want in markets from Japan, to Australia, to anywhere in Europe. And then, for the long tail, this is where AI and stablecoins come together, you have places like Vietnam or Argentina where the preferred payment methods may actually be paying directly to stablecoins.
Turner Novak:
Do think, do these people ultimately want U.S. dollars, probably, and the stablecoins is the way that you go from Vietnamese dong, or Argentinian peso, I hope I got that right, into being able to converge to U.S. dollars; is that ultimately what stablecoins get at?
Will Gaybrick:
Yeah, Iām guessing the sort of need behind the need is just a stable currency, and one that they can use with any counterparty because they will accept it. But what that amounts to typically is USD.
Turner Novak:
I have a lot of friends who theyāre in Brazil, or what you maybe consider an emerging market, but they kind of have the institutional knowledge, or they behave just like your friend in San Francisco or New York, but theyāre kind of trapped in, my wages are in Nigerian naira or something, and the government is deflating the currency, or I guess inflating the currency a hundred percent every year and I canāt use it. Thereās that desire to get out. What other interesting thing are you seeing in the data on all the AI stuff thatās happening on Stripe?
Will Gaybrick:
So the companies are growing incredibly quickly. We look at that top 100. Again, the median company among top 100 AI companies on Stripe is reaching $10 million in revenue in 10 months. If you look at the top 100 SaaS companies on Stripe, the median reached 10 million in revenue in three years. So theyāre just monetizing faster.
Turner Novak:
So itās really like a little over three times as fast, three-and-a-half times as fast.
Will Gaybrick:
So just monetizing incredibly quickly.
Turner Novak:
Do you know what it is? What are they monetizing faster? Is it because thereās this self-serve, easier to get value quickly type of thing?
Will Gaybrick:
Yeah. Listen, in many ways, just LLMs is almost like just this alien technology landing on earth, being able to create experiences that you just never thought were possible, and weāre seeing a lot of entrepreneurs leveraging that to create new types of software and new types of products that we all find very, very useful. Itās interesting because these companies have a bunch of new needs.
Turner Novak:
What are you seeing specifically?
Will Gaybrick:
So inference is expensive, and so with SaaS, you could have just subscription-based pricing models, super high margins.
Turner Novak:
And itās pretty simple, too, just right. Me and Will are both going to sign up two seats, itās 20 bucks a month or whatever.
Will Gaybrick:
Exactly. Whereas the most popular business model weāre finding for this new cohort of AI companies is this hybrid business model where you have a SaaS subscription. Say youāre buying a hundred dollars a month of whatever service, Lovable, Cursor, so on, and then that SaaS subscription gives you an entitlement to a certain amount of usage. You burn down that entitlement. In many cases, you go over it. And so, you need to think about how you calculate overages. And so, this hybrid subscription and usage model, and then you think about adding new products and how that sort of factors into your existing entitlements and your existing subscriptions. And so, the component torques of commerce complexity here is really interesting. So what weāre seeing and what weāre focused on building is... Maybe give a bit of background. Stripe Billing is one of our major product areas. Over half of startups going live on Stripe use Stripe Billing, and Stripe Billing is sort of a layer on top of the core payments infrastructure used to orchestrate payments to model your business.
Turner Novak:
So it says if you have fixed or any kind of customer relationship, itās like how you know what to charge them, essentially.
Will Gaybrick:
Exactly. Subscriptions, invoicing, how you model your product catalog, how you assess taxes, how you do revenue recognition, all of that. And this has always been an impediment to just customer growth. One of the adages we have internally is that every company has a billing system, and very few companies are enthusiastic about their billing system. So with billing, weāre really focused on changing that.
Turner Novak:
So they historically use a different billing system than Stripe? Was it like Google Sheets, usually, or?
Will Gaybrick:
The real competitor is the in-house build. Basically, everyone ends up building this whole board of infrastructure. Stripe itself has this whole board of infrastructure, a big team on it. We are actually migrating onto billing ourselves service by service by service.
Turner Novak:
Thatās very Meta, the billing infrastructure for the billing infrastructure company.
Will Gaybrick:
Exactly. Yeah, and itās interesting, you can see the billing revenue for Stripe Billing in the Stripe dashboard using billing. But you really do see with these AI companies the complexity of their business models because those usage, subscriptions, hybrid is slowing them down. So browser base on Stripe uses billing and usage-based billing, and by experimenting with things like free trials and discounts, they were actually able to increase their growth by about 17% in relatively short order. And itās just giving you that business model agility, is something that these AI companies are looking for. Another related point on the note of free trials is youāre seeing new fraud vectors.
Turner Novak:
Yeah, I was going to say.
Will Gaybrick:
Because you obviously want to enable your customers to go live quickly, but there is a cost. Unlike SaaS, thereās a cost to people using your product and then disappearing. And so, we have Stripe Radar, which has always been focused on mitigating fraud at the time of transaction. Now weāre looking up-funnel with users and saying, āHey, is this customer coming to you before they even process a payment? Are they fraudulent? Are they likely to convert to a paying customer?ā Things like that.
Turner Novak:
Are there any common things that that is currently capturing now? Like common AI scams, I guess we could say, or most common that youāve seen of ways people are making a quick buck?
Will Gaybrick:
We see fraud rings where fraudsters are creating a ton of free trial accounts and then selling them. So itās, hereās an account with this service, it gives you this much usage, and weāre packaging these all up. And here now you can test a bunch of your AI workloads on-
Turner Novak:
Oh, interesting.
Will Gaybrick:
... Shadeform, or Cerebras, or any of these types of companies. Not speaking about any of them in particular, but that type of business model where youāre hosting models. I think weāre seeing a lot of fraudsters basically aggregating usage models and then selling it.
Turner Novak:
Interesting. So it would basically be, if you were to go and sign up, you hit the limit and you have to start paying. But we could go and say, āOh, wait a second, this person has a hundred or a thousand-ā
Will Gaybrick:
Yes.
Turner Novak:
ā... extra additional free trials that we can just tap into.ā
Will Gaybrick:
Exactly. Yeah.
Turner Novak:
Interesting.
Will Gaybrick:
Yeahā.
Turner Novak:
And you can see how they would lose a lot of money on that pretty quick.
Will Gaybrick:
Exactly. Yeah, yeah. Yeah, the four major areas weāre focused on in AI at Stripe are, one, building economic infrastructure for AI. So this is everything weāre just talking about. The commercial infrastructure to implement AI business models, these hybrid models, subscriptions and usage. Two is the infrastructure to enable agentic commerce. I think everyone is really excited about this.
Turner Novak:
Yeah, you guys just did something. Whatās the thing that you just announced?
Will Gaybrick:
So a couple weeks back, we announced that we were partnering with OpenAI on ChatGPT Instant Checkout, which for anyone who hasnāt used it, it is a great experience.
Turner Novak:
I should have actually tried to use. I have not used it yet. I should have before we did this, but-
Will Gaybrick:
Yeah, yeah, yeah. No, itās fantastic. Incidentally, the day before we announced it, I was on Etsy buying a new dog collar for my dog. And the next day, the demo was shopping in Etsy in ChatGPT. And theyāre both great experiences, but I was just struck by how I really do think this modality of the iterative shopping within a model is going to work.
Thereās a lot of hubbub and promise to social commerce, and itās had some mixed success. I think TikTok is having some good success there. But I really do think chat-based commerce mediated by LLMs is going to be powerful. So we announced Instant Checkout with ChatGPT.
And the other thing we announced is what we call the Agentic Commerce Protocol, which is a relatively thin protocol, but just a way of standardizing how agents should interact with checkouts. So checkouts have a lot of logic built into them. Which payment methods can you use? What discounts are there? How are shipping rates calculated based on basket size? Is this fraudulent or not?
And so, we had done a lot of experimentation with pure browser-based automation, and we found that the technology just isnāt there yet. And we thought the world would really benefit from an open protocol. So itās not just a Stripe thing, but an open protocol to just enable the ecosystem to grow a whole lot faster.
Turner Novak:
Is this the shared payment token? Is that another word for it or...
Will Gaybrick:
Yeah, thatās how itās implemented on Stripe.
Turner Novak:
Okay.
Will Gaybrick:
And basically, a shared payment token is a way of securely passing credentials between the point of purchase-
Turner Novak:
So this would be Etsy or-
Will Gaybrick:
Exactly. So passing it from the point of purchase, which is, say, ChatGPT, to Etsy. And the point is that you donāt actually have to expose the underlying credentials to the merchant.
Turner Novak:
Okay. And I know there was a specific reason that you did that, which I feel like I should know. What was the reason that you did that? Is it so that ChatGPT is not in the flow of funds or is in the flow of funds? Iām not sure.
Will Gaybrick:
Well, I think in a lot of cases when weāre designing these things, weāre thinking not only just how do we very proximately solve a problem for just this use case, but also, over time, whatās good for the world? And we felt like the idea of agents just slinging underlying payment credentials to every merchant on the internet didnāt-
Turner Novak:
It could be dangerous.
Will Gaybrick:
Yeah, it didnāt really pass security and privacy muster. And so thatās why we designed it to allow for this secure processing, where you can collect the credentials in the chat interface and then use them to process without having to vault them with all the merchants.
Turner Novak:
So Stripe doesnāt actually own the payment then with the underlying merchant, right? Is that fair?
Will Gaybrick:
Yeah. Again, itās open protocol. You donāt have to build Agentic Commerce Protocol on top of Stripe. We donāt have to process the payment. You can use Stripe for it, and we make it very turnkey and seamless. But we were pretty inspired by MCP, Model Context Protocol, that Anthropic released, I guess it was last year, from the standpoint of it being a very simple protocol, something that everyone could get behind. And again, creates sort of this shelling point for how weāre all going to work together on what we think is a very big opportunity across the industry.
Turner Novak:
And thatās sort of how you think about product at Stripe too, is very modular. To your point, you donāt have to use billing.
Will Gaybrick:
Yes.
Turner Novak:
You probably donāt have to use Stripe Issuing. You could use something else if you want. Why do you do that? If Iām a company, come to Stripe, Iām a startup, give me everything Stripe. Why is it so important to be modular?
Will Gaybrick:
So Iād say this is a big inflection point in our thinking about how we develop products at Stripe. There was a time in which we thought, āHey, itās going to be the best experience if we are relatively closed, and Apple-like, and just everything will be tightly stitched together.ā
And then we just learned from our users that thatās not what they wanted. I remember a conversation I had with one of the largest retailers in Europe when I was in Paris. And they were looking for help in decreasing the size of their payments team, because they had a couple hundred people on it.
Turner Novak:
On a retailer in France.
Will Gaybrick:
Yes. And they were...
Turner Novak:
Thatās like a Stripe at a Series D stage, right? A couple hundred people...
Will Gaybrick:
Yeah, exactly. And just on their payments.
Turner Novak:
Yeah, just on the payments team, yeah. They have a like a whole Stripe internally.
Will Gaybrick:
Exactly. And actually, weāve seen this. I think this was what people would build. And one of the things we try to do is say, āHey, listen, we can take a lot of that off your plate.ā
Turner Novak:
Yep.
Will Gaybrick:
And I remember them saying, āWe would love help decreasing the footprint, the number of payments processors we work with, and some help in orchestrating across them.ā And so I asked them, āWell, how many processors do you work with?ā And I expected them to say something like six. They said 83.
Turner Novak:
Okay. So who are 83 payment processors?
Will Gaybrick:
So you have a bunch of big names. Listen, I didnāt ask them for all of the names. We have a bunch of big names who are offering card processing across the world. And then you have a bunch of local players, because theyāre selling all over the world. And so theyāre integrating to local processors in Indonesia, or the Philippines, or Malaysia. And so, navigating this, the logic across all of these is just a huge workload for them.
And so for us, we could either ignore that and say, āHey, what you really want is just to run everything on Stripe.ā I think they would just say ānoā. Or we could say, āHm, interesting. We really think we can help with that.ā And so weāve been leaning into, how do we help you orchestrate your commerce infrastructure?
Like you said, you donāt have to use Stripe payments with billing. You can use billing to run subscriptions, send invoices, and then process them with a third party. And I think weāve just learned, particularly as we go upmarket and work with more enterprises... now more than half of the Fortune 100 uses Stripe... that users want to be met where their systems are today. And thatās what weāre focused on.
Turner Novak:
And they probably have different pace of adoption, different things. There be some companies that are like, āFor the first time, we are accepting payment online.ā
Will Gaybrick:
Yes.
Turner Novak:
And they are not ready for everything else that you can do.
Will Gaybrick:
Yeah, itās interesting because the S&P 500, the average tenure is something like 15, 20 years. And so you could say, āWell, 10 years hence, are all these people just processing only on Stripe or are many of them processing only on Stripe?ā Itās possible.
But we also do see companies growing up on Stripe needing to add other payment processors because our coverage in a given geography isnāt good enough yet, or they have a board mandate for redundancy. We see that that latter one has been waning. Last year we had 99.9996% of requests on our critical APIs succeeding. Itās actually less than one minute of downtime for the entire year.
Turner Novak:
I was going to ask you how much time that was, yeah.
Will Gaybrick:
Yeah, this is pretty interesting because-
Turner Novak:
Was it all at once or was it seconds spread out?
Will Gaybrick:
Yeah, itās actually not hard down for the API overall. Itās just looking at individual failed requests and then sort of imputing a time from that.
But the fascinating side, the other side is weāre actually also deploying about 1400 times a day. So youāre actually pushing to production roughly every minute, and youāre down for less than a minute the whole year. And this has been just a concerted investment in reliability over the years.
Turner Novak:
What are you updating 1400 times a day at Stripe?
Will Gaybrick:
Yeah.
Turner Novak:
Whatās changing?
Will Gaybrick:
Any number of services. And a lot of them are internal services, a lot of them are bug fixes. And then of course, some of them are bigger pushes.
Turner Novak:
So this might be an interesting... When youāre talking about updating bugs, fixing things, pushing so many times, are you guys using AI internally at Stripe? To what extent are you using it?
Will Gaybrick:
Yeah, weāre using it a lot. And weāre using it in targeted ways where we are really focused on getting real value out of it. Thereās the Jim Collins Good to Great framework of, I think itās crawl, walk, run in terms of adopting new technologies. And I guess his framework is, the best companies are first discovering how this can really advance how they operate and create value for the customers. Then they start to scale that and they sort of go all in. There was the HBR paper recently around āworkslopā, I donāt know if you saw that, in September?
Turner Novak:
I donāt think I did. What was it?
Will Gaybrick:
Yeah, itās a study that asserted that 95% of companies are actually getting negative ROI on using AI internally.
Turner Novak:
I saw that headline.
Will Gaybrick:
That headline. Exactly.
Turner Novak:
Itās always funny, thereās these ebbs and flows of, āItās over. AIās done. It was all a scam.ā
Will Gaybrick:
Yes. Yes, yes. Thatās certainly not what we believe. But we really donāt want that to be the case. So weāre not just sort of like, āHey, everybody, you have to use an LLM now for everything.ā And so we have found some very powerful targeted use cases. So I mentioned Minions before, but also just writing code in general. And particularly for our most senior engineers, giving them leverage to go faster, because theyāre often thinking a lot about architecture. And keeping them from needing to write every line of code. Instead, putting them in a position where theyāre sort of...
We think about new grads and one level up as doing a lot of the actual code writing at Stripe. Empirically, they do write the most lines of code at Stripe. And now senior engineers almost have a dedicated team of level ones, level twos at Stripe because of LLMs, which can assist them in actually code creation. And then theyāre almost in a position of code review of the code that theyāve created themselves via an LLM. So depending on how you measure it, you could say more than 50% of code at Stripe is written by LLMs.
Turner Novak:
Mm, interesting.
Will Gaybrick:
Or you could say itās less.
Turner Novak:
Whatās the highest youāve seen? Didnāt Salesforce say like 75 or 90%? What was his number?
Will Gaybrick:
Yeah, I think Google recently quoted, or maybe it was Amazon, 70%. Iām always interested as to how people count this. Because we do, again, a lot of bug fixes and easy deploys via tools like Minions internally. And then we do a lot of code writing, as in edited, by people internally. And so did he write that code? It was more of a collaboration.
Turner Novak:
Yeah. Because a lot of times you just think itās like AI agents just running wild, adding a new field to Salesforce or whatever.
Will Gaybrick:
Yes. Yes.
Turner Novak:
And I feel like when some of the CEOs go on, theyāll do a CNBC clip, and thatās kind of what they make the audience think is happening. But itās still, thereās very much humans that are involved in this.
Will Gaybrick:
Yes, very much humans involved. Thereās sort of a dynamic where LLMs are doing a ton of work supervised by humans. So a good example would be actually risk mitigation. So not just anyone can process payments on Stripe. You need to be a bonafide business. Thereās certain categories of business that are prohibited or deemed highly risky. Those need some stepped-up review with partners, financial institutions, banks.
And so thereās a very long document that if you are a human risk reviewer at Stripe, you need to deeply understand around whether or not a business is supportable for given use cases. LLMs are incredibly good at ingesting rules and then applying them.
So we have seen incredible returns on assessing supportability on Stripe, both in accuracy and in efficiency, by saying, āHey, model, take a look at this website, take a look at all their product pages, take a look at everything thatās said about them on the internet, and let us know how confident you are that this business is supportable.ā And if things are flagged, then humans get involved and can provide that last 10% oversight that can be dispositive.
So thatās been incredibly powerful. The other application that weāre really excited about is augmenting our sales development team, our SDR team. Just the early signs are that outbounding, which is sort of a new motion for Stripe.
So much of our go-to-market motion has been driven by inbound. But outbounding is going to become not just more efficient, but also more effective when we let LLMs go deeply understand a business on the internet, and then craft a message about how Stripe might be useful to that business. And then SDRs get involved to edit that, send it, and then track the lead from there.
Turner Novak:
So when you say that that might be interesting, are you indicating that itās not quite there yet? Because I think everyone listening to this gets those really bad, AI-generated sales emails.
Will Gaybrick:
Yeah, weāre actually seeing so far that they are getting higher response rates.
Turner Novak:
Really. Okay. Is there anything specific that the higher response rates get? Are they shorter emails? Do they use numbers? Do they have misspellings? Do you know?
Will Gaybrick:
I donāt know offhand. I was actually reading some of these last week.
Turner Novak:
Oh, really? Okay. What seems to be the general vibe of these good AI emails?
Will Gaybrick:
They seem very personalized. So they seem like, āI really deeply understand what youāre doing.ā And again, I think this is because theyāve been able to go leverage the base model that OpenAI, or Anthropic or other has created that has sort of compressed the internet.
And so they can take that compression, go look at a website, and come back with a lot of really interesting facts about that company that would take a long time for just a human to aggregate.
Turner Novak:
Yeah. It might be interesting in non-tech related or even non-email denominated jobs.
Will Gaybrick:
Yeah.
Turner Novak:
So if you run a manufacturing firm or something and you get AI-generated, really good email about some manufacturing-related software or something. Iām actually... Maybe this is a terrible example, because maybe they get a ton of emails. But when I think of mine as a VC, Iām constantly getting emails, and a lot of them are AI generated. So I wonder if thereās in those less email-denominated industries or something where thereās... Or even like... Actually, I think this might be illegal, but texting. I actually donāt think you can text people. Thereās a lot of rules on that. But even-
Will Gaybrick:
Yeah, I donāt know.
Turner Novak:
... LinkedIn, I think thereās a lot of LinkedIn automation that people do.
Will Gaybrick:
Yeah, I have anecdotally seen a lot more inbound texts. So, illegal or not, it seems to be happening.
Turner Novak:
I get a lot from loan companies. I think itās because if you have a credit card balance, you show up on a list and theyāre like, āYou qualify for our personal loan, like $63,000. Consolidate your debt.ā Iām not interested. Stop calling me eight times a day.
Will Gaybrick:
Yeah, yeah. I get a lot of the fixed-term employment offers.
Turner Novak:
Oh, thatād be nice. Yeah. Like, āHey, we heard youāre looking for a job.ā
Will Gaybrick:
Yes. Exactly.
Turner Novak:
Do you ever get those? I get a lot from Indeed recruiters. Like, āWe found your resume on Indeed or LinkedIn, and Iām Patricia from Geico,ā or something, āand weāre hiring remote workers.ā
Will Gaybrick:
Yes. Yes, yes.
Turner Novak:
Iāve actually heard that there is some shady stuff that goes on with that. They pitch you as like youāre running payroll for a company, but youāre actually almost like a drug money mule in a way, where youāre facilitating those funds. Itās actually a pretty common tactic that a lot of people use for that, which is... Does that show up on Stripe at all?
Will Gaybrick:
Not that I know of offhand. But broadly, some of the best advice that I ever received, and I think should be as a company have ever received, came from Robin Vince. Heās the CEO of BNY Mellon. At the time he was the CRO. Maybe this actually was back when he was the treasurer at Goldman Sachs. And I had recently joined Stripe. I joined as a CFO.
And I asked him, among other things, about being a CFO. Like, āWhat is the single most important thing that you think Stripe needs to get right?ā This is way back in 2016. And he said, āYou all will need to become conspicuously good at risk management.ā And it always stuck out to me as plausibly the case, but I wasnāt so sure why.
Turner Novak:
Yeah, because you probably think, āWe need to build a beautiful product.ā
Will Gaybrick:
Exactly.
Turner Novak:
That thatās the most important thing.
Will Gaybrick:
Exactly. And I think, now Stripe is processing a lot more than...
Turner Novak:
Yeah, whatās the last public number?
Will Gaybrick:
Actually, I guess the last public number is 1.3 trillion or 1.4 trillion of payment volume per year. So youāre between one and 2% of global GDP. And so at this scale, youāre becoming an actor that is systemically large.
Turner Novak:
Yeah. Someone takes you down, 2% of the economy doesnāt run.
Will Gaybrick:
Exactly. Yeah. And so, the investment in reliability is so critical. Investment security is so critical. And also, you become a huge target for not just fraud risk, but you mentioned a second ago scams or transaction laundering. Youāll see a lot of times on Stripe what looks like a very bonafide merchant. Hereās the domain, and you can actually place purchases. Itās real-looking products. And it is in fact sort of a mule for you to buy guns on a different website.
Turner Novak:
Oh, interesting.
Will Gaybrick:
You go to the other website, you see what the gun costs. You buy a product that costs the same amount on here.
Turner Novak:
Like youāre buying a makeup kit, or something completely unrelated.
Will Gaybrick:
Exactly. Exactly. But this is actually where AI is, again, extremely powerful. Weāve always had a bunch of ML models in production. About 150 models in production at Stripe. And some of these are DNNs, deep neural nets. Some of the more sophisticated ones are.
But we launched mid-year this year, the first ever Payments Foundation Model. And basically what the Payments Foundation Model is, is these giant vectorized embeddings where you take just a ton of Stripe data, you tokenize it, you throw it against the Transformers, and you then end up with this clustering.
Itās a encoder-only model. So youāre not actually saying that weāre then going to decode it and make a lot of extrapolations about exactly what the data means. But weāre able to see the clustering of different merchants, and these transaction-laundering merchants just show up together in the giant embedding space.
And so, itās been fascinating to see. I think one of the great discoveries or revelations the past few years in AI has been that you take less opinionated model architecture, a ton of compute, a ton of data, and that tends to work better than the really minute architectural thinking around models.
Now, thatās only partially true because thereās still a lot around model architecture that has been the unlocks. Things like inference-time compute. But what weāve seen is that we had a lot of hand-curated features in our models in the past. And by just creating a really big model with all of our data, weāre able to mitigate risk a whole lot better than we were able to by using our intuition around how you would discover these companies.
Turner Novak:
And itās just because you keep getting a bigger data set or more tests, and then it solves itself? Is that how it works?
Will Gaybrick:
Yeah, itās sort of the magic of DNNs, they discover their own features. And so the features that you may intuitively think will lead you to whether or not a customer is laundering firearms transactions through what looks like a kidsā toys website, may not be the most telling features. And a giant foundation model can discover those.
Turner Novak:
Interesting. So you said you donāt really use it necessarily, but you made it to see things. Is that a fair way to describe it?
Will Gaybrick:
We do use it. Yeah, yeah.
Turner Novak:
Oh, you do use it. Okay.
Will Gaybrick:
Yeah. Yeah. And primarily for risk mitigation.
Turner Novak:
Okay.
Will Gaybrick:
So card testing. Again, the card testers show up as a cluster, or transaction launderers show up as a cluster.
Turner Novak:
Whatās a card tester? Is that like youāre testing to see if the card will work-
Will Gaybrick:
Yes.
Turner Novak:
... if you want to just run a bunch of money through it?
Will Gaybrick:
Youāve stolen a ton of cards or other payment credentials. You want to resell them. You need to know which ones you are actually usable. Maybe you want to use them yourself. Yeah.
Turner Novak:
Yeah. I think the surprising thing, I had a guest probably about a year or two ago, her name was Natasha. The company is called Cable. They help just financial institutions fight crime, basically. Financial crime. And itās like a couple percentage points of GDP is just like people-
Will Gaybrick:
Totally.
Turner Novak:
... doing crime, which is fraud, drugs, other even more horrible things. So itās pretty interesting just how big that world is. Itās kind of scary.
Will Gaybrick:
Yeah. Yeah, one of the pillars of our company strategy for the past three years running is to become guardians of the global financial ecosystem. And that has meant by turns, getting even better at mitigating risk on Stripe while creating a surprisingly great user experience. And on the back end, how do we create tools and platforms that we can actually sell to other financial institutions or to our users to mitigate risk on their side?
Turner Novak:
I think you were pretty involved in this product. Itās connected to ChatGPT right now and how you do payments is called Link. Whatās the story with Link? What is it and whatās the scale of it now at this point?
Will Gaybrick:
Itās our only consumer product, and itās effectively a very thin layer that we created to increase conversion and broadly improve online payments. It now has over 200 million consumers in the network. Itās growing very, very quickly. And it was predicated upon the belief and now I think confirmed belief that we could create better payments experiences by having the ability to set this little cookie with consumers and have a thin relationship with them than we could otherwise. So for example, our customers have for Stripeās entire existence wanted bank account driven payments.
Turner Novak:
So ACH you mean?
Will Gaybrick:
ACH, SEPA in Europe, Bacs in other parts of the world, in Australia and New Zealand and so on, or Bacs I guess in the UK and the Bacs in Australia and New Zealand. And the problem is bank account payments are... They donāt have a lot of the magic of credit card payments. I think the schemes, Visa, Mastercard, Amex, have built some really powerful infrastructure.
They have root access to program the money in your bank account. They can say, āHey, you just bought this coffee. Iām going to hold $5 or in New York $8, whatever, in your bank account and Iām going to then settle it to the merchant at the right time.ā You donāt have that with ACH. So you actually see that something like 4% of transactions fail five days after the transaction happened.
And so we built Link for many reasons, but one of them was, hey, how can we actually make a bank account payment method that works the way you want it to? So one of the things that customers want is instant confirmation that theyāll get the cash. They donāt want to wait five days and whether or not it went through.
Turner Novak:
A merchant that is using check.
Will Gaybrick:
Exactly. So if you actually can go into Uber and you can see this supported right in the Uber app, Uber wants to know at the time of booking the ride, is this payment going to go through or not? So if you use instant bank payments on Link, you get instant confirmation that itās going to go through. And so thereās a whole lot of things we have to do to enable that, but one of the biggest ones is just the underlying ML model to say, well, we know all of this about you and your transaction history and youāve linked your bank account to Stripe, and so we can automatically confirm that payment.
Thereās also just the pure convenience of Link where youāre showing up at a new merchant and you can pack in your 16 digits and your zip code and everything else, or you can just get a quick OTP on your phone to say, āAt a new AI service,ā I say that because Link is overwhelmingly popular with AI services, āAnd I would rather not have to go through the whole checkout flow again. I just want to get started quickly.ā
Turner Novak:
And you started pretty quickly, right? It was kind of this initially just I think it was the remember me feature, like you click a button or something like that?
Will Gaybrick:
Yes, thatās right.
Turner Novak:
Did you not expect it to just keep growing? Whatās the story there?
Will Gaybrick:
So we technically launched Link, it wasnāt called Link, back in I think 2014, and it was a ability to remember your payment credentials on what was then called Stripe Checkout. And then when we launched the new version of Stripe Checkout, which was in 2018, it didnāt have remember me built into it.
Turner Novak:
Oh, so you removed it?
Will Gaybrick:
Yeah, we removed it and there were millions of consumers that were using remember me, but it didnāt have a name. There was no sort of identified consumer thing. It was just a feature on Stripe. And then we woke up one day, weāre like, why did we get rid of that? That was really useful to merchants.
And so Link was the sort of reincarnation of remember me. It started only a few years ago. And as I mentioned, now over 200 million consumers and becoming a big deal. And we actually see that for a lot of merchants, particularly the new cohort of merchants joining Stripe, itās 50-60% of their transactions are Link transactions.
Turner Novak:
Oh, interesting. Is there a benefit to them whereas itās like speed, cost, anything like that? Or is it just higher conversion because people are saved?
Will Gaybrick:
Thatās the principle one, higher conversion. Customer experience, just consumers love Link. Itās a very subtle experience today, so actual brand awareness is relatively low. Weāre going to be changing that in the near future, but just go through Twitter, you see people say, āI love it when I see Link.ā
Turner Novak:
Yeah, yeah, itās really sick. My favorite Link story personally is when I first signed up, I put the wrong phone number in. I think I put a six instead of a nine at the end of my phone number. And so it didnāt work for me for a while. And anytime a checkout had Link, I couldnāt use it because I couldnāt change my phone number in the thing, which I think you guys have since updated and I can use it again, but I just remember thinking like, ah, thatās the only time I ever put my phone number in wrong. I couldnāt use the product.
Will Gaybrick:
Yeah, it definitely was a launch and iterate quickly experience on the consumer side, so thank you for bearing with us, but it goes beyond cards and bank account payments. Weāre always thinking at Stripe about obviously our users and they are the primary focus for Stripe, but also just other constituents in the payments ecosystem. And so weāve partnered with Klarna recently to create what I believe is the best BNPL purchasing experience on the internet, which is LinkXKlarna.
They serve a bunch of custom APIs for us. And if you use buy now pay later services, youāll find that theyāre incredibly powerful for increasing conversion in some ways, but theyāre also very frictionful. You actually have to get bounced out to Klarna or otherwise, put in a bunch of information, wait for a spinner for a while to say whether or not youāre eligible to make the purchase. We do all of that underwriting in the background in Link. So by the time you actually get to Link checkout, if Klarna appears, itās just one click to check out.
Turner Novak:
Oh, interesting. I didnāt realize that, because you always see those... Youāve probably seen the memes. You remember back probably, itās probably 2021, there would be like literally 10 options. Thereād be a Stripe type checkout button. Thereāll be Amazon Pay. Thereād be Shopify Pay, Fast. Thereās a couple of these startups that were doing one click checkout, and then Klarna. Thereās like four BNPL options. Itās just a funny observation, but I feel like weāve moved past that. Is there consolidation in that checkout button space?
Will Gaybrick:
I mean, thereās been some consolidation, Cash App bought Afterpay, but there hasnāt been a whole lot of consolidation, and this is actually one of the areas that we really focus on for consumers, sorry, for customers, is when one of your shoppers hits the checkout page, for that particular shopper, which payment methods should you show?
Turner Novak:
Oh yes. You source the certain one higher.
Will Gaybrick:
Yeah. We see this drives conversion a lot.
Turner Novak:
So one question, thereās probably people who have been listening to this, I donāt know, weāre like an hour in, thereās probably people that are kicking me like, āDude, you got to ask him this question. Why havenāt you asked this yet?ā Stripe going public. I know people always comment on it. Anything to say?
Will Gaybrick:
Not a ton to say. I think these incumbent perspective is why arenāt you going public?
Turner Novak:
Or why are you staying private?
Will Gaybrick:
Yeah. And I think the way that John, Patrick, Steffan, our CFO, and I think about this is why are you going public? Itās a bunch of work. Itās a whole different way of operating. We already operate with the controls, the rigor, the time to close of a public company. And so itās like whatās the incremental benefit of going public? When I joined as CFO a long time ago, we had this principle of we always wanted to operate cashflow positive.
And to this day, we have actually never burned a dollar of investor money. There were a couple years in the early 2020s where we burned cash, but we didnāt burn below the amount that we had accrued on our balance sheet. Stripe has always been financially independent in that sense. Stripe is now extremely profitable, and so thereās a question of what is the reason to go public?
Turner Novak:
It seems like a lot of people, you need to give retail investors the ability to participate. Thatās a common reason that people give.
Will Gaybrick:
Yeah, and I guess thatās a belief in creating equity between institutional investors and retail investors. Iām not sure that that is the thing that should guide our decision making here so much as...
Turner Novak:
Thereās billions of people that need Stripe equity in their retirement portfolios.
Will Gaybrick:
Yeah. I would say being a mission-driven company, it is how do we grow the GDP of the internet as quickly as possible? How do we stay just resolutely focused on our users? Weāre already a highly regulated global financial institution. If the day comes where we say, āHey, there really is a good reason to do this,ā then I suspect we will.
Turner Novak:
Yeah, I feel like some of the benefits of being public is your stock is more liquid, so in theory you have a lower cost of capital. But then you can argue on a company per company basis how much of a benefit does a specific company get from sliding into that, you may not get much of a benefit. I think another interesting thing when I think about this is just the incentives of your investors.
So if youāre managing a public market assets or funds, when you think about how do you make money, you charge management fees and you get carry and you get paid carried quarterly if itās public, you get paid on the exit if itās private. But typically, I mean, thereās been a lot of fee compression in public markets. So if youāre managing public market money, you might be getting one point a half percent management fee, maybe less.
You may be charging on average like 15% carry versus in the private markets 2% management fee. Some people do two point a half or 3%. And then also you get the carry, the percentage is also higher. Thereās a certain element of certain investors are like, I think itās just perpetually continue to shift from public to private because why wouldnāt you?
Will Gaybrick:
Yeah, and thereās clearly a blurring between public and private investors at this stage.
Turner Novak:
Yeah. I think in Johnās podcast, Cheeky Pint, with Dan and D1, heās like a public market investor, but I think he said 70% of their AUM is private. I may be getting this number wrong, but I think I remember him saying that more was actually on the private side, more AUM.
Will Gaybrick:
Yes.
Turner Novak:
Which you wouldnāt have thought about that.
Will Gaybrick:
Yeah. I think the principle reasons why companies go public besides just inertia and the idea that youāre supposed to do it are, one, access to capital. Stripe has issued bonds, investment grade company, highly, highly cash flowing. Just thatās not an acute need today. Creating employee and liquidity. Weāve been able to run tender offers. We canāt promise to always run those, but weāve returned many billions of dollars to investors, to employees, and so on.
And then the third is maybe something around marketing and branding, where by being public youāre more trusted or seen as more enterprise grade. That may linger around in certain corners, but I think weāve mostly crossed that transom. Half of the Fortune 100 using Stripe. Stripe being one of the major processors for companies like Amazon, moving $1.3 trillion in money per year. So just none of those three really seem to be pushing us down the IPO river.
Turner Novak:
I think thereās another element too of early people with illiquid stock want liquidity. And so if you just go public, itās like just get your liquidity and just deal with it. But you definitely see this with some larger private companies where they just do... I think SpaceX does quarterly liquidity or every six months. Itās just a planned... You can sell whatever you need to sell.
We do this event. So I think thatās something thatās kind of evolved too. Well, some of the business models, some of the private market funds also is like we are on the cap table. We will give you some of our access to our shares that we have. Thatās another model thatās emerged.
Will Gaybrick:
Yeah. Listen, the market, I donāt mean the stock market, I mean just the market changes quickly. Every year we write a consolidated company strategy. You just canāt over invest in strategic clarity for yourself and for your team.
Turner Novak:
You said you cannot over invest?
Will Gaybrick:
You canāt over invest in it. Itās one of my biggest learnings as a leader is that what may seem obvious to you just is not obvious in all the minds of the company. You really want to emphasize strategic clarity. Actually, itās fascinating, in 2021, we do an internal Stripe stat and we ask about a bunch of different things. One of them was like Stripe has a clear strategy.
I think itās something like 50% of respondents agreed with that. Weāre now above 90%, and itās just a concerted investment in canting the most important things, what you focus on and why. This year with these twin revolutions in stablecoins and AI, itās not that weāve thrown the strategy out the window, but we have had to be very, very nimble in thinking about deploying internal resources and some of our best people against new opportunities.
And so being private, having the flexibility to think about doing that without wondering about any major knee-jerk reactions that might happen from investors and activists and so on, I think itās a luxury. And again, it may be that at some point there is a good reason to go public, but today we ask ourselves why do it versus why not.
Turner Novak:
Thereās also to your point about activist investors. Thereās just annoying component is like if youāve got $2 billion or your company is worth X amount and you can get a percentage of it and make demands. I think I had Aaron from Vox on the podcast and he was like, they had to deal with some stuff where people were trying to kick him out and force him to sell, and they got through it, but it was just not what he wanted to do.
Will Gaybrick:
And all these are tractable problems and Aaron is amazing and he navigated that so well. But man, it probably wasnāt the way he wanted to spend his time and probably wasnāt the highest and best for how he spent his time.
Turner Novak:
His hairās a little more grayer than it was. You age, for sure.
Will Gaybrick:
Yeah.
Turner Novak:
So I want to ask you, so back in college you did this thing called Hack Yale. What was it?
Will Gaybrick:
Actually I was in law school.
Turner Novak:
In law school?
Will Gaybrick:
I was in law school. It was this very fun incidental mission where I was at Yale Law School. I was actually also working in the city as an engineer at the time, and I had been an undergrad at Harvard and had seen this incredible culture.
Turner Novak:
What year was this, like 2010 ish?
Will Gaybrick:
For undergrad?
Turner Novak:
Yeah.
Will Gaybrick:
I guess I graduated in 2007, so it was in the early to mid 2000s. This is the Facebook Meta gestation era.
Turner Novak:
Did you have a Hot or Not account? What was the original one called? Were you a real member?
Will Gaybrick:
I honestly canāt remember. I know there was Friendster, but that was separate.
Turner Novak:
But you were a member of TheFacebook.com? Okay, back in the day.
Will Gaybrick:
I was on The Facebook when it was Harvard only. And there was this incredible culture of entrepreneurialism as an undergrad, and I was actually really struck when I got to law school that there just wasnāt. And the curriculum for computer science was OpenGL and some legacy technologies and learning about how computers worked in a backwards looking way versus just like, hey, how can you build things quickly?
There wasnāt really this builder mentality. And so Hack Yale was about how do we create this builder mentality on campus? And so I decided to just... I had all these undergrads coming to me, I guess through word of mouth saying, āI have this idea. Can you work with me on it? Can you build it with me?ā
Turner Novak:
Oh, they knew you knew how to code?
Will Gaybrick:
Exactly.
Turner Novak:
You were an engineer. Itās like that classic, āI got an idea, I just need to build it.ā
Will Gaybrick:
And the answer was, no, I canāt. I have a lot to read for law school and I have a job, but maybe I can teach you. And so it started with just a handful of kids the first week, and I was just thinking, okay, Iām going to teach JavaScript end-to-end so that you can think of server side and client side and a single language. This was early days of Node.js. And just going to get people from zero to on., they can just start building their own at least demo apps.
So it started with probably nine kids the first week or something like that. Second week it was standing room only, but in a 30 person room. Third week it was over a hundred. There was no fourth week. I shut it down and said, āOkay, letās see what this could be.ā So I just poster it around campus we were doing this and allowed students to apply. And in a week, a third of the Yale undergrad student body applied for the class, which was pretty amazing.
Turner Novak:
But this wasnāt actually a class. It wasnāt an official class.
Will Gaybrick:
No. In fact, the computer science department hated me for doing it. It was actually quite controversial.
Turner Novak:
Really? Okay.
Will Gaybrick:
I ended up working with some amazing people on it. Miles Grimshaw, whoās an incredible partner at Thrive was one of my TAs, Big Rose, who runs a great startup and was at Google before, was a co-teacher with me. And so we turned it into a proper curriculum. It was all pro bono, and it lives on to today and actually went to multiple campuses, but it was really rewarding and fun just trying to change culture on campus and actually finds a leader.
Just culture is one of the hardest and most fun things to work on. I often find that companies have too many operating principles, and this was even true of us until recently where itās like, hereās the 15 things that we reward and think about prioritizing as a company. Weāve really distilled ours to just a handful. And I think about the people who are the most successful at Stripe. They are the most user-focused.
They are the most impatient and fast. Theyāre most demanding of quality in what they do. So weād say be meticulous about your craft, focus on craft and beauty. Theyāre the most humble, the most collaborative, and itās really that handful of things that makes people successful at Stripe. Just always thinking about culture, how are you modeling it, how are you messaging it. I think itās one of the most important things you can do as a leader.
Turner Novak:
And then around this same time, I canāt remember if you joined Thrive when you were at school or if you had graduated, but how did that come about, because that was a pretty big move back in the time?
Will Gaybrick:
Yeah, I was still in school. Josh Kushner, whoās Thriveās founder and runs Thrive, reached out to me, called me. We didnāt really know each other, and I was working as an engineer at one of his portfolio companies. At this point, Thrive was a really, really small fund. It was Josh and Jared, a little bit of individuals money. I donāt know actually. I think it was some of the general Catalyst partners were involved, and he reached out because he wanted a CTO for the fund.
And so the idea is that I would join the fund as a CTO. And actually I remember hacking together like a CRM in the early days so we could track conversations and start thinking about tracking information about companies. And then it turned out what we really needed was just people to do deals. And so I started doing that and was fortunate and very excited to be on the journey from Thrive being about a $5 million fund to being a few billion by the time I left for Stripe about five years later.
Turner Novak:
And so I think Claire asked me to ask you this. You wrote a memo to do the investment. How is that memo played out? I donāt know if thereās significance of what was in this thing, but she wanted me to ask about it. How does that all kind of happened between...
Will Gaybrick:
Oh, I see, the investment memo at Thrive to invest in Stripe?
Turner Novak:
Was there a legendary memo that sheās mentioning, or maybe sheās just wondering what did you think about Stripe at the time, what it could be, and how itās evolved?
Will Gaybrick:
I always think that in B2B you really want to focus on problem areas that have an enormous market size, some degree of universality, like everybody needs the mousetrap youāre building, and then a notion that people really care about the quality of the mousetrap. They want the best one. I sort of think of this as itās being core to the purchaser. They really will care about their choice.
So an example of a company I invested in that is universal, but maybe not core, was a great company, a company that Stripe uses, called Greenhouse for applicant tracking. Everyone needs an applicant tracking system, but thereās a few on the market and people donāt necessarily agonize over which one. Part of the reason Iām so excited about Stripe was itās such a macro opportunity. You could just see the focus of the company, the mission of the company being to grow GDP.
Turner Novak:
That was the mission back at...
Will Gaybrick:
That was the mission back then, yes, to accelerate the base of globalization. So quite inspiring. John and Patrick are very inspiring people.
Turner Novak:
Yeah, thatās great marketing at the time from getting people excited to join, invest, et cetera.
Will Gaybrick:
Yes. It had the right customer base. Back in the day, Iām not actually sure that people thought of the merchant acquirer they were choosing as being super core so much as just which oneās the cheapest. And developers as customers really wanted the thing that could help them move the fastest. Growth was extraordinary. I felt a ton of respect for John and Patrick and really liked working with them as an investor.
And yeah, I think in terms of the question of how itās played out, I donāt think Iāve projected Stripe out 10 years. Itās been almost 10 years. I could go back and look. But last I checked, Stripe was ahead of my projections on volume and slightly below my projections on margin. And I donāt mean OpEx margin, but just margin in terms of gross margin.
Turner Novak:
Like the take rate essentially?
Will Gaybrick:
Exactly. Yeah, the take rate, net of network fees.
Turner Novak:
So Stripe did not turn in a big corporate behemoth extracting value as fast as you thought they would. Youāre still very much... Youāre giving a lot to the ecosystem.
Will Gaybrick:
Yeah. One belief I have that maybe is a bit contrarian is that it is great to be a low margin business and that people over index on margins. For example, public market investors will really care what your gross margin is. To a large extent, that depends on how you report your revenue. So a lot of payments companies report net revenue, and then off of that, their gross margin will look really high.
Theyāre pulling out all the payment systems costs. If you look at gross revenue, then theyāre a lot lower margin. The reason why I think being a low margin business is great is it just keeps you obsessively focused on customer needs. Thereās so much leverage in the next thing you can build for customers that they will fervently, enthusiastically adopt, because the core of what youāre selling them is a low margin business. If you build that additional service, you might be able to augment your margin by 20%, by 10%.
Turner Novak:
But if youāre already super profitable, the new thing might just not even move the needle youāre saying, or it might actually decrease your margins or something like that.
Will Gaybrick:
Exactly. And so I think if you did some sort of regression or PCA or something like that, youād probably find that for the most part, companies that have lower margins, at least to begin, are the most customer obsessed. Amazon obviously being a great example.
Turner Novak:
Yeah, I think Amazonās an interesting example too, because thereās such a long period of time where just the general consensus was Amazon is not profitable, so just internet commerce isnāt going to work.
Will Gaybrick:
Yeah, exactly.
Turner Novak:
Yeah, and I think thereās a crazy story with the way Walmart ran the business and compensated. People like as a store GM, you are compensated based on how your store performed. And so if somebody in a customer of your store were to order something online, you were incentivized for that to not happen because you get paid less. So Walmart itself, because of this almost meme of, oh, Amazonās not profitable, we donāt need to worry about e-commerce. Itās got to the point where itās like 2014, youāre like, oh, wow, we have never focused on this and Amazon kind of beat us and we will never fully recover. We kind of lost our footing that we have.
So I think itās an interesting way to continue to operate under the surface, maybe in a way too. People just donāt fully pay attention. Because another example is you just look at... You ask someone whatās the most interesting thing right now? And itās like AI. Whatās the fastest growing, most profitable thing, and itās like ChatGPT. All right, we need an AI strategy. So everyone is attracted to this light and they forget about, oh, thereās a low margin also really good business that no one cares about because AI is the thing.
Will Gaybrick:
Yes. Yeah. There was definitely a consensus in the early days that Stripe wasnāt a good business because margins were low. Actually it wasnāt a consensus, but there was a cohort of people who felt that strongly, and I think others who saw more potential. But on the note of just how it keeps you focused on users, stable coins are I think an incredible opportunity for our customers who have low margins. Because if you think about marketplaces, maybe they take three, five, 10% off of the transaction, the GMB that theyāre creating. If youāre looking at then the payment method costs, be that cards or BMPLs or otherwise or wallets, that can chew up two thirds of the take on those payment methods.
Turner Novak:
Instantly boost your profits 30, 50%.
Will Gaybrick:
Yeah. If you can create a high converting global payment method using stable coins, that is really going to boost the margins of these companies.
Turner Novak:
So then how did you decide to join? You had been at Stripe, you were at Thrive, and you probably were pretty happy as an investor, Iām assuming. How did it kind of come about that also you joined this one as CFO?
Will Gaybrick:
Yes.
Turner Novak:
I would probably not have you peg as the top on paper choice a CFO of a company at the time.
Will Gaybrick:
So I think I was an accidental VC and then sort of an accidental CFO. I loved working at Thrive. It was an amazing place. So I love the team. I donāt think that being a venture capitalist was the thing I wanted to do when I grew up. Just technologist when I have my hands in what weāre building, be very close to it. One night I got a text from Patrick saying, āHey, do you want to come be CFO?ā
Turner Novak:
That was it.
Will Gaybrick:
Yeah, that was it. And I think I replied, I probably have the text way back somewhere, āThat seems like a really bad idea.ā And Patrick being very convincing, said something along the lines of, āWell, John and I think we could do it.ā And I was sort of like, all right, challenge accepted. And I wasnāt planning at the time to leave Thrive by any means. I think my sort of incumbent plan was, this is great. Iām going to do this for a while longer. Iām going to go start a company. I think thatās always a journey that I had expected to be on, but it was Stripe and John and Patrick are just incredible people.
I was really motivated by the mission. I thought the product was excellent. And then in terms of CFO, we talked earlier about going public. At the time we were certainly nowhere near that. And so there was I think the ability to bring in a CFO without public market experience. And I think there were a lot of ways I could contribute as CFO. Iām not sure that I wouldāve been the sort best CFO to take the company public. And a few years in, I talked to Patrick and just CFO wasnāt really the journey that I wanted to be on. So today we have an incredible CFO in Stefan Tomlinson whoās worked at a lot of public companies, and I think heās a better profile for Stripe at CFO than I wouldāve been.
Turner Novak:
Yeah, thatās fair. And then how do you think you earned their trust as somebody from the outside? Obviously there was a reason that he texted you this. It probably wasnāt like, āHey, what do you think about this idea?ā Itās probably made up his mind and was like, come do this. What do you think was the reason that they trusted you?
Will Gaybrick:
Thatās a good question. We all believe strongly that just by far the most important input into hiring is back channels and references. I think people really over-index on the several times that you spend 30 minutes with a highly consequential hire.
Turner Novak:
And just how well can you read them?
Will Gaybrick:
Exactly.
Turner Novak:
You look in their eyes and the guyās a killer or something like that. Okay.
Will Gaybrick:
Yeah, I tend to look at sustained performance at excellent companies as being a key indicator and then back channels and references. So Patrick and I knew each other pretty well at that point. Weād known each other for a couple of years and I think he had talked to a lot of people and felt like it was a good bet to make.
Turner Novak:
Interesting. Yeah, thatās fair. And then Claire also mentioned you put a really high sort of priority on the strategic finance function. I think I sort of know what that is, but some people might not really know what that means. So what is strategic finance and why was that such a big deal to do at the time?
Will Gaybrick:
Yeah, I think maybe the industry term for it would be FP&A, but the part of FP&A that is closer to product development and go-to market. Thereās sort of corporate, which is consolidation and closing the books and so on. And then thereās the part of it partnering deeply with the business and with product development teams. When I joined, I think we had five people on the finance team, and most of those were certainly the accounting side, we didnāt even have a treasury team, we had a small tax team and we really just didnāt have a good understanding of the business. And so in the early days, just wanted to get to having a baseline model for Stripeās trajectory and how we make money today and how we can expect to do so tomorrow. I remember one of the early inflection points, that doesnāt sound like an inflection point, but it actually was in our thinking was when we launched Radar. Radar is our fraud mitigation tool. It was the first time that we launched a product that was not payments.
Again, this probably sounds incredibly obvious in retrospect, but the incumbent perspective to that point was Stripe would be the product and it would be you buy Stripe and Stripe processes your payments and does a bunch of other things for you and youāre going to pay for Stripe. And this was the first unbundling of products. And I think that was guided significantly because of some of the early investigations we were doing and seeing that as our customers were getting larger, they wanted to see more unbundling of value and they wanted to see, well, Iām paying this amount for payments, Iām getting competitive bids from others saying, weāll give you lower payments rates, but we knew our fraud product was really, really good. So okay, well, letās give you a better rate on the core payments, but letās also sell you fraud mitigation and demonstrate that value as being two dimensions as opposed to one.
Turner Novak:
So itās an interesting kind of evolution then. So then how do you think about the go to market of the Stripe enterprise of products today? When you said youāre seeing results on SDR, AI, outbound or whatever, do you figure out a specific pain point you try to have as the entry point? Or is it always payments?
Will Gaybrick:
So itās not always payments and I would say just a big learning for me, and I think just something I would advocate to founders, CEOs, product leaders, is just get incredibly close to go to market and to sales. Because when you step back, sales is reality. It is that moment in front of the customer, are they interested or not? Why are they interested? What can you learn from that, that is just so viscerally educating as to where you should be going as a company. And so today I think we see a bunch of different entry points into companies. One is around core payments. You have for enterprise customers, in most cases, a pretty significant enterprise... Sorry, payments infrastructure they built. And you might have more of a land and expand motion there where itās like, okay, adopt Stripe for part of your payment stack. See how we perform.
Off rates tend to be a very key part of the calculus. Or you want to add a payment method that you donāt have today, you can use Stripe from that. The really interesting thing that weāve been seeing on the core payment side is customers using what we call our optimized checkout suite, which is the hosted interfaces I referenced earlier. And there was this incumbent perspective that people would never outsource their checkout to Stripe or to any third party. And now we see that selling in the optimized checkout suite and saying, you can use any payments processor you want behind it, but itāll convert better and youāll have a better customer experience has worked very well. But we see billing as being a huge independent sale at Stripe. Connect, which is our infrastructure for marketplaces and platforms. Stablecoins, a really, really big one at this point, particularly for companies that have a very global footprint, increasingly radar and fraud tooling as an independent sale.
So we have our go-to-market machine structured as a frontline of a few hundred account executives, which is still very, very small. And then an overlay of specialized sellers, technical sellers for these individual product areas. And one of the really valuable learnings for me over the past couple of years is just how impactful it is to think about how you organize your sales team as almost a chipset. You have to design it really, really carefully. You ship it, you canāt really change it that much for a year. And so being so intentional about how sellers are compensated, what youāre pushing, what your sales plays are, do you have hunters and growers or is it all just one type of a has been-
Turner Novak:
So hunter is like new customer, grower is like, hey, hereās some more stuff you should try thatāll help you grow as a customer.
Will Gaybrick:
And hereās how weāre just making you successful even on your current product set.
Turner Novak:
Did you mess that up at some point? It sounds like maybe thereās some scar tissue around how you designed or-
Will Gaybrick:
I would say less scar tissue and more just didnāt pay enough attention to it.
Turner Novak:
So maybe you got lucky ish that it kind of worked?
Will Gaybrick:
Well, weāve been really under invested in sales and marketing. Even today, sales and marketing together are under 10% of the company, which is well below-
Turner Novak:
110% of headcount or revenue youāre saying?
Will Gaybrick:
Of headcount.
Turner Novak:
Headcount. Okay. Interesting. And so the one thing you mentioned maybe and interesting topic just semi related, you mentioned Stripe. We like the lower margin is weāre totally okay with it. Why are the margins low? Do you just have a lower take rate than competitors or thatās... Because probably another question that people always think is, come on, you got to ask this question.
Will Gaybrick:
Yeah, itās funny because youāll sometimes see in startup board decks, the margin bridge where thereās this big chunk that just has Stripe. I promise you we are not taking 3% of your revenue. We would be a very differently shaped business if we were. The vast majority of that is going out to payment systems and in particular to banks via interchange to Visa, MasterCard, Amex for scheme fees. And so particularly in the enterprise, your pricing that is sort of the underlying payments costs plus a small number of bits, basis points on top of it. Thatās what Stripe takes. But then in that small number of basis points, youāre paying your infrastructure costs, youāre paying a bunch of variable costs related to serving customers and so on. So core enterprise payments is not a very high margin business.
Turner Novak:
Do you do anything differently than other people that adds more value to customers that ends up having lower margins? Is that a fair thing to say?
Will Gaybrick:
Well, I think we do a lot of things differently that adds value. I mean, I mentioned earlier reliability just being a critical one. And I do think that some of our infrastructural workloads are pretty fat in order to enable that.
Turner Novak:
So you spend more on your own infrastructure to have a higher uptime.
Will Gaybrick:
And Iād say both in terms of operating expenditures and in terms of just core compute consistency promises and our API are pretty expensive, meaning that the data is updated atomically in the API across a bunch of different services. For startups, we give away fraud protection for free. We think about this as being a core part of our value proposition, but also critical just to the overall ecosystem. You sort of have to buy fraud from Stripe at some level because otherwise just fraud rates in payments are going to be too high.
Turner Novak:
And plus, if Iām a fraudster and Iām thinking about who is the best company to go after, is it a big corporation or is it a small startup that doesnāt have capabilities? Theyāre probably my first target if Iām thinking through that totally.
Will Gaybrick:
Yeah. You see, fraudsters are infinitely creative. Card testing used to be a matter of taking a bunch of cards, choosing one merchant and attacking them, just running a ton of transactions through them via scripts. A couple of years ago we saw this new type of attack where card testers were using Stripe Checkout, which is the fully hosted payments page we offer. So itās actually hosted on stripe.com.
Turner Novak:
They would make their own checkout page?
Will Gaybrick:
Actually, they wouldnāt make their own checkout page, but they found a way to identify thousands and thousands and thousands of checkout pages. And then they would dribble their tests across all of these. So youād see a few transactions. They were using very generic names, I canāt remember what they were, but it was like Jack Smith or something like that. And you would then see on Twitter these startups saying, I have this Jack Smith customer whoās clearly fraudulent, like Stripe, what are you doing? And it was just very hard to detect these. It was only a few transactions on each. Itās actually pointing back to the foundation model by aggregating so much data into one big model and then clustering in this very high dimensional space. Itās super easy to solve that today. But yeah, fraudsters are very entrepreneurial and clever.
Turner Novak:
Yeah, I mean ,you have to be, you got to evade. You got to not only evade them customers and the merchants, but like law enforcement, you got to stay on top of a lot of stuff. One, actually, probably one of the last questions. I feel like Stripe is pretty easy to use just generally speaking. Itās probably one of few financial services products where you can, maybe this has changed a little bit over time, but go to the website, sign up. I think you can just start using it right away if Iām remembering right. Why do you think so many financial services products are just hard to use? You got to talk to a salesperson, you got to wait so long.
Will Gaybrick:
Itās hard. One of our core strategic principles is to make our products in Stripe just generally instantly self-serviceable. What customers want, they donāt want to wait. Some of them do. Some of them want to do a deep assessment, but startups in particular donāt, they just want to get going as quickly as possible. And to the point of becoming conspicuously good at risk management, thereās a whole lot you have to do to enable that. And so as youāre going through this at Stripe, we are underwriting you in the background. Weāre running a bunch of different models. Weāre deeply understanding your site. Weāre looking at our foundation model to see is there anything thatās sort of popping up there? And a lot of customers, not the majority actually donāt know what percentage offhand, but probably 20% or so are stepped up for review, meaning weāll do human review before you can move your first dollar. But even then we can get you live within a day.
Turner Novak:
You can still probably get the API half Stripe in the product, but you canāt use it yet. But get it. You can start using it, but you canāt transact yet.
Will Gaybrick:
Yeah, majority of users can start moving money within minutes. Some percentage of users, it maybe takes a day and then some percentage takes longer because we need to do some back and forth.
Turner Novak:
So this is just an element of the... Youāve built the risk management fraud detection using AI and LMS to automate that stuff. So itās like when I first sign up the first page of the application process, itās a couple pages, I enter my company name and click enter, and then youāre immediately the spinning wheel that I might be seeing at the end of the application, youāre doing it throughout.
Will Gaybrick:
A core dimension of product development at Stripe is finding a way to responsibly, steward of the ecosystem, bend like partner and regulatory space-time. This notion that partners will always insist upon this data or the consensus is that you have to do these things. If you step back and say, what are people solving for? They tend to be solving for trust and safety, risk management on their side. And what we found is we really started investing in risk management was that it wasnāt that others were really good at it. In many cases, they were printed PDFs being reviewed in basements it. They were just taking a much more conservative stance than we were in thinking about what it would require to be good at risk management. And so the idea was you have to do this, and the fact of the matter was, with the right technology, you donāt. So I actually feel like in most cases weāre better at managing risks than a lot of the partners that we work with, but we have taken a different route together.
Turner Novak:
Yeah. Do you have... Last question. Do you have a favorite CEO, founder, business that you just learned from over time or gotten inspiration from over time? This could even be going back to Roman Empire type or more recently, just like any time throughout history?
Will Gaybrick:
Itās a guy named Alan Mulally who was CEO Boeing and then Ford. And I think his system and his operating model and his way of thinking has always been very inspiring to me. It is this focus on what he refers to as relentless implementation, but in the context of sort of egoless collaboration and clarity for everyone in the team. I think what he did with Ford was incredible. He turned around during the financial crisis if I remember, I think he joined Ford in 2007 or something like that. And I think the share price bottomed out. I think I should look this up. Itās somewhere around $2 a share. And I think when he left order of eight years later, something like $22 a share, something like that. So I think just an incredible operator and very inspiring people leader.
Turner Novak:
This is a lot of fun. Anything you want to plug on your end? People follow you on the internet anywhere?
Will Gaybrick:
Best way to follow me is to follow everything weāre doing at Stripe. Thereās a lot more coming. Just in the past month, weāve launched the agent commerce protocol, instant check out with ChatGPT, something called Open Issuance at Bridge, which is a platform to issue your own stablecoin. So for example, Phantom, maybe the most beloved crypto wallet in the world is launching Phantom Cash, which is their own stablecoin, open loop combined and defy pools built on top of open issuance. Weāve launched, as I mentioned earlier, Klarna in Link. A whole lot more. So stay tuned to everything coming out of Stripe. Always eager for feedback. You can slide into my DMs or email me. Iām WG at Stripe and thanks for having me. Itās been really fun.
Turner Novak:
Yeah. Yeah, this was awesome. Thanks for doing it.
Will Gaybrick:
Thank you.
Stream the full episode on YouTube, Spotify, or Apple.
Find transcripts of all other episodes here.