Ramp Signals US Economic Acceleration as Corporate AI Spend Surges

企业AI支出激增，Ramp预示美国经济呈加速态势 企业财务的快速自动化正在将商业支出从人力薪资转向软件Token，同时让高增长企业相比传统竞争对手获得巨大的资本效率优势。 Ramp已从一家企业信用卡服务商发展为多产品财务自动化巨头，其非卡类软件业务目前贡献了大部分毛利润。该公司利用AI Agent以99%的准确率每天处理超过10万笔报销费用，在宽松的信任文化与严格的合规要求之间搭建了桥梁。对话还探讨了软件工程领域的格局变迁——AI正在打破传统团队边界并催生自愈式代码库，同时深入剖析了Capital One的风险管理传统历史以及自动化企业资金管理的未来前景。 Ramp已拓展为多产品平台，预计账单支付、资金管理和采购等非卡类业务线将占其整体业务的绝大部分。 (0:00) 高增长企业倾向于宽松且基于信任的报销制度，Ramp利用LLM实现自动化，以99%以上的准确率每天审核10万笔费用。 (4:27) B2B账单支付依然陈旧且充满博弈，因为财务主管在经济上有动力拖延付款以赚取浮存金利息。 (11:06) AI正在模糊Ramp传统的软件工程角色边界，允许非工程师部署代码，并大幅缩短从发现问题到修复问题的时间。 (16:13) 软件开发正从确定性的代码路径转向以结果为导向的指令，为能够消除历史技术债务的自愈式代码库铺平了道路。 (18:14)

Ramp co-founder Eric Glyman argues that artificial intelligence and automation are fundamentally transforming corporate finance by shifting the core value proposition from providing capital and rewards to saving customer time. He contends that as cognitive costs plummet, software defensibility will rely on proprietary historical data and complex operational edge cases rather than visual code. Automated Policy Enforcement and Time Savings (4:27) Glyman emphasizes that modern spend management competes on selling time rather than rewards points. Ramp utilizes large language models to agentically review over 100,000 transactions daily with over 99% accuracy, satisfying Sarbanes-Oxley separation-of-duty requirements without manual managerial oversight. Rather than enforcing rigid bureaucratic rules, contextual AI evaluates expenditures against commercial outcomes, balancing corporate compliance with operational flexibility. Ramp claims its controls and spend intelligence reduce client expenses by 2% to 10% annually. Programmable single-use cards, automated merchant blocks, and strict spend caps replicate the security of paper checks in digital form. Furthermore, by benchmarking seat-level software pricing data across millions of transactions, the platform proactively identifies when customers are overpaying for SaaS contracts by margins as high as 20%. Proprietary Data and Dark-Matter Moats (25:18) While AI coding tools enable cross-functional individuals to push code to production, the speakers view predictions of a wholesale software collapse as overstated. Basic utility software faces rapid commoditization, but enterprise defensibility resides in proprietary historical data and accumulated "dark matter"—the institutional knowledge and thousands of edge cases embedded in mature corporate workflows. Bootstrapped legal provider vLex, for instance, scaled annual revenue from $20 million to $100 million in a single year by pairing language models with centuries of exclusive digitized legal records. Corporate expenditures are subsequently expected to shift away from human payroll toward software and token infrastructure. The speakers argue that companies will continue outsourcing specialized administrative functions under the principle of comparative advantage, because the ongoing token and engineering expense required to independently rebuild and maintain complex systems like multi-entity accounting outweighs the subscription cost. B2B Payment Inefficiencies and AI Spend (32:15) Business-to-business transactions remain distorted by an adversarial accounts receivable and accounts payable dynamic, wherein large enterprises deliberately slow payments to capture interest on cash float, forcing small vendors to borrow at steep rates. To overcome cross-border settlement hurdles, spend management platforms are adopting stablecoin-backed card rails. Meanwhile, AI agents are reducing the transaction cost of disputing fees and negotiating invoices to near zero, threatening traditional retail breakage models that rely on consumer inertia. Transaction data across Ramp's 55,000 business customers also contradicts conservative official statistics regarding economic health and AI usage. While Census Bureau surveys indicated single-digit adoption rates, Ramp's spend data reveals a clear majority of businesses actively paying for AI subscriptions and agentic tools. Operating on average profit margins of 8%, American businesses find expense reduction twelve times more impactful than top-line revenue growth, supporting observed domestic economic acceleration. Algorithmic Cash Management and Banking Inertia (66:32) Drawing a direct lineage from Capital One's early Information-Based Strategy and credit-curve modeling to modern fintech leadership, the discussion highlights an impending shift toward programmatic corporate treasury. Major commercial banks currently pay an average yield of 0.07% on business checking deposits despite elevated benchmark interest rates. Automated treasury systems correct this disparity by sweeping excess operating funds into higher-yielding instruments, transferring money back only when obligations like payroll are due. The long-term vision entails automated capital allocation where intelligent systems continuously route liquidity toward its highest-return internal or external application. However, speakers note that traditional banks retain significant market share due to administrative friction and behavioral inertia, as the hassle of updating linked recurring accounts or waiting for uncashed paper checks to settle discourages businesses from fully severing incumbent relationships. While AI automation and real-time treasury promise to eliminate administrative waste and optimize corporate capital efficiency, the transition remains constrained by switching inertia, deeply entrenched banking relationships, and the operational complexity of enterprise systems.

Ramp联合创始人Eric Glyman探讨了企业支出与软件生态的技术演变。他指出，随着AI降低软件构建与知识工作的边际成本，企业成本正从薪资向软件及算力转移；而现代财务平台的核心护城河，已从单纯的发卡与返现竞争转向基于专有数据、深层系统集成和全自动化现金管理的效率网络。 业务多产品化与AI代理支出合规 (4:27) 成立于2019年的Ramp在六年多内实现了年化收入超10亿美元。公司正从单一的企业信用卡模式转型为多产品平台，其两年前推出的软件业务年化收入已超1亿美元，非发卡业务预计将在年底贡献大部分整体业务。根据平台统计，客户使用Ramp后平均削减约5%的年度开支，客户自身营收平均增速达16%，显著高于约5%的全美平均水平。 在支出审批方面，企业常在完全信任与官僚式硬性规则之间拉锯。Ramp引入大语言模型（LLM）驱动的AI代理，每日实时审核逾10万笔费用，声称准确率达99%以上。该系统在满足萨班斯-奥克斯利法案（Sarbanes-Oxley）关于职责分离的合规要求的同时，免除了人工逐笔审核的负担，并能结合下游业务产出（如昂贵差旅是否促成关键交易）进行情境化合规裁量。 账单支付困境与应收应付利差脱节 (11:06) 尽管支付技术不断演进，企业间（B2B）账单支付仍深陷于PDF发票手动录入与跨系统核对的落后状态，这与资产折旧和即期SaaS费用等复杂的会计处理规则直接相关。在高利率环境下，应收账款（AR）与应付账款（AP）之间存在天然的博弈：付款方财务部门倾向于利用汇款浮动期（float）拖延付款以赚取利息，使得纸质支票因邮寄周转时间而在财务上更具优势。 对话指出，大型企业往往凭借极低的融资成本发行长期债券，却对小微供应商尽可能延期付款；而小供应商为了弥补应收账款滞后的现金流缺口，被迫承受高达18%至20%的借贷利率。这种资金成本断层暴露了当前财务系统的隔离，行业需要类似“支付DNS”的中央清算网络以及更深层的数据对接来解决错配。 研发边界消融与“暗物质”护城河 (25:18) AI开发工具（如Cursor和Claude Code）正在打破传统产品经理、工程师与设计师的组织边界。在Ramp内部，设计师与客服人员已能直接向生产环境提交代码，界面微调的部署周期缩减至数分钟。虽然代码编写门槛降低可能导致软件冗余，但对谈嘉宾预测，未来的软件架构可能演化为仅定义预期目标的结果导向型系统，由AI每年自动重写底层代码以清除技术债，形成自我修复代码库。 针对“SaaS大毁灭”的市场担忧，对话认为易被AI克隆的主要是缺乏深度的单一功能应用。复杂企业软件的核心防线在于“暗物质护城河”——即历经数年运转累积的技术债与海量极端边缘案例（edge cases）。此外，专有历史数据构成了不可复制的门槛，例如法律科技公司vLex凭借自1492年以来的独家数字化判例库实现收入爆发，DomainTools长期聚合的历史WHOIS数据也无法被后发者倒序抓取。 支出结构演进与企业集采议价规模 (30:21) 在宏观层面，轻资产企业的成本中心正从传统人力薪资向软件工具和AI Token转移。尽管美国人口普查局的调查称仅有个位数比例的企业应用AI，但Ramp来自55,000多家客户的实际支出数据显示，绝大多数企业已在为ChatGPT、Anthropic或Cursor等AI工具付费。 针对美国企业平均约8%的净利润率，对谈强调节省1美元在数学上相当于增加12美元的营收。Ramp通过单次虚拟卡、商户拦截与席位级比价等硬性控制，帮助企业规避非必要支出。随着Ramp处理全美超2%的企业和中小商业卡交易，公司探索借鉴Costco或医疗团购组织（GPO）模式，通过汇聚数万家企业的采购需求直接向供应商索取折扣，实现由“贩售资金”向“销售时间与自动化”的战略转型。 动态财资管理与智能化资本配置 (66:32) 企业财资管理目前面临严重的利差分配失衡。全美企业商业活期账户的平均存款利率仅为0.07%，传统商业银行截留了美联储高基准利率带来的绝大部分收益。现代财务管理平台通过自动化手段，将除两三个月日常周转金以外的闲置资金自动归集至高收益资产，并在工资发放日精准转回，重塑了现金管理效率。 这一思路延续了Capital One早期的创新经验。1990年代，Capital One创始人打破了传统的信贷准入二元论，通过“基于信息的战略”（IBS）在连续的风险曲线上进行动态实验与风险定价。嘉宾预测，未来AI将使“资金具备思考能力”，全天候自主寻找最佳回报渠道；然而在现实中，传统银行的黏性依然受到用户惰性、未兑现支票以及繁琐的账户绑定等日常摩擦的有力支撑。 对谈总结认为，企业财务平台正在从依靠刷卡费率的金融工具演化为驱动资金与工作流自动化的效率中枢。尽管AI极大地压缩了软件研发与知识审核的边际成本，但真实商业环境中的制度合规、不可复制的专有历史数据、深层系统累积的边缘处理经验以及传统银行业的转换摩擦，依然构成了限制纯技术颠覆的长期现实屏障。

[16:13] Eric Glyman: My AR is their AP. [21:58] Alex Rampell: It's like the start of 'The Dark Knight'... they're all wearing the Joker mask. [32:15] Eric Glyman: A penny saved is equivalent to 12 pennies of revenue earned. [48:17] Eric Glyman: We will sell your expenses, done. [51:31] Alex Rampell: The blue bottle of expense reports. [53:47] Alex Rampell: The marginal cost of arguing has gone down to zero. [53:47] Alex Rampell: Your agent will talk to my agent. [60:11] Alex Rampell: Guys in windbreakers. [69:10] John: This kid is keeping you at Chase.

[0:00] I don't know if we need, if we want to do this, but— [0:02] I already look like that naturally, so. [0:05] Oh, this is Kevin from "The Office." [0:07] Yeah, yeah. [0:09] Eric Glyman is the co-founder and CEO of Ramp, [0:11] the corporate card and finance automation platform. [0:13] Founded in 2019, Ramp has taken the industry by storm. [0:17] Oh, cheers. [0:18] Cheers. [0:21] Can you describe maybe a good framing, [0:22] just the Ramp business today? [0:24] What's the biggest part of the business [0:25] in terms of where you make money? [0:27] What are the new growth lines? [0:28] Any metrics you can share on the scale? [0:30] Alex and I have a million questions, [0:32] but maybe you can start by just framing it up for us. [0:34] Yeah, of course. [0:36] So first, just the pace that this has come together [0:41] has been pretty remarkable. [0:43] Seven years, yeah. [0:43] You know? [0:44] I think by the time we were, six years and change, [0:48] the company had passed over a billion a year in revenue. [0:51] The largest portion of that is card. [0:55] And so, that might be a classic [0:56] kind of interchange-based model. [0:59] But behind it— [1:00] So this is people, [1:00] in their business they have spend cards, [1:02] everyone walking around the company has a Ramp card. [1:03] And you run interchange on those. [1:05] That's right. [1:07] Next, you can think about bill payments and software. [1:13] Software, it's a [1:14] two-something-year-old business line, [1:17] just about two years and some months. [1:19] That's over 100 million a year [1:20] business line in and of itself. [1:21] And that can be advanced functionality to maybe manage [1:25] lots of entities, to automate aspects of accounting, [1:28] maybe aspects of procurement, bill payments. [1:31] So this can be sending checks, wires, ACH [1:34] and that's predominantly a float [1:36] as well in some cases foreign exchange transaction business. [1:39] Treasury. [1:41] That's a product that is about a year old, [1:43] several billion dollars of deposits. [1:45] Some of that is checking-like products. [1:47] Some of that is more of an investment [1:49] and money ladder type product. [1:52] And then the last, the other ones are procurement [1:55] and then travel, which is a bit of an in-kind. [1:58] And what's been so interesting is that if you break down [2:00] and look towards maybe let's say contribution profit, [2:03] gross profit of the business, [2:06] a few years ago it would've been [2:08] 90 plus percent card. [2:10] I think by the end of this year, the second, [2:13] third, fourth, fifth lines of business [2:15] will comprise and aggregate [2:18] the majority of Ramp's business. [2:20] And so, it's evolved into this platform by which [2:24] if you're trying to operate your company, [2:26] it's just a lot more efficient. [2:29] You spend less. I think people know Ramp for, [2:33] we help the average company cut their expenses [2:35] by about 5% per year. [2:37] The thing that's been really fascinating [2:40] is people are starting to use these products in aggregate. [2:42] Not only are they not paying [2:43] for five, six sets of point solutions, [2:46] but they're also not wasting time. [2:48] They're growing faster. [2:49] The average customer last year on Ramp [2:51] grew their revenue by, I think it was 16%, [2:56] which compared to, in the United States, [2:59] the average business— [3:00] Much higher than that. [3:01] Yeah, I think it's 5% in the US. [3:01] And so, in some sense what we're trying to do [3:05] is be not just a better platform, [3:08] but a better, almost kind of digital brain for organizations [3:12] to allocate resources and make sure spend is not wasted. [3:15] Yeah. [3:16] So, you start by selling, [3:21] say, spend cards into a business. [3:23] The CFO likes it, the employees like it, [3:26] and then that gives you permission to go sell [3:27] some of this other functionality. [3:29] And so it's kind of a classic multi-product cross sell. [3:32] Is it always starting with spend cards [3:34] or do you now have multiple front doors [3:36] into the business? [3:37] It's been the big story of the past year. [3:39] I think there were, [3:40] over the last quarter, [3:41] thousands of businesses that came in just for bill payments. [3:45] There's accounting firms that are becoming into, [3:50] they say, "We're a Ramp shop here [3:52] if you want to work with us. [3:54] We'll bring in bill payments, [3:55] treasury..." [3:58] We use Ramp to go power that [4:00] and they can manage 10, 100, 200 clients [4:04] all through one platform. [4:07] And we're excited. [4:07] I mean, probably the fastest growing line [4:10] of business is procurement. [4:12] Or, they're single clients that are looking [4:13] to kind of bring in and do full-scale [4:16] purchase orders across, [4:19] whether it's single [4:20] or a dozen plus entities, all the way through. [4:23] And so, that's been [4:24] the really fun story of the past year. [4:27] Do you have a view on what the right policy is [4:30] for employee spend within a business? [4:32] I love that you ask this, yeah. [4:34] 37signals or someone will say, [4:36] "Companies are so lame. [4:38] you just give everyone a credit card [4:39] and trust people and it's fine." [4:40] And then meanwhile, large companies are like, [4:43] "Well, this hotel is in a Tier 2 city [4:46] and therefore there's this dollar limit [4:47] and it must be booked 14 days in advance." [4:49] And so, you see companies operate at really [4:51] two ends of the spectrum in terms of [4:53] how much flexibility they give people. [4:55] What is correct for a company, or does it just vary by size? [4:57] I love this question. [5:00] And the interesting part, if you really kind of [5:04] dig into Ramp's businesses [5:05] we have ways of back testing [5:07] and actually understanding [5:10] based on how strict your policy is, [5:13] how things are running. [5:14] Does that have an impact on how much time [5:17] your employees are, let's say doing expenses, [5:20] how quickly you're growing. [5:22] And you can start to actually compare [5:24] based off of the operating hygiene of the company… [5:27] What are end margins? [5:30] What is a pace of growth that occurs at the company? [5:33] And I bring this back to what— [5:34] So you can do a correlational study [5:36] between expense policies [5:38] and company growth. [5:39] Yeah. [5:41] And look, it's pretty interesting. [5:43] I mean, the first, there's an aspect of, [5:47] in most companies, particularly high growth companies, [5:49] it tends to resemble something closer to [5:51] the "no rules rules" Netflix approach [5:54] of we're going to trust but verify and shine a light on it. [5:57] Spend tends to be reasonably permissive. [6:00] But then part of the breakthrough of the past year [6:02] is you could basically take an expense policy, [6:05] let's say it's written [6:07] in plain English. [6:09] "If the flight is more than five hours, [6:10] you can take business. [6:11] If it's under..." Things that would've been horrible to go [6:14] and have your managers go and verify this stuff, [6:18] now you can run that through an LLM. [6:20] I think that we're processing over 100,000 expenses a day [6:24] that are being reviewed agentically. [6:27] This is a very fast-growing subset of the business. [6:30] So you give the Ramp agent [6:33] this company's expense policy and then it applies it [6:35] to the transactions that are coming through. [6:37] Exactly. [6:38] And so you can start to do things [6:39] like, I think about the episode you'd had with Susan Li. [6:43] It was incredibly instructive where, [6:46] she sort of says, it would feel so silly [6:51] and it's not a great use of my time in some sense [6:52] to be a very expensive machine learning algorithm to do, [6:55] to review expenses. [6:56] But yet somehow a lot of people in companies [6:58] are reduced to that thanks to Sarbanes-Oxley [7:00] and the rules around it. [7:02] Now you can have an agent functionally that does that, [7:04] that takes that aspect of the job. [7:06] And so, it has access to the full set of data [7:09] that you would, all the transaction-related metadata, [7:12] the receipt data, the timing data, the policy data. [7:15] It then can go in real-time review. [7:19] Was this in or out? [7:20] Can have the full audit trail explaining its reasoning. [7:23] And today it's over 99% accurate, [7:25] which turns out it's much more accurate than people are. [7:28] So people don't know the expense policy [7:29] and expenses are pretty automated. [7:31] And I find this thing super interesting, in part [7:35] because I don't think it should be anyone's job [7:39] to go and review people's expenses, [7:40] but somehow, while it's in no one's JD, [7:43] the law kind of requires everyone to waste an hour [7:45] of their time every month, if you're a manager. [7:47] And does the law allow for it to be reviewed agentically [7:49] and then just signed off on by the human? [7:51] It does, yeah. [7:52] The goal of this is separation of duties. [7:55] That you yourself [7:57] are not reviewing your own expenses. [7:59] There needs to be [8:00] a set of procedures that govern [8:03] and actually are effective through the action of, okay, [8:06] the rule says this. [8:08] Did you take some steps to do this? [8:09] It can be a system— [8:10] Yeah, you can't self-certify. [8:12] The funny thing is that, I mean, one way of doing this… [8:15] Actually, we used to do this is [8:16] why not just post your expense reports [8:18] in the public eye, right? [8:20] So, and the reason to do this is actually it could go wrong. [8:22] You could say, "I'm going to spend... [8:23] You stay at the Four Seasons, [8:24] I'm going stay at the Five Seasons." [8:26] You could go that direction. [8:27] But what you actually want is a moral code. [8:30] It's like, [8:30] if you were the owner of the business [8:32] it is your own money. Yes. [8:33] So if you were staying at the Five Seasons, [8:35] I've just made that hotel chain up, [8:37] it's like, you're— [8:38] Sounds pretty nice. [8:38] You're stealing from yourself, why would you do that? [8:41] The problem is that once you get [8:43] to tens of thousands of people, [8:45] how do you impose that moral code? [8:46] And actually like the two [8:47] tent poles that you just mentioned, they're both bad. [8:51] Because if you have to stay at a crappy hotel [8:54] and you're about to sign a two million dollar contract [8:57] and you didn't get any sleep. It's the Motel 6 [9:00] and you take a two-hour taxi, [9:01] it's like, all right, you saved the company money. [9:03] That does not make sense. [9:04] Stay at the Four Seasons. [9:06] But you can't really embody that [9:08] in, "if this, then that." [9:09] Because what you want is this moral code. [9:12] Like that's what you want. [9:14] And like how do you actually instill that? [9:16] And there's almost a behavioral psychology way, [9:18] which is like, we'll just [9:19] show what people are spending. [9:21] If you're at the top of the leaderboard, [9:22] we might trust but verify and verify a lot more. [9:25] But that's the hard thing to do [9:27] and actually preserve as the company goes from [9:28] the founder, to the founder plus the founder's brother [9:31] to many, many more people. [9:33] How do you keep that same esprit de corps [9:36] or this informal code. [9:39] I think it comes back to… [9:43] What was the name of the movie? [9:44] "12 Angry Men," I think it was [9:46] where, as you're watching the film, [9:49] you get some level of detail [9:51] and someone seems very guilty. [9:53] It's a dead shut case. [9:55] And as more evidence emerges, [9:56] you realize actually, this may be different. [9:59] And you need more context [10:01] in order to arrive at maybe a moral conclusion. [10:04] Did it relate to closing a project? [10:07] How does it relate to an outcome [10:11] that maybe you'll see emerge one month, two months [10:13] after you stayed at the Five Seasons [10:15] and closed the deal that changed the company, [10:17] whatever the example is. [10:18] I would argue in favor of tools [10:22] that have more context, that do more jobs of work. [10:26] Because not only can you do the narrow job better, [10:30] but you can start to get at the answer [10:31] of what actually is right for shareholders. [10:34] Where should we be allocating capital? [10:35] It's so interesting to hear [10:36] because every company approaches this differently. [10:40] Like I'm on the board of Wise [10:41] and you know, Kristo flies coach everywhere, [10:44] everybody flies, that's just the policy [10:47] but it might be penny-wise, pound-foolish. [10:50] Again, not for me to say, [10:51] but I also really admire the corporate code. [10:54] It's just like you, actually, that is cultural. [10:57] It's kind of weird [10:58] to say expenses are cultural, but they are. [11:00] For sure. [11:00] Yeah, it's shared— [11:03] Shared belief system. [11:03] It's a shared belief system. [11:05] It's like, what is the culture of the company? [11:06] You talked about your expansion into Bill Pay. [11:10] It seems like bill payment [11:13] has proven, over the past 40 years, [11:16] uniquely resistant to automation and modernization. [11:20] If you look at how a typical business [11:23] pays their rent or if you order a keg of Guinness, [11:27] you'll get a PDF emailed to you [11:30] by some person who has a relationship with you [11:33] and there will be bank account details [11:34] which hopefully are correct. [11:36] And you send a payment to those. [11:39] And even bill payment products, [11:44] you guys have a bunch of technology here, [11:45] but it's like, we scan the PDF in an automated way [11:49] or we ensure that the bank account details [11:50] weren't mistyped or something. [11:52] But it's kind of "putting a little bit of lipstick [11:54] on the pig" of the whole system [11:56] being super antiquated. [11:58] Why is the system so antiquated [11:59] when it comes to bill payment in particular? [12:02] I think this is some of what Alex was bringing up. [12:05] It's the constraints of programming things [12:08] in an "if this, then that" kind of world are very heavy. [12:11] There's a lot of complexity. [12:12] And when you think about the nuance [12:15] and algorithms that govern [12:18] why do companies spend money under some circumstances, [12:21] how hard it can be to record. [12:24] I wouldn't overlook... [12:27] Let's say that you're a manufacturer [12:30] and you're buying some asset which you're going to use [12:32] for five years and it's going to depreciate. [12:34] It's a very different accounting treatment [12:35] versus I'm buying a pay-as-you-go SaaS app. [12:39] All the complexities around that [12:41] might govern whether or not you decide to spend. [12:43] And then finally, once you get all this detail, [12:46] how do you review this [12:47] and decide where to allocate your next marginal dollar [12:50] is very complicated. [12:52] But many systems are in place upgraded [12:55] despite the fact that that's pretty complex. [12:57] So credit cards started [12:59] with no real-time authorization system, [13:01] which is crazy. [13:02] And you just hoped that they were good for— [13:03] Carbon copy. [13:06] Exactly. The machine with the pleasing sound. [13:08] They'd do great in the current ASMR environment. [13:10] That's true. [13:12] But then they added, [13:13] you could call up and get an authorization [13:15] and then they obviously added [13:16] to the current systems we know where— [13:18] The modem, you forgot the modem. [13:19] Exactly. And then the modem. [13:20] And so, in place credit cards were upgraded [13:24] with much better capabilities. [13:25] And similarly, if you look at checks [13:27] and how they work, [13:28] even though we think of checks as antiquated, [13:30] it used to be the case that physical checks had [13:32] to be flown all around the country [13:33] to be physically settled. [13:35] And then there was the Check 21 Act in, [13:37] what was it, the year 2000? [13:38] Where they said a scan of a check is good enough [13:41] and so they could be digitized by the banks [13:43] and then shredded. [13:44] And then there was the, [13:45] you can take a photo of a check with your phone. [13:47] And so we've managed to take this super [13:49] old-timey check system [13:50] and despite the fact it's super old-timey, [13:52] in some ways it has actually been meaningfully upgraded. [13:54] And if you looked at the bill pay system from the outside, [13:57] you would say we should have DNS for companies. [14:00] So rather than a company [14:01] like sending you their bank account details, [14:03] you should look them up in some central clearinghouse. [14:06] And that way you confirm that you're not being phished. [14:08] A lot of spear-phishing attacks work that way. [14:10] Although you referenced this. [14:11] So there's actually an indifference point [14:13] that you can graph. [14:14] If interest rates go up high enough, [14:16] if I send you a check, [14:18] which is like a paper mail check [14:20] and that takes five days [14:22] and hopefully the USPS goes on strike [14:23] for two more weeks so I've got the postmark. [14:26] I actually benefit… It's actually cheaper for me. [14:29] You legally paid, right? [14:30] It's crazy. [14:31] No, it's crazy. [14:32] This is the problem. [14:32] Like credit cards, everybody... [14:33] I am in the store, I want the TV. [14:35] You want to sell me the TV. [14:37] We both want this done right away. [14:39] And in the knuckle-crunching, ASMR, [14:41] carbon-copy days, [14:43] they might not have been willing to sell that to me [14:46] because they didn't know if I was good for the money. [14:47] They'd never seen me before. [14:49] And the interesting thing is that [14:51] AR and AP are almost an adversarial process. [14:53] You mentioned this. [14:54] What do you want to do as a controller? [14:57] Well, you want to pay as late as possible [14:59] and you want to collect as early as possible. [15:01] And then you have this weird thing where checks, [15:04] it's not just an accident [15:06] that they stick around for a long time. [15:08] It's like there's a financial incentive for one party. [15:10] So that's true on payment timing, I agree with that [15:14] but there are some things that are just a dead weight loss, [15:16] like the ability to fat-finger a bank account. [15:20] No one benefits from that. [15:21] Everyone has a horrible time [15:22] if that happens. No, that Nigerian guy. [15:23] There's some people that benefit. [15:25] Nigerian princes. [15:26] Let's think of everyone. [15:28] I will posit that many networks [15:30] have been upgraded in place. [15:32] But somehow the loose network of businesses [15:34] paying each other via PDF invoices [15:37] has proven very resistant to in-place upgrades. [15:40] Also, I love your notion of DNS [15:43] for payments or for expenses. [15:45] So, I don't know if you saw this, [15:46] Google wanted to issue a 100-year bond. [15:49] Yes. [15:50] So, why is that interesting? [15:51] Imagine that I am owed money by Google. [15:54] I am one of the small businesses that you reference. [15:56] Their job is to pay me as late as possible. [15:59] Yes. [16:00] They can borrow money for a hundred years [16:02] at like 3% or whatever it is, [16:04] cheaper than the US government probably. [16:06] T plus a hundred, yeah. [16:07] I mean it's a little more expensive [16:09] than the US government, but they are borrowing... [16:10] That's crazy. [16:13] I can borrow money at 20%, [16:15] but why do I need to borrow money? [16:17] Because Google's paying me late. [16:19] Those jerks. [16:20] So, I should be able to borrow money [16:22] kind of at the rate at which Google is borrowing money [16:26] because my AR... [16:28] My AR is their AP. [16:30] But the problem is, unless you put all this together, [16:32] you would just look at me and you're like, [16:34] "Well, you're just a little schmuck. [16:35] I'm going to charge you 18%." [16:37] It's like, yeah, but it's Google. [16:39] They could borrow money at T plus [16:40] whatever, you know, SOFR plus a hundred. [16:43] 100%. [16:44] That's not fair. [16:45] But the only way to really solve that [16:47] is with more data and actually entangling... [16:49] Not entangling, but just kind of [16:50] connecting these things together. [16:51] For sure. [16:52] What's your vision for [16:54] once AI is doing a lot of software engineering, [16:56] what does that look like in three or four years time? [17:00] Is the competitive equilibrium similar, [17:03] but the expectations for [17:07] the amount of software and functionality businesses [17:09] are shipping is much higher? [17:12] Are software teams similar looking, different looking? [17:15] Again, you're doing a lot of AI engineering. [17:18] What does the future hold? [17:19] It's all blurring. [17:21] It's totally crazy. [17:22] There's a designer shipping code. [17:28] Marketing at Ramp reports into [17:30] our CTO Karim, my co-founder [17:34] who's doing some of like the best marketing [17:36] I've ever experienced and seen. [17:38] We have customer support agents [17:39] like shipping code to production too. [17:41] I just think that the half-life of you see a problem [17:47] to how long it takes for you to go fix it [17:49] and do something about it is shrinking immensely. [17:52] Or if you want to change a color of a button, [17:54] you can just say, "Hey at Ramp Inspect, [17:57] can you change the color of this button?" [17:59] And it goes and spins it up [18:01] and it does it within minutes. [18:02] It verifies and validates it. [18:04] And so, I actually think in some sense, [18:07] these classic barriers [18:09] that software businesses had [18:11] are clearly going to erode. [18:14] Yes but, there's the old joke of [18:16] "It's much more fun to write code than read code," [18:18] which explains a lot of software engineer's behavior. [18:21] And it becomes easier to change the button. [18:23] Is that another instance of it being more fun [18:26] to write code than read code? [18:27] Where lines of code are a liability and not an asset [18:31] because they're something [18:32] you have to reckon with. [18:33] And so, are companies at some level incurring [18:36] some tech debt now where they're adding a bunch of code, [18:39] which will maybe be harder to reason with later? [18:42] Or do you just get bailed out by [18:44] the models getting better? [18:46] It's really interesting. [18:47] And by the way, one of the other interesting [18:49] sub-conversations I'm hearing a lot, too, is [18:52] when you think about where tech debt comes from, [18:57] there's a set of conditions [18:58] and trade-offs you make, [18:59] you write things in a discreet [19:01] and deterministic way. [19:03] Under these circumstances, follow this code path. [19:05] Under this circumstance, follow this other one. [19:07] In a world where there's LLMs [19:09] and the models themselves are improving, [19:11] I think it's entirely possible that the way [19:13] that code is written is you say, [19:15] "Here's what I'm solving for. Under [19:17] these kinds of conditions, here's what I want to occur. [19:19] Go write the code that drives this outcome." [19:23] And maybe with the models such as they are today [19:27] can get it done in sort of a [19:28] spaghetti code written fashion. [19:30] But it kind of works even though it's some kind of [19:32] Rube Goldberg machine underneath the hood. [19:35] But as these models get much smarter, [19:37] you might write your code base in such a way where you say, [19:39] "Every year, rewrite the underlying code, [19:43] but here's the outcome I can drive." [19:44] And today you accomplish my outcome 90% of the time, [19:47] 95, 98, 99, 100. [19:49] And do you just have kind of self-healing [19:51] and writing code all the way through? [19:53] And this notion of like underlying code does goes away [19:55] because we're writing things in this different manner. [19:58] Obviously, there's real conditions in which [20:01] that's not going to… That won't occur. [20:02] Like, if you're writing code that needs to have, [20:05] four nines of accuracy and uptime, [20:08] like that's probably not the methods you're using. [20:11] But if you're like a growth engineering team, [20:14] that's probably what you're doing today [20:16] if you're kind of on the leading edge of this stuff. [20:18] And so, I find this stuff very fascinating. [20:23] And when I think about [20:24] the deeper implications of this stuff, [20:26] I think that if you are completing this small amount [20:31] of cognitive work... [20:33] Let's say you are an expense app. [20:36] The spend has occurred, [20:39] it needs to go get the... [20:41] Write it down somewhere and get someone's approval. [20:44] And that's all the knowledge work that you're doing. [20:46] That's very few tokens in order to accomplish that, [20:49] both to create the infrastructure in order to facilitate it, [20:51] like that probably evaporates. [20:54] Anyone can probably custom write that kind of app. [20:56] Whereas if you're doing much deeper kind of work, [20:59] such to underwrite a company, [21:02] provide financing, automate areas of accounting. [21:06] I think the fitness function for companies [21:09] becomes can you actually do things in such a way [21:11] where even if you could spend tokens on it, [21:15] it would take more tokens to create the thing [21:16] or do that work than the system maybe that you've built [21:19] to kind of drive that outcome. [21:22] I just think the rules of what it means [21:24] to be a software company are changing. [21:25] I think network effects and what you're building [21:27] is more important than ever. [21:29] Like, if you think about the classic set of moats [21:32] that are there, I think that this question of [21:35] where are their moats in a world where... [21:39] I like Dario's way of putting this. [21:41] If you have a country of geniuses— [21:43] Data center, yeah. [21:46] Living somewhere and they're writing code [21:48] and they're incentivized to compete with you, too. [21:51] If you're not really following [21:53] what are the classic four moats [21:54] and building towards that, [21:55] like, I think life gets a lot harder. [21:57] Yes. [21:58] Do you guys have a house view on [22:00] the new competitive equilibrium [22:05] with a lot of AI engineering really working? [22:08] Well, I think there are a couple things. [22:09] I think Marc talked about this a little while ago, [22:12] but what is an EPD team? [22:14] Engineering product design team. [22:17] Product development team. [22:17] You normally have one product manager, [22:19] maybe you have five to eight engineers and one designer. [22:22] And now everybody has Cursor and Claude Code. [22:25] So the designer is like, "I don't need [22:26] any of these bozos." [22:28] The product manager says, "I don't need any of those bozos." [22:30] Each of the engineers are like, [22:32] "I don't need any of these bozos." [22:34] And they're all— [22:34] It's like the start of "The Dark Knight." [22:35] And they're all right. [22:36] It really is, it's… Exactly. [22:38] And they're all right. [22:39] It's like they're all wearing the Joker mask. [22:40] Or, the Joker mask [22:42] is Claude Code or Cursor. [22:44] But the problem is, because I was talking to one of my CEOs, [22:48] this is a pretty scaled company, lots of revenue. [22:51] And they use all these tools, [22:53] they're not more efficient. [22:54] And he is like, "Why aren't we more efficient?" [22:56] And it's two things. [22:57] Number one, it's a real company, [22:58] they have real customers. [22:59] They can't just push things and hope for the best. [23:02] But it's like you actually have [23:03] an HR problem [23:05] in that if you're building from scratch, [23:07] granted you have nothing, you have no tech debt, [23:09] you don't really have to worry about customers, [23:10] you don't have any, [23:12] but you wouldn't have an eight person EPD team. [23:15] You would probably just have each person... [23:18] We have eight different product managers [23:20] slash engineers slash whatever. [23:21] Each one is a pod unto themselves. [23:24] So that's one thing. [23:27] I totally agree on the data moats [23:28] where that has become more important. [23:31] I'll tell you a story that you might like. [23:32] I met this company, vLex. [23:34] It's a 25-year-old company. [23:36] This guy Alex bought up every legal record in Spain, [23:40] like probably going back to 1492. [23:42] Buys the Ferdinand and Isabella... [23:45] Here's the document, [23:46] I'm going to take a picture of that, [23:48] I'll sell it to every law firm. [23:50] And he built like [23:51] a $20 million a year SaaS business, [23:54] bootstrapped for 25 years. [23:55] And then it went to like a hundred million in one year. [23:58] Now why is that? [23:59] Because he used to sell this document [24:01] or he used to sell a subscription [24:03] to Kirkland & Ellis and Latham & Watkins [24:04] and all these big law firms. [24:06] And a paralegal would take that, [24:08] they would turn it into a document. [24:10] They'd charge the client $10,000 [24:13] and ChatGPT can't do that. [24:15] GenAI can't do that. [24:18] But you know who can do that is vLex. [24:20] Another good example of this, [24:23] of who has proprietary data? [24:25] A lot of times, the data's free. [24:27] This is the really cool thing. [24:28] Or it's sitting, I wouldn't call your data free, [24:31] but it's not like for-sale data sets. [24:34] Do you know domaintools.com? [24:36] It's my favorite business. [24:37] So DomainTools [24:38] runs a cron job every day on every single internet website. [24:43] They do a WHOIS lookup [24:46] every single day. [24:47] And they build a historical [24:47] record of that which no one has. [24:48] So if you want to see who owned stripe.com in 1999, [24:52] well, it was free in 1999. [24:54] If you invented time machine, go run that WHOIS query [24:56] in your terminal but you can't do that. [25:00] So you have to pay them. [25:01] But all that data is free [25:02] or FlightAware with ADS-B data. [25:04] So you have a lot [25:05] of these weird businesses. [25:06] And of course this is not a new idea, [25:08] like FactSet, Bloomberg... [25:09] You have aggregators of data, [25:11] but that becomes so much more valuable. [25:13] Totally. [25:14] Because your 10,000 geniuses cannot do that. [25:17] 100%. [25:18] You always experience this if [25:20] you hire someone extraordinarily talented, [25:23] like top of the class at MIT [25:26] and you say, "All right, join the engineering team [25:29] and go ship the code." [25:30] And they ship something kind of crazy [25:32] on the first day. [25:33] And it's sort of slow. [25:34] And it's like, [25:35] you can have someone who is intellectually [25:38] an absolute giant, [25:39] but learning kind of how a company does things [25:43] the way it does. [25:44] Learning the procedures and nuances [25:46] of an organization takes time. [25:48] It's part of what makes working with interns [25:50] or junior engineers fun. [25:52] I think the learning curve is obviously immensely steep. [25:54] And I think that the speed at which, [25:56] with sufficient access to the data [25:57] and the procedures of a company, [25:59] they're going to get there very, very fast. [26:01] But these moats are real, you know? [26:04] Well, the other moat [26:05] I would argue is the dark-mattered moat. [26:07] Yes. And what I mean is [26:08] you know, like the Milky Way, [26:11] nobody could figure out why the mass is the mass [26:13] because it's like all the observed stars [26:14] and planets and everything else. [26:15] It's, no, actually the vast majority of the matter, [26:18] or energy, because they're the same thing, [26:20] is dark matter, dark energy. [26:22] And what does that mean? Nobody the hell knows. [26:23] But it's just like, we don't know it. [26:25] And for products, it's like you built, [26:26] if I tell Claude, "Go clone Ramp." [26:31] Because this is the SaaSpocalypse [26:33] that's happening right now. [26:34] It's like, oh, of course. [26:35] Well, I just went to the website [26:36] and now I have a website that looks like Ramp. [26:37] No, no, no. There are like nine million edge cases. [26:40] The tech debt sounds bad [26:41] because it's a pejorative, debt = bad. [26:43] But actually it's good [26:45] because what you've done is you've uncovered [26:47] every single problem that can go wrong. [26:49] The fact that, I mean, a great example, [26:51] remember when the power went out in San Francisco [26:53] a month and a half ago? [26:55] So we have this amazing thing called Waymo. [26:57] Waymo had not figured out [26:59] this corner case [27:00] of what if the power goes out? [27:02] It was a thundering herd problem with the customer, [27:03] or the human overrides, yeah. [27:05] But they had no idea. [27:07] And it's actually great that that happened. [27:09] You need these problems to happen. [27:11] For sure. [27:12] And just because my background, [27:13] I used to write shareware, [27:14] this try-before-you-buy software. [27:16] Shareware was a big thing. [27:17] That's a blast from the past. [27:18] It's a blast from the past. [27:19] So there's a website that's part of CNET— [27:21] That might come back, you know? [27:22] Well, but it's freemium. [27:24] This is where I'm going with this story [27:25] and you'll like this. [27:26] So Download.com was the preeminent site [27:29] for downloading shareware. [27:30] It was like... [27:31] CNET was one of the most popular properties in 1999, 2000. [27:35] So, all of the top downloadable [27:36] software products were there— [27:38] Download.com sounds like a company [27:39] that had a Super Bowl ad. [27:41] They would've. [27:41] I bet they did, actually. They probably had two. [27:43] I'd be shocked. They probably had five. [27:45] Right between Pets.com and something else. [27:47] So Download.com had all the popular products [27:50] and then this site called Elance shows up [27:53] and Elance is now called Upwork. [27:55] This connected you with work that you would want done [27:59] by very smart people, basically in the developing world. [28:02] So I want a software product built. [28:03] Here's everybody in Romania and Russia and India. [28:06] I want translation, I want graphics. [28:08] So what ended up happening [28:10] is people would look at the Download.com top list [28:12] because before it was a very cottage industry [28:13] of who wrote shareware and it was very profitable. [28:15] Like ID software shareware company, [28:18] McAfee shareware company. [28:20] Actually Cybersource started off processing payments [28:23] for shareware companies. [28:24] That's how they got McAfee. [28:26] That's funny. [28:27] Very interesting history to this [28:28] we can talk about some other time, [28:29] but basically what happened is people found Elance, [28:31] like, "Oh, I could pay anybody $500 [28:34] to clone anything on this list. [28:36] Everything on this list makes [28:37] tens of millions of dollars a year. [28:39] Let's go." [28:40] And the cloning never worked. [28:42] It functionally worked. [28:43] Because if I say "I'm going to clone Eric." [28:45] I might not know that you have a pancreas. [28:47] Right? [28:48] I'm sure you do. [28:49] I might not know that you have two kidneys. [28:51] So the problem is that cloning just, it goes skin deep [28:55] and like that's the embedded advantage of these companies. [28:59] Do you then think [29:00] the SaaSpocalypse is kind of irrational? [29:02] I think it depends. [29:03] I think some of it is very rational. [29:04] Because if I want, if I have a feature [29:06] that became a company, [29:07] and I know this sounds [29:08] like a very negative thing to say, but [29:10] I want to get paged if my server goes down. [29:13] It's a very logical reason [29:15] to build a company called PagerDuty. [29:18] That now has a lot of stuff around it [29:21] but I might just say, "Hey, whenever my servers..." [29:24] There are corner cases, [29:25] there is the proverbial the power goes out in San Francisco, [29:28] do something differently. [29:29] That's very, very different than NetSuite, [29:32] which is, I have a saying that I like, [29:33] which is the best companies have hostages, not customers, [29:35] at least in enterprise software. [29:36] Not for you because you actually have INPS. [29:38] But that they're much harder to go rip that out [29:42] and a lot of it is like the "Goldilocks zone" [29:45] that you operate in. [29:46] If you're too expensive, [29:48] of course I'm going to try to rip that thing out. [29:50] If it's so cheap then I forgot nobody even used it. [29:54] So you have to be in this Goldilocks zone, [29:56] you have to have enough complexity. [29:57] And ideally the front end is different than the back end. [30:00] So, Workday I think is... [30:03] Nobody's going to get rid of Workday. [30:06] And I actually credit David Ricardo with this. [30:09] Like, comparative advantage. [30:10] Sure, you could grow your own food. [30:14] You could plumb your own plumbing. [30:17] But you're not going to do that. [30:18] And somebody could do your own HR system. [30:20] Yes. [30:21] The one maybe pitch for Stripe [30:25] and others that I would say here is, [30:29] I do think that that is right. [30:32] And there is a lot of complexity [30:34] and local tax law for payments [30:37] and things that make Workday [30:39] an extraordinary business. [30:41] There was this overarching macro question of [30:43] how does work get done? [30:45] If you believe that. [30:46] You know, work can get done through tokens [30:50] and models. [30:52] And presumably [30:54] it doesn't go through payroll, [30:55] probably goes through a credit card or check [30:58] to these types of companies. [30:59] And so, the share of, [31:01] it's almost this sort of simple math [31:03] that if you X-rayed the P&L of most companies, [31:06] the majority of it classically, [31:07] at least for asset-light firms, is payroll. [31:10] You're paying for people to do things [31:12] and SaaS is maybe a small percentage [31:15] of a company's income or a company's cost base. [31:23] Does suddenly that share grow to high single digits, [31:28] low double digits, significant double digits? [31:31] And does the share actually move where, [31:33] the payroll economy itself, [31:35] even though maybe it grows, becomes less relevant? [31:37] I think this is the really funny part of even [31:39] as this has been going on, [31:40] what do I know as a private company [31:42] other than seeing our own payment data. [31:44] But you start to see these companies [31:47] beating their earnings [31:48] in a way they never have before where [31:51] the terminal value perhaps is falling out. [31:53] But they're saying, [31:56] "Shareholders, we're reaccelerating." [31:58] We haven't seen this since 2021. [31:59] We're going faster and you're going to have [32:01] multiple progressive quarters of this [32:03] actually occurring for certain types of businesses. [32:04] And so, it's a really interesting time to... [32:08] I very violently agree with your view [32:11] is like yeah, it depends. [32:13] But it's doing really well. [32:15] That's my brilliant comment, it depends. [32:16] Yeah, you mentioned... [32:19] Very Howard Marks of you. [32:23] As you're hearing from Eric, [32:24] Ramp has become the default way [32:26] a lot of American companies manage spend. [32:28] Stablecoins allowed that functionality to work [32:30] in many countries all at once. [32:32] With Ramp's stablecoin-backed corporate cards, [32:34] businesses can fund a balance with stablecoins, [32:37] issue cards against those balances instantly, [32:39] and allow employees to spend anywhere cards are accepted [32:41] without the business having to think about [32:43] stablecoins all the time. [32:45] Same card experience, same controls, [32:48] just a much more global set of rails underneath. [32:50] This is one of the many practical ways [32:52] we're seeing businesses on Stripe use stablecoins [32:54] by launching card programs in many more countries [32:57] and doing so in much less time [32:58] than it would've taken otherwise. [33:00] If you're thinking about using stablecoins to expand, [33:02] Stripe can help. [33:06] You mentioned your spend data. [33:08] What does the spend data that Ramp sees [33:11] tell us about the economy? [33:14] I think it is stronger [33:17] than many people understand in lots of ways. [33:23] I mean first, I'll go back to one of the things [33:25] that perplexed our team for a long time. [33:28] Our economists on the team [33:31] saw this data [33:32] even as recently as last year [33:34] were the Census Bureau would do these periodic surveys. [33:38] They'd go out and ask [33:41] "How much is your business using AI [33:44] to produce goods or services?" [33:45] A very refined economic way [33:48] of wording the questions. [33:49] And they would come back [33:51] with these pronouncements saying [33:52] a single digit percent of businesses in the US [33:55] have adopted AI. [33:56] And we looked at our data and [33:59] we support over 55,000 businesses. [34:02] We lean a little bit towards tech but not heavily. [34:05] It kind of resembles the distribution of businesses [34:07] you would see in the States. [34:09] And well, the majority of businesses have used AI. [34:14] You look at businesses whether they're paying for— [34:16] As in they subscribe to ChatGPT or Anthropic [34:18] or something like that. Exactly. [34:19] Or, maybe a business that is a true [34:23] agentic… Cognition [34:24] or a Cursor, something like that [34:26] where this is kind of vertical application. [34:29] And so, one, there is this disconnect between [34:32] the use of tools if you look at [34:34] how quickly businesses are adopting [34:36] and responding to these new tools [34:39] versus what maybe people report on. Next: growth. [34:43] I think it's been clear over the last few quarters [34:46] is the US itself is reaccelerating, [34:48] GDP growth is gone from maybe the one to 2%, [34:52] area to four to five [34:54] and you could argue how much of this is, [34:55] a little artificial with subsidies, [34:57] but it's been pretty significant. [34:59] But subsidized by what? [35:00] Some of the Big Beautiful Bill [35:02] as well as some of the tariffs. [35:05] And where that's been reallocated, [35:06] I think some people, I would argue [35:08] on the whole maybe unfairly [35:09] said this. [35:11] I think that there's more durable growth [35:14] inside of the businesses. [35:15] But I would say [35:18] overall business health is much stronger. [35:21] And I think that the really interesting thing [35:22] and again some of this is specific to the data that we see, [35:25] but I think that businesses are generally [35:27] getting more and more savvy about finding tools [35:30] that help them be more efficient. [35:32] And I think what skews our data [35:34] is maybe the savings we drive for people. [35:39] Seems the way of explaining it is like people know the, [35:42] Ben Franklin would say this phrase [35:44] of "a penny saved is a penny earned" [35:45] and this is awesome aphorism [35:47] but the average American business [35:49] has an 8% profit margin. [35:51] And so mathematically speaking, a penny saved is equivalent [35:55] to 12 pennies of revenue earned. [35:57] And so, if you save businesses a lot, [35:59] you end up with these outcomes [36:01] of the average Ramp bases materially outgrowing [36:04] the regular US average. [36:06] And we can clearly see this [36:09] and demonstrate this for our own set of customers. [36:11] I suspect these types of things [36:13] are occurring at businesses adopting [36:15] these new sets of technologies, [36:17] I would argue probably faster, [36:19] very clearly, than what the US [36:20] is maybe, the classic sources are seeing. [36:23] So I think it's much healthier. [36:25] When a business saves money on Ramp, what happens? [36:27] The CFO comes in [36:31] and says, "Oh my god, we're spending— [36:33] Confetti. "Way too much money, [36:35] on confetti. [36:36] We don't need this much confetti." [36:37] Who's buying the confetti? [36:39] "We have huge stockpiles that we don't need." [36:41] What happens when a business saves 5%? [36:43] What are they actually cutting? [36:45] So there's two pieces of it primarily. [36:48] One is like these hard-dollar cost savings [36:50] and then some of this is time, [36:51] and I'll start with the time because it's the squishy one, [36:53] but I think that it's, in some cases, [36:55] actually much more important. [36:56] Oh sure, they save time on their expense reconciliation. [36:59] That makes sense. [37:01] So they're no longer going through line by line [37:02] and therefore there's [37:03] one person who is freed up to go do something else. [37:06] Yeah. [37:07] And I think one of the classic biases [37:10] that people forget is people chronically [37:12] undervalue their own time. [37:14] As a principle of business, [37:17] everyone has an hourly rate. [37:21] Within a business, human time is incredibly expensive. [37:23] I think the lean companies got this right [37:25] and eliminated the bottlenecks there. [37:27] So, okay, that's one form. [37:28] What else? [37:29] Next, all right, when you start to have [37:31] these fine-tuned kind of controls. [37:33] This part has been I think probably [37:36] the biggest lion's share of this. [37:39] Let's go back to one of the earlier conversations [37:41] of why do people use checks, [37:42] this horrible system and purchase orders. [37:45] Well, one of the advantages of checks [37:46] is unless you mail it to someone and you sign it, [37:50] no money can leave your company. [37:52] And what every person knows [37:54] who's ever had a credit card and gone to a gym once [37:57] and had a good New Year's resolution. [37:58] But is you go a couple times [38:00] and you're like, "Goddammit, this gym is charging me." [38:02] The difficulty of sending a check is a feature, not a bug. [38:04] Yeah, exactly. [38:05] But it's part of what Ramp [38:08] was the first to build. One-time use cards. [38:10] Single-use cards, but also more than that, [38:12] merchant blocking. [38:14] So you can, whether it's on one card [38:16] or 10,000 cards at once, [38:18] build in a kill switch to say, "Okay, [38:21] we've signed this new deal with this merchant. [38:24] We are only on Uber, [38:26] we are no longer taking Lyfts. [38:27] And anytime someone goes and tries to go on merchant A or B, [38:30] you can do this." [38:31] Or you sign a new contract and you say, [38:33] "I'm going spend $10,000 only with Salesforce [38:37] and on the 10,001th dollar they try to charge you, [38:40] it declines." [38:41] And what gets to happen, you get to have a conversation [38:43] with their sales team who really wants to make their budget. [38:46] And if you look at this mathematically, [38:48] for most companies that come over a low end, [38:51] maybe a very hygienic company [38:53] that is giving out a few cards [38:55] and the old world [38:57] could durably cut their expenses [38:58] about 2% a year by doing this. [38:59] Some of these very laissez-faire businesses [39:02] are saving just 10%, just through these things. [39:06] Next, you start ending up with these more fine-tuned [39:08] kind of controls where you can say [39:09] under these circumstances [39:11] I want to go and have kind of charges occur. [39:14] Maybe you have an engineer stay till 8:00 PM at night, [39:17] you can go and buy a meal. [39:20] But if you take an Uber on a Saturday, that should turn off. [39:22] And so you have the cards, [39:23] if it's like an Uber on a Saturday auto-decline, [39:27] but you can text and say actually it's for work. [39:29] You send back a yes, it turns on. [39:31] And so it's all these little tiny paper cuts [39:33] that actually start to go into run. [39:34] Next, it's vendor data. [39:36] One of the unique assets of Ramp [39:39] is anytime you spend money, you upload a receipt, [39:42] maybe you upload an invoice or an MSA. [39:44] This takes us back to the Paribus days. [39:46] We know across hundreds of millions of purchases every year, [39:51] not just that you spent money at a software vendor, [39:54] but what did you spend per seat? [39:57] And so if you're a customer on Ramp [39:58] and you're about to go and pay this large bill [40:01] on a random SaaS vendor, [40:03] we can show you in real-time [40:04] before you send the funds, [40:06] you are paying 20% more than the rest of the market. [40:10] Here's kind of the cost curve [40:12] of what others are paying. [40:13] And actually by going and empowering your procurement team— [40:16] Do you have enough data to do that? [40:18] Because my AWS bill is not like your AWS bill. [40:20] And so, don't you need to know [40:22] how many instances we're running [40:23] to know if this is a good deal or not? [40:25] You can, but you can still kind of get down to the level [40:30] of is there a bulk discount or not? [40:31] And start to know of, is there a negotiated aspect? [40:34] There's other types of things where I'd argue [40:39] maybe a lot of classic software- [40:41] as-a-service ends up in this way. [40:42] It's like your pricing is really dependent on, you know, [40:44] did you sign that deal on January 1st [40:47] or on December 31st when the sales team [40:49] really needs to hit the quota. [40:50] And you can actually see down to the seat level, [40:53] you're paying $30 per seat. [40:55] It's a great time to be signing a Salesforce contract. [40:57] Exactly. [40:59] And so you can actually get down to the seat level and say, [41:01] okay, you're about to auto-renew. [41:04] The market price has actually moved for the set of service. [41:07] And so it can be very, very useful to procurement teams [41:10] to know where they compare with the market. [41:13] That was actually my question for you, [41:15] which is, so if you remember Groupon— [41:17] Yes. The much beleaguered Groupon, [41:19] but Groupon started off as a site called [41:22] ThePoint.com/groupon [41:24] because originally it was like, [41:25] "I don't like the fact that Starbucks uses paper cups. [41:28] If I get 10,000 other people to donate a dollar, [41:30] we'll fly a plane in front of Howard Schultz's house," [41:33] which also emits carbon, a bad example. [41:35] But you only want, it's a collective action thing. [41:39] So, and then because I met [41:41] Andrew Mason and Brad Keywell [41:43] when they were first starting this business, [41:44] it was like eight people. [41:45] It was ThePoint [41:46] and downstairs there was a pizza restaurant or something [41:48] like, "Oh, if get like 50 people [41:50] to agree to buy pizza, maybe they'll give us a lower price." [41:54] The cause became discounts. [41:55] Yes and then, but then the thing is [41:57] Groupon wasn't known for this "it has to tip" [42:00] because they always had the demand. [42:02] So they never had to worry about it [42:03] because before it's like you've got supply, [42:05] you've got demand. [42:06] I guess my question for you is, [42:08] can you be the arbiter of pricing? [42:11] Can you aggregate the demand? [42:12] And the same way that like Costco, who shops at Costco? [42:14] Every small restaurant shops at Costco. [42:17] And Costco uses that collective bargaining power, [42:19] if you will, to say, "Hey Coca-Cola, [42:21] give us a very, very low price." [42:23] They take very, very little of that [42:25] because right now I'm using Ramp to buy stuff. [42:28] Yes. [42:28] But I would imagine if you could say, [42:31] "Hey, I have $100 billion." [42:32] I didn't know it was that high, that's amazing. [42:34] "I have a hundred billion dollars of spend. [42:35] I can hopefully direct it this way or that way. [42:38] Please give me a 20% discount." [42:40] Can you do that? [42:41] Or is that further in the funnel [42:44] of the purchase process? [42:45] There is this large swath of businesses [42:49] that I've been fascinated by [42:51] called group purchasing organizations. [42:53] A lot of these [42:55] came out of the healthcare world [42:57] where it's a very small set [42:59] of— APMs, yeah. [43:00] Yeah, exactly, supplier. [43:01] But if you can go and aggregate demand, [43:03] you could say, okay, for these type of purchases, [43:05] you're going to offer this procedure [43:06] at this account for this type of equipment, [43:08] you're going to give this level bulk kind of discount. [43:10] And you've gone and you've done that [43:11] and I think these are very popular now [43:13] in the private equity world. [43:14] Even when I look at Ramp data today, [43:16] there are dozens of, [43:18] and more and more every year, of merchants where, [43:20] you know, we were sending billions of dollars. [43:24] It is truly, it is the client's money, [43:26] they are going where they're going. [43:28] But we have a sense of actually, [43:30] okay, where is this actually going? [43:31] And can you go and say, you know, across the Ramp buyer base [43:36] over the next 12 months, [43:38] this is how many dollars will go towards you. [43:40] Can we negotiate a discount? [43:41] The other way which maybe starts to... [43:45] Sometimes these businesses veer closer to advertising, [43:47] but where it gets really interesting [43:49] is we see the fastest-growing businesses. [43:53] What are the businesses [43:54] that people are getting really excited [43:55] and are adopting really quickly? [43:58] And we might have a signal of [43:59] these are actually good businesses and good tools [44:01] and actually most businesses should move towards that. [44:04] Could you go and actually start to say, [44:07] "Hey, your renewal is coming up in 90 days. [44:11] Have you considered this other business [44:13] who has provided a 20% complimentary welcome [44:17] lower price to you?" [44:18] And so, there's multiple forms that can take place. [44:20] But the short version of it is, I think [44:22] one, it's yes, I think you can offer a bulk discount. [44:26] But two, you end up almost getting closer to, [44:30] back to this other theme of can you start to go [44:33] and maybe show businesses [44:36] that directing their next marginal dollar [44:39] will lead to this outcome. [44:41] Well, it feels like there's even like a third thing. [44:44] Which is, [44:45] I would call it merchant-specific balance. [44:48] So imagine that it's December 31st, [44:50] I'm buying something at Whole Foods. [44:52] And then I see an offer, not donate a dollar [44:54] to the, you know, whatever charity they're pushing. [44:57] But it's like, why don't you commit [44:59] $2,000 of spend for 2026 for $1,800? [45:05] And the nice thing there is that they lock in [45:08] that spend for me. [45:09] They actually collect it. [45:10] I mean I know there's a stupid law [45:11] and all this complicated stuff. [45:12] But the key thing is it's almost a risk management [45:15] and a loyalty play for them at once. [45:19] Because the problem is that they don't know [45:20] how much celery to buy. [45:22] They don't know how many... [45:23] So, it kind goes back to the AR/AP thing [45:25] that we were talking about earlier. [45:26] You nailed it, yeah. [45:27] Because the other thing is not just saying, [45:28] "Hey, I'm going to group bargain [45:31] on behalf of all the different customers that I work with." [45:34] But it's an ancillary thing, which is you might [45:37] want to just pre-commit your spend to a particular vendor. [45:40] And that vendor would love to have that spend pre-committed, [45:43] even if they're Amazon [45:44] and their cost of capital is very low. [45:46] Because they know they don't have to worry [45:48] when they're planning their budget for the next year. [45:50] 100%. [45:51] That you're going to churn in month nine. [45:52] And that seems like another, [45:54] like the closest that you can get to that is gift cards. [45:57] Strangely enough which I'm sure you know well [45:58] from your— [45:59] It's gift cards [46:00] for business. Yeah because it's like [46:01] you go to Costco, this is actually a big business at Costco. [46:03] You go to Costco, they sell gift cards. [46:05] We were just talking about this. [46:06] So they sell gift cards. [46:07] You can buy $100 of Starbucks gift cards for 80 bucks. [46:10] Is Starbucks dumb? [46:11] Like why are they, if they're worth $100? [46:13] No, because they basically have, in the ecosystem, [46:16] money that is going to be spent at Starbucks. [46:19] And yeah, there's breakage and other stuff, [46:20] but it's actually a quite, [46:21] it's quite a valuable value proposition [46:24] for the receiving merchant [46:26] because they know that you're going to spend [46:27] your money with them. [46:28] But I want to go back to your previous idea [46:30] where McDonald's every five years or whatever [46:33] does a big bake-off between Coke and Pepsi [46:35] and they always renew with Coke [46:36] and they have a long-time… [46:36] Maybe a movie theater chain is better example. [46:39] All the big chains, like they don't serve Coke and Pepsi, [46:41] they only serve one and they do a big bake-off [46:44] between the brands every five years. [46:47] And you have lots of small businesses [46:49] that are buying from Coke or buying from Pepsi. [46:51] And so, it feels like you could just do a bake-off [46:53] between them saying Ramp is going to have a preferred vendor [46:56] and you can obviously use a Ramp card on either, it'll work, [47:00] but we have preferred rates with Pepsi. [47:03] Is that a thing Ramp should do? [47:05] It's a really interesting question. [47:07] The short is I think over the long duration of time. [47:10] Like I'm kind of bit of a crazy person. [47:13] We should try everything is my actual answer. [47:16] Claude Code. Yeah, Claude Code. [47:17] Let's go, let's do it, [47:18] build it. [47:19] Pepsi. Make money from Pepsi. Go and negotiate. [47:20] The cost of a phone call, [47:21] a negotiation has never been lower, but… [47:24] And there's this almost philosophical question of like, [47:27] all right, where is the differentiation? [47:30] What makes Ramp different? [47:32] But isn't that increasingly the scale? [47:34] It is, and I think that you're right in that [47:39] at the very beginning of the company, [47:41] the answer was very easy of like, look, [47:43] our cost of capital is nowhere close to a bank's. [47:48] The interchange that we could make is nowhere close to it. [47:51] We have all the disadvantages of scale [47:54] and kind of the only thing that we could do [47:56] to compete with these people was we could… [47:58] Everything was slow at [48:00] the bank that I worked at before, [48:02] it was hard to ship things. [48:04] We need to move fast. [48:06] We need to kind of have this emphasis around velocity [48:08] to ship product and just kind of blitzkrieg faster [48:10] than maybe others could respond. [48:12] What it's evolved into [48:14] and I think is today probably [48:17] the largest point of differentiation at the meta level [48:19] between Ramp and maybe the financial service providers [48:25] that we compete with is they sell money [48:28] and we sell time. [48:31] They'll sell you a lower, [48:32] like a loan at a lower cost of capital, [48:35] rewards at wherever they set it. [48:37] We will sell your expenses, done. [48:40] And so, are you naturally skeptical of this idea [48:42] because it's selling money rather than selling time? [48:44] I wouldn't say I'm naturally skeptical. [48:46] I would say that— [48:47] It's okay. We're not precious. [48:48] We have lots more ideas. [48:49] No, no, no. It's good. [48:50] There is this question of, [48:52] I think back to our roots, like, I don't know. [48:54] How do you build a product that is 10 times better [49:00] than other folks? [49:02] And I think when you get really large, [49:03] actually just a line drive is good [49:05] and a 2x or a 3x is better kind of thing is actually good. [49:09] But how do you find these sets of things [49:11] that other your product offering is so different [49:14] than what you can find elsewhere? [49:16] And I think that we can conceivably, [49:17] and I know this, you can negotiate with [49:20] the large logistics companies. [49:22] A deal with FedEx [49:24] that would be better than what most companies are paying [49:26] and I think that that's interesting. [49:28] Then you have all the sublevels of how do you get people [49:29] to know about the offer, to sign in to the offer, [49:31] to deal with all these sub things. [49:33] Or you could just go and spend that next marginal hour [49:36] to go and say like, let's just automate [49:39] all of accounting for these customers. [49:41] Let's go and start doing financial work for these customers. [49:46] No financial institution is able to compete [49:48] on that kind of vector. [49:50] And so, I would say it [49:51] in the fullness of time, we want to do both. [49:54] Like I think that you want to save people [49:57] the maximum amount of time, [49:58] the maximum amount of money that you can, [49:59] but I think part of what makes us so different [50:01] in this kind of world that we operate in [50:04] is we have this obsession of [50:07] sources of drag of the things that slow down purchases [50:10] that lead you to overspend versus other companies. [50:13] But just to push on that, [50:15] I feel like the differentiation for companies [50:18] often has to change as time goes on [50:21] because they start in one competitive equilibrium. [50:25] And then as time goes on, [50:26] they're in another competitive equilibrium [50:28] because it's dynamic. [50:29] That's fair. And it feels to me [50:31] that Ramp got its start with very fast product velocity [50:35] and having this great product experience. [50:37] And as you grow up, you can just build a scale [50:39] into the product and scale… [50:41] Scale-derived product advantages. [50:43] Not just like we're big, [50:44] but like Costco, [50:46] the advantage comes from the scale [50:48] and they really passed it on to the customer. [50:50] And it feels like there could be [50:52] a second stage to this rocket [50:54] where you get going with faster product velocity. [50:56] But then there's actually a pretty different set [50:59] of product differentiations as you scale up. [51:03] I think it's true. [51:04] I think it's well said [51:05] and there's this almost question of like, [51:07] in every line of business that we're building, [51:10] where are we in the S-curve? [51:13] How far are we… [51:16] Is this a product that is serving [51:18] these tinkerers, early adopters? [51:20] Are we in the early majority, are we in the late [51:22] where we need to shift the business to kind of harvesting [51:25] contribution, optimize price, not quantity in it [51:29] or towards the end? [51:31] And the thing which is so shocking [51:33] and I think maybe as part of why I love this business [51:37] and I think even too, I think about Stripe [51:40] and many of the great businesses [51:41] and payments that have been built. [51:44] I believe today Ramp powers more than 2% of all corporate [51:49] and small business card transactions in the United States. [51:52] Which is amazing for a business that, [51:55] the product, you couldn't even sign up for it six years ago. [51:59] And yet that's a really fancy way of saying [52:02] 98% of spend is not on us. [52:05] On Ramp yeah. [52:07] And let's say we do it again. [52:09] We grew faster last year [52:11] than we did the year before. [52:12] I feel very good about [52:14] the way this year is starting off [52:16] and maybe it's closer to four. [52:19] 96% is still big. [52:21] 92% is still big. [52:23] It is so large. [52:24] And by the way, there's more ways [52:25] to buy things than just cards. [52:27] You think about bill payments. [52:28] And so, I think that there is an aspect of, [52:33] for most people, there's so much drudgery around... [52:37] I think people are doing artisanal expense reports. [52:40] You know, it's fun to be like a hipster [52:41] and spend an hour making coffee, [52:43] but like, it's a little crazy that— [52:44] This is the blue bottle of expense reports. [52:46] You know but unfortunately you're in Concur [52:48] and you have like this spinning wheel. [52:49] It's a pour-over expense report. [52:51] It's a pour, we love it that way. [52:53] Single origin. [52:54] You try it. It's the experience. [52:56] You're the Nespresso. [52:58] It's much more efficient. [52:59] That's right. [53:00] But like, what I would say is this aspect of time [53:04] is most people have not experienced [53:08] this thing you know, [53:12] is just a fact of life [53:13] for building a business that's been hard [53:15] doesn't need to be hard. [53:17] Doing your books doesn't need to be hard. [53:18] And so, I still think in some sense [53:20] in where we are on the S-curves of our business, [53:23] I lean that way. [53:24] But you're totally right when you're at this level [53:27] where in the not too distant future… [53:30] I mean, you're there processing trillions of dollars [53:32] per year in economic activity. [53:35] And we aspire to be there in the not too distant future. [53:38] Sounds like you'll be there pretty soon. [53:40] We're working our hardest. [53:41] Exponential growth [53:42] is a powerful thing. [53:43] Yes, it's definitely given me good things [53:46] to think about and work towards. [53:47] So I was thinking about when we met, [53:49] when you were running Paribus. [53:52] And it kind of feels like one of the first [53:54] or second or third order effects of AI is the marginal cost [53:57] of arguing has gone down to zero. [54:00] I'm serious. [54:01] It's true. It's just like, [54:02] I mean, my wife got into an argument with some company [54:05] and it's like she used ChatGPT to argue with them, [54:07] they used it to argue back with her. [54:09] And it's like, I just saw the future. [54:11] This is incredible. [54:12] And like, we won, [54:13] which is great. Your agent will talk [54:13] to my agent, yeah. [54:14] But you think about this [54:16] from a chargeback where it's like, [54:17] "Hey, this price went down." [54:19] I mean this is how we, I remember like, [54:21] this is such a good idea. [54:22] It's like the price went down, [54:23] you have these breakage models that the card networks had. [54:27] And Best Buy has to write you a check [54:28] or Visa has to write you a check [54:30] because that TV fell by 20%. [54:31] But nobody does that. [54:34] And now everybody's going to do that. [54:36] Or it's like you do a chargeback for $5, [54:39] it's just not worth it. [54:40] It's like, [54:41] I thought it was Pepsi that I was buying at McDonald's. [54:43] It was Coke. Chargeback, right? [54:45] Nobody's going to dispute that because the cost is too high, [54:49] but now the cost goes to zero. [54:50] So I'm just... [54:51] What do you think, I mean, it's hard to, [54:54] picture five years from now [54:56] what the actual full effects of this are. [54:57] But if you kind of think about that [54:58] as the macro abstraction of it's like [55:00] marginal cost of [55:01] arguing goes to zero. [55:03] What changes? [55:04] I think to be really explicit, [55:06] I think what's happening is the marginal cost [55:08] of time, of knowledge. Yes. [55:09] That is a much better abstraction. [55:11] Is what's going down so rapidly. [55:14] And I think about our customer base. [55:16] Most of our customers don't have a single software engineer, [55:19] let alone a software engineer for their finance team. [55:23] And if what we are very good at doing [55:26] is selling functionally sets of work, [55:32] maybe it's embedded in a financial operating platform, [55:35] but expenses done, accounting done, [55:37] some type of knowledge work done, [55:39] and you can deliver that. [55:40] That is immense high-leverage value for these customers. [55:44] Let's say it costs $5 before, [55:46] they didn't do it, now they can get $5 of value. [55:49] But maybe the cost of… It's a software type cost, [55:53] maybe it's pennies of tokens to actually go and do that. [55:57] That is a great business to be in [55:59] because it's very high customer value. [56:01] We can capture just a small amount of that [56:03] and build this business. [56:04] And when you go back to kind of [56:05] the original insight of Ramp, [56:08] we entered into this industry where it was very profitable, [56:11] but not only was it misaligned, [56:13] but people were fighting over like basis points. [56:17] Every last dollar that went into rewards [56:20] could have meant like tens or hundreds [56:21] of millions of dollars in profit for the business, [56:25] for them. [56:26] But if you think about for a customer, [56:31] let's say that in order to make one extra basis point [56:34] as a business, you would try to incent them [56:36] to spend a hundred more dollars. [56:37] For that customer and let's say they buy something [56:40] that gym membership they didn't need, [56:41] or that subscription keeps going, [56:43] they have lost $100, it's gone out the door [56:45] and maybe they go to the gym, maybe they don't. [56:48] But that is out of their bank account. [56:50] And if you just try to say, [56:51] I'm going to have a better rewards program, [56:54] sure, maybe you can get that customer [56:56] a dollar or $1.10 back or some amount. [56:59] That pales in comparison to helping them just [57:02] not spend that $100, cancel the subscription. [57:05] The economic leverage of that activity is much higher. [57:08] You sold your last business Paribus to Capital One. [57:11] Capital One is one of the biggest [57:15] founder-run financial firms [57:17] that people in Silicon Valley don't talk about. [57:20] What should we all be... [57:21] And has been extraordinarily successful. [57:23] What should we all be learning from Capital One [57:26] in their success? [57:26] Actually, I should say, how did they? [57:28] What product really broke out for them? [57:31] Just what is Capital One success? [57:34] I think there's a lot [57:35] that makes them amazing as a company. [57:37] Both the founders are excellent. [57:39] Rich Fairbank, Nigel Morris. [57:42] I think it's worth people reading up on their stories. [57:43] It was founded in '90s, [57:47] is that right? [57:48] So, not quite. [57:51] From a legal standpoint— [57:52] I'm very wooly on my Capital One history. [57:54] I think that legally the incorporation in some sense [57:57] for Capital One, it was, I think it was in 1994, [58:01] but the actual start traces back to the '80s. [58:05] At the time Rich and Nigel had met, they were consultants [58:10] and they had this insight that [58:14] the business of credit cards was… [58:18] While it was lucrative, was not serving most of the country. [58:22] The way you could kind of think about [58:24] credit cards back then, you know [58:26] Diners Club, maybe others have heard of it. [58:28] It was was a way for rich business people [58:31] to get together and have lunch. [58:32] And like they put the card down [58:34] and the restaurant would pick up the tab [58:37] and they could go and pay the restaurant back later. [58:41] And they had kind of navigated that [58:42] and the simplification is [58:45] if you had a very high credit score, [58:47] you could get one of these cards [58:49] with the benefits that it had. [58:51] And if you didn't have a credit score that met that, [58:54] you could have a debit card. [58:55] And that was basically it. [58:56] It was a very linear cutoff. [58:59] And what Rich and Nigel… The insight that they had had [59:03] is there must be some curve. [59:07] We should be able to test this. [59:08] Maybe we give people cards that have [59:10] above an 800 credit score or something like that then. [59:12] But what about 790? [59:15] There might be people there that can go and pay for this. [59:18] If and the way we could go and take on the cost of this, [59:21] maybe we charge them a somewhat higher interest rate [59:24] until they prove their efficacy. [59:26] If 790 works, let's go to 780. [59:28] So it's kind of the BNPL of its day. [59:30] There is a population who are, for whatever reason, [59:32] just below the threshold where banks [59:34] are giving them good access to credit [59:36] where you can actually very profitably lend to them. [59:38] This is right. Exactly. [59:40] And so, they, in the '80s, were functionally pitching [59:44] all these different banks to say, [59:46] "We should go on this exploration. [59:50] We'll even run this for you. [59:52] Let us go and run this." [59:53] And they went door to door to door to door to door— [59:55] The banks didn't buy it. [59:56] They didn't buy it. [59:57] And then finally, there was a bank in Virginia, Signet Bank, [60:01] that said, "Fine, we'll let you run it. [60:03] You can come join. [60:05] We'll give you this group [60:07] and some resources to go build this." [60:08] So they built this as this division inside of it. [60:11] And it started to work. [60:13] As this was going on, [60:14] the computer revolution was taking off [60:17] and it was sort of an unfortunate acronym, [60:20] but they called it IBS. [60:23] Irritable something something. [60:26] No, it's Information-Based Strategy. [60:29] And they said, "We could use different pockets of data [60:31] and run these." [60:33] It was before there was big data, [60:35] there was this kind of data of we could go and test. [60:38] You know, maybe this person has a balance [60:40] of $8,000 at this bank. [60:42] We could send them an offer at this interest rate [60:44] or this kind of [60:46] you don't need to pay back interest [60:48] for six months or 12 months. [60:49] And they could start to go [60:50] and see the mathematical return against this. [60:53] And this division started really working [60:55] and it got so profitable [60:57] it became a problem for the business [60:59] and they convinced them to spin it out. [61:01] And so in 1994, that is the founding as people know [61:06] Capital One today. [61:07] But it actually was I think almost a decade [61:10] in building to actually go and do this. [61:13] I think that there was a lot that they got really right. [61:16] One, I think that they took much more [61:19] of a first-principles view of the business. [61:22] Whereas others said, "This is a profitable product. [61:25] I'm going to do it like the other person [61:26] and use scale and distribution advantages." [61:28] They said, "Well, let's think about [61:29] the product nature itself." [61:31] There's different, everyone's making money on interchange. [61:33] Maybe there's lending could work [61:35] and you could kind of have different strategies [61:37] for different types of populations. [61:39] They did this so large [61:40] where I think at one point they were the largest customer [61:42] of the US Postal Service. [61:44] They would send out so many offers [61:45] I think until Amazon took them over. [61:47] There was a period of time that they did this. [61:49] So it's incredibly experimental. [61:52] They kind of carved the path [61:53] and showed how do you go from credit facilities [61:55] for fintech founders, [61:57] eventually maybe you become a bank [61:59] or buy a bank and carve that path. [62:01] So there's some tactical lessons. [62:02] I also think they've done a great job [62:04] of building a great and durable brand. [62:06] One of the things that Capital One has done very, very well [62:09] I think has focused on like, [62:10] what is the consistent visual message? [62:13] What is the aesthetic and how do you stand out [62:16] in this busy and different world? [62:19] And so, there's a lot of pieces [62:21] to what's made that company work, but— [62:23] And they've also remained focused, [62:25] which a lot of banks haven't. [62:27] I would say so. [62:28] Well, I think that is a rewrite of history. [62:33] They were doing like cell-phone financing, [62:37] healthcare financing. [62:38] They were a part of this thing. [62:39] They have come back to focus after wandering in the woods. [62:41] They had some JV, it was like America One, [62:44] I think it was called, to compete with Amazon. [62:46] They were actually the most experimental [62:50] business ever in a lot of ways. [62:53] In, I think it was Hibernia Bank in Louisiana. [62:58] And this was after Hurricane Katrina. [63:00] I think it was maybe during it [63:01] or just after they were able to buy. [63:03] And it was like, it was an existential question for them. [63:05] And I think that many people worried about, [63:08] you knew this like, well too from the BNPL businesses [63:10] where if you're kind of a lending-based business, [63:14] and it's fine when interest rates are low, [63:17] but if there's a crisis [63:18] and people don't want to lend to you, [63:20] it can just break your business. [63:22] And so they had these kind of scares [63:24] and they said we need a stable kind of cost base. [63:26] Eventually they bought a bank, [63:27] which was in a lot of ways very good. [63:29] They had kind of the deposits as this durable, stable, [63:32] low cost of, or consistent capital base. [63:36] But the flip side of buying it [63:39] was they had to reconcile their culture [63:43] of like crazy experimentation of trying everything to [63:47] we're a regulated bank now. [63:49] We got to be bank people. [63:50] And it's more than just an attitude thing. [63:53] If you are a nationally chartered bank— [63:55] It's real strictures, yeah. [63:57] You know, look, it's to a level of, [64:01] let's say the bank fails on Monday night. [64:05] There is personnel. [64:08] These guys in windbreakers in your office the next day. [64:10] And they've probably been working there [64:11] for some time before. [64:13] You know, and so it's... [64:15] I think there was a change. [64:18] In some sense, there is experimentation [64:21] I think within certain constraints [64:24] would be how I would describe [64:26] the Capital One post buying a bank versus before [64:30] would be my read of it. [64:32] But when you look at them in the '90s and 2000s, [64:34] it was like them and in some ways, [64:37] like name the high flying tech company. [64:39] They were right up there in terms of share price growth. [64:44] They started re-achieving it I think in the late 2010s. [64:46] But it's a fascinating company. [64:49] I mean, I think the talent pool. [64:50] So at Affirm, our first like really good risk person, [64:54] but this is true for every fintech company. [64:56] It's just like, actually this is kind of cool. [64:57] Billy Alvarado, our first COO grew up at Capital One. [64:59] I had played the piano [65:01] and my teacher was taught by somebody [65:03] who was taught by somebody. [65:04] There was actually a chart, you can look this up. [65:05] Czerny and Beethoven taught everybody [65:08] who ever taught anybody. [65:10] So like Lang Lang or Yuja Wang, [65:12] like pick any famous pianist today. [65:14] They can always [65:16] trace their lineage [65:16] 100% of the time to Czerny, Carl Czerny or Beethoven. [65:22] Capital One is like the Czerny. [65:23] Our Head of Risk Srinath Srinivasan, he's amazing, [65:29] came from Capital. [65:30] We had the same boss 10 years apart actually, [65:33] in some sense. [65:34] And all the heads of risk for all modern fintechs [65:38] came from Capital One. [65:39] Well, because it's one of these things where [65:40] you know that you don't want to just hire for intercept, [65:43] you want to hire for slope. [65:44] And this is the problem is like you find, [65:46] not to pick on Bank of America [65:48] or Chase or somebody, but it's like, all right, [65:49] this person clearly knows how to do their job. [65:53] The bank that they work in is worth a lot of money. [65:55] But they don't necessarily have, [65:57] how do I put this delicately, the slope. [65:59] They won't know [66:00] what that means because they don't know [66:01] what slope means. [66:01] But seriously, the guy that we hired [66:03] just so… He was very, very smart. [66:06] And that was the key thing about Capital One [66:08] is that they didn't hire banking people, [66:10] they just hired smart people. [66:12] And it became known as the place where smart people went. [66:14] And actually the fact that companies would poach people [66:17] from Capital One actually added to the alert. [66:20] It's like, I want to work at McKinsey [66:22] because at the end of McKinsey, [66:23] I'm going to get a better job somewhere else. [66:25] Capital One job market. [66:25] Capital One actually had, [66:27] it did have that imprimatur. [66:28] And I think part of it is [66:29] it's the only founder-led institution. [66:32] Last question, Eric. [66:33] We're talking about banking here. [66:34] You have a treasury product. [66:35] In five years, [66:36] where do businesses keep their money? [66:38] Is it with, I mean statistically mostly, [66:40] you mentioned 2% on Ramp, [66:42] it's even higher today in traditional banks [66:45] like a Chase or a Bank of America. [66:46] There's also neobanks, you know, [66:52] I mean kind of bank-like entities like Mercury [66:54] or Revolut or Monzo or things like that. [66:58] There are companies like Ramp [67:00] where you started in spend cards, [67:02] but maybe moving more into treasury. [67:03] How do you think that shakes out as a market? [67:08] So, as a macro point, [67:11] I think there is an incredible amount of money [67:14] made by institutions who are enjoying the profit pool [67:19] and sharing very little with their end customers. [67:22] You know, if you think about the implications [67:24] of like the federal overnight funds rate. [67:26] This is basically saying, [67:28] hey, if you want to go and— [67:30] Just it's insane that there's so little yield sharing [67:32] in the current market. [67:32] It's crazy. [67:33] I think that the national average for businesses, right, [67:36] these are sophisticated entities with personnel. [67:39] They're supposed to be able to manage [67:40] and put their funds in some place that's more high yield. [67:43] I think the national average [67:46] on checking accounts is 0.07% in the US. [67:49] And so I'm guessing you don't believe [67:51] that Bank of America and JP Morgan and all these folks [67:53] will wake up more altruistic one day. [67:56] And so, what is the competitive process [67:59] by which you think we'll get there? [68:00] I think that new businesses, [68:04] whether it's they have their own charter [68:05] or they work with banks too that, [68:08] this is not a monopoly. [68:10] Like they have, everyone has competitors [68:13] and it is a market that evolves. [68:14] I think that that rate will go up, [68:17] that the easier it is for people to, [68:19] whether it's to create depository institutions, [68:23] create accounts at these institutions, [68:25] or create stores of value [68:27] maybe even outside of these systems, [68:29] maybe in a stable currency. [68:31] And so, I think that rate goes up. [68:32] I think part of what's driven the extremely rapid growth [68:36] of Ramp Treasury is it's just [68:38] a vastly better deal for customers. [68:41] Maybe you keep three months [68:43] of just walking around money in your checking account, [68:47] but if you know the funds you're going to receive [68:49] and what you're paying out, [68:51] well, we can move funds on the day payroll [68:54] is coming due into that account [68:57] and then every other day make sure the funds are earning [68:59] the highest rate possible. [69:01] And so, I think from a macro perspective, [69:03] these things tend to go up in terms of the relative yield. [69:05] I think there's another question of slack in the system. [69:10] If you have more and more... [69:13] Money can think, right? [69:15] If the dollars in your company [69:18] have some level of intelligence [69:21] like it's able to kind of determine [69:26] when can it be spent, under what circumstances? [69:29] It's increasingly recorded in real time [69:33] and there's some reasoning around this, [69:35] some ability to kind of opine [69:36] of where should the next marginal dollar go? [69:39] And you have systems that are able [69:41] to think infinitely about these things [69:43] even at 3:00 AM in the morning [69:44] when most of your team is asleep. [69:47] Well, once you determine, [69:49] like maybe you'll determine I should keep the funds [69:51] in some level in like a 2% or a 4% yielding account. [69:56] But I think more of those dollars will go in flight [69:59] actually to go spend to, if you have a business [70:01] that makes an 8% profit margin a year, [70:03] that's a lot higher than what you can earn [70:05] in the overnight rate. [70:06] And so in some sense, I think you have the dual effect. [70:08] Smarter capital allocation. [70:09] I think more dollars will be put to work. [70:11] I agree with Alex's macro point of [70:15] if you kind of understand counterparties, [70:17] you understand more information. [70:18] I think that one, the cost of financing should go down [70:21] and I think more dollars [70:23] should be in the system in some sense. [70:25] It's actually a waste for everyone [70:26] to have your dollars just sitting in a bank account. [70:28] Who does that benefit? [70:29] But it goes back to your time point. [70:30] And it's like the reason why [70:33] I got mad at Chase a while ago, [70:35] and I still have a Chase account. [70:36] And like why is that? [70:37] It's like I have a life insurance policy. [70:39] I don't remember the login for it. [70:40] It's just too much work. [70:42] I wrote, this is a true story. [70:43] Isn't this an issue if you die? [70:44] Yeah, I mean. [70:45] Well, but they'll pay my wife. [70:46] I just don't know [70:47] how to log in and change the bank account. [70:49] So, they'll send her a nice letter I'm sure, some flowers. [70:53] But, you know, why... [70:55] Well actually, this is a true story. [70:56] I wrote a check for somebody's bar mitzvah. [70:58] Mazel tov. [70:59] Has not been deposited. [71:01] Do I want to be the schmuck who has the bounced check? [71:04] No, I don't. [71:06] So it's like I have to— [71:07] This kid is keeping you at Chase. [71:09] Well, great to see you. [71:10] You too. Eric, thank you. [71:12] Thanks for the Guinness. [71:13] It was a great time.