Ramp Pivots Pricing Model as AI Agents Disrupt Seat-Based Software

AI Agent颠覆席位制软件，Ramp调整定价模式 大语言模型带来的巨大效率提升将迫使SaaS公司放弃传统的按席位授权商业模式，转向按可量化、可衡量的价值收费。 在本次炉边对话中，Eric Glyman详细介绍了Ramp如何从2019年利用早期机器学习进行收据匹配，演进到深度将大语言模型产品化以实现自动化会计和价格情报。他探讨了跨主要AI提供商进行多路复用的架构策略，并概述了SaaS定价模式从按席位收费向按价值收费的必然转变。此外，Glyman还分享了关于金融科技创业、领导力从匮乏心态向充裕心态转变，以及通过小型自治团队保持快速产品迭代的见解。 Ramp的设计肩负着双重使命，即帮助企业客户节省资金与时间，从根本上使其财务工具与为客户省钱保持一致，而非鼓励过度消费。 (0:04) 在成立的前四年半里，Ramp成功为客户节省了超过6亿美元，并减少了850多万小时的人工工作。 (0:04) Ramp自2019年成立起就采用了机器学习技术，最初将其应用于极其繁琐的结构化数据工作流（如自动收据匹配），以消除手动费用报告。 (2:50) 商业化LLM的发展使Ramp得以推出“价格情报”和“会计智能”工具，用于解析SaaS合同等非结构化文本，并实现总账分类的自动化。 (2:50) 一项重大的技术变革已经发生，商业化LLM能在不到一分钟内准确解析并结构化处理大量非结构化文档，例如长达30页的合同。 (5:56)

Ramp co-founder and CEO Eric Glyman argues that fintech software is shifting from passive payment tools into automated, agency-based systems that save businesses time and money. He contends that success in modern software relies on leveraging commercial AI models, value-based pricing, and disciplined growth rather than chasing peak market valuations. Aligning Financial Workflows with Efficiency (0:04) Glyman views traditional corporate cards and card issuers as fundamentally misaligned with their clients, relying on rewards like points to encourage excess spending. Drawing from his background at a consumer savings startup acquired by Capital One, Glyman designed Ramp around a dual metric: minimizing the money and time businesses spend. Over its first four and a half years, Ramp reported saving clients over $600 million and eliminating more than 8.5 million hours of manual administrative tasks. This workflow-oriented model extends across corporate cards and bill payments. Glyman notes that product development is anchored directly to customer pain points rather than defending single product categories, with internal development executed through autonomous teams capped at under ten people to prevent bureaucratic slowdowns. Unstructured Data and the Shift to Agency Software (2:50) While Ramp utilized classic machine learning for receipt matching, underwriting, and fraud detection as early as 2019, Glyman highlights that the commercialization of large language models resolved the longstanding challenge of processing unstructured documents. Earlier systems struggled with tasks like parsing 30-page vendor contracts; modern LLMs perform these extractions reliably in under a minute, enabling features like price intelligence, which analyzes SaaS contracts across aggregate purchasing data, and accounting intelligence, which automates general ledger categorization. Glyman connects these capabilities to an emerging category of agency-based software that executes operational workflows directly on behalf of users. Recalling his 2015 startup Paribas, which parsed 200 million consumer emails daily to claim price-drop refunds through fragile custom scrapers, Glyman emphasizes that modern LLMs make executing complex, data-driven tasks trivial in structured domains like corporate accounting. Model Architecture and SaaS Pricing Realities (10:17) Addressing AI infrastructure, Glyman cautions founders against training proprietary foundational models to compete with well-capitalized providers such as OpenAI, Anthropic, and Meta. Instead, Ramp implements an open architecture that multiplexes across different foundational models, dynamically selecting providers according to latency, response quality, and compute costs. Glyman observes that model performance in complex domains can be sensitive to prompting nuances, and he expects specialized sub-language models to emerge for technical areas like double-entry bookkeeping. As AI agents assume repetitive analytical tasks, Glyman anticipates pressure on conventional per-seat software licensing models. If employees become substantially more productive through automation, software companies will likely need to transition from selling individual access seats to billing based on quantifiable, delivered business value. Fintech Infrastructure and Capital Discipline (15:58) Glyman identifies significant upside for new fintech entrants because legacy payment systems historically shut out pure software companies, and vast transaction volumes remain conducted via cash and paper checks. Post-financial crisis reforms, including the Dodd-Frank Act, opened banking infrastructure to modern software interfaces, allowing lean engineering teams to launch financial platforms without massive balance sheets. Reflecting on market volatility across fintech and crypto cycles, Glyman advises founders against accepting peak valuation offers during funding booms. To preserve room for error and maintain healthy investor alignment, Ramp intentionally turned down top-tier valuations, prioritizing long-term enterprise sustainability over short-term paper pricing. While modern AI enables software platforms to process complex financial documentation and assume autonomous tasks, the long-term impact on enterprise software business models remains evolving as companies determine how to shift from seat licenses to measurable value capture.

Ramp联合创始人兼首席执行官Eric Glyman指出，现代金融科技的核心机会在于颠覆传统金融机构诱导企业过度消费的错位激励，利用大语言模型对非结构化文档的高效解析，推动财务工具向直接代行操作的代理式软件演进。他同时强调，初创企业应采用多模型复用架构而非押注对抗底层基础模型，并在AI显著提升人效的背景下，推动SaaS软件从席位计费向价值计费转型。 重塑金融激励与代理式软件落地 (0:04) Eric指出，传统信用卡发卡机构依赖积分和返现诱导企业增加非必要支出，其商业利益与客户诉求根本脱节；Ramp则将自身定位为工作流与效率公司，致力于帮助客户节省金钱与时间。在运营的四年半时间里，Ramp称已累计为客户节省超过6亿美元资金，并削减了850万小时的手工操作工时。得益于金融危机后《多德-弗兰克法案》（Dodd-Frank Act）带来的金融基础设施开放，如今小型技术团队无需投入巨额资本与庞大人力即可发卡或提供金融服务，推动软件直接涉足资金流转环节。 伴随底层设施的成熟，软件形态正从提供辅助建议向直接代表客户执行任务的“代理式软件”（Agency-based Software）转变。Eric回顾了其2015年创立的初创公司Paribas，该软件曾通过扫描收件箱追踪价格波动并自动为用户索取差价退款，峰值期每日索引2亿封邮件；尽管当时受限于技术实现困难且错误率高，但在大语言模型普及的今天，这种直接代表客户执行具体业务操作的代理模式，在会计核算等垂直领域已具备高度可操作性。 大模型突破：非结构化数据与会计智能 (2:50) 针对AI的发展路径，Eric澄清Ramp自2019年创立起即运用经典机器学习处理收据匹配、反欺诈与信贷承保。过去经典机器学习擅长处理高度结构化的数据，以解决美国国税局（IRS）对75美元以上交易需附凭证的合规要求；但面对长达30页的非结构化商业合同时，传统技术提取手段繁琐且准确度极低。近几个月商业化大语言模型的成熟打破了这一壁垒，系统能够在1分钟内快速解析长篇非结构化文本并输出高精度分析。 基于这一能力，Ramp推出了两项核心产品：一是通过分析数百亿美元年度采购数据、评估SaaS供应商合同价格是否公允的“价格智能”（Price Intelligence）；二是自动进行总账分类以替代传统人工录入或死板规则的“会计智能”（Accounting Intelligence）。此外，Eric分享了工程实践经验：通过更具鼓励性与严谨设定的提示词能显著提升模型输出表现；但他同时指出，现存通用大模型多基于通用文学语料训练，未来仍需出现针对复式记账等专业规则优化的“子语言模型”。 多模型复用策略与席位计费模式危机 (10:17) 在技术选型策略上，Eric建议初创企业切勿押注与OpenAI、Anthropic或Meta等巨头在基础模型层面竞争。Ramp选择构建开放式架构，在多家基础大模型之间实施多路复用（Multiplexing），根据响应质量、延迟、执行速度和成本进行动态平衡。他认为，终端客户本质上只关心产品能否改善财务表现并提高效率，而不在意底层具体运行的是哪一款大模型。 在大模型推动的人效跃升面前，传统企业软件的商业模式正在经受考验。Eric认为，AI的核心价值在于成倍放大员工能力，使初级财务人员也能获得行业经验支持。随着知识工作被AI大幅提效，传统SaaS按席位付费（Seat-based Pricing）的逻辑将难以为继；尽管高效率可能促使企业为了扩张而雇佣更多高产能员工而非直接裁员，但软件供应商未来必须转向根据可量化的实际交付价值进行定价。 周期防御策略与小团队的高敏捷迭代 (23:59) 结合早年参与加密金融公司Tagomi（后被Coinbase收购）并经历2021年市场狂热的经验，对话就AI领域的融资繁荣向创业者提出了警示。Eric强调，企业构建是一场需要持续多年的马拉松，在资本过热期，Ramp主动放弃了过高的估值要约，以此维护健康的投资者关系并保护长期企业价值，避免估值倒挂导致后续发展丧失回旋余地。在领导风格上，他也从早期因资金与签证压力造成的“匮乏心态”转变为以高标准招募顶尖人才的“充沛心态”。 在业务扩张与产品敏捷性方面，Eric指出大部分用户并不关心抽象的技术理念，营销的关键在于通过真实用户故事激发共鸣。为了在扩张过程中避免官僚化阻滞，Ramp始终依赖人数不足10人的小型自主团队进行研发，通过赋予团队实质性的决策权，让产品功能在数月内完成迭代上线，从而保证技术创新能迅速转化为市场端的产品力。 本场对话阐明了金融底层开放与大语言模型如何共同驱动企业财务向智能化代理演进，但通用大模型在严格遵循复式记账等专业合规体系上仍面临局限，需等待垂直模型补充；同时，随着AI大幅压缩手工工时，企业软件从席位计费向价值计费过渡的实际计量标准与推行节奏，仍是全行业尚未完全解决的核心议题。

[0:04] Eric Glyman: ultimate platform for finance teams [0:04] Eric Glyman: helping customers spend less money and spend less time [2:50] Eric Glyman: George Costanza wallet [5:56] Eric Glyman: You are the best contract analyst there ever was, you're really thoughtful, and you check your answers twice [10:17] Eric Glyman: very challenging time to be like implicitly betting against the mega models [10:17] Eric Glyman: multiplier on people [15:58] Eric Glyman: wore top hats [17:09] Eric Glyman: door kind of blew open [17:09] Eric Glyman: more rejects than truly... great learning [21:35] Eric Glyman: all a company is is a collection of people [23:59] Eric Glyman: serious relation that you can't really break up [27:45] Eric Glyman: people don't care about your product [27:45] Eric Glyman: No one likes it when they have to go check for like 30 people to like change the color of a button

[0:04] Hi, everyone. [0:04] Thanks for joining us. [0:06] I'm Mark from General Catalyst. [0:07] I think you guys all know Eric from Ramp as well, [0:10] and we're very appreciative of you coming in today. [0:13] Thank you. [0:15] I guess there's no big sign in the back that says applause or whatever. [0:18] So it'll be all organic and authentic. [0:21] But I appreciate you guys. [0:22] I think it looks like a laugh track in here. [0:25] Exactly. [0:27] So, Eric, we're going to run through a bunch of questions. [0:29] Obviously focus on AI, how AI [0:32] has impacted Ramp and fintech. [0:34] But for anyone who might not be familiar with it, can you start off [0:38] by telling us what is Ramp and what you do for your customers? [0:41] Yeah, 100%. [0:42] And even before that, first, just want to say [0:44] thank you to the team at Stripe and everyone years putting on this event [0:47] gathered everyone and frankly also to even mean [0:52] what we're doing at ramp possible [0:53] so thrilled to be here and just want to say thank you [0:55] for all the effort that went into organizing. [0:58] So thank you. [1:00] Next ramp, [1:04] we are trying to build what we believe is the [1:06] you know, we want it to be the ultimate platform for finance teams. [1:09] We focus and what we're known for is for delivering [1:13] what is the fastest growing corporate card in the US. [1:16] Fastest growing bill payment network. [1:19] And every product that we build is designed with the intent [1:22] of helping our customers spend less money and spend less time. [1:26] We've saved our customers over $600 million [1:30] in the last four and a half years that we've existed as a company [1:34] in over eight and a half million hours of manual work. [1:38] And so really, we're trying to design in organizations [1:42] that have software payment rails that are focused on helping companies [1:45] be more efficient, build better businesses. [1:47] And the motivation of this a lot of this came from [1:51] in our background, we had started a company [1:53] focused on helping consumers save before we were acquired by capital One, [1:57] one of the largest banks and credit cards in the US. And, [2:01] you know, the longer that I had spent, you know, getting to know [2:04] the credit card industry, the more it just deeply felt it [2:08] wasn't totally aligned with customers. [2:10] A lot of people were thinking really hard, and we're fantastic [2:13] at getting customers to spend more money than maybe they intended to. [2:17] Customers, I believe, aren't looking really for cash back or points. [2:21] They're frankly looking for more in their bank account, more time in the day. [2:26] And we wanted to build a company that was focused on exactly that [2:30] spirit deeper. But in many ways, [2:32] we're much more of a workflow company. [2:34] We're much more thinking about how do you actually make [2:37] organizations more efficient, spend less money and time. [2:41] Very cool. Yeah, I was early. [2:42] Lucky enough to be an early investor in Ramp and obviously John is [2:46] and ramp as well and in Stripe so that's very exciting. [2:50] I remember asking you, you know, how has it changed your business in the last year? [2:55] And you kind of shot me down and said, well, we've been using A.I. [2:58] for four years since you started. [3:00] So I'm kind of curious. [3:01] Maybe you can explain that a little bit [3:03] to the audience, like, where were you using A.I. [3:05] at the beginning, the first couple of years? [3:08] How has that maybe changed in this last year? [3:10] Has it really exploded as well? [3:12] Maybe take us through that narrative. [3:14] Yeah. [3:14] And I mean, it's certainly accelerated a lot over the past past year [3:19] for sure, with the advent of large language models getting commercialized. [3:22] But I think going back to, you know, our fundamental premise is how do you [3:25] how do we give our customers leverage? [3:27] And we're constantly thinking [3:28] how can we apply technologies to make us more efficient? [3:32] One of the first workflows when we're following that [3:35] mission of helping customers spend less money and time [3:38] is, you know, customers need receipts for the IRS. [3:42] It's required for transactions over $75. [3:45] Very painful for people to have to go have a George Costanza wallet full overseas, [3:51] have things around, match it to the right transactions [3:54] and it was a perfect use case where we had the structured data coming through from [3:59] from the network. [4:01] We knew we authorized the [4:02] transaction is much easier to match in image [4:07] versus, you know, blank slate, read it, figure out what it is to match it. [4:11] And even back in 2019, pre launch [4:15] I one of our first priorities simple receipt matching was leveraging [4:18] machine learning. [4:19] Over the years [4:20] there's a lot of classic use cases even where you saw advancements in today [4:24] from the Stripe team around underwriting fraud, risk detection, [4:28] using large signals and we're trying to go and understand our own activities [4:33] and applying machine learning to it to make that simpler. [4:36] But I think really over the past year [4:39] in a big way, over the past six months, but certainly [4:41] even over the last nine is creeping and accelerating [4:45] is looking at ways where you can productize as deeply into workflows. [4:49] And so, you know, even we made a large announcement around this in May, [4:55] you know, [4:55] but just yesterday we scaled price intelligence to all of our customers. [5:01] You know, that can give you the insight and the, you know, the benefit of, [5:06] you know, tens of billions of dollars per year of purchases, of knowing, [5:10] are you getting a great rate from your SAS vendor or are you getting charge more? [5:15] You know, are you in charge less [5:18] accounting intelligence? [5:20] There are several ways one could say this Facebook [5:24] transaction should be accounted for in my general ledger, [5:28] but it's not rocket science. [5:30] There's surely a right way. [5:32] And you can learn from those patterns. [5:34] And so starting to really take [5:35] really large data sets and being able to leverage it into [5:40] not air based accounting, but just automated accounting. [5:44] Now I am, you know, price intelligence, but alerts [5:48] when there's better rates available to actually save is suddenly been possible. [5:51] So lots of products use cases in lots of internal use cases as well to talk about. [5:56] So walk me through that. A year and a half ago, [5:59] why couldn't you have done this price intelligence product? [6:01] And you know what changed where you can do it today? [6:04] Totally. I mean, so I mean, first [6:08] we tried to see just large language models, but, [6:10] you know, when is highly structured data, [6:13] I think a lot of classic machine learning techniques certainly could have worked. [6:16] But if you want to take, you know, one shot's like a 30 page [6:20] contract in deliver it all to me. [6:23] Upload it and say, is this a good rate? [6:24] Are these terms standard or non standard, [6:27] you know, at least in a broadly commercialized manner? [6:30] There was no I, I think the accuracy was quite dubious [6:35] and the level of building you would need to, to deliver in order to say [6:38] is this a good or a bad rate would be immense and very, very hacky. [6:44] You know I think the out in it really commercialization scaling [6:47] largely was models you could take 30 pages of tax upload it [6:52] a minute later really truly have [6:54] in a structured way and you can ask questions like [6:56] even something that has been making the rounds in a [6:59] in an enormous way around ramp over the past week, [7:03] you know, as part of our our commitment to [7:06] to try to improve the way we sell, the way we support customers, [7:09] we record calls, we learn from it, we try to pull insights out of it. [7:14] So we have a huge database of [7:15] actually what customers are telling us that they want. [7:19] What's changed suddenly, as you can ingest in mass, [7:23] what's actually going to start learning from and asking questions [7:25] and saying what our customers asking for, what are they interested in, [7:29] and how do we actually become better products thinkers in learning [7:32] and not building whatever it is that we come up with, but really having [7:35] add scale insights from our customers, [7:38] functionally taking that, you know, completely unstructured [7:41] data and structuring, it would have been very, very challenging. [7:44] And it turns out that can be something that's done with [7:48] certainly very skilled engineers. [7:50] But that done in days to weeks, not years. [7:53] And we've certainly seen, you know, open air in chat. [7:56] CBT has got like a five on biology and history, but maybe not done as well [8:01] on map. [8:02] Where have some of the limitations be where, you know, you've said [8:06] we really want this product, but the air is [8:07] just not there yet or the text not there yet. [8:10] What's kind of cutting edge? Where is it falling short? [8:12] Where in your product would you like to do more? [8:15] Well, there's a lot I mean, I'm sure everyone in this audience [8:18] is playing with these models, using it in various applications. [8:21] And there are a lot of things that are very like surprising [8:24] as you're actually endeavoring to build with it. So [8:28] I think everyone here Friday experiences, [8:30] if you're more encouraging in your prompt, you can get a very different output [8:35] if you're saying like, hey, you know, can you get is this a good price or not? [8:39] You get one level of prompt of like versus like. [8:41] You are the best contract analyst there ever was. [8:44] You're really thoughtful. [8:45] You're spending time in it and you're you're checking your answers twice. [8:49] You know, now go do this task. [8:52] You get a very different level of quality. [8:54] And in some sense, like that seems very bizarre. [8:57] But in others, like, right, this is taking language based context. [9:01] And so even thinking about how you're structuring and writing the prompts or [9:05] I think people are still wrestling and grappling with [9:08] what is it mean to be cutting this mean or were really the training base is [9:12] is language and the way you prompt and encourage in some sense [9:15] can affect the quality of the output. [9:17] I think it's just been interesting [9:19] for folks to be learning and wrestling with that. [9:21] I think that the other [9:25] thing that for me is interesting, a lot of these large language models [9:28] are generally on on on vast bases of text, [9:32] in literature, in various languages. [9:35] And I think for us, when you think about, you know, double entry accounting [9:41] numbers, like they are like they're our own systems of rules and languages. [9:46] And I think some of the large models are trained on [9:49] these specific types of use cases I believe will very quickly emerge. [9:54] But in the meanwhile, you know, you can use it. [9:57] And a lot of these are surprisingly good, but I think that they are going to be [10:00] vastly better iterations that are fit to specific, you know, [10:04] sub languages that aren't just, [10:06] you know, along the order of languages people speak, [10:09] but the languages that professionals use in various capacities, [10:13] you know, I [10:13] think there's a lot that can get done, [10:14] but I think that's going to improve a lot quickly. [10:17] And how much of this are you doing in-house at RAM versus leveraging models [10:20] that are already out there? [10:22] And maybe you could speak a bit to which companies [10:24] or TAC you found really impressive for other founders here [10:27] who might be thinking who to partner with or what tech to leverage. [10:31] I mean, I think that this the piece of improvement is so quick. [10:37] There are real use cases where you want to be using internal models. [10:41] But I think it's a very challenging time to be like [10:44] implicitly betting against the mega models. [10:47] Like the amount of resources being invested by these companies is so vast [10:52] that I think organize like positioning your own architecture to benefit [10:56] and let the world the machine learning benefit. [11:00] What you were you were doing is first is I think in general being open [11:03] and frankly learning and multiplexing and seeing, [11:07] measuring the quality of response, latency, [11:12] speed, cost, all of it looking and comparing. [11:14] I think, you know, Openai is certainly at the forefront [11:18] in many ways, but I think Anthropic has been phenomenal, [11:22] you know, to see to even, I think what would Facebook is doing [11:25] and you know, more specific models are just fascinating. [11:29] And so being urban, having an architecture if you can, that implicitly is testing [11:33] it is in some senses can see where it's better is worthwhile [11:37] because at the end of the day now for anyone building products, [11:40] your mission, I believe, is to serve your customers. [11:45] And, you know, I think in some sense, if you have poor architectural [11:48] and infrastructure choices, they will feel it. [11:51] But if you're making great choices of the day, like I believe our customers [11:55] actually care about, are they more successful? [11:57] Are they spending less or are they more profitable? [11:58] Not which particular model that one use. [12:01] So that's I think the first piece. [12:03] You know, I think that the second kind of base where we think about it is [12:08] when you're [12:08] using the macro models, you can certainly do, you know, in context and learning. [12:12] We're effectively you're not, [12:13] you know, sending data back to the model for learning and think about privacy. But [12:18] in any workflow company [12:20] or any kind of SAS based tool, you're going to have your own set of data. [12:23] And so thinking about where you get the benefit of the workflows that you're, [12:27] you're creating for customers, the problems you're solving your own data [12:30] sets and leveraging the learnings of large language models to, you know, take [12:35] general context, general understanding and apply it into your own workflows. [12:40] Data sets. Gotcha. [12:42] So you use multiple providers at the end of the day for a lot of the technology. [12:47] Certainly ends at least in some level of testing. [12:50] That makes a lot of sense. [12:52] To ask you a little bit about this, a tech, [12:54] would you see it as something more that grows your revenue or cuts your costs? [12:58] So specifically, is it an opportunity to, [13:01] you know, help with customer support and reduce your cost function? [13:04] Or do you see it more as a new revenue stream? [13:07] People use the products more, you make more money. [13:09] Which of the two does it fall under more? [13:12] I mean, [13:13] I think so much of what's happening is so profound [13:16] and the capabilities are so vastly improved. [13:20] Over a year ago. I think it hasn't fully been answered. [13:22] I think in many ways I think it's a multiplier on people. [13:27] You know, when you think about like I thought would be nice, [13:29] I don't have the time to watch 50,000, you know, customer calls. [13:33] They try to be on as many as they can. [13:34] But, you know, there is a level of superintelligence [13:36] that you can achieve of learning from the broad based of calls. [13:40] And I think, you know, [13:42] I think it lends itself to products that we simply could not carry before. [13:46] Line price intelligence is extraordinary, I think, at counting intelligence [13:49] that you can be benefiting from the wisdom of [13:52] what the market is is seeing in the proper way to do. [13:55] Even if you're you're a junior accountant, you have really kind of [13:58] the the most involved of learning is benefiting you is exciting. [14:02] And I think it creates [14:04] and it expands the ability for our customers to be productive, [14:09] to be [14:09] strategic, to be not spending time on like tracking down receipts [14:13] and tagging individual transactions, but instead thinking strategically about [14:18] how do they improve the facets of their own businesses. [14:22] So I think that's really profound and I think it creates a lot of certainly [14:26] not just customer benefit but, you know, revenue opportunity [14:30] in a real way. [14:32] You know, I think [14:35] I think there are going to be [14:36] those use cases on cost reduction as well. [14:40] Like, I mean, even in terms of like document [14:43] parsing and adjusting, yeah, [14:47] you could go to services that or [14:51] you know, using lots of, you know, regular expressions [14:53] to go into it and maintaining it where you suddenly have, [14:56] you know, extraordinary large language model [14:58] that can do the job much better and cheaper. [14:59] And so you have those opportunities as well. [15:02] But I think of it really as a [15:04] as a multiplier on people and creating new models. [15:07] I think there are interesting questions that are come out of it. [15:09] Like I think a lot of folks who are working on SAS companies [15:12] price on your dating or on like you have users in a seat [15:16] and they're doing work and you're charging and you've done this rally over years [15:20] like what happens you know when suddenly am [15:25] do you have machines that are doing knowledge worker based work or augmenting [15:30] or amplifying and you quite simply don't need so many people in the same task. [15:34] I think you make them more productive. [15:35] I think in a broad sense, like it's like if you have a salesperson [15:38] who's suddenly 100% more productive, you're like, I want more of those. [15:41] Like, and you try to hire a lot more instead of like, [15:44] you know, cutting is my, my take. [15:45] But like, I think even the the way that businesses [15:49] think about their overarching strategy is going to change a lot. [15:52] And I think thinking about how do you deliver [15:55] measurable, quantifiable value is going to get even more important. [15:58] And you're a two time founder of both companies and fintech, [16:02] as you think about for new founders, you know, how can they leverage A.I.? [16:06] There's a pretty strong argument that the leaders are folks like Microsoft [16:10] Openai they're folks like Google. [16:12] Even if you look at Adobe or Zapier, we heard from earlier, [16:15] the incumbents are moving really, really quickly to leverage AI. [16:19] What kind of opportunity is left for founders? [16:22] What what criteria should they use when they think about starting your company [16:25] and maybe especially talk about that if they're interested [16:28] in starting a fintech company since we're here at the stripe for. [16:32] Yeah, I mean, so first just just fintech broadly speaking, [16:36] like I think it's an it's a really fascinating time, right? Like [16:41] said this before in some context. [16:43] But I think that our actual competitors and most of their founders knowledge [16:49] so quite literally like like they wore top hats, like they like lived in the 1800s. [16:55] They're around like a very long time ago. [16:56] And unless you were a large bank, truly, [17:00] until very recently, you could not move money. [17:05] You had to be a bank. [17:06] And if you were a software provider, you think is like the world went from like [17:09] no phones to flip phones to iPhones to, you know, virtual reality. [17:14] Like your credit card basically stayed the same. [17:16] The payment terminal, you know, stripe is a different case. [17:19] But for for others who are not Stripe kind of the same. [17:22] Right. Is is what it always was. [17:25] And I think the [17:27] door kind of blew open in many ways. [17:30] Great regulations coming out of the financial crisis, Dodd-Frank, [17:34] you know, creating a whole set of infrastructure [17:36] and providers where you can quite literally, [17:40] you don't need to have tens of thousands of people, [17:42] billions of dollars in order to, you know, create your own bank and buy that. [17:46] But you can actually, you know, out of an apartment [17:50] created useful set of software, [17:52] issue a credit card and be on your way to building great products. [17:56] And that is such a profound difference in terms of the costs of doing it [18:00] and just the ability and accessibility for folks [18:02] to to start businesses that I think it's an amazing time in fintech. [18:06] I also think too, when you think about the premise that [18:10] I think that software was historically very adjacent to money movement, [18:14] and I think that great software companies that also move money [18:18] are finally possible and really coming to the forefront. [18:23] Most money, too, is move through, [18:25] you know, you know all the ways checks, cash, very simple ways. [18:31] And I think we're still very early in the digitization. [18:33] So broadly speaking, yeah, I think there's a ton of opportunity [18:37] all over broadly in moving, moving money [18:42] next to when it when I think about, [18:45] you know, the question you had asked, [18:48] does this benefit the incumbents [18:50] or individual businesses more or startups? [18:56] I, I actually think look, there, [18:59] if you're competing on infrastructure, maybe there's a few folks in the audience [19:02] or in the world who can compete in that way, [19:04] you know, fundamentally better algorithms. [19:05] I think that's that's very hard. [19:07] And but look, I think the ability to be sincere, [19:11] to notice problems in the world, understand and be thinking about [19:15] where can we be applying really disruptive technologies to new spaces, [19:19] I think is still as large as it's ever been. [19:22] And if you can digitally call [19:25] brilliant, you know, functionally knowledge workers [19:28] through an API call in order to apply to operational workflows is interesting. [19:33] And I think loss, like many software businesses, could create [19:36] software that assisted people. [19:38] And I think if the agencies improve, you might be able to do operational tasks. [19:43] And I think that's a very interesting framework change for businesses [19:46] that can digitally do work on behalf of whether it's consumers, businesses. [19:53] You know, Paribas my my last [19:55] company when we started back in 2015, it was very funny. [19:58] It was it was basically trying to save customers money on online purchases. [20:03] It was an early Gmail app. [20:05] The idea is you link your email, we'd scan your inbox for receipts, [20:09] we would check prices on what you bought. [20:11] And if a store guaranteed you [20:13] like, basically if you bought something a week [20:14] later dropped in price, you could get the difference. [20:16] And we built it was early and help. Right. [20:20] It's a peak. [20:20] We were, you know, indexing more than 200 million emails per day. [20:25] And we would try to go and, you know, read it. [20:28] It was more rejects than truly, like, great, [20:31] you know, the learning. [20:35] But it's kind of early hacky versions of that. [20:38] We would send notes, you know, construct them and read the responses. [20:43] That was incredibly hard. [20:44] Now it's basically trivial to go and do that. [20:47] And when you think about like the purpose of the business, by the way, [20:50] if we save you money, we would charge a percentage. [20:52] It was an agency based model. [20:54] I think it was like an early prototype of it. [20:56] I think agency based business models are now much [21:00] more possible and feasible to do work on people's behalf. [21:03] I think there's a lot of fields to add, [21:06] you know, that are large structure data sets like [21:10] I think accounting is a language, it is a standard. [21:13] You can create a lot more efficiency in that world for folks. [21:16] And so there's different verticals, there's different style of businesses [21:19] that I think are more possible and used too. [21:20] And frankly, like I think among early stage startups and for companies, look, [21:25] not everything works inherently like this is a very challenging thing. [21:29] I think it's never been more exciting [21:31] and I think the ability to augment people's potential has never been higher. [21:34] So I'm very fired up. [21:35] And you talked a bit about Paribas and Ramp. [21:38] You know, [21:38] has your CEO and leadership style was that really different [21:41] between your first company and your second company? [21:43] How have you sort of evolved as a CEO? [21:46] I are still learning, still try to figure stuff out. [21:48] I mean, we it's interest. [21:51] I mean, so the first company was very different. [21:53] We raised in total about $2 million, [21:55] which was like more money than we could fathom at that, you know, at some point. [21:58] But like it was like now it it was very much like [22:03] scarcity mindset and we were like freaked out [22:06] that like we're going to run out of money. [22:08] Like, my co-founder is from Beirut, Lebanon is on a visa [22:11] and like, you know, like it was like real stakes to this stuff. [22:14] And I think the first time around was watch every penny, everything, [22:21] and don't spend it unless, you know, and I think there's a lot of goodness. [22:23] Like you actually avoid a lot of mistakes. [22:25] You learn in real ways what the job actually is. [22:29] You know, you know, I think it's easy. It creates this. [22:31] We just need to hire this person or [22:32] if only we spent on ads would be growing faster. [22:34] And I think it taught very hard prioritization and internalizing. [22:39] I think we'll see a lot of our our personal leadership is like [22:43] do the work, be sincere, like actually like care and be involved with it. [22:47] I think that's always I think that never goes out of style. [22:50] I think what's the next go around? [22:54] The the thing that we reflected on [22:58] a lot was all a company is is a collection of people. [23:01] There are some people you hire that [23:04] multiply the I think, the company's ability [23:07] to succeed and to, you know, create extraordinary products. [23:11] And so the second time around, we were, you know, what [23:13] we wanted to be thoughtful about money, but it was much more an abundance of [23:16] there are extraordinary people. [23:18] And if we're able to create in many ways, Stripe was an inspiration for this where [23:22] I think [23:24] there were such extraordinary people who came to straight that it became, [23:28] you know, in some ways self fulfilling people wanted to come to straight [23:31] because there was an extraordinary quality of folks. [23:34] There was a question of how do we create that intentionally early on [23:37] in building company, an interest in, you know, extraordinary engineering, [23:41] extraordinary design, syntactic sales [23:45] and be thinking about that very early and creative place [23:47] where people wanted to come in and be a destination. [23:49] And so I spent a lot more time really on on recruiting, spending time [23:53] with the team, diving into products deeply like I think that's been an evolution by [23:57] you know [23:59] turn it to bumper had too often but we do. [24:01] And we were kind of both part of this I started a company [24:04] with Juniper called to go me which we ended up selling to Coinbase [24:08] and obviously that started growing a lot afterwards as well. [24:11] And so we were both part of fintech and crypto were very much [24:14] the hype cycle just a couple of years ago. [24:17] Now it very much feels like maybe that's a fair case to make about A.I. [24:21] as well. [24:21] The size of the rounds, the valuations pre product, [24:24] are there any learnings from sort of the 2021 era and FinTech crypto [24:29] that maybe you would give advice to AI founders, whether it's, you know, [24:32] raise it a reasonable valuation for the next round [24:35] or take less money at first and prove it out. [24:37] Or, you know, are there some kinds of learnings [24:39] from that cycle that you've gone through it [24:41] you've been one of the great survivors, keep growing. [24:44] You know, what kind of advice would you give to maybe some of the folks [24:48] who are going through some of that cycle themselves? Right. [24:51] I mean, I think first and foremost, [24:55] businesses take a lot of years, like a lot, a lot, a lot of years [25:00] in order to really do been achieve what you set out to do. [25:04] And I [25:07] when an investor invests in joins your company [25:11] like in some ways like it's longer you know, and then [25:16] it's like series relation that you can't really break up [25:19] like you're on it, your hits together for like a long time. [25:21] And you know, I think it's both thinking about, [25:25] you know, first, do you want to work together? [25:27] And I think spending time and actually investing in relationship is important. [25:30] But I think more than that, [25:33] I think it's it's very easy when you're like very hot to get [25:37] just like lose perspective on stuff that like [25:39] it might be like an extraordinary time right now, [25:41] but in three months or six months or something is cool off. [25:44] You have the rest of your life ahead of you. [25:46] And I think some of the things that really helped ramp, [25:49] you know, as much as we grew was like we never took the highest price. [25:53] We asked for advice and thought about [25:56] how do we solve problems to better together versus [26:00] trying to raise funds directly and in the right people showed up. [26:05] People who were like really interested in the problem space with you [26:08] responded to emails or you'd say, I'm thinking about this. [26:11] And they come back like it's like a very DC question. [26:14] I was, how can I be helpful? [26:15] And occasionally some were like, they actually feel like [26:17] because of instructions they thought about different questions. [26:19] And I think you figure out [26:20] who do I actually want to work with and you know, last by like not over [26:24] raising, you know, it's exciting for people. [26:28] You know, I think people, you know, people are human want to know like, [26:32] you know, we could have been a I could have been charged 20, 30% more. [26:36] But, you know, founder read less ticking more simple. [26:39] And I actually feel like now like my investment [26:43] is already worth 20% more than what it could have been like. [26:46] And people talk about it and it's exciting. [26:48] And I think it's good not just for, for that, [26:50] but like the actual optimization is not the, the, [26:54] the price of any specific round, but it's the long term enterprise value. [26:58] And I think in some ways by being more reasonable over time, [27:01] it's a better partnership with investors, it's a better value proposition for [27:05] employees, too. [27:07] And I think it [27:08] structurally allows you to look at you, knock it out of the park. [27:12] Great. [27:12] You know, you've built a more valuable businesses. [27:15] You can raise up [27:16] if it goes a little sideways or not quite exactly what you want to do. [27:20] You've built in some room, and I think that actually does [27:25] help over the years. [27:29] Oh, from the audience. [27:30] Oh, okay. We Oh, sure, yeah. [27:33] If anyone has questions, I will take one or two. [27:38] All right. I guess, Gray, back to me. [27:40] I'll give you one final question and then if anyone on the audience thinks of one. [27:43] In the meantime, we'll go to you [27:45] both, you and Kareem, the two co-founders, very technical product people. [27:49] But I'm always surprised to see ramp all over my Twitter [27:52] feed and billboards in New York and huge partnerships. [27:55] So maybe talk to us about, you know, if you have a technical product team, [28:00] how did you guys get really good to go to market? [28:01] Was it hiring the right people? Was it doing it early? [28:04] Was it spending more like what [28:06] do you learn there that maybe you could impart on folks [28:08] who are starting out have excellent products, excellent technical jobs, [28:12] and would love to have the same sort of branding and distribution ramp has got. [28:16] I mean, it's been a [28:18] while. [28:18] I think we're still very much learning like I don't know, I think about [28:23] frankly like the like the launch Shopify had yesterday. [28:26] I thought like that was an amazing learning moment where, you know, [28:31] it was a really simple production. [28:33] It was SEO, it was a camera and a demo. [28:37] And he was telling you about his snowboard shop and how, you know, [28:41] he wanted to leverage in some ways was a liberating item [28:44] to make product better, [28:44] but it was actually just a person trying to make their business better. [28:48] There are great lessons to learn all the way through. [28:51] I think a lot of what helped us was, [28:54] I don't know, I think we're a little bit curious [28:56] about like people and curious about like what makes really great [29:00] marketing and stories people want to write about. [29:02] And I think I think a lot of people's first impulse in marketing is like, [29:05] I have a message, I have a story, I need people to know about it. [29:07] So they like buy our product is like very natural [29:10] and a lot of books are written that way and everything. [29:13] And I think it's just internalizing like people don't care about your product, [29:17] nothing personal. Like it's not like that. [29:19] Like people are very busy, they have their own lives, [29:21] they have their own stresses, their own things. [29:23] They're thinking about. [29:24] And I think rather than thinking about how do I describe [29:28] my company and market it, it's how do I be interesting? [29:31] Like, how do we create things that people want to talk about? [29:34] And I think part of, you know, right now is some of the most popular things [29:38] that people talk about are rants, [29:39] you know, just how we've saved them, but how we recognize different companies, [29:42] how we're highlighting different businesses, how people who are [29:45] a part of the story want to cheer the company on. [29:49] And I think the starting place of that is really seeking from the customer first. [29:54] I think that's part of what made the video yesterday from from Shopify. [29:57] Great. [29:57] But I also think, you know, just being curious and spending time [30:01] reading about the companies [30:02] you really admire and trying to figure out why that helps. [30:05] Now that makes a ton of sense. [30:07] I was early at Airbnb and they said the same thing. [30:09] A lot of the marketing was [30:10] How do we get the hosts and guests to talk about being hosts and guests? [30:13] So that certainly resonates well unless we have any other questions. [30:17] Oh, one question right here. [30:18] Oh, I like. [30:20] It's like. At all it starts to look austere [30:24] and this kind of curious, like, how do you guys think about balance? [30:27] You know, I'm like the spoiled investors are more at me and I'm sort of part of [30:33] super a good question. [30:35] I mean, I think in general, [30:38] when there is an intentionality around our mission statement of we, [30:42] we want to maximize is the amount of money and time [30:45] that we're saving or it's not we want to like win [30:49] the credit card market or have the largest bill payment provider. [30:53] Like I'm not in the business of like selling products. [30:56] Like I'm much more interested in like what is the end problem that a customer's [31:00] trying to use and being generally curious about it. [31:02] And if you start the premise, customers will tell you like [31:05] these are actually the problems in business [31:06] and some are really idiosyncratic [31:08] in like particular of their business, but some like keep coming up [31:12] and there are certain vectors where there's a high degree of pain [31:15] in what we're doing over time can start to lend structural advantages. [31:18] And if we can do that, there's deep pain. [31:20] We have a unique ability to win. [31:21] We can build a vastly better product. [31:24] We want to do that. [31:25] There is areas around the fringes where we might be. [31:27] We might be integrating when people like going deeper, but really [31:30] starting with the users problem and letting that guide in a lot of ways [31:33] a product strategy. And if you do that, you get much more input. [31:36] Not of did they buy the product or not, do they like this thing? [31:39] You know, how do you price it? But like what are their problems? [31:41] And it becomes an engine to start discovering and thinking of [31:44] how do we build it? [31:45] And we're constantly like the next half of the year, [31:47] like we're deeply excited about the product velocity [31:50] and like we're building things up in like months, you know, in some cases [31:54] like a year before, like it's coming out to market and it's sort of, [31:58] you know, coming back from that central problem. [32:01] I think the other piece of it for velocity, you know, [32:06] no one likes it when they have to go check [32:07] for like 30 people to like change the color of a button or like, [32:10] you know, get everyone's input. Like, what do we call this thing? [32:13] Or like different design decisions. [32:15] Like, I think it's still just like great things can be done with small teams [32:19] in a very compressed timeline [32:21] and you know, remembering that is the scale is super important. [32:24] Like even some of our largest teams today or like less than ten people, [32:28] we have lots of small teams with real ownership. [32:31] We try to do as much as we can, [32:34] you know, say like, what does success actually ultimately look like? [32:38] And so those are some of the principles. [32:40] Awesome. [32:41] Well, thank you so much, Eric, [32:42] for coming here today and spending your time with us. [32:44] I'm sure a lot of founders. Appreciate it. [32:45] So thank you so much. [32:47] Thanks, Mark. [32:47] And thanks, everyone, for coming today.