Tony Xu of DoorDash: Surviving 1,000 Days of Startup Hell

以下是内容的中文翻译： 托尼·徐（Tony Xu）分享了他在DoorDash早期阶段的见解，强调了一种植根于不懈实验、深刻客户理解和注重行动的理念。 **1. 最小可行产品（PaloAltoDelivery.com）：** * **起源：** 仅用43分钟、花费9美元（域名别名当时正好可用）搭建而成。这比先进的AI工具出现早了12-13年。 * **目的：** 旨在快速测试一个想法：从那些传统上不提供送餐服务的餐馆提供送餐。创始人们曾怀疑在2013年人们是否真的*需要*送餐服务，假设其缺失意味着需求不高。 * **机制：** 一个静态网页，展示了八家当地帕洛阿尔托餐厅的PDF菜单。订单通过一个Google Voice号码下达，该号码会拨打四位创始人的手机。其中一位创始人会接单、下单、取餐，然后送餐。收款则使用Square读卡器在现场进行（徐在之前实习时用过的）。 * **早期市场：** 当时送餐市场“门户大开”。美国一百万家餐馆中，只有2万到2.5万家提供送餐服务（主要是在大城市中心的披萨店和一些中餐馆）。现有的“送餐公司”本质上是潜在客户开发服务，他们将订单传真给餐馆，餐馆再自行处理送餐。DoorDash开创了为没有自有配送团队的餐馆提供第三方物流网络的概念。 **2. 起源故事与物流网络战略：** * **小企业焦点：** 徐的个人背景让他对小企业主产生了深刻的敬意，他作为移民的母亲曾打三份工（包括中餐馆服务员）。他将小企业主视为“我们所居住所有城市的GDP”。 * **发现：** 创始人们采访了300多家湾区企业。一位面包师给他们看了一个三英寸厚的活页夹，里面全是*被拒绝的*外卖订单，因为她无法满足这些需求。这凸显了一个明显未被满足的需求。 * **垂直行业选择：** 为了建立一个高效、快速（30分钟弹性）的物流网络，他们需要“网络密度”。他们选择了餐馆，因为有100万家（相比之下，杂货店只有几十万家），这提供了最高数量的门店，可以开始建立一个密集的网络，以便最终“运送其他一切物品”。 **3. 帕洛阿尔托与市中心：“通过实践获得的宝贵洞察”：** * **意外的开始：** 在帕洛阿尔托开始业务并非有意识的选择，而是因为他们当时是那里的学生。 * **关键实验：** 他们比较了帕洛阿尔托和旧金山的送餐时间。出乎意料的是，帕洛阿尔托的送餐速度*更快*。 * **为何帕洛阿尔托更快/更好：** 停车更方便，多层公寓楼更少（减少了在大堂/电梯导航的问题）。帕洛阿尔托代表了大多数美国城市的“轴辐式模型”（主街商业，住宅区为辐条），这使得物流效率更高。 * **客户洞察：** 旧金山的客户可以轻松步行到餐馆；而帕洛阿尔托（斯坦福附近）的客户则需要步行数英里。这表明在密度较低的地区消费者需求更高。 * **目标受众：** 早期的客户通常是带着小孩的妈妈，她们寻求便利和节省时间的解决方案。这些地区“需要喂养的家庭成员更多”，且多为独栋住宅，简化了送餐过程。这种自然的契合对于早期实现产品与市场契合至关重要。 **4. 早期运营、融资和Y Combinator：** * **精益运营：** 创始人零薪资，9美元的网站。“寻找我的朋友”应用程序用于追踪。没有营销预算。尽管业务量小，但银行账户没有减少，这给了他们信念。 * **YC焦点：** 在Y Combinator期间，主要目标是回答三个问题：1）消费者是否愿意支付6美元的送餐费？2）餐馆是否愿意以15%的佣金率合作？3）他们能否向配送员支付可持续的工资？ * **“开着本田送鹰嘴豆泥”：** 徐回忆道，当同学们都在度假时，他却“开着本田送鹰嘴豆泥”，这凸显了早期创业的艰辛和他们解决核心问题的决心。 * **创始人的专业知识：** 创始人们都没有物流或餐饮经验。他们必须亲自送餐，才能了解应该如何运作，这让他们意识到需要构建四个相互关联的系统：消费者网站、餐馆应用程序、配送员应用程序和调度系统。 **5. “被误解”的业务与隐藏的复杂性：** * **消费者认知：** 大多数人认为DoorDash只是一个提供食物配送（午餐/晚餐）的“魔法按钮”。 * **现实：** 这种“魔法”在于其看不见的复杂性：Dashers（配送员）的体验、运营效率、质量控制、成本管理、订单准确性以及为商家消除摩擦。 * **竞争对手洞察（Wolt）：** 欧洲配送公司 Wolt 的首席执行官米奇·库西（Miki Kuusi）向徐坦承，尽管获得了10亿美元的融资报价，但他觉得“无法击败”徐，因为DoorDash拥有深厚的、不为人知的运营专业知识，最终选择将 Wolt 出售给 DoorDash 并加入团队。 * **“数万次实验”：** DoorDash的成功源于此。95%的实验在触达客户之前就失败了。物理世界是混乱、无序、不断变化的，并且没有组织化的数据（例如，丢失的苹果、想家的员工）。DoorDash建立了系统来快速检测、预防和响应这些现实世界的问题。 **6. 实验与持续改进：** * **每年数千次：** DoorDash的目标是每年进行数千次实验，由一个优先考虑学习的系统驱动。 * **过程：** 从“做那些无法规模化的事情”开始，识别反复出现的问题，提出假设，进行实验，然后将成功的解决方案工程化为产品。这形成了一个紧密、高效的学习循环。 * **北极星：** 在多个维度上“为客户做得更好”：更广泛的选择、更实惠的价格、更快的送达、零失误、准时服务以及当出现问题时提供卓越的支持。 * **“不变的事物”：** 人们将永远需要更多的选择、实惠的价格、速度和便利。这构成了长期战略的稳定基础。 **7. 首席执行官的日常客户支持：** * **目的：** 徐亲自参与日常客户支持（邮件、聊天、电话）。 * **可观测性：** 客户提供“免费的宝贵意见”——有价值的反馈。“沉默是企业最大的杀手。” * **文化：** 随着公司发展，防止公司与客户之间产生距离。提醒员工“在这家公司，最重要的工作和唯一的信条就是为客户解决问题。” * **数据与轶事：** 轶事常常揭示“边缘案例”（分布的尾部）。虽然数据有助于确定优先级，但改进产品通常意味着解决这些边缘问题。他关注高频用户和新用户，因为他们的体验通常最有价值，即使它们“与数据不符”。 * **调试：** 他分析详细的客户（尤其是Dashers）邮件，使用内部调试工具追踪订单，识别错误来源（物理世界、系统、产品界面），并生成改进假设。 **8. “永恒的使命”：** * **使命宣言：** “发展并赋能本地经济”。这是一个“永恒的”使命，因为它是一场为小型、中型和大型企业成功的持续斗争，这些企业创造就业、贡献GDP，并定义了社区。 * **物理世界的复杂性：** 这项使命本质上是复杂的，因为物理世界是无序、不断变化的，并涉及许多人类参与者（消费者、Dashers、商家）。 * **替代方案是可怕的：** 一个由一两个大公司主导的未来将扼杀本地经济，并消除社区独特的“个性”。 **9. 为商家和未来产品利用数据：** * **结构化数据：** DoorDash正在将以前未经整理的物理世界信息结构化。 * **赋能商家：** 为商家提供关于其自身业务的洞察（缺货商品、最佳定价、捆绑销售机会），以帮助他们成长。徐引用了“贝佐斯‘不可或缺’”的比喻：让商家觉得合作的价值巨大，以至于“不合作就是不负责任”。 * **商业伙伴：** 目标是成为“任何企业、任何问题时的第一个求助电话”，而不仅仅是配送。例如： * **自动化：** 为商家调整菜单价格、运行促销活动。 * **增长：** 帮助小企业从一家店发展到多家店，甚至分销其产品（例如，一位饼干烘焙师利用DoorDash的网络通过其他企业销售其饼干）。 * **“运送一切”：** 长期愿景是运送城市中的所有物品（目前只占一小部分）。 * **Dashbar 履约中心：** 为大型零售商（例如 Kroger、CVS）运营仓库和库存管理，以实现更快、更准确的配送。 * **自动驾驶汽车（DoorDash Dot）：** 开发了自有专用自动驾驶汽车（因为合作伙伴不愿建造他们所需的东西），用于最后一英里的配送。DoorDash Dot 在道路、人行道和自行车道上行驶，旨在解决“最后十英尺的问题”。这个历时六年的开发项目目前正在菲尼克斯/斯科茨代尔运营。 **10. 人才与文化（“结合了罗德学者和海豹突击队成员的特质”）:** * **所需特质：** 聪明、高处理能力（罗德学者）与强大的行动偏好和责任感（海豹突击队）相结合。 * **独特的面试流程：** * **非工程师：** 进行20分钟的讨论，然后提供20美元和8小时时间去获取100个客户。这测试的是行动而非分析。 * **工程师：** 最后一轮面试包括与徐一起开着本田送货，讨论如何将改进产品化。这寻求的是超越编码的“解决问题的能力”。 * **可观察的行为：** 寻找： * 首席运营官克里斯托弗·佩恩（Christopher Payne）在被录用前亲自送货并撰写了一份3000字的分析报告。 * 总裁兼首席财务官候选人普拉比尔（Prabir）在咖啡聊天时带来了一份数兆字节的财务模型。 * 注重行动、注重细节、能够兼容对立观点、强大的追随力、对持续改进的执着（例如，最佳卡拉OK歌手）。 **11. “注重行动”与“千日炼狱”：** * **争议解决：** DoorDash的文化是通过行动和实验来“解决争议”，而不是无休止的讨论。 * **2016年融资危机：** 经过三年的增长，一次度假（三年来的第一次）导致一份投资意向书因公开市场崩盘而被撤回。这开始了“三年”艰难的融资时期，面临负面媒体报道（“亏损企业”、“有毒行业”），而内部指标（可重复性、单位经济效益、自然增长）却表现强劲。 * **掌控心态：** 1. **思想上的坦诚：** 与团队分享所有指标（包括不断减少的现金余额），解释外部叙述与内部进展之间的脱节。 2. **专注于可控：** 为团队设定一个“且”函数：“持续增长、持续抢占市场份额、提高盈利能力，*并且*不耗尽现金。” 3. **真诚的关系：** 强调在工作中有“真正的朋友”以及这项使命的集体“冒险”，以维持意志力。 4. **个人习惯：** 在混乱中保持锻炼（马拉松）和约会之夜作为不变的常态。 * **股价：** 徐明确表示不关注股价，认为它超出他的控制范围，也不是他的动力。他优先关注内部指标和解决客户问题，这呼应了杰夫·贝佐斯在亚马逊互联网泡沫破裂时的做法。 **12. 两种管理体系：** * **核心业务：** 不断革新和改进现有成功业务（就像为一架大型飞机进行“空中引擎更换”）。 * **新业务：** 同时启动“纸飞机”——寻找产品与市场契合度的新项目。这些项目需要不同的指标、人才、资源和时间表。 * **阶段性门控方法：** 运作方式类似于内部风险投资系统，项目根据已证明的客户价值和进展，在资源受限的情况下逐步获得更多资源。 **13. 向同行和导师学习：** * **YC同行：** 与Airbnb、Stripe、Coinbase等公司一同成长，从他们的共同经验中学习。 * **马克·扎克伯格：** 作为Meta董事会成员，徐钦佩扎克伯格“永远学习新事物”和“重塑自我”的意愿，尽管受到批评，但他仍在新的平台（VR/AR、AI）上大胆押注，即使他是一个“新手”。 **14. 巴西柔术与商业哲学：** * **“身体上的国际象棋”：** 巴西柔术要求兼容对立面（坚定/放松、有意/灵活）。 * **持续改进：** 专注于在微小细节上“每天进步1%”。精英练习者通过复合式的小改进而非灵丹妙药获得成功。这与DoorDash精益求精的方式如出一辙。 **15. AI的影响：** * **编码与学习循环：** AI智能体（大型语言模型）正在加速编码过程，让个人能够更快地进行原型设计、实验并向小群体发布，从而大幅缩短学习循环。 * **背景与记忆：** 大型语言模型擅长处理大量信息。挑战在于向它们输入*正确*的信息，以改进目前仍需手动完成的活动，从而提高效率和效果。 * **数据与行动：** 尽管DoorDash收集了大量数据，但其价值在于将这些数据与可操作的步骤结合起来，以解决端到端的客户问题。 **结论：** 徐回顾了DoorDash从传真机到人工智能的快速演变，这得益于一个核心信念：“成为专家的最好方法就是亲身实践。”这种亲力亲为、解决问题的方法一直是DoorDash成功的核心，也是其持续追求永恒使命的关键。

Tony Xu shares his insights on building DoorDash from its earliest days, emphasizing a philosophy rooted in relentless experimentation, deep customer understanding, and a bias for action. **1. The Minimal Viable Product (PaloAltoDelivery.com):** * **Genesis:** Built in just 43 minutes for $9 (domain alias was available). This was 12-13 years before advanced AI tools. * **Purpose:** To quickly test the idea of offering delivery from restaurants that didn't traditionally provide it. The founders questioned if people even *wanted* delivery in 2013, assuming its absence meant low demand. * **Mechanism:** A static webpage displaying eight PDF menus of local Palo Alto restaurants. Orders were placed via a Google Voice number that rang the four founders' cell phones. One founder would take the order, place it, pick it up, and deliver it. Payments were collected on-site using Square card readers (from Xu's previous internship). * **Early Market:** The delivery market was "wide open." Out of a million US restaurants, only 20,000-25,000 offered delivery (mostly pizza, some Chinese, in large city centers). Existing "delivery companies" were essentially lead generation services that faxed orders to restaurants, which then handled their own deliveries. DoorDash pioneered the concept of a third-party logistics network for restaurants without their own delivery fleet. **2. The Origin Story & Logistics Network Strategy:** * **Small Business Focus:** Xu's personal background as an immigrant whose mother worked three jobs (including as a Chinese restaurant waitress) instilled a deep appreciation for small business owners, whom he views as the "GDP for all the cities we live in." * **Discovery:** The founders interviewed over 300 Bay Area businesses. A baker showed them a three-inch binder of *turned-down* delivery orders because she couldn't fulfill them. This highlighted a clear unmet need. * **Vertical Selection:** To build an efficient and fast (30-minute flexible) logistics network, they needed "network density." They chose restaurants because there were 1 million of them (compared to hundreds of thousands of grocery stores), offering the highest count of stores to start building a dense network to eventually "deliver everything else." **3. Palo Alto vs. City Centers: An "Earned Secret":** * **Accidental Start:** Starting in Palo Alto was not a conscious choice but due to being students there. * **Key Experiment:** They compared delivery times in Palo Alto vs. San Francisco. Counter-intuitively, Palo Alto deliveries were *faster*. * **Why Palo Alto was Faster/Better:** Easier parking, fewer multi-story apartment complexes (lobby/elevator navigation issues). Palo Alto represented the "hub and spoke" model of most US cities (main street commerce, residential spokes), allowing for efficient logistics. * **Customer Insight:** SF customers could easily walk to restaurants; PA customers (near Stanford) had to walk miles. This indicated higher consumer demand in less dense areas. * **Target Audience:** Early customers were often moms with young children, seeking convenience and time-saving solutions. These areas had "more mouths to feed" and single-family homes, simplifying delivery. This organic fit was crucial for early product-market fit. **4. Early Operations, Funding, and Y Combinator:** * **Lean Operations:** Zero salaries for founders, $9 website. "Find My Friends" app for tracking. No marketing budget. The bank account not depleting despite small volume provided conviction. * **YC Focus:** During Y Combinator, the primary goal was to answer three questions: 1) Would consumers pay $6 for delivery? 2) Would restaurants partner for 15%? 3) Could they pay drivers a sustainable wage? * **"Hummus in my Honda":** Xu recalls delivering hummus in his Honda while classmates vacationed, highlighting the non-glamorous early days and dedication to solving core problems. * **Founders' Expertise:** None had prior logistics or restaurant experience. They had to do the deliveries themselves to learn how it should work, leading them to realize the need to build four interconnected systems: consumer website, restaurant app, driver app, and dispatch system. **5. The "Misunderstood" Business & Hidden Complexity:** * **Consumer Perception:** Most people see DoorDash as a "magic button" for food delivery (lunch/dinner). * **The Reality:** The "magic" lies in the unseen complexity: Dasher experience, operational efficiency, quality control, cost management, order accuracy, and friction removal for merchants. * **Competitor Insight (Wolt):** Miki Kuusi, CEO of European delivery company Wolt, confessed to Xu that despite having a $1 billion funding offer, he felt he "can't beat him" due to DoorDash's deep, hidden operational expertise, choosing to sell Wolt to DoorDash and join the team. * **"Tens of thousands of experiments":** DoorDash's success stems from this. 95% of experiments fail before ever reaching the customer. The physical world is chaotic, unstructured, constantly changing, and has no organized data (e.g., missing apple, homesick employee). DoorDash builds systems to detect, prevent, and respond rapidly to these real-world issues. **6. Experimentation and Continuous Improvement:** * **Thousands Yearly:** DoorDash aims to run thousands of experiments annually, driven by a system that prioritizes learning. * **Process:** Starts with "doing things that don't scale," identifying recurring problems, forming hypotheses, running experiments, and then engineering successful solutions into products. This creates a tight, efficient learning loop. * **North Star:** "Better for customers" across multiple dimensions: wider selection, affordability, faster deliveries, no mistakes, on-time service, and excellent support when things go wrong. * **"Things That Don't Change":** People will always want more selection, affordability, speed, and convenience. This forms a stable foundation for long-term strategy. **7. Daily Customer Support as a CEO:** * **Purpose:** Xu personally engages in daily customer support (emails, chats, calls). * **Observability:** Customers provide "freebies" – valuable feedback. "Silence is the greatest killer of a business." * **Culture:** Prevents distance between the company and its customers as it grows. Reminds employees that "the number one job and the only religion at this company is to solve problems for customers." * **Data vs. Anecdotes:** Anecdotes often highlight "edge cases" (the tail of the distribution). While data informs prioritization, improving the product often means addressing these edges. He focuses on power users and new users, as their experiences are often the most valuable, even if they "disagree with the data." * **Debugging:** He analyzes detailed customer (especially Dasher) emails, using internal debugging tools to track orders, identify error sources (physical world, systems, product interface), and generate hypotheses for improvement. **8. The "Eternal Mission":** * **Mission Statement:** To "grow and empower local economies." This is an "eternal" mission because it's a constant fight for the success of small, medium, and large businesses, which create jobs, contribute to GDP, and define neighborhoods. * **Complexity of the Physical World:** This mission is inherently complex because the physical world is unstructured, constantly changing, and involves many human actors (consumer, Dasher, merchant). * **Alternative is Terrifying:** A future dominated by one or two large players would stifle local economies and remove the unique "personality" of neighborhoods. **9. Data Utilization for Merchants and Future Products:** * **Structured Data:** DoorDash is structuring information about the physical world that was previously unorganized. * **Merchant Empowerment:** Provides insights to merchants about their own business (out-of-stock items, optimal pricing, bundling opportunities) to help them grow. Xu uses the "Bezos Prime" analogy: make it so valuable for merchants that it's "irresponsible not to partner." * **Business Partner:** Aims to be the "first phone call for any business, any issue," not just delivery. Examples: * **Automation:** Adjusting menu prices, running promotions for merchants. * **Growth:** Helping small businesses scale from one store to many, or even distribute their products (e.g., a cookie baker using DoorDash's network to sell their cookies through other businesses). * **"Deliver Everything":** The long-term vision is to deliver all items within a city (a small fraction today). * **Dashbar Fulfillment:** Operating warehouses and inventory management for large retailers (e.g., Kroger, CVS) to enable faster, more accurate delivery. * **Autonomous Vehicles (DoorDash Dot):** Developed their own purpose-built AVs (after partners didn't want to build what they needed) for last-mile delivery. DoorDash Dot travels on roads, sidewalks, and bike lanes, designed to solve the "last 10 feet problem." This six-year development project is now operating in Phoenix/Scottsdale. **10. Talent & Culture ("Rhodes Scholars that meet Navy SEALs"):** * **Desired Traits:** Smart, high processing power (Rhodes Scholar) combined with a strong bias for action and accountability (Navy SEAL). * **Unique Interview Process:** * **Non-Engineers:** A 20-minute discussion followed by $20 and 8 hours to acquire 100 customers. This tested action over analysis. * **Engineers:** Final round interview involved doing deliveries together in Xu's Honda, discussing how to productize improvements. This sought "problem-solving prowess" beyond just coding. * **Observable Behaviors:** Looked for: * Christopher Payne (COO) doing deliveries and writing a 3,000-word analysis before being hired. * Prabir (President/CFO candidate) bringing a multi-megabyte financial model to a coffee chat. * Bias for action, attention to detail, ability to hold opposing ideas, strong followership, obsession with continuous improvement (e.g., best karaoke singer). **11. "Bias for Action" & "Thousand Days of Hell":** * **Debate Resolution:** DoorDash's culture is to "settle debates" through action and experiments, not endless discussion. * **2016 Funding Crisis:** After three years of growth, a vacation (first in three years) led to a term sheet being pulled due to a public market crash. This began "three years" of struggling to raise capital, facing negative media narratives ("money-losing business," "toxic sector") while internal metrics (repeatability, unit economics, organic growth) were strong. * **Controlling Psychology:** 1. **Intellectual Honesty:** Shared all metrics (including decreasing cash balance) with the team, explaining the disconnect between external narrative and internal progress. 2. **Focus on Control:** Set an "AND" function for the team: "keep growing, keep taking share, get more profitable, AND don't run out of cash." 3. **Genuine Relationships:** Emphasized having "genuine friends at work" and the collective "adventure" of the mission to sustain willpower. 4. **Personal Routines:** Maintained exercise (marathons) and date nights as constants amidst chaos. * **Stock Price:** Xu explicitly doesn't focus on the stock price, viewing it as outside his control and not motivating. He prioritizes internal metrics and solving customer problems, echoing Jeff Bezos's approach during Amazon's dot-com bust. **12. Two Management Systems:** * **Core Business:** Continuously reinvent and improve the existing successful business (like "mid-air engine transplant" for a large airplane). * **New Ventures:** Simultaneously launch "paper airplanes" – new projects in search of product-market fit. These require different metrics, talent, resources, and timelines. * **Stage-Gated Approach:** Operates like an internal venture capital system, where projects earn their right to more resources based on demonstrated customer value and progress, often starting resource-constrained. **13. Learning from Peers & Mentors:** * **YC Peers:** Grew alongside companies like Airbnb, Stripe, Coinbase, learning from their shared experiences. * **Mark Zuckerberg:** Serving on Meta's board, Xu admires Zuckerberg's "willingness to always learn new things" and "reinvent themselves," making bold bets on new platforms (VR/AR, AI) despite criticism and being a "rookie" in those domains. **14. Jiu-Jitsu & Business Philosophy:** * **"Physical Chess":** Jiu-Jitsu demands holding opposites (firm/relaxed, intentional/flexible). * **Continuous Improvement:** Focus on "1% better every day" in tiny details. Elite practitioners succeed through compounding small improvements, not silver bullets. This mirrors DoorDash's approach to mastering its craft. **15. AI's Impact:** * **Coding & Learning Loop:** AI agents (LLMs) are speeding up the coding process, allowing individuals to prototype, experiment, and ship to small groups faster, significantly collapsing the learning loop. * **Context & Memory:** LLMs excel at processing vast amounts of information. The challenge is feeding them the *right* information to improve activities that are currently done manually, increasing efficiency and effectiveness. * **Data & Action:** While DoorDash collects immense data, its value is in combining that data with actionable steps to solve end-to-end customer problems. **Conclusion:** Xu reflects on DoorDash's rapid evolution from fax machines to AI, driven by the core belief that "there's just no better way to be an expert than to just do the work." This hands-on, problem-solving approach has been central to DoorDash's success and its continuous pursuit of its eternal mission.