Back to Details

b2b-ai-startup-levie

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Aaron Levie: Why Startups Win in the AI Era

Aaron Levie：为什么创业公司能在 AI 时代取胜

Strategic frameworks and tactical guidance for building B2B AI startups during the 2024-2027 window.
面向 2024-2027 年窗口期内构建 B2B AI 创业公司的战略框架与实战指南。

Core Thesis

核心论点

AI creates a once-in-a-decade window for startups to build transformative companies by targeting enterprise work that was previously uneconomical to automate. This window closes approximately 2027.
Key insight: Target work categories where AI fundamentally changes economics, not incremental "X with AI" improvements to existing software that incumbents will address.
AI 为创业公司创造了一个十年难遇的窗口期，让它们能够通过解决过去因经济性不足而无法自动化的企业工作来构建具有变革性的公司。这个窗口大约会在 2027 年关闭。
关键洞见：应瞄准那些 AI 能从根本上改变经济性的工作类别，而不是对现有软件做渐进式的“X with AI”改进——后者 incumbent 也会着手解决。

The Opportunity Framework

机会框架

Structured vs Unstructured Data

结构化数据与非结构化数据

Data TypeExamplesHistorical AutomationAI Opportunity
StructuredCustomer IDs, invoice numbers, revenue figuresFully automated by traditional softwareMarginal improvement
UnstructuredDocuments, contracts, presentations, marketing assetsNever automatedMassive opportunity
Action: Focus AI efforts on unstructured data workflows where software never could automate before.
数据类型示例历史上的自动化程度AI 机会
结构化数据客户 ID、发票编号、收入数据传统软件已完全自动化边际改进空间
非结构化数据文档、合同、演示文稿、营销素材从未实现自动化巨大机会
行动建议：将 AI 投入集中在非结构化数据工作流上，因为这些工作过去 软件根本无法自动完成。

The Nouns and Verbs Exercise

Nouns 与 Verbs 练习

List all human activities (eat, sleep, travel, watch, read, write, analyze) and identify:
  1. Which problems technology has already solved
  2. Which remain unsolved
  3. Which AI now makes economically viable to solve
列出所有人类活动（吃、睡、旅行、观看、阅读、写作、分析），并识别：
  1. 哪些问题已经被技术解决
  2. 哪些问题仍未解决
  3. 哪些问题如今借助 AI 已具备经济可行的解决方案

Market Timing Assessment

市场时机评估

The Window (2024-2027)

窗口期（2024-2027）

2008-2014: Consumer/enterprise "nouns and verbs" solved
2024-2027: AI startup window open ← WE ARE HERE
Post-2027: Markets saturated, harder to enter
Evaluate timing with:
  • Is this problem newly economical to solve with AI?
  • Would this have been possible 2 years ago?
  • Will incumbents address this within 18 months?
2008-2014：Consumer/enterprise 的“Nouns 与 Verbs”已被解决
2024-2027：AI 创业窗口开启 ← 我们在这里
2027 年之后：市场趋于饱和，进入难度更高
评估时机时问：
  • 这个问题是否因 AI 而首次变得具有经济可行性？
  • 两年前是否有可能解决？
  • incumbent 会在 18 个月内着手解决吗？

Competitive Positioning

竞争定位

Core vs Context Framework (Geoffrey Moore)

Core 与 Context 框架（Geoffrey Moore）

TypeDefinitionWho Builds ItExamples
CoreDifferentiates the companyIn-house or customTrading algorithms, proprietary analytics
ContextNecessary but non-strategicBuy from vendorsHR systems, expense reporting, document management
Strategic insight: Enterprises will NOT build custom AI for "context" functions due to maintenance burden and liability. They only build for "core" differentiating activities.
Action: Target "context" functions—enterprises will buy, not build.
类型定义构建方式示例
Core形成公司的差异化能力内部构建或定制交易算法、专有分析能力
Context必需但不具备战略意义从供应商购买HR 系统、费用报销、文档管理
战略洞见：企业不会针对 Context 职能构建定制 AI，因为维护负担和责任风险过高；它们只会为 Core 差异化活动自行构建。
行动建议：瞄准 Context 职能——企业会选择购买，而不是自行构建。

Incumbent Analysis Workflow

Incumbent 分析工作流

  1. List competitor capabilities (be generous in assumptions)
  2. Assume they execute perfectly on AI integration
  3. Identify remaining gaps:
    • Speed to market (your advantage)
    • Organizational constraints (their disadvantage)
    • Technical debt (their disadvantage)
    • Incentive misalignment (their disadvantage)
  4. Design strategy that wins even if their AI agents are excellent
Example analysis for competing with Workday:
Workday strengths: Existing customer base, data access, brand trust
Workday constraints: Can't cannibalize seat revenue, slow product cycles
Your opportunity: Consumption-based model for work Workday doesn't automate
Win condition: Target workflows Workday has no incentive to automate
  1. 列出竞争对手的能力（在假设时尽可能宽泛）
  2. 假定它们能够完美执行 AI 集成
  3. 找出仍然存在的差距：
    • 上市速度（你的优势）
    • 组织约束（它们的劣势）
    • 技术债务（它们的劣势）
    • 激励错位（它们的劣势）
  4. 设计即使它们的 AI agents 表现出色，你的策略仍能取胜的方案
与 Workday 竞争的分析示例：
Workday 优势：现有客户群、数据访问权限、品牌信任
Workday 约束：无法自我蚕食席位收入，产品迭代周期缓慢
你的机会：针对 Workday 尚未自动化的工作采用 consumption-based 模式
获胜条件：瞄准 Workday 没有动力去自动化的工作流

Pricing Model Design

定价模型设计

Seat-Based vs Consumption-Based

Seat-Based 与 Consumption-Based

ModelCharacteristicsConstraintsBest For
Seat-basedPer user/licenseLimited by job function demographicsTraditional SaaS
Consumption-basedPer unit of work processedScales with usageAI products
模型特征约束最适合
Seat-based按用户或许可证计费受职位群体特征限制传统 SaaS
Consumption-based按处理的工作单元计费随使用量扩展AI 产品

Recommended AI Pricing Structure

推荐的 AI 定价结构

Base: Subscription floor (predictable revenue)
Variable: Consumption above baseline (captures growth)
Margin target: 80-90% gross margin
Token-to-Value Stack Assessment:
Raw AI token cost: $X
Your price: Should be >> 2X token cost
Software value above tokens: This determines your margin
Warning signs of price compression:
  • Margin approaching 2x token costs
  • No proprietary workflow above AI layer
  • Easily replicable with raw API calls
Action: Build substantial software layers above AI tokens to maintain margins.
基础：订阅费下限（收入可预测）
可变费用：超出基准的 consumption（获取增长）
利润率目标：80-90% 毛利率
Token-to-Value 堆栈评估：
原始 AI token 成本：$X
你的定价：应远高于 token 成本的 2 倍
token 成本之上的软件价值：这决定你的利润率
价格压缩的警示信号：
  • 毛利率接近 token 成本的 2 倍
  • AI 层之上没有专有工作流
  • 通过原始 API 调用即可轻松复制
行动建议：在 AI token 之上构建 substantial software layers，以维持利润率。

Startup Idea Evaluation

创业想法评估

Quick Assessment Checklist

快速评估清单

  • Does AI fundamentally change the economics? (Not just "faster/cheaper")
  • Is this unstructured data or context work? (Not already automated)
  • Would incumbents face disincentives to build this?
  • Can you build 80%+ margin above token costs?
  • Is the timing right? (Not too early, not too late)
  • AI 是否从根本上改变了经济性？（不只是“更快/更便宜”）
  • 这是非结构化数据还是 Context 工作？（尚未被自动化）
  • incumbent 是否会因缺乏激励而不愿构建它？
  • 能否在 token 成本之上实现 80%+ 毛利率？
  • 时机是否合适？（不能太早，也不能太晚）

Red Flags

危险信号

  • "X with AI" positioning (incremental improvement)
  • Targeting structured data already in databases
  • Competing directly with incumbent's core product
  • Thin wrapper over AI APIs with no proprietary workflow
  • Targeting "core" enterprise functions (they'll build in-house)
  • “X with AI”定位（渐进式改进）
  • 瞄准已存入数据库的结构化数据
  • 与 incumbent 的核心产品直接竞争
  • 仅是 AI API 的薄封装，且没有专有工作流
  • 瞄准企业的 Core 职能（它们会自行构建）

Green Flags

积极信号

  • New category of work now economically viable
  • Unstructured data transformation
  • "Context" function incumbents won't prioritize
  • Clear consumption-based monetization path
  • 18+ month lead time before incumbent response
  • 新的工作类别如今已具备经济可行性
  • 非结构化数据转换
  • incumbent 不会优先处理的 Context 职能
  • 清晰的 consumption-based 变现路径
  • 在 incumbent 作出响应前拥有 18 个月以上的领先时间

Founder Preparation

创始人准备

Required Reading (Complete Before Starting)

必读书籍（开始创业前完成）

  1. Innovator's Dilemma (Clayton Christensen)
    • Key takeaway: Successful companies fail to adopt disruptive tech serving niche markets
    • Application: Identify where incumbents are structurally unable to respond
  2. Crossing the Chasm (Geoffrey Moore)
    • Key takeaway: Gap between early adopters and mainstream requires different strategies
    • Application: Plan distinct go-to-market for each phase
  3. Blue Ocean Strategy
    • Key takeaway: Create uncontested market space rather than competing in existing markets
    • Application: Define category where you don't compete head-to-head
  1. Innovator's Dilemma（Clayton Christensen）
    • 核心要点：成功公司往往会因未采用服务利基市场的颠覆性技术而陷入困境
    • 应用：找出 incumbent 在结构上无力回应的领域
  2. Crossing the Chasm（Geoffrey Moore）
    • 核心要点：早期采用者与主流市场之间的鸿沟需要不同的策略
    • 应用：为每个阶段规划不同的 go-to-market 策略
  3. Blue Ocean Strategy
    • 核心要点：创造无竞争的市场空间，而不是在现有市场中竞争
    • 应用：定义一个无需与现有产品正面交锋的新品类

Team Composition

团队构成

  • Find a co-founder even if not technical
  • AI enables small teams to act like large companies
  • Prioritize great design in enterprise software (differentiation opportunity)
  • 寻找一位即使不是技术出身的联合创始人
  • AI 让小团队能够像大公司一样行动
  • 重视企业软件中的优秀设计（差异化机会）

AI Impact Mental Model

AI 影响力心智模型

What AI Does NOT Do

AI 不会做什么

  • Eliminate jobs wholesale
  • Make all enterprise software obsolete
  • Enable enterprises to build everything custom
  • 整体消灭岗位
  • 让所有企业软件过时
  • 让企业能够定制一切

What AI DOES Do

AI 会做什么

  • Frees human time for strategic work
  • Makes previously uneconomical work viable
  • Shifts value capture from seat count to work volume
  • Creates leverage for small teams
Reframe: "AI is coming for jobs" → "AI eliminates non-strategic activities humans shouldn't be doing"
  • 把人的时间解放出来用于战略工作
  • 让以前不具备经济可行性的工作变得可行
  • 将价值获取从席位数转向工作量
  • 为小团队创造杠杆
重新定义：“AI 正在取代岗位” → “AI 消除了人类本就不应承担的非战略活动”

Quick Reference: Decision Trees

快速参考：决策树

Should I Build This AI Product?

我是否应该构建这款 AI 产品？

Is the work currently automated by software?
├─ Yes → Likely incremental improvement, incumbents will address
└─ No → Continue evaluation
   │
   Is this "core" or "context" for target customers?
   ├─ Core → They'll build in-house, risky market
   └─ Context → Continue evaluation
      │
      Can you build 80%+ margin above token costs?
      ├─ No → Thin wrapper, will face price compression
      └─ Yes → Strong candidate, assess timing
这项工作目前是否已由软件自动化？
├─ 是 → 很可能只是渐进式改进，incumbent 会着手解决
└─ 否 → 继续评估
   │
   这对目标客户而言是“Core”还是“Context”？
   ├─ Core → 它们会自行构建，市场风险较高
   └─ Context → 继续评估
      │
      你能否在 token 成本之上实现 80%+ 毛利率？
      ├─ 否 → 薄封装，将面临价格压缩
      └─ 是 → 候选较强，继续评估时机

How Should I Price This?

我应该如何定价？

What's the natural unit of work?
├─ Documents processed
├─ Queries answered
├─ Workflows completed
└─ [Define your consumption unit]
   │
   Set subscription floor at: Expected base usage
   Set variable rate at: Captures 80%+ margin above token cost
   Validate: Revenue grows with customer value, not headcount
这项工作的自然计量单位是什么？
├─ 已处理文档
├─ 已回答查询
├─ 已完成工作流
└─ [定义你的 consumption 单元]
   │
   将订阅费下限设为：预期基础使用量
   将可变费率设为：可在 token 成本之上实现 80%+ 毛利率
   验证：收入随客户价值增长，而不是随员工人数增长

Examples from Box's Journey

Box 发展历程中的案例

Cloud Transformation Parallels

云转型类比

Cloud Era (2005-2015)AI Era (2023-2027)
Had to convince people cloud was comingEveryone already believes AI is coming
Mobile + cloud created new IT architectureAI + agents create new work architecture
Freemium → enterprise pivot workedConsumption + subscription hybrid emerging
Competed by being cheaper/faster than incumbentsCompete by automating what incumbents can't/won't
云时代（2005-2015）AI 时代（2023-2027）
必须说服人们云即将到来所有人都已相信 AI 即将到来
移动 + 云创造了新的 IT 架构AI + Agent 创造了新的工作架构
Freemium → 企业转型取得成功Consumption + Subscription 混合模式正在兴起
靠在成本和速度上优于 incumbent 展开竞争通过自动化 incumbent 不能或不愿自动化的工作展开竞争

Key Lesson from Box

Box 的关键经验

Box pivoted from consumer to enterprise because:
  1. Consumer platforms would give away storage free
  2. Couldn't monetize against bundled offerings
  3. Enterprise had clear value prop: cheaper, faster, easier than incumbents
AI application: Don't compete where AI is commoditized. Find enterprise workflows where your AI solution creates clear, monetizable value above raw AI capabilities.
Box 从消费者市场转向企业市场，是因为：
  1. 消费者平台会免费赠送存储空间
  2. 无法通过与捆绑式产品竞争来实现变现
  3. 企业市场具有清晰的价值主张：比 incumbent 更便宜、更快速、更易用
AI 应用：不要在 AI 已经商品化的领域竞争。应寻找企业工作流，让你的 AI 方案在原始 AI 能力之上创造清晰且可变现的价值。