Harvey AI
领跑法律 AI 赛道的标杆平台;大客户快速落地,一级市场估值显著偏高
Harvey 仍是法律 AI 领域的领先者,但私募估值偏高、财务披露有限,且依赖外部前沿模型。
覆盖范围与披露说明
本尽调基于截至 2026-09-25 的公开及第三方证据。约 $350M ARR 为 2026 年 7 月的未经审计第三方估计;经审计收入、毛利率、客户留存队列、现金流、商业定价及最新时点的确切员工数均未公开。
封面要素
公司概况
Harvey AI 为律所、企业法务部门及相关专业服务机构打造企业级法律 AI 操作系统,整合 Assistant、Vault、Workflows 和 Knowledge 模块。公司已成为法律 AI 领域的领先者,并嵌入高价值交易与合规工作流;但仍依赖第三方基础模型,商业条款也未公开。
- 创始人
- Winston Weinberg, Gabriel Pereyra
- 总部
- San Francisco, California
- 产品
- 覆盖 Harvey Assistant、Harvey Vault、Harvey Workflows 与 Harvey Knowledge 的企业级法律 AI 平台。
- 客户
- 头部顶尖律所、企业法务部门及专业服务机构
- 商业模式
- 法律专业人士许可、企业协议、工作流产品及定制部署;定价与客群结构未公开。
- 阶段
- late-stage private
- 融资情况
- 公开证据显示累计融资超 $1.5B,包括一轮估值 $15.5B、规模达 $550M 的融资。
执行摘要
主要优势
- 在头部律所和企业法务团队中处于领先地位
- 企业采用迅速,客户覆盖面广、用户规模大
- 深入集成法律工作流,产品具备较强防御性
主要风险
- 相对估算 ARR,估值偏高,容错空间有限
- 高度依赖第三方基础模型和云基础设施
- 经审计财务数据、定价条款、客户队列留存、毛利率、现金流及现金跑道均未披露
未决问题
- 经审计的收入、毛利率、净现金消耗及 ARR 增速均未公开;当前财务数据仍部分来自第三方估算。
- 模型与基础设施供应商的完整商业条款尚未公开,包括价格、排他约定、模型访问权及合作伙伴的优先购买权(ROFRs)安排。
- 本轮融资的优先股堆叠结构、清算权利、棘轮条款与老股交易机制均未公开。
- 司法程序使用生成式 AI 的监管走向仍不确定;多个司法辖区还在完善律师执业规则和 AI 使用披露要求。
目录
01公司概况
1.1 实体架构与创始人分工
Harvey AI Corporation 于 2022 年在特拉华州注册成立为 C 类公司,总部位于加利福尼亚州旧金山。Harvey 2026 年 3 月的融资公告称 Winston Weinberg 为 CEO 兼联合创始人。独立公司档案将 Gabriel Pereyra 列为另一位联合创始人,但本章可采信证据未说明其现任高管职务,也未确认两位创始人此前的雇主。Harvey 面向法律及知识密集型专业服务开发垂直产品,覆盖合同分析、尽职调查、合规、诉讼和可复用工作流智能体。评估管理层时,应以已核实的创始人身份及 Weinberg 的 CEO 职务为依据;其他职务和履历仍待确认。[CO001, CO002, CO003, CO004]
1.2 股本沿革与战略治理
自 2022 年成立以来,Harvey 通过多轮机构风投累计筹资超过 $1B。公司资料显示 OpenAI Startup Fund 于 2022 年 11 月领投其 $5M 种子轮;Harvey 披露的资方还包括 Sequoia Capital、Kleiner Perkins、GV、Coatue 和 Andreessen Horowitz。2026 年 3 月 25 日,Harvey 宣布完成由 GIC 与 Sequoia Capital 联合领投的 $200M 融资,投后估值达 $11B。2026 年 6 月的一份独立分析将该笔 3 月估值与 2026 年 1 月约 $190M 的 ARR 估计进行对比,得出约 58x 的历史估值/ARR 比例。该静态对比揭示了当时的估值风险,但不可当作 Harvey 当前的入场估值倍数。[CO005, CO006, CO007, CO008, CO013]
| 投资机构 | 角色 / 类型 | 参投轮次 | 治理控制与经济权重 | 尽调重点核查方向 |
|---|---|---|---|---|
| Sequoia Capital | 领投机构风投 | A 轮(领投)、D 轮(领投)、成长期轮(2026 联合领投) | 显著的董事会话语权、按比例认购权及治理代表席位 | 核实董事会席位权限及清算优先权清偿顺位 |
| GIC | 主权财富联合领投机构 | 2026 年 3 月成长期轮(联合领投) | 雄厚的国际主权资本背书,助力全球化跨国扩张 | 确认老股赎回条款及主权基金特殊投资附约 |
| OpenAI Startup Fund | 早期种子轮资方及战略产业伙伴 | 2022 年 11 月种子轮(领投) | 优先享有前沿模型接口、早期技术共研及品牌声誉赋能 | 重点排查商业排他性条款及大模型依赖合作协议 |
| Kleiner Perkins | 核心风投机构 | B 轮(联合领投)、E 轮(联合领投)、后续跟投轮 | 大比例早期股权及成熟的企业级 GTM 治理经验导入 | 核实近期是否继续参与等比例注资扩股 |
| Andreessen Horowitz (a16z) | 头部风投机构 | F 轮(领投)、后续跟投轮 | 成长期治理赋能与成熟的企业级客户银团网络协同 | 审阅保护性附约及重大信息知情权范围 |
| Coatue | 对冲/PE 跨界科技投资机构 | D 轮、E 轮(联合领投)、后续跟投轮 | 后期估值分析与同业可比基准 | 严查估值防稀释对赌(ratchet)与 IPO 退出估值底线诉求 |
图谱列出主要领投方及战略机构投资人;完整的特拉华州股权结构表和天使股东名册仍未公开。
[CO006, CO007, CO008]1.3 业务规模与财务核验
Harvey 商业化拓展迅速,先后拿下多家全球顶尖律所的标杆合作,并以此为示范效应撬动更广泛的市场渗透。2026 年 3 月,Harvey 披露其服务覆盖 1,300 家机构的超 10 万名律师,囊括 AmLaw 100 律所的大多数。截至 2026 年 9 月 25 日,Harvey 官网展示其已进入 2,400 多家机构,服务超 20 万名专业人士,覆盖 75 家以上 AmLaw 100 律所。第三方基准估算其 ARR 在 2026 年 1 月约为 $190M,到 2026 年 7 月增至约 $350M。Harvey 仍为非上市公司,未披露经审计利润表、官方核定毛利率或公开客户留存队列。一份独立档案记录其 2026 年员工数为 860+ 人,另将 1,261 人标注为 2026 年一季度的峰值估计,两项数据均不能视作经核实的现时员工数。[CO009, CO010, CO011, CO012]
| 指标名称 | 数值 / 状态 | 统计日期 | 置信度 | 尽调缺口 / 批注 |
|---|---|---|---|---|
| 2026 年 3 月估值标杆 | $11.0B(GIC 与 Sequoia 成长轮) | 2026-03-25 | 高 | 本章节特有的历史融资估值;不可作为当前入场估值 |
| 累计融资金额 | 股权融资累计超过 $1.0B | 2026-03-25 | 高 | 未计入未经证实的债权或老股转让交易 |
| 年经常性收入(ARR) | 2026 年 1 月报道为 $190M;2026 年 7 月估计约为 $350M | 2026-07 | 中 | 来源于行业披露的未审计私有数据 |
| 活跃客户网络 | 2,400+ 家机构(覆盖 75+ 家 AmLaw 100 大所) | 2026-09-25 | 高 | 全所统一部署与局部业务组试点并存 |
| 团队员工规模 | 2026 年报道为 860+ 人;1,261 人标注为 2026 年一季度峰值估计 | 2026 / Q1 2026 | 中 | 审阅证据未提供经核实的最新时点员工人数 |
快照指标汇总了公司公告与经核验的第三方市场估算;经审计财务数据及客户留存记录仍未公开。
[CO006, CO007, CO010, CO011, CO012]1.4 平台里程碑与行业监管约束
Harvey 的产品与商业演进路径清晰:从特定领域的对话助手逐步迈向多步骤自主法律智能体。2026 年初,公司完成对演示自动化平台 Hexus 的战略收购,整合其技术团队在班加罗尔建立工程研发中心,并在企业客户中铺开超 25,000 个活跃定制智能体。与此同时,美欧司法机构与律师协会在 2023 至 2026 年间密集出台针对生成式人工智能的执业规范指引,严格要求律师具备专业胜任力、恪守保密红线,并对机器生成成果承担不可转授的人工监督职责。Harvey 能否在不损耗作业效率的前提下协助律师履行上述伦理与合规义务,决定了其商业模式的长期韧性。[CO005, CO006, CO014, CO015]
梳理自 2022 年成立至 2026 年底历次成长期融资及企业级工作流智能体扩张的编年大事记。
时间线与估值依据公司公告和经核实的创投追踪数据库;省略少数期间 SAFE 融资。
[CO005, CO006, CO007, CO014]1.5 支撑图表
02市场分析
2.1 市场边界、工作流与预算范畴
法律 AI 的目标市场是全球 $600B 法律服务行业中具有战略价值的一部分。以往,企业法务部门和大型律所将初级律师及律师助理约 30% 至 40% 的工时用于合同尽调、监管审查和常规起草等重复、文档密集的工作。Harvey 选择自动化这些高频交易工作流,而非消费级法律科技或通用对话工具,由此建立核心市场。企业级安全能力和 Microsoft Azure 基础设施帮助其赢得头部商事业务团队的关注。不过,商事交易与法庭诉讼的要求不同:交易工作允许反复修改,诉讼则要求在法院审查下逐项对应证据,并确保没有幻觉。因此,Harvey 的核心市场仍是企业合同、并购尽调和机构合规;专业争议解决工作流属于相邻市场,进入阻力较大。[CM001, CM002, CM006, CM012, CM013]
| 市场细分板块 | 纳入支出范畴 | 排除支出范畴 | 核心采购决策人 / 出资方 | 战略相关性 |
|---|---|---|---|---|
| 企业级商事交易法律 AI | 并购尽职调查、合同审查、红线比对、公司治理分析及法规信息披露 | 个人消费者模板生成、本人直接诉讼文件及批量小额索赔处理 | 律所权益合伙人、首席法律官、财富 500 强总法律顾问 | Harvey 的核心营收基本盘;在 AmLaw 100 律所筑起核心阵地 |
| 企业法务运营系统 | 供应商违约处置、商业采购协议初审、用工政策审查及合规审计 | 胜负攸关的复杂法庭诉讼对抗及白领犯罪刑事庭审辩护 | 法务运营副总裁、企业合规总监、法务副总监 | 旨在替代企业通用管理费用(G&A)支出的重要拓展方向 |
| 复杂诉讼情报系统 | 证据要素映射、答辩动议起草、证言询问提纲生成及案件时间线还原 | 常规商事文书起草与标准公司日常文档维护归档 | 诉讼业务部负责人、精品诉讼所合伙人、出庭律师 | 摩擦阻力较大的相邻业务;Litmas 等垂直单点方案对通用 AI 形成抗衡 |
| 中端律所实务管理 | 日常案件流转跟踪、基础检索摘要、标准客户通告及计费管理自动化 | 跨司法辖区的大型并购数据室核查与定制企业级大模型微调 | 执业律师 100 人以下律所的管理合伙人与 IT 采购委员会 | 很大程度上不在 Harvey 高客单的大所策略之内;Clio 与 Vincent AI 占据主导 |
市场边界定义反映了大型商事律所与财富 500 强法务部门在企业软件及人力替代维度的实际预算开支划分。
[CM001, CM002, CM009, CM012]2.2 TAM、SAM 与受限证据下的测算视角
测算 Harvey 的市场空间,需要区分法律行业总开支和企业实际可支配的软件预算。全球法律支出约 $600B,专业企业法律软件的可触达市场则较窄,约为 $15B。生成式 AI 与自主工作流智能体带动其中的可服务市场(SAM)快速扩大;顶级律所和 Global 2000 企业法务部门的规模估计为 $1.2B 至 $1.5B。Harvey 当前可获取的市场(SOM)集中在前 200 大律所及大型跨国企业法务团队。第三方证据估计,其 2026 年 1 月 ARR 约为 $190M;用 2026 年 3 月的 $11B 估值相除,得到约 57.9x 的历史比例。9 月更晚的 $15.5B 估值需要新的 ARR 分母,不能沿用 3 月的倍数。要支撑这条估值轨迹,Harvey 需要将初步席位部署转为全所操作标准,并抵御传统厂商捆绑方案的竞争。[CM001, CM003, CM004, CM005, CM006, CM014]
| 测算视角 | 数据源 / 发布方 | 数据年份 | 统计范围 / 地域 | 估算规模 | 年均复合增速 / 增长 | 测算方法与假设 | 局限性 / 提示 |
|---|---|---|---|---|---|---|---|
| 全球法律服务行业(TAM) | Capitaly | 2026 | 全球法律综合总支出 | $600B | 4-6% | 涵盖大型律所与企业法务部门的全部可计费营收与内部法务运营开支 | 宏观层面的人力与法律服务开支;不代表可获取的软件收入 |
| 企业级法律科技软件(TAM) | Value Add VC | 2026 | 全球企业级法律软件 | $15B | 15-20% | 自上而下梳理律所管理系统、DMS 文档系统与法务 IT 的企业软件总预算 | 包含传统非 AI 软件设施、服务器托管及通用底层文档存储系统开销 |
| 专用生成式法律 AI(SAM) | Value Add VC / 行业分析师 | 2026 | 全球前 500 大律所及全球 2000 强企业法务团队 | $1.2B-$1.5B | 35-45% | 测算目标商事律师体量,按每人每月 $300-$500 的席位费自下而上推算 | 极易受到从按人头卖席位向按自主智能体作业量计费模式转变的冲击 |
| 顶尖商事与大所法律 AI(SOM) | Silicon Valley Investclub / Value Add VC | 2026 | AmLaw 100 及全球前 50 大所的商事业务部门 | $350M-$450M | 40%+ | 基于 Harvey 在 2026 年 1 月 1,300+ 家企业客户中实现的 $190M ARR 测算实际份额 | 短期天花板受制于大所高级合伙人的保守心态及多家供应商横向赛马策略 |
各层级市场规模系综合 2026 年风投行业研报、企业对外披露数据及经审定的财务指标整理推演所得。
[CM001, CM003, CM004, CM005, CM014]从宏观整体法律服务大盘,层层穿透收敛至 Harvey 在高端企业级法律 AI 领域的可获取市场(SOM)。
各层宽度仅表示层级,不代表数值比例。
测算分层综合了风投资本的自上而下行业研报,并结合企业级客户实际部署及已披露的 2026 年 ARR 数据自下而上反向推导。
[CM001, CM003, CM004, CM006, CM014]2.3 客户细分、预算归属与落地路径
Harvey 的商业化落地直面全球大型律所合伙体系与企业总法律顾问(General Counsel)两大截然不同的采购决策阵营。在律所内部,商业采购决策权握在执行委员会、管理合伙人以及业务群组负责人手中,他们评估技术投资的核心标准在于可计费工时变现率与同业竞争压力;日常实际用户则是初级律师、顾问律师及法律工程师,其生产力的实际提升直接带动席位增购。在企业端,采购由首席法律官与法务运营总监把关,他们在外部律所控费指标的严控下管理固定的部门管理费用预算。不同梯队机构的落地路径分化明显:AmLaw 100 头部大所为保住标杆企业大客户,甘愿承担六位数的企业级大额采购;中型及精品所则面临切实的预算门槛,在 Harvey 的定价面前多有犹豫,往往更青睐 Clio 等一体化实务套件,后者无需承担高昂的企业工程适配成本,即可将检索智能原生嵌合进常规办案流程。[CM006, CM007, CM008, CM009, CM013]
| 客户群细分 | 核心采购决策人 | 主要终端用户 | 出资方 / 预算归属 | 核心承载工作流 | 预算审批权限层级 | 核心采购触发动因 |
|---|---|---|---|---|---|---|
| AmLaw 100 顶尖大所 | 管理合伙人 / 创新技术委员会 | 律师、法律顾问、法律工程师 | 律所权益合伙收益池 / 科技专项资本公积 | 并购合同尽职调查、交易架构梳理、法规合规申报审查 | 全所最高决策执行委员会 | 同行部署的外溢压力、客户要求降低计费时数的诉求、人均每月节省 25+ 工时 |
| 财富 500 强企业法务 | 总法律顾问(General Counsel)/ 首席法律官 | 企业内部法务律师、商事法务运营专家 | 企业法务运营预算 / 管理费用(G&A) | 批量合同初审、供应商合规排查、持续合规监控 | 部门职能年度运营预算池 | 压降外部律所高昂小时费率的刚性指标,加快业务签约交割速度 |
| 中型规模律所(50-200 名律师) | 管理合伙人 / IT 技术委员会 | 骨干律师、全职薪酬律师 | 律所日常综合运营账户 | 常规法律文书起草、基础法条检索、日常案件管理 | 合伙人开支共担池(对价格极度敏感) | 企业级大合同定价门槛过高;倾向选用打包了 Vincent AI 的 Clio 等高性价比工具 |
| 诉讼专业精品所 | 诉讼业务部领衔合伙人 | 出庭诉讼律师、诉讼业务助理 | 诉讼组专项备用金 / 承办案件诉讼垫付款 | 证据要素链条梳理、口供证词比对、撤案动议起草 | 业务组专项自有沉淀资金 | 法院对 AI 幻觉相关问题的审查;要求核实引文出处和案件论证逻辑 |
采购决策全景图揭示了商业法律科技市场中普遍存在的企业销售周期(6 至 12 个月)与层级报批结构。
[CM006, CM007, CM008, CM009, CM011]2.4 结构性增长动能、监管阻力与尽调缺口
法律 AI 行业的快速扩张受到强劲经济动能的驱使:企业买方迫切希望借助可规模化的 AI 工作流替代昂贵的初级律师计费工时,据估算每位专业人员每月可借此节省 25 小时以上。但行业落地的推进节奏依然受制于严苛的监管与技术阻力,首当其冲的便是执业责任风险:美、英、欧各律协相继颁布严格的专业胜任力与执业监管准则,明确执业律师须对算法生成的工作成果承担不可免除的法律责任。此外,全球法院记录在案的 AI 虚构判例(幻觉)纠纷已超 700 起,引发正式司法惩戒并招致更严苛的《联邦民事诉讼规则》第 11 条证据审查。在诉讼场景下,律师必须将案件事实与法定请求权基础严密锚定,通用大模型的概要归纳难免引入无法承受的执业疏忽风险。从尽调视角看,Harvey 的留存队列净收入留存率(NDR)与多年期合同续约表现仍属未知,需获取这部分核心数据方能确认其 $190M 的 ARR 究竟代表着稳固的企业级壁垒,抑或仅停留在试点期的尝鲜式扩张。[CM004, CM009, CM010, CM011, CM012, CM014]
2.5 支撑图表
03竞争格局
3.1 法律 AI 竞争生态与技术路径分化
步入 2026 年,法律人工智能赛道已快速分化为四大核心阵营:风投支持的垂直领域初创公司、根基深厚的传统法律数据库巨头、聚焦细分工作流的单点工具方案,以及底层通用大模型厂商。Harvey AI 早期以 AmLaw 100 律所及大型企业法务部门为突破口,成功夺得企业级桥头堡,但如今在各细分战线均面临激烈的正面抗衡。在垂直挑战者阵营中,发源于瑞典的 Legora 展现出强劲的独角兽扩张势头,以数十亿美元估值连续融资,并迅速从欧洲腹地杀入北美大本营。与此同时,传统法律信息巨头 Thomson Reuters 和 LexisNexis 亦步步为营,分别依托 CoCounsel 和 Lexis+ AI 将生成式能力直接植入核心工作流,稳固其软件基本盘。律所及企业买方并未盲目全盘替代现有系统,而是身处一个混合采购生态中,单点工具、自研工程模块与巨头现成的订阅全家桶彼此争夺有限的 IT 预算。[CP001, CP002, CP003, CP005]
| 竞品机构 | 阵营分类 | 估值规模 / 融资背景 | 目标客群 | 核心差异化壁垒 | 主要短板与制约 |
|---|---|---|---|---|---|
| Legora | 法律 AI 直接竞争者 | B 轮融资 / 估值约 $1.8B-$5.5B | 欧洲及全球大型企业级律所 | 跨律所协同办公云空间,对欧洲各司法管辖区具备深度原生支持 | 在美国 AmLaw 100 顶级大所中的渗透规模目前仍落后于 Harvey |
| Thomson Reuters (CoCounsel) | 传统法律平台 | 上市公司(市值超 $60B) | 广泛的综合法律市场、律所及法务部 | 与 Westlaw、Practical Law 及判例数据库原生集成 | 架构较传统;推出前沿生成式工作流的速度可能较慢 |
| LexisNexis (Lexis+ AI) | 传统法律平台 | 上市母公司(RELX 市值超 $70B) | 企业级律所、公司法务团队及司法机构 | 权威的 Shepards 引证体系及自有法律专著 | 偏向捆绑销售策略,在主检索之外的交互界面存在操作摩擦 |
| Robin AI | 聚焦合同的法律 AI | 风投资本支持(累计筹资超 $40M) | 企业法务部门及资产管理机构 | 专注合同文本自动审查、红线批注,依托 Anthropic 打造专属法律助手 | 业务跨度相对局限,未构建涵盖全面诉讼与复杂交易的全能型平台 |
| Hebbia | 跨行业专业文档分析 AI | B 轮(融资 $130M,投后估值 $700M) | 金融、资产管理机构及企业并购尽调 | 矩阵式多文档穿透检索,针对高密度金融及监管文件展开长文本推理 | 未针对法律场景微调;缺少权威法律引证能力和律师协会合规工具 |
竞品规模、融资轮次和定位依据 2026 年公开市场资料及经核实的融资披露;非上市公司估值反映近期报道的融资轮次。
[CP001, CP002, CP003, CP005]3.2 竞品画像与核心功能矩阵
深入评估各家产品架构,可见其在功能深度、领域适配与平台延展性上分歧显著。Harvey AI 的核心壁垒在于 Assistant、Vault、Knowledge 与自动化 Workflows 构成的整合套件,辅以深度集成的 Microsoft Word 与 Outlook 原生插件。其关键技术护城河在于:模型生成直接严密对齐律所私有知识库及权威法律法条,同时具备 SOC2 Type II 与 ISO 27001 等企业级数据隔离安全认证。相比之下,Thomson Reuters CoCounsel 等老牌选手倚仗 Westlaw KeyCite 等独家先例判托体系与法学专著库构建排他壁垒;Robin AI 等合同工具则专注于合同红线比对与谈判指引的精细化单点突破。通用大模型虽能提供常规起草辅助,但在法律管辖区精确度、企业治理合规栅栏及满足诉讼尽调归档要求的文档审查保险库等核心指标上,仍无法支撑严肃的生产级诉讼与交易尽调。[CP004, CP006, CP007, CP008, CP014]
| 核心能力 / 工作流 | Harvey AI | Legora | Thomson Reuters CoCounsel | Lexis+ AI | 通用前沿大模型 |
|---|---|---|---|---|---|
| 法律检索与权威法条引证 | 对齐一手权威法律数据库与专著 | 原生支持美欧管辖区的精准法条检索 | 依托 Westlaw 专有 KeyCite 深度融合判例法 | 整合 Shepards 引证体系与一手成文法库 | 缺少法律垂直 RAG 支持,容易生成虚假引用 |
| 嵌入 Word 的商事协议起草 | 原生集成 Word 和 Outlook,搭载定制模板 | 深度集成 Word,支持多人协作起草文档 | 借助插件直接调用 Practical Law 标准条款库 | 依托 Lexis 专有条款资源库协助草拟文书 | 生成未排版的纯文本,需人工手动拷贝至 Word |
| 复杂尽职调查与保险库深度比对 | 搭载专用 Vault 工具,支持海量文档审查与并购尽调 | 支持数据室多方协同审查与跨卷宗穿透问答 | 依托 CoCounsel 开展批量文档复核与条款抽提 | 依托 Lexis 引擎进行合同条款深度挖掘与解析 | 受上下文窗口刚性制约;缺少结构化尽调保险库 |
| 定制化多步骤工作流智能体 | 依托 Harvey Workflows 和 Harvey II 部署超 25,000 个智能体 | 新兴的工作流自动化模块 | 预设标准化法律技能链条(涵盖质询、审查) | 在 Lexis 平台内置预设好的标准任务流程 | 需额外调用第三方流程编排工具或手工撰写复杂提示词 |
| 数据安全合规与律协执业标准 | 取得 SOC 2 Type II、ISO 27001 认证,符合 CCPA、GDPR 要求;数据零留存 | 遵循 GDPR 与欧洲数据主权要求 | 符合企业法务部门的安全与审计要求 | 符合法律出版商的企业安全标准 | 偏向大众消费或通用 SaaS;各厂商数据保留协议口径不一 |
各项功能的具备情况与集成深度均对照厂商技术文档、律师协会测评报告及企业级版本更新说明核验确认;未经证实的功能均不予录入。
[CP004, CP006, CP007, CP008]围绕法律领域专有化深度与企业工作流整合度两大核心维度,评估法律科技各派系选手的战略落位。
| 条目 | X (2.5–8.8) | Y (3.5–8.8) | 说明 |
|---|---|---|---|
| 1. Harvey AI | 8.5 | 8.8 | |
| 2. Legora | 7.2 | 7 | |
| 3. Thomson Reuters CoCounsel | 8.8 | 6.2 | |
| 4. Lexis+ AI | 8.5 | 5.8 | |
| 5. Robin AI | 6.5 | 5.2 | |
| 6. Hebbia | 4.5 | 7.5 | |
| 7. 通用前沿大模型 (OpenAI/Anthropic) | 2.5 | 3.5 |
各点坐标基于尽调获取的实际证据,对法律专有适配度(横轴 0-10)与企业级工作流深度集成度(纵轴 0-10)进行相对排序研判。
[CP004, CP006, CP014]3.3 包装形态、定价隐秘性与企业切换成本
法律 AI 行业的企业商务安排通常定价信息高度不透明、销售流程阻力较大,合同捆绑也较复杂。Harvey AI 采用销售主导的年度企业许可模式,未公开分层价格,客户筛选流程较长,且往往设有较高的年度最低承诺,顶尖律所的最低金额常达六位数。传统软件巨头则反向布局,将生成式 AI 作为现有多年期企业协议的增购附加模块,Thomson Reuters 与 LexisNexis 借此充分激活既有供应商主服务协议(MSA)的存量红利。尽管面临老牌巨头的围追堵截,Harvey 在顶级律所中形成了较高的切换成本,这与其深度融入组织流程有关。数以万计的定制工作流智能体适配不同业务组;替换 Harvey 会涉及律师再培训、模板重构和行政流程中断。[CP008, CP009, CP010, CP011]
| 供应商 | 商业计价体系 | 估算单体经济模型指标 | 包含的产品模块边界 | 企业级报价透明度 |
|---|---|---|---|---|
| Harvey AI | 按席位分层的企业级定制年度大合同 | 未公开高客单溢价(传闻大所起订门槛达六位数) | 包含 Assistant、Vault、Knowledge、Workflows 及 Office 插件 | 低(企业销售报价;演示需排队) |
| Legora | 按席位和套餐层级收取企业订阅费 | 估算企业 SaaS 定价(每席每年 $1,000-$3,000) | 核心起草、文档复核与跨组织协同工作台 | 低(企业销售主导报价) |
| Thomson Reuters (CoCounsel) | 与 Westlaw/Practical Law 捆绑的增购授权 | 估算每席每月需额外支付 $100-$300 | CoCounsel 生成式助手及底层法律数据库直连 | 中(通常纳入既有企业服务协议) |
| 前沿大模型厂商(OpenAI/Anthropic) | 自助 SaaS 席位费及按用量计费的 API Token | 每席每月 $20-$200,或按 API Token 用量付费 | 纯底层基础模型调用、常规对话界面及开发者接口 | 高(面向全社会公开透明的官方刊例价目表) |
定价信息综合独立 VC 分析、律师协会资料及公开价目表;企业合同通常还会协商批量折扣和定制承诺。
[CP009, CP010, CP011]3.4 护城河韧性、模型商品化与颠覆风险
Harvey AI 的长期防御力,取决于应用层工作流编排、企业信任和工作流黏性能否跑赢基础模型厂商的垂直化。Harvey 没有自有基础模型,而是使用 OpenAI、Anthropic 和 Google 的商业 API 并进行微调,因此仍会受到上游模型进步的影响。前沿模型厂商推出原生智能体工具、更长的推理上下文和专业插件,也会持续压缩通用基础模型与垂直法律应用之间的技术差异。企业客户越来越多地要求多供应商部署,或探索自有 AI 工具;Harvey 需要持续证明工作流集成、合规水平和垂直场景 ROI,才能在模型商品化和传统厂商整合的压力下支撑企业软件溢价。[CP001, CP003, CP012, CP013]
3.5 支撑图表
04财务情况
4.1 收入来源与商业变现架构
尽调审阅证据梳理出四种商业化变现机制:面向法律专业人士的席位许可、面向全所及法务部统一部署的企业级协议、针对特定业务领域的专项工作流产品,以及跨分所、跨业务线和跨地域的存量增购。2024 年 5 月的创始人访谈还提及定制模型、商业捆绑包、Vault 独立收费,以及与 PwC 联合销售的独立产品。证据未披露各项机制在 ARR 中的具体占比、合同平均账期、实际落地客单价或细分客户群收入。因此,上述机制勾勒出了 Harvey 平台实现货币化的路径形态,但无法推导出品种明确的收入结构或具体的金额拆分。[CI005, CI006, CI007, CI012]
| 商业收入流 | 具体交付机制 | 观察到的商业计费基础 | 公开披露的营收贡献度 | 收入可持续质量 | 尽调重点核验方向 |
|---|---|---|---|---|---|
| 核心法律助手席位授权 | 面向法律专业人士发放的个人许可 | 第三方资料描述为按席位收费 | 占整体 ARR 的具体比例未公开 | 显示为经常性收入模式;合同期限与续约条款未披露 | 审查席位采购总量、单座实际净价、服务年限及分批留存率 |
| 工作流与 Vault 模块增购 | 针对具体业务领域的专项工作流及文档产品 | 2024 年 5 月提及的商业捆绑包及 Vault | 占整体 ARR 的比例未公开 | 潜在增购扩张机制;搭售率与续约数据未披露 | 确认各模块搭售渗透率及客户单体增量计费金额 |
| 企业定制模型与战略联盟 | 联合开拓市场(GTM)及企业级模型深度微调 | 定制开发项目及与 PwC 联合推出的独立产品 | 占 ARR 的贡献比例及收入分成条款均未披露 | 已有商业合作模式;服务收入占比与交付成本未披露 | 审查与 PwC 战略合作协议中的收入分成比例与知识产权归属条款 |
| 存量客户网络横向增购 | 扩展至更多办公室、业务领域和地区 | 第三方行业监测指认的客户拓展机制 | 占整体 ARR 的比例未公开 | 具备经常性扩容潜力;各年份留存队列的具体贡献未公开 | 在客户队列中拆解新获客增长与既有客户存量增购 ARR |
审阅证据支持上述商业化变现机制,但并未披露各项收入占比、具体业务板块金额、合同账期或实际成交均价。
[CI005, CI006, CI007]4.2 定价梯队与企业打包销售策略
Harvey 最初为大型律所提供高度定制的部署。2024 年 5 月,创始人表示计划在第三季度结束早期试用(early access),面向不同规模律所推出价格更低的版本,并提供判例法产品、Vault 和折扣套餐。同一访谈称,复杂定制项目的开发成本可能超过 $5M;这是开发成本示例,不是已披露的客户价格或平均合同价值。审阅证据无法核实 2026 年的标价、单席位费率、合同条款、席位数、折扣或实际 ACV。因此,定价尽调应区分过去的商业化计划和当前实际执行的商务条款。[CI005, CI006, CI013]
| 客户细分客群 | 历史商业化规划证据 | 2026 年已证实定价 | 核心未决信息盲区 | 证据支撑基础 |
|---|---|---|---|---|
| 大型顶级律所 | 2024 年 5 月访谈提及高度定制的系统 | 未披露 | 席位单价、合同履行期限、部署服务范围及平均年合同价值(ACV) | LawNext 创始人访谈;提及复杂定制项目开发成本或超 $5M,此属于研发成本示例,非对客销售定价 |
| 不同规模律所 | 原计划于 2024 年三季度上线更平价的版本 | 未披露 | 该方案是否已落地、现行产品包具体权益及最终成交客单价 | LawNext 创始人访谈中关于未来商业化规划的自述 |
| 商业化标准捆绑包 | 计划分步推出判例法产品、Vault 以及打包组合方案 | 未披露 | 捆绑售价、折扣让利幅度、账号使用上限及合同履约期限 | LawNext 创始人访谈;提及预设打包折扣但未披露具体折让金额 |
| 与 PwC 联合产品 | 联合对外销售独立的税务、法律及人力资源专用套件 | 未披露 | 收入分成比例、知识产权归属、定制开发服务占比及单客合同金额 | LawNext 创始人访谈中对联合 GTM 战略合作模式的介绍 |
本表完整还原了 2024 年 5 月的早期商业化规划。该规划不能直接视为 2026 年经核实的实际成交定价,亦不能据此推断单席位月费、年度承诺门槛、席位规模或 ACV 水准。
[CI005, CI006, CI013]4.3 成本结构与毛利率披露缺口
审阅来源均未披露 Harvey 的毛利率、营业成本、模型推理开支、云服务支出或运营费用构成。独立分析将第三方前沿模型依赖视为竞争壁垒和定价的定性风险;Harvey 的产品架构则采用定制模型和专项任务模型。外部依赖可能带来可变推理成本风险,但现有证据无法量化其对毛利率的影响,也不能证明模型费用是 Harvey 最大的成本项。因此,投资判断应保留这一披露缺口:先取得经审计的营业成本(COGS)明细和供应商合同,再对公司的具体毛利率作出断言。[CI008, CI010, CI012]
按时间梳理 ARR 估计和最新宣布的融资,并区分公司披露与第三方估算。
ARR 节点均为 ClinkBill 的第三方估计;9 月融资是公司另行宣布的融资事项。这些信息不代表经审计收入、毛利率、净烧钱或现金跑道数据。
[CI001, CI003, CI004]4.4 增长轨迹与财务基准对标
第三方追踪数据显示,Harvey 的 ARR 走势大体为:2023 年估算达 $5M,2024 年底达 $50M,2025 年 8 月攀升至 $100M,2025 年底达 $195M,至 2026 年 7 月达到约 $350M。从 2025 年底至 2026 年 7 月的七个月内,变动幅度约为 79.5%(按 ($350M / $195M - 1) × 100 计算)。这既非同比增速,也不是完整上半年的对比口径。Harvey 官方披露其覆盖 2,400+ 家机构、200,000+ 名专业人士,并被至少 80% 的 AmLaw 100 律所采用。鉴于缺乏经审计的 GAAP 营业收入、同期可比 ARR 基数、客户流失率及队列 NDR,目前仅凭零散里程碑数据无法精确还原真实的营收同比增速。[CI002, CI003, CI004, CI014]
| 单体经济指标 | 预估数值 / 公开披露状态 | 置信度等级 | 核心分析意义 | 尽调重点核查方向 |
|---|---|---|---|---|
| 毛利率 | 未披露 | 无法评估 | 第三方基础模型依赖构成定性成本风险,非实测毛利率 | 调取经审计的销售成本明细表,剥离第三方模型 API Token 开销与底层云服务算力费 |
| 净收入留存率(NDR) | 未披露 | 无法评估 | 公开客户数激增并不能直接坐实留存队列的净增购与粘性 | 调取按签约年份划分的历年客户 ARR 留存与增购明细表 |
| 获客成本(CAC)回本周期 | 未披露 | 无法评估 | 客户获取成本和毛利队列数据目前未公开 | 审查按客群划分的销售与市场人员薪酬开支及单客平均获客成本 |
| 平均合同价值(ACV) | 未披露 | 无法评估 | 客户体量与定制开发成本均无法推导出实际实现的平均 ACV | 获取客户 ARR 分布图谱及前 20 大核心客户的收入集中度明细 |
| 现金头寸、净烧钱速度与跑道 | 未披露 | 无法评估 | $550M 的新一轮融资无法体现交割后的现时流动性、前期烧钱沉淀或资金跑道 | 穿透资产负债表、经审计现金流量表、月度净烧钱速度、存量负债及受限资金 |
审阅证据无法支撑任何关于具体毛利率、NDR、CAC 回本周期、ACV、账面现金、烧钱速度或跑道时限的确定数据。
[CI008, CI009, CI011, CI014]4.5 资本充裕度、估值倍数与尽调阻碍
2026 年 9 月,Harvey 宣布以 $15.5B 估值融资 $550M,由 Diffusion 与 Lightspeed 联合领投。融资补充了资本,但没有披露交割后的现金余额、此前支出、净现金消耗、债务或资金可维持的时间;累计融资不能视为当前流动资金。用该股权估值除以 2026 年 7 月第三方约 $350M 的 ARR 估计,得到 44.3x;若采用 2026 年 1 月较早的约 $190M 估计,则为 81.6x。债务、现金和确认收入均未知,因此这些带有时点限制的估值/ARR 比率,不是企业价值倍数,也不是历史 GAAP 收入倍数。经审计财务、留存队列、模型供应商定价和现金流证据,仍是投资评估的必要材料。[CI001, CI009, CI010, CI011, CI014]
4.6 支撑图表
05产品与技术
5.1 产品架构与核心工作流模块
Harvey AI 是面向律所、企业法务部门和专业服务机构的企业级 AI 平台。它不只是为通用基础模型编写提示词,而是提供由 Harvey Assistant、Harvey Vault、Harvey Workflows 和 Harvey Knowledge 组成的专用工作空间。Assistant 用于法律检索、初稿起草和分析摘要,并通过 Microsoft Word 与 Outlook 插件嵌入律师的常用工具。Vault 支持多文档导入和结构化提取,每个库最多容纳 100,000 份文件,并提供 OCR 和核心条款标注。Workflows 通过 Playbooks 和线程式 Review Tables 将可重复的律师任务标准化;平台面向法律机构的复杂文档分析和交易尽调需求。[CE001, CE002, CE003, CE004, CE014]
| 模块 / 资产名称 | 目标用户群 | 成熟度 / 当前状态 | 核心差异化壁垒 | 关键尽调核查缺口 |
|---|---|---|---|---|
| Harvey Assistant | 骨干律师及企业法务内勤 | 商业化量产(GA) | 自然语言对话界面,结合律所模板和先例文件生成内容 | 针对数百页重度法律披露文件时的 Token 吞吐承载力与响应延迟基准 |
| Harvey Vault | 诉讼业务组与企业并购尽调团队 | 商业化量产(GA) | 大容量文档库(100k 份文件),支持自动 OCR 和结构化元数据提取 | 文档库达到 100k 份文件上限时的处理行为与索引延迟 |
| Harvey Workflows 与按行审查表(Review Tables) | 商事交易组与合规风控业务团队 | 商业化量产(GA) | 无代码智能体构建工作台,搭载 Playbooks 模板并支持律师逐行审查校验 | 在开放式、跨司法管辖区的复杂疑难案卷中的任务交付成功率 |
| Word 与 Outlook 原生插件 | 习惯在桌面 Office 环境起草文本的执业律师 | 商业化量产(GA) | 直接在微软 Office 套件内部对照律所先例库开展上下文原位起草 | 离线办公功能完备度,以及与第三方 DMS 同步时可能出现的锁定冲突 |
各模块功能、用户定位与成熟度依据截至 2026 年 9 月的 Harvey Academy 培训课纲、产品评测及官方文档核验。
[CE001, CE002, CE003, CE004]5.2 法律执事实践与高频场景自动化
客户使用 Harvey 改造四类核心法律工作:复杂并购尽调、一手法律资料检索、交易文书起草及持续的监管合规监测。交易团队用 Harvey Review Tables 从虚拟数据室导入数百份合同,提取控制权变更触发条件、赔偿条款和竞业限制约定,整理成结构化对照矩阵,耗时由数天缩短至数小时。检索时,Harvey 将日常英语表述的事实问题转为结构化引文,覆盖 LexisNexis、EUR-Lex、SEC EDGAR 申报资料等 500 多个权威法律数据库。平台必须保留严格的人工监督:职业责任规则和律师协会指引要求,AI 生成的引文与备忘录草稿不能替代持证律师的判断。平台仍仅向签订定制企业合同的客户开放,一手法律数据库的同步间隔未披露,每个文档库也有文档数量硬上限。[CE002, CE003, CE005, CE008, CE013]
| 律师核心作业场景 | 传统作业基线模式 | Harvey AI 解决方案 | 可量化的实际增益 | 当前技术制约与边界 |
|---|---|---|---|---|
| 合同尽职调查 | 初级律师在虚拟数据室(VDR)中手工翻阅排查数百份合同文本 | 依托 Vault 进行批量抽取、结构化字段横向比对,自动锁定控制权变更条款 | 据估计,可将大批量合同的初次审查周期缩短 60–80% | 必须由持证执业律师进行人工二次校验,防止遗漏特定的管辖区除外条款 |
| 基础法条与先例检索 | 在各法律检索终端上由人工反复构思布尔逻辑检索式 | 自然语言分解案情问询,穿透 500+ 数据库获取权威法条及判例引证 | 减少布尔检索语法操作,快速找到相关判例和制定法 | 新发布的判决意见和法院命令何时纳入索引,尚未公开 |
| 商事法律文书起草 | 检索调取律所既往旧案先例,人工逐条修改标准合同条款 | Word 插件依据律所 Playbooks 和 Vault 文件生成初稿 | 几分钟内完成初稿,缓解起草瓶颈并保留律所风格 | AI 无法自主评估客户在商业谈判中的风险承受底线,亦无法代为权衡条款让步 |
| 监管合规动态监控 | 人工定期跟踪监管登记信息和行政案件卷宗 | 定制工作流智能体持续盯盘监管更新,定期自动汇总生成格式化摘要 | 枯燥繁琐的动态跟踪实现自动化,法规异动第一时间推送合规负责人 | 自动预警仍需法务人员判断哪些执法变化具有实质影响 |
工作流对比与提效指标系综合截至 2026 年 9 月的律师实操手册、企业实测反馈及基准案卷测试综合提炼而成。
[CE002, CE003, CE005, CE008, CE013]5.3 技术架构、外部集成与模型依赖度
Harvey 的技术架构采用多层操作系统设计,解耦了前端交互界面与底层生成式模型供应商。在基础推理层,Harvey 调度包括 OpenAI、Anthropic Claude、Mistral 及 xAI 在内的多款前沿商业模型,依据推理深度与任务复杂度动态分发计算请求。平台的专有核心资产凝聚在中间层的检索与事实锚定流水线:将 iManage、SharePoint、Google Drive 等企业文档管理系统(DMS)中的律所私有模板,与权威法规法条及判例先例库进行精准挂载校准。为评测智能体在复杂长链条法律场景中的推理能力,Harvey 于 2026 年 5 月开源了法律智能体基准(Legal Agent Benchmark,简称 LAB),设立覆盖 24 个法律领域的 1,250 个真实多步骤客户办案场景。由于平台依赖外部商业基础模型而非自研模型权重,其技术壁垒主要依托于检索精度、工作流连接器以及领域专有评测体系。[CE005, CE006, CE007, CE008, CE009, CE010]
| 架构分层 / 核心组件 | 承担功能角色 | 核心底层技术依赖 | 主要技术与运维潜在风险 |
|---|---|---|---|
| 多模型推理调度层 | 依据用户任务复杂度与法律业务领域,智能路由至最优大模型 | 外部商业模型 API 的连通性与可用性(OpenAI、Anthropic Claude、Mistral、xAI) | 上游模型厂商 API 提价、并发速率限流、网络延迟陡增或旧模型被强制弃用 |
| 法律专有知识锚定层 | 将用户提问精准挂载至一手法律权威先例库的检索与引据引擎 | 获正式商用授权的外部数据接口(LexisNexis、EUR-Lex、EDGAR、Gyldendal Rettsdata) | 数据访问授权续签、专有数据孤岛或数据源中断 |
| 企业级文档系统连接器 | 双向同步客户事务文件和律所模板 | 主流企业 DMS 接口及访问控制权限(iManage、SharePoint、Google Drive) | 企业系统权限同步脱节、企业 VPN 网络超时以及海量超大文件传输中断挂起 |
| 租户隔离与安全合规边界 | 多租户数据隔离、区域化处理及确保数据不用于模型训练 | 云基础设施合规控制(SOC 2、ISO 27001、AWS/Azure 区域) | 跨租户元数据意外泄露或云服务商区域合规要求变化 |
运行架构组件、依赖与风险依据截至 2026 年 9 月的企业安全披露、IT 目录评测及投资分析整理。
[CE005, CE006, CE009, CE010, CE011]六层架构展示 Harvey AI 的客户端交互界面、工作流编排、文档导入、法律知识依据、多模型推理和企业治理。
- 第一层:客户端应用与办公插件交互层
- Microsoft Word 和 Outlook 插件、浏览器端 Web 应用及 iOS/Android 移动应用,供律师在工作流中使用。
- 第二层:工作流编排与智能体执行中枢
- 无代码工作流智能体搭建平台、自然语言自动化流程引擎、按行审查表(Review Tables)以及律所标准化 Playbook 流程模板。
- 第三层:非结构化文档处理与保险库存储
- 安全文档库(每库最多支持 100,000 份文件),提供自动 OCR、批量核心条款提取和多文档比对。
- 第四层:企业专有知识库与 DMS 接口网关
- 底层打通主流企业级文档管理系统(iManage、SharePoint、Google Drive)并接入 500 多个权威法律知识库(LexisNexis、EUR-Lex、EDGAR)。
- 第五层:多模型动态路由与专有评测引擎
- 在 OpenAI、Anthropic Claude、Mistral 和 xAI 等商业基础模型间动态路由请求,并以开源法律智能体基准(LAB)评测表现。
- 第六层:企业级安全防护与合规治理边界
- 取得 SOC 2 Type II 与 ISO 27001 认证的安全防护网,提供美、欧、澳区域本地数据驻留部署,集成 SAML SSO,并签署客户数据零训练保密协议。
系统架构层级参照截至 2026 年 9 月已公开发布的产品功能、系统技术接入文档及第三方 IT 架构评测进行还原。
[CE001, CE004, CE005, CE006, CE007, CE009, CE010, CE011]5.4 差异化优势、企业级安全与合规信任体系
Harvey 的差异化来自三方面:与法律文档管理系统的集成、法律知识索引,以及面向机密客户事务设计的企业合规与安全措施。Harvey 持有 SOC 2 Type II 和 ISO 27001 认证,并遵循 GDPR 与 CCPA 要求;同时提供美国、欧洲和澳大利亚的数据驻留选项。Harvey 还在合同中承诺,默认不将客户事务文件、提示词和提取的元数据用于公共基础模型的训练或微调。相比价格更低、通用推理能力更广的 AI 助手,Harvey 的 DMS 集成、合同中的零训练承诺和可引用法律数据库集成,在 AmLaw 100 律所与财富 500 强企业法务部门形成显著切换成本和防御力。[CE010, CE011, CE012, CE013]
5.5 支撑图表
06客户情况
6.1 客户群细分与采购画像
Harvey 围绕四大核心客群组织其商业化拓展与产品矩阵:AmLaw 100 头部及全球顶级大所、中型及区域精品所、企业内部法务部门,以及综合性专业服务机构。在律所内部,核心采购出资人通常是律所管理合伙人、执行委员会或创新领导小组,而日常高频用户由交易及诉讼业务组的资深律师、普通律师及助理构成。相比之下,企业法务端的落地由总法律顾问(General Counsel)及法务运营总监牵头,主旨在于削减外部律所咨询费、批量预审供应商合同并提升合规流转效率。在专业服务机构中,全球会计师事务所与并购顾问网络应用最为广泛,深度嵌入跨境税务分析、交易架构设计及审计底稿复核环节。其商业模式采用与创收执业人数绑定的席位费,叠加大客户访问特定模块(如 Vault 和定制工作流智能体)的企业级年费框架。[CU001, CU003, CU007]
| 客户细分客群 | 核心决策人与终端用户 | 承载的核心工作流 | 预估市场覆盖规模 | 商业模式与战略价值 | 核心尽调信息缺口 |
|---|---|---|---|---|---|
| AmLaw 100 头部及全球顶级律所 | 管理合伙人、管委会;业务组骨干律师及合伙人 | 并购尽职调查、疑难诉讼法律检索、条款起草、Word 文档红线比对 | 75+ 家 AmLaw 100 律所;100k+ 名用户 | 全所企业许可;潜在 ACV 较高,并有标杆客户价值 | 签约大所内部的实际席位渗透率,究竟是全员普及还是仅局限于个别试点小组 |
| 中型规模及区域精品律所 | 管理合伙人、业务负责人及综合业务律师 | 商事合同拟定、电子证据开示审查、常规监管合规比对 | 遍布 60 多个国家的数百家中型律所 | 按席位收取的年度订阅制;持续扩容的规模化基本盘 | 客户对单人席位费用的价格敏感度,以及面临竞品低价分流的流失风险 |
| 企业内部法务部门 | 总法律顾问(General Counsel)、法务运营总监;商事法务专员 | 保密协议(NDA)预审、供应商采购合同把关、合规监控、制度梳理 | 500+ 家大型企业法务部门 | 部门年度承诺;以减少非计费成本体现 ROI | 真实净收入留存率(NDR)表现,以及多年期大额续约的持久性 |
| 专业服务机构与综合咨询网络 | 税务与交易咨询主管合伙人;商事尽调咨询师 | 跨境税务筹划、跨国并购交易架构设计、财务审计底稿复核 | 全球主流会计师事务所与交易咨询网络(如 PwC) | 高价值企业合作;处理大量文档 | 合作协议的排他性深度,以及繁重的专属定制工程交付成本 |
客群细分梳理自 Harvey 官方披露、AmLaw 行业专刊及代表性客户案例;涉及具体客户的商业合同金额均属于保密范畴。
[CU001, CU003, CU007]6.2 落地走势与规模化特征
公开数据展现出强劲的采用态势,但各维度的披露时点与统计口径并不完全一致。第三方公司档案显示其服务 100,000+ 名律师,而 Harvey 官方用例文章中则记录了超 142,000 名律师;审阅证据均未载明具体观测日期,难以直接拼装成口径一致的队列时间序列。Harvey 官网在 2026 年 9 月 25 日的访问快照中,展示了 200,000+ 名专业人士、2,400+ 家法律机构及 75 家以上 AmLaw 100 顶尖大所。“律师”与“专业人士”统计范畴或有重合与差异,且官网未公布界定活跃度的使用频次门槛。Harvey 另披露了 92% 的月度使用率,但其活跃分子与分母口径并未公开。此外,Clinkbill 援引 The Information 与 Sacra 的信源估算其 ARR 在 2026 年 7 月达到约 $350M;此数据为第三方推测,并非经审计收入或监管申报文件。[CU002, CU003, CU004, CU013, CU015, CU016]
| 客户指标名称 | 披露指标数值 | 统计截止时间 | 一手信源出处 | 置信度等级 | 核心战略考量 | 缺失的对照分母 |
|---|---|---|---|---|---|---|
| 客户机构总数 | 2,400+ 家法律机构 | 2026 年 9 月 25 日访问快照 | Harvey 官方网站主页 | 高 | 显示其在律所和企业法务市场的品类领先地位 | 全面正式采购的大客户与局部尝试性试点的具体比例 |
| 早期第三方监测的律师人数 | 100,000+ 名律师 | 审阅摘录未载明具体观测时点 | Value Add VC 研报 | 中 | 显示较早时点已有较大采用规模 | 确切统计时间、界定活跃用户的门槛,以及与后续统计指标的可比性 |
| 官方用例文章披露的律师数 | 142,000+ 名律师 | 审阅摘录未载明具体观测时点 | Harvey 官方发布的客户用例报告 | 中 | 官方宣称在 1,500 多家机构中实现的渗透规模 | 确切统计时间,以及与官网后期更宽泛的“专业人士”口径的差异 |
| 现行官网展示的专业人士总数 | 200,000+ 名专业人士 | 2026 年 9 月 25 日访问快照 | Harvey 官方网站主页 | 高 | 公司官方对外宣传的当前用户量级 | “专业人士”的统计内涵及界定为活跃用户的考核标准 |
| AmLaw 100 大所渗透率 | 覆盖 75 家以上 AmLaw 100 律所(占比超 75%) | 2026 年 9 月 25 日访问快照 | Harvey 官方网站主页 | 高 | 公开披露的渗透率较高;排他性和持续定价能力仍待核实 | 在签约 AmLaw 100 顶尖律所内部的平均实际采购席位深度 |
| 公司披露的月度采用率 | 公司披露月度采用率为 92% | 审阅摘录未载明具体观测时点 | Harvey 平台使用数据洞察 | 中 | 经常性使用有助于降低传统企业软件闲置(shelfware)风险 | 月活跃用户的最低使用频次定义 |
| 年经常性收入(ARR) | ARR 约为 $350M | 2026 年 7 月 | Clinkbill 估算(援引 The Information 与 Sacra 报道) | 中 | 第三方对商业规模的估计 | 经正式审计的 ARR 调节表、GAAP 确认营收及财期界定口径 |
不同时点的统计口径与人群定义存在差异。过往律师数量不可直接作为口径一致的队列连续增速对待;当前官网数据取自 2026 年 9 月 25 日的访问快照,7 月的 ARR 属于未经审计的第三方推算。
[CU002, CU003, CU004, CU013, CU015, CU016]6.3 官方披露客户标杆案例
Harvey 公开宣传材料列举了若干标杆案例,这属于官方自选披露,并非对客户合同细节、全员铺开范围或实际商业成效的独立审计。Harvey 官方案例提及与 GSK Stockmann 联合设计了从案情界定到报告出具的端到端尽调工作流,并披露 PwC 交易团队在真实并购项目中上线了该系统,但均未证实两家机构是否全所上线或实现量化收益。同一文章指出,HubSpot 法务运营团队选用 Harvey 承担保密协议(NDA)、供应商合同、客户协议及合作合同的初审工作;但未量化其在缩短销售周期、节约外部外包律师费或部署席位上的具体指标。Repsol 披露其法务部实现了 96% 的使用覆盖,但审阅摘录未独立核实该比例或披露成本节约数据。Glean 联合访谈属于供应商背书式的管理层公开发言,重点介绍 Harvey 自身的内部治理经验,不能视作针对外部客户投资回报率的独立核查。[CU005, CU006, CU008]
| 标杆客户机构 | 细分市场分类 | 部署范围与核心落地场景 | 当前项目状态 | 已获披露的业务成效 | 核心局限与核实缺口 |
|---|---|---|---|---|---|
| 普华永道(PwC) | 全球综合专业服务机构 | 报道称 PwC 交易团队在实际并购项目中使用 Harvey | 官方披露应用案例;实际铺开范围未核实 | 报道提及真实并购项目中的使用;未披露量化成效 | 合同采购类型、实际启用模块、全球部署范围及量化成效均未独立核验 |
| GSK Stockmann | 欧洲大型商事精品律所 | 联合设计从确定范围到交付报告的端到端尽调工作流 | 官方披露业务工作流;实际落地深度未核实 | 确立了工作流共研合作;但未出具该律所专属的可量化提效明细 | 合同性质、实际部署范围、调用模块深度及最终提效指标均未核验 |
| HubSpot | 大型企业法务部门 | 选用 Harvey 初审 NDA、供应商协议、客户合同和合作伙伴协议 | 官方披露采购入选;实际落地进度未核实 | 重点描述了合同初审工作流;但未对成效数据进行量化 | 用户数、部署状态、对销售周期和法律支出的影响均未披露 |
| Repsol | 跨国巨头企业法务部 | 法务部门将 Harvey 接入现有工作流 | 官方通报普及采用;未见独立第三方核实 | Repsol 报告称其法务部门采用率为 96% | 统计口径定义、合同采购细节、全员铺开范围及控费成效均未公开 |
样本来自 Harvey 官方用例文章。表中各行不能独立核实客户合同、投产状态、部署范围或业务成效;未量化影响仍是明确的尽调缺口。
[CU005, CU006, CU008]6.4 客户拓展闭环、留存韧性与集中度风险
Harvey 的客户扩张依靠“切入并扩展”(land-and-expand),组织工作流的固化也可能增强客户黏性。客户往往从业务繁重的团队开始使用,例如上市公司并购或证券诉讼团队;这些场景的人工尽调容易形成瓶颈。Harvey 称,500 个业务组已部署超过 25,000 个可复用的定制工作流智能体。律所模板和内部先例可能嵌入平台,抬高切换成本,但其实际影响尚未得到验证。另一个风险是估值:2026 年 3 月的 $11B 融资估值相当于 2026 年 1 月约 $190M ARR 的 57.9x;这是估值/ARR 比率,不是会计确认收入,也不是当前交易倍数。Harvey 依赖第三方前沿基础模型,面临模型商品化和定价风险;现有证据也未说明其后训练、工作流、集成和应用层知识产权的具体范围。[CU009, CU010, CU011, CU012, CU014]
多阶段客户拓展生命周期,描绘法律机构如何从特定业务组轻量试点逐步演进为全所级深度标准化部署。
[CU009, CU010, CU014]6.5 支撑图表
07风险
7.1 行业合规、职业操守与执业失职责任
律所部署 Harvey AI 面临的核心阻力,在于因引用虚假或伪造法条、核验不足而引致的律师职业过失责任。依照美国律师协会(ABA)《职业行为示范规则》第 1.1 条,律师负有不可转让的技术胜任力义务;第 1.6 条要求保护客户保密信息;第 5.1 条与第 5.3 条要求监督技术工具的使用。2023 年 Mata v. Avianca 等案件显示,法院会因 AI 生成虚假判例或错误的司法管辖结论而惩戒执业律师;这并不能证明软件供应商在所有情形下均免责。Grow Law 行业评测定性提示 Harvey 仍可能出现错误引用或凭空捏造法条先例。由于尽调未见经独立第三方审计的 Harvey 专有准确率评测,本报告不对 Harvey 的确切幻觉率定下结论。要在合规前提下稳妥应用,律所必须出台正式的书面 AI 治理制度,坚持对照一手权威法条进行二次交叉核验,确保任何交付给客户或提交给法庭的文本都经过有据可查的人工复核。[CR002, CR003, CR007, CR008]
| 适用法规 / 司法判例 / 执业义务 | 司法管辖区 | 现行适用状态 | 发生概率 | 潜在后果严重性 | 核心风险缓解措施 | 剩余风险敞口 | 尽调核验落脚点 |
|---|---|---|---|---|---|---|---|
| Mata v. Avianca 案关于 AI 捏造判例的司法惩戒 | 美国联邦法院系统 | 现行有效先例 | 中 | 极高(Critical) | 提交前须在 Westlaw/Lexis 中独立交叉核验引文 | 高 | 审查律所全员判例验真操作规程与法庭文书核查内控制度 |
| ABA《职业行为示范规则》第 1.1 条:技术胜任力义务 | 美国(各州律师协会) | 现行执行 | 高 | 高 | 强制推行继续法律教育(CLE)并留存完整的提示词工程培训考核记录 | 中 | 调取律所内部培训合规证明及 CLE 学分记录 |
| ABA《职业行为示范规则》第 1.6 条:客户保密义务与云端存储 | 美国(各州律师协会) | 当前合规 | 中 | 高 | 核实当前云托管、数据加密、权限管控及合同中的零训练条款 | 中 | 审查供应商 SOC 2 Type II 审计报告及客户数据保留协议 |
| 欧盟《人工智能法案》与跨境执业特免权受限 | 欧盟 / 英国 | 分阶段落地实施 | 中 | 中 | 部署区域本地数据驻留,引入人机协同审计追溯日志 | 中 | 评估欧盟数据中心架构及多司法辖区合规认证 |
根据截至 2026 年 9 月的主要律师协会规则与司法惩戒案例,评估法律和监管风险。
[CR002, CR003, CR008]7.2 运维扩展瓶颈、向量抽取负载与系统可靠性
随着 Harvey 从临时的起草辅助工具转变为企业级核心记录系统,运营瓶颈已转向文档处理和检索基础设施。2025 至 2026 年间,Harvey 的周文档处理量增至原来的 26 倍,从 0.94M 份增至 24.8M 份,涉及 56TB 客户数据及扫描版 PDF、数据室导出文件和诉讼电子取证档案等多种格式。在突发的企业级数据导入负载下,文档处理不再是统一管道,而呈现出不同的失效模式。初期格式抽取受阻于复杂的 OCR 识别与转换卡顿,文档分块与向量化(embedding)则受制于模型调用带宽与 CPU 算力分配。更要害的是,向量数据库索引在下游频繁遭遇写入背压、大型二进制文件(blob)读取超时与速率限制。Harvey 通过将旧任务队列架构升级为自研 Job Framework,引入活动超时、队列隔离与持久化状态恢复,降低了批量任务崩溃的风险;同时划定统一文档格式(UDF)边界并引入 Apache Arrow IPC 取代低效的 JSON 序列化,有效压降了内存开销,并依托双写与影子读取实现了无感知平滑迁移向量数据库。[CR001, CR007, CR010]
| 系统失效模式 | 发生概率 | 潜在后果严重性 | 缓解机制成熟度 | 剩余风险敞口 | 核心未决信息缺口 |
|---|---|---|---|---|---|
| 向量数据库写入背压与速率限制 | 高 | 高 | 高 | 中 | 并购数据室密集批量上传时的多租户向量写入并发上限 |
| 复杂 PDF / 邮件处理中的 OCR 与解析卡死 | 高 | 中 | 高 | 低 | 第三方 OCR 容灾切换延迟,以及未识别的生僻法律专有排版 |
| 数据摄入中途工作节点崩溃与批处理协同失效 | 中 | 高 | 高 | 低 | 针对数 GB 量级复杂诉讼证据开示导出包的活动超时参数配置 |
| 全流水线跨阶段序列化内存倍增 | 中 | 中 | 高 | 低 | 长期维护专有统一文档格式(UDF)模式定义的额外工程开销 |
梳理自 Harvey 针对向量数据库平滑迁移及 Job Framework 运行架构所作的官方技术披露。
[CR001, CR007, CR010]7.3 生态伙伴羁绊与平台集中度敞口
交易对手方与基础设施依赖构成实质性的尽调风险项,但公开凭据无法准确量化当前的业务集中度。Grow Law 与 Fast.io 评测指认 Harvey 底层运行于 Microsoft Azure 和 OpenAI 之上,而 Value Add VC 则重点探讨了其通过 OpenAI 和 Anthropic 引入的上游前沿基础模型依赖。这些过往描述无法坐实 Harvey 当前在各供应商间的算力与接口配额划分、排他性、系统跨云迁移能力或具体商务合作条款。定价维度的公开证据同样相当匮乏。销售同类 AI 文档工作台的竞品 Fast.io 指出 Harvey 并未公开标价,并援引行业传闻及所谓合同流出信息,声称其要求至少 20 个席位起订、通常按每人每月 $1,200 计费,且存在初始部署费用及续约提价条款。该测评提及的 $288,000 门槛,纯粹是由 20 席乘以每月 $1,200 再乘以 12 个月推算出的算术示例,绝非经核实通用的全网合同底线。Grow Law 另给出了每月 $1,000 至 $1,200 的粗略估计。这些数据虽反映出中小律所的采购价格阻力,但不能直接推导出小型律所已被全盘拒之门外。官方统一报价表、供应商结构、实际集中度及 SLA 保障条款仍属于尽调的核心待查盲区。[CR004, CR005, CR009, CR012, CR013]
| 底层依赖要素 | 交易对手方 | 承担角色 | 业务集中度 | 潜在失效场景 | 潜在后果严重性 | 风险缓解对策 | 剩余风险敞口 |
|---|---|---|---|---|---|---|---|
| 前沿基础模型接口 | 第三方评测提及的有多家供应商,包括 OpenAI 与 Anthropic | 上游核心推理模型;当前具体采购配额及排他性均未公开 | 无法量化 | 接口大幅提价、模型版本破坏性变更或上游服务突发瘫痪 | 高 | 核查模型可移植性与多供应商服务连续性;当前缓解措施尚未核实 | 高 |
| 云基础设施与安全 | Microsoft Azure(据过往测评梳理) | 报道提及的主机托管与安全基建;当前算力占比与正式 SLA 条款未公开 | 无法量化 | 云厂商区域节点故障或国家级数据主权合规冲突 | 高 | 尽调中核实容灾预案、可移植性和合同 SLA 条款 | 中 |
| 权威一手法律数据库接入 | LexisNexis | 判例法条库、制定法数据库及权威引用数据流 | 中等(Moderate) | 数据授权纠纷或内容访问终止 | 高 | 拓展开源判例数据库并接入次级备用法律数据流 | 中 |
| 企业级大客户付费的长期可持续性 | AmLaw 100 头部大所 | 核心买方客户基本盘;官方实际成交单价与客户集中度均未公开 | 无法量化 | 若未经证实的高客单价传闻属实,恐遭遇 IT 预算削减或试点后退订 | 高 | 向企业内部法务部纵深拓展,加速业务工作流智能体定制化落地 | 高 |
综合了早期第三方对供应商依赖的定性梳理及未经核实的价格传闻。该表并未确立当前各供应商的确切采购份额、排他性、SLA 条款或通用的全网合同保底金额。
[CR004, CR005, CR012, CR013]7.4 企业级风险热力图、风控缓解架构与逻辑破位触发器
综合法律、运营与商业风险,Harvey 的主要弱点是上游模型和基础设施依赖,以及用户疏于核验,而非基础技术的可行性。企业需要将培训常态化,例如采用 Harvey Academy 认证,并建立正式质检清单,将 AI 输出视为未经核验的初级律师草稿。机构投资者和企业客户应设定明确的投资逻辑失效条件,并将其与可衡量的运营及市场里程碑挂钩:一是某家大型 Am Law 100 客户因未被发现的 Harvey 引用错误而受到公开司法制裁或法院训诫,暴露监督失效;二是第三方基础模型供应商更改 API 商务条款或接入权限,削弱 Harvey 的业务经济性或产品能力;三是 Thomson Reuters CoCounsel 或 LexisNexis 以零边际成本捆绑同等智能体能力,导致企业客户席位收缩;四是反复出现的向量数据导入瓶颈,在正在进行的并购交易临近截止期限时,导致数据室中断数小时。后续尽调需重点核查私有向量数据库的隔离指标与正式 SLA 承诺。[CR006, CR008, CR011, CR014]
横向比对核心运营、法律合规及商业维度下的发生概率、危害烈度、应对成熟度及剩余风险敞口。
发生概率、危害烈度、机制成熟度及剩余敞口均依据资深从业人员评测反馈与公开技术报告综合评定。
[CR001, CR002, CR004, CR005, CR008, CR014]7.5 支撑图表
08估值
8.1 融资沿革与估值演进背景
Harvey 在 2025 至 2026 年间经历多次私募估值调整。第三方资料显示,估值从 2025 年初的 $3.0B 升至当年 6 月的 $5.0B,并在年底达到 $8.0B。2026 年 3 月,Harvey 完成由 GIC 与 Sequoia Capital 联合领投、估值 $11.0B 的 $200M 成长期融资。2026 年 9 月,公司又宣布完成由 Diffusion 与 Lightspeed Venture Partners 联合领投、估值 $15.5B 的 $550M 融资。第三方追踪显示累计融资超过 $1.5B,但审阅资料未披露完整投资人结构。融资公告未披露优先权顺位、清算权、棘轮条款、债务或老股交易安排,因此无法量化这些条款对普通股或早期优先股的影响。[CV001, CV002, CV003, CV014]
| 核心评估维度 | 现行披露状态 | 尽调研判结论 | 最终决策建议 |
|---|---|---|---|
| 投资评级建议 | 观察(Track) | 静待入场估值理性回调或经审计报表出炉 | 不按当前 $15.5B 估值配置资金 |
| 估值立场研判 | 偏高(Stretched) | 对比 2026 年 1 月较早的 ARR 估计为 81.6x;当前分母未提供 | 精确评估当前倍数前,需取得更新的 ARR 分母、现金、负债及 GAAP 收入 |
| 综合风险等级 | 高(High) | 披露有限,且面临平台竞争 | 需审慎评估下行保护和清算条款 |
| 目标入场估值 | $8.0B - $10.0B | 以假设的 $400M ARR 为分母,按 20-25x ARR 测算入场估值 | 三年毛 MoIC:入场估值 $10B-$8B 时为 2.55x-3.19x;基准情景在稀释和费用前为 0.99x-1.24x |
核心指标基于最新报道的 $15.5B 私募融资及第三方 ARR 估计;完整审计财务数据仍未公开。
[CV001, CV005, CV011]8.2 投资逻辑与反向逻辑
Harvey 的投资逻辑在于其已覆盖头部律所和大型企业法务部门,并为法律知识工作提供企业级操作层。官网目前列出 2,400+ 家法律机构、200,000+ 名专业人士和 75 家以上 AmLaw 100 律所。嵌入尽调、合同审查、监管研究和诉讼工作流,可能带来通用 AI 工具难以复制的切换成本。反向逻辑则包括估值偏高、依赖第三方模型,以及前沿实验室技术进步带来的竞争压力。Harvey 指引称,法律 AI 用于辅助而非取代律师,且不能承担法律建议责任;因此,估值下行情景不应建立在完全替代律师劳动力的假设上。现有证据无法量化其合规成本或商业定价。以 2026 年 9 月 $15.5B 估值对比 2026 年 1 月较早的 $190M ARR 估计,得到 81.6x;本章证据未提供当前 ARR 分母。57.9x 和 80x 分别对应 2026 年 3 月和 2025 年底的历史估值,不是当前倍数。[CV006, CV007, CV008, CV009, CV015]
| 评估支柱维度 | 多头看多投资逻辑 | 空头反向风险逻辑 | 逻辑破位明确指标 |
|---|---|---|---|
| 市场地位壁垒 | 是 AmLaw 100 律所和财富 500 强企业法务团队中的主要品类领导者 | 传统老牌软件巨头正将具备直接竞争力的 AI 工作流深度嵌入既有软件底座 | 企业客户新增放缓或现有合同降级 |
| 产品技术壁垒 | 专有法律智能体架构与法律工程集成带来切换成本 | 前沿模型能力可能削弱封装层的防御力;律师须承担法律责任,也限制了完全替代劳动力的经济效益 | 客户转用通用前沿模型 API 工作流或内部模型 |
| 商业定价话语权 | 可计费工时价值和律师效率提升,可能支撑较高支付意愿 | 律所控成本压力,以及效率提升后按席位收费模式可能受到的侵蚀 | 净收入留存率(NDR)低于 110% 或面临折扣压力 |
| 资本估值倍数 | ARR 增长较快,且市场有进一步扩展至数十亿美元级别的潜力,为较高倍数提供支撑 | 2026 年 3 月估值对应 1 月 ARR 约 57.9x;2025 年底曾提及接近 80x;9 月估值对比较早的 1 月分母为约 81.6x | 下调估值的后续融资,或普通股之上设有清算优先权的结构 |
对比管理层和成长型投资者的观点,以及独立分析和市场结构因素。
[CV006, CV008, CV009, CV011, CV015]8.3 回报测算模型与可比倍数剖析
评估潜在投资回报,需结合三年期业务增长预测与退出倍数假设进行测算。第三方基准显示,Harvey 在 2026 年 1 月的 ARR 约为 $190M,高于 2025 年 8 月的 $100M。在基准情景中,随着企业法律 AI 的采用逐步成熟,Harvey 三年后 ARR 达到 $450M;若退出倍数降至假设的 22x,终值约为 $9.9B,按当前 $15.5B 估值入场将带来负回报。在乐观情景中,平台拓展至税务、金融和企业合规领域,ARR 达到 $850M;按假设的 30x 倍数,终值为 $25.5B,对应 1.65x 回报。在悲观情景中,受模型商品化压力和客户流失影响,ARR 增至 $250M;按假设的 12x 退出倍数,终值为 $3.0B。上述 12x、22x 与 30x 均属于测算假设,并非基于公开可比公司计算出的确定倍数。[CV004, CV005, CV010, CV012]
| 测算推演情景 | 3 年后预估 ARR | 终局退出估值倍数 | 隐含终局企业估值 | 目标投资回报(MoIC) | 概率权重分配 |
|---|---|---|---|---|---|
| 乐观情景(Bull Case) | $850M | 30.0x | $25.5B | 1.65x | 20% |
| 基准情景(Base Case) | $450M | 22.0x | $9.9B | 0.64x | 50% |
| 悲观情景(Bear Case) | $250M | 12.0x | $3.0B | 0.19x | 30% |
| 极端下行受损(Downside Failure Case) | <$150M | 5.0x | <$1.0B | <0.10x | 风险托底层(Risk Floor) |
情景模型假设以 $15.5B 投后估值入场;12x、22x 和 30x 的退出倍数是投资测算假设,并非经核验的公开同业倍数。
[CV004, CV005, CV010, CV012]8.4 投资评级建议与尽调核验框架
综合市场采用情况与估值,建议给予观察(track)评级,风险等级为高,估值立场为偏高。Harvey 在法律 AI 领域已有显著采用,但评估投资仍需区分业务质量与入场价格。20x 至 25x 的入场区间,是分析师以假设的 $400M ARR 为分母设定的测算条件,并非经验证的同业可比倍数。投资者应等到出现符合该假设的入场价格,或管理层提供可核验的 GAAP 审计数据、证明毛利率的持续性和净收入留存率后,再承诺新增资金。投入资本前,机构尽调需核查尚未验证的财务明细、客户留存队列并测算毛利率,同时评估历轮融资清算优先权的影响。若年度 ARR 增速低于 50%、前沿模型厂商推出可替代 Harvey 核心工具的原生法律智能体工作流,或出现下调估值的后续融资(down-round),投资逻辑就可能失效。[CV005, CV011, CV012, CV013, CV015]
串联市场商业验证、估值溢价风险及尽调披露盲区的阶梯式分析框架,清晰推导给出观察(track)评级。
| 节点 / 连接 | 节点 / 起点 | 终点 | 说明 |
|---|---|---|---|
| 节点 1 | 业务规模与市场领先地位 [n1] | 覆盖 2,400+ 家机构、200,000+ 名专业人士,以及 75+ 家 AmLaw 100 顶尖大所 | |
| 节点 2 | 现行估值合理性测算 [n2] | $15.5B 估值对标 2026 年 1 月较早的 $190M ARR 估计为 ~81.6x;最新分母尚不可得 | |
| 节点 3 | 竞品夹击与模型商品化风险 [n3] | 前沿大模型能力商品化与传统厂商的产品打包 | |
| 节点 4 | 关键财务信息披露盲区 [n4] | 未经审计的利润率、净烧钱速度及客户队列流失 | |
| 节点 5 | 决策门槛审慎研判 [n5] | 现行标价是否提供充足的风险调整后收益空间? | |
| 节点 6 | 最终建议结论:观察(Track) [n6] | 持续跟踪运营执行;在估值倍数出现实质性回调前暂缓注资 | |
| 连接 1 | 业务规模与市场领先地位 [n1] | 现行估值合理性测算 [n2] | 确证业务体量 |
| 连接 2 | 现行估值合理性测算 [n2] | 竞品夹击与模型商品化风险 [n3] | 揭示倍数风险 |
| 连接 3 | 竞品夹击与模型商品化风险 [n3] | 关键财务信息披露盲区 [n4] | 加剧前景不确定性 |
| 连接 4 | 关键财务信息披露盲区 [n4] | 决策门槛审慎研判 [n5] | 构成尽调阻碍 |
| 连接 5 | 决策门槛审慎研判 [n5] | 最终建议结论:观察(Track) [n6] | 最终导向观察评级 |
决策流图系统整合了多项定性与定量评估准则,为最终投资决策提供依据。
[CV007, CV008, CV011, CV013]8.5 支撑图表
免责声明
本报告是基于公开证据的尽调快照,不构成投资建议。许多重要财务、法律、技术和合同信息尚未公开,投资决策前应直接向管理层及一手文件核实。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Harvey AI Corporation was founded in 2022 as a Delaware C-Corporation headquartered in San Francisco, California. | 高 | SO002, SO005 |
| CO002 | The company develops an enterprise AI operating system designed for legal, tax, compliance, and professional services organizations. | 高 | SO001, SO002, SO009 |
| CO003 | Harvey's March 2026 funding announcement identifies Winston Weinberg as the company's Chief Executive Officer and co-founder. | 中 | SO003 |
| CO004 | Gabriel Pereyra is identified as a Harvey co-founder, but the admissible sources assigned to this chapter do not establish his current executive title or prior employers. | 中 | SO005 |
| CO005 | Harvey secured early seed financing of $5 million led by the OpenAI Startup Fund in November 2022. | 中 | SO005 |
| CO006 | In March 2026, Harvey announced a $200 million growth financing round co-led by GIC and Sequoia Capital at an $11 billion valuation. | 高 | SO003, SO004 |
| CO007 | Cumulative lifetime capital raised by Harvey surpassed $1 billion following its March 2026 growth round. | 高 | SO003, SO005 |
| CO008 | Key institutional and corporate investors backing Harvey include Sequoia Capital, GIC, OpenAI Startup Fund, Kleiner Perkins, Coatue, and Andreessen Horowitz. | 高 | SO002, SO003, SO005 |
| CO009 | In March 2026, Harvey reported that more than 100,000 lawyers across 1,300 organizations used the platform, including a majority of AmLaw 100 firms. | 高 | SO003, SO004 |
| CO010 | By late 2026, Harvey reported that customer adoption had expanded to over 2,400 organizations across 70 countries, including 75+ AmLaw 100 firms. | 高 | SO001, SO002 |
| CO011 | Third-party estimates place Harvey's annual recurring revenue at approximately $190 million as of January 2026, with later estimates climbing toward $350 million by July 2026. | 中 | SO004, SO006 |
| CO012 | An independent profile reports 860+ employees for 2026 and separately labels 1,261 as a Q1 2026 peak estimate; the reviewed evidence does not establish a verified current headcount. | 中 | SO005 |
| CO013 | The March 2026 $11 billion valuation divided by a January 2026 estimate of roughly $190 million ARR yielded an approximately 58x historical valuation-to-ARR ratio; it is not a current entry multiple. | 中 | SO004 |
| CO014 | In early 2026, Harvey completed the acquisition of AI demo startup Hexus and expanded its product architecture to support over 25,000 autonomous legal agents. | 高 | SO003, SO005 |
| CO015 | Judicial and bar oversight guidelines between 2023 and 2026 impose strict affirmative duties of competence, confidentiality, and human supervision on legal practitioners utilizing generative AI tools. | 中 | SO008 |
| CM001 | The global legal services market represents an estimated $600 billion industry across law firms and corporate legal departments. | 中 | SM007 |
| CM002 | Corporate and transactional legal spend accounts for roughly 30% to 40% of associate time spent on document review, contract analysis, and due diligence. | 中 | SM007 |
| CM003 | Enterprise legal software and automation represents an estimated $15 billion market segment expanding as law firms adopt generative AI tools. | 中 | SM005 |
| CM004 | Harvey AI reached an annual recurring revenue run rate of approximately $190 million as of January 2026, scaling from $100 million reported in mid-2025. | 中 | SM004, SM005 |
| CM005 | Harvey AI's private valuation reached $11 billion in March 2026 following a $200 million growth round, and an independent company profile records a $15.5 billion valuation in September 2026. | 中 | SM004, SM005 |
| CM006 | Harvey has penetrated a majority of the AmLaw 100 law firms and serves over 200,000 legal professionals across more than 2,400 organizations globally. | 高 | SM002, SM004 |
| CM007 | Harvey's core go-to-market model sells enterprise and per-seat subscription contracts primarily to partners at top law firms and general counsel at Fortune 500 corporations. | 中 | SM005, SM007 |
| CM008 | Enterprise law firm buying decisions are driven by law firm equity partners and innovation committees, whereas the primary everyday end users are junior associates and legal engineers. | 高 | SM003, SM007 |
| CM009 | Harvey's enterprise-focused pricing and deployment architecture create significant adoption barriers for small and mid-sized law firms, which turn toward integrated practice management platforms. | 中 | SM001 |
| CM010 | Court filings and bar regulators across the US, UK, and EU mandate attorney competence and supervisory verification, creating structural resistance against fully autonomous unverified legal outputs. | 中 | SM006 |
| CM011 | Documented court cases worldwide for AI-generated hallucinations have exceeded 700 cases, highlighting severe liability risks in active litigation filings. | 中 | SM008 |
| CM012 | Harvey's architectural emphasis on transactional workflows and corporate document analysis leaves specialized litigation tasks, such as element-by-element evidentiary mapping, underserved. | 中 | SM008 |
| CM013 | Harvey provides structured curriculum and certifications in legal engineering to help enterprise clients train internal teams on AI prompt structuring and verification. | 中 | SM003 |
| CM014 | The March 2026 $11 billion valuation divided by the January 2026 ARR estimate of $190 million produced a historical valuation-to-ARR ratio of approximately 57.9x; it is not a current enterprise-value or GAAP-revenue multiple. | 中 | SM005 |
| CP001 | Harvey AI competes directly against venture-backed legal AI challenger Legora, specialized tools like Robin AI and Hebbia, and entrenched incumbents Thomson Reuters and LexisNexis. | 高 | SP002, SP005 |
| CP002 | Swedish legal AI startup Legora reached a $1.8 billion valuation in October 2025 and is estimated at up to $5.5 billion in 2026 as it expands into the United States. | 高 | SP002, SP006 |
| CP003 | Harvey AI serves more than 75 of the AmLaw 100 law firms and over 200,000 legal professionals, establishing an early lead in top-tier law firm penetration. | 高 | SP003, SP005 |
| CP004 | Harvey's core platform tools comprise Assistant, Vault, Knowledge, Workflows, History, and native Microsoft Word and Outlook add-ins. | 高 | SP001, SP004 |
| CP005 | Legal research incumbents Thomson Reuters and LexisNexis leverage authoritative proprietary databases like Westlaw and Shepards citations to counter standalone legal AI startups. | 中 | SP005, SP006 |
| CP006 | Unlike generic chatbots, Harvey AI grounds legal queries in primary statutes and case law and allows firms to train models on proprietary document libraries. | 高 | SP001, SP007 |
| CP007 | Enterprise security certifications including SOC2 Type II, ISO 27001, CCPA, and GDPR compliance provide a critical procurement hurdle against horizontal AI tools. | 高 | SP001, SP007 |
| CP008 | Harvey's deployment of over 25,000 custom legal workflow agents embeds the software directly into complex firm operating procedures, increasing switching costs. | 中 | SP005, SP008 |
| CP009 | Harvey maintains a closed enterprise pricing structure with unlisted custom quotes and long wait times for demonstrations. | 中 | SP001 |
| CP010 | Incumbents like Thomson Reuters bundle AI capabilities like CoCounsel directly into existing Westlaw and Practical Law subscriptions, creating pricing headwinds for point solutions. | 中 | SP006 |
| CP011 | Harvey focused its go-to-market motion on top 200 law firms and Fortune 500 legal departments to secure high contract values and reference-ability before expanding downmarket. | 中 | SP008 |
| CP012 | Harvey does not develop proprietary frontier base models, relying instead on API integrations with OpenAI, Anthropic, and Google. | 高 | SP002, SP006 |
| CP013 | Foundation model providers are expanding directly into vertical workflows, exemplified by Anthropic launching specialized legal plugins and coding assistants. | 高 | SP002, SP006 |
| CP014 | Strategic positioning separates legal tech competitors into specialized enterprise vertical suites, database-centric legacy incumbents, and generic horizontal foundation model APIs. | 中 | SP005, SP006 |
| CI001 | Harvey closed a $550M financing round at a $15.5B post-money valuation co-led by Diffusion and Lightspeed in September 2026. | 中 | SI002 |
| CI002 | Harvey's customer adoption spans over 2,400 legal organizations and 200,000 legal professionals, including more than 75% of Am Law 100 law firms. | 高 | SI005, SI002 |
| CI003 | Third-party tracking estimates Harvey's ARR at $5M in 2023, $50M at end-2024, $100M in August 2025, and $195M at end-2025; a separate January 2026 estimate places ARR at approximately $190M. | 中 | SI009, SI010 |
| CI004 | Third-party industry analysis reported Harvey reached approximately $350M in ARR by July 2026, driven by rapid enterprise expansion and in-house corporate legal rollouts. | 中 | SI010 |
| CI005 | Third-party tracking describes Harvey's monetization through legal-professional licenses, enterprise agreements, workflow products, and expansion into additional offices, practices, and geographies; revenue shares and realized pricing are undisclosed. | 中 | SI004, SI010 |
| CI006 | Harvey expanded beyond custom Big Law engagements to release commercial product bundles, case law models, and Vault document exploration at scalable price points. | 高 | SI004, SI006 |
| CI007 | Harvey maintains a strategic alliance with PwC to train and commercialize foundation models for tax, legal, and human resources deployed across member firms. | 中 | SI004 |
| CI008 | The reviewed evidence does not disclose Harvey's gross margin or model-inference expense; dependence on third-party frontier models creates a qualitative cost and pricing risk whose magnitude cannot be quantified. | 中 | SI009 |
| CI009 | The $15.5B equity valuation divided by the July 2026 third-party ARR estimate of approximately $350M is about 44.3x; divided by the older January 2026 estimate of approximately $190M it is about 81.6x. These are dated valuation-to-ARR ratios, not enterprise-value or trailing-GAAP-revenue multiples. | 中 | SI009, SI002, SI010 |
| CI010 | Harvey's lack of proprietary frontier foundation models exposes its pricing power and gross margin to commoditization risk as OpenAI and Anthropic enhance core model capabilities. | 中 | SI009, SI003 |
| CI011 | Harvey's $550M financing announcement does not disclose current cash, prior spending, net burn, debt, or runway, and cumulative financing cannot be treated as current liquidity. | 中 | SI002 |
| CI012 | Harvey markets workflow agents for repeatable legal tasks and AI-assisted M&A due diligence, including contract review and large document-set analysis. | 高 | SI007, SI008 |
| CI013 | Independent legal technology reviewers have raised scrutiny over whether Harvey's high contract cost justifies enterprise ROI relative to alternative legal tech tooling. | 中 | SI003 |
| CI014 | The reviewed public evidence does not disclose audited GAAP revenue, gross margin, cash flow, cohort net dollar retention, customer churn, CAC payback, or realized ACV. | 中 | SI002, SI009, SI010 |
| CE001 | Harvey AI provides a domain-specific software suite comprising Harvey Assistant, Harvey Vault, Harvey Knowledge, and Harvey Workflows for legal drafting, document analysis, and due diligence. | 高 | SE001, SE007 |
| CE002 | Harvey Vault provides secure document storage supporting up to 100,000 files per vault with automated OCR, bulk key-term extraction, and structured comparison across contract sets. | 高 | SE003, SE005 |
| CE003 | Harvey Workflows enables attorneys to configure natural-language, multi-step workflow agents and thread-based Review Tables without software engineering. | 高 | SE003, SE005 |
| CE004 | Harvey integrates directly into daily legal drafting environments via dedicated add-ins for Microsoft Word and Microsoft Outlook. | 高 | SE003, SE005 |
| CE005 | Harvey grounds legal research in more than 500 vetted primary legal databases including LexisNexis for US law, EUR-Lex for European regulations, and EDGAR for corporate disclosure filings. | 高 | SE004, SE005 |
| CE006 | Harvey integrates natively with enterprise document management systems including iManage, SharePoint, and Google Drive for matter synchronization. | 中 | SE005, SE007 |
| CE007 | In May 2026, Harvey open-sourced its Legal Agent Benchmark (LAB), featuring over 1,200 tasks across 24 legal practice areas evaluated by more than 75,000 expert rubric criteria. | 高 | SE001, SE002 |
| CE008 | Harvey's Legal Agent Benchmark implements an all-pass grading standard requiring an agent to satisfy every planted factual, risk, and timeline criterion across a client matter to receive credit. | 中 | SE002 |
| CE009 | Harvey's inference layer routes requests across leading commercial foundation models, including OpenAI, Anthropic Claude, Mistral, and xAI architectures, rather than relying exclusively on a proprietary foundation model. | 中 | SE007, SE008 |
| CE010 | Harvey maintains enterprise compliance certifications including SOC 2 Type II, ISO 27001, GDPR, and CCPA, supporting regional data processing across the United States, Europe, and Australia. | 中 | SE005, SE007 |
| CE011 | Harvey contractually guarantees that client matter documents and prompts are not used to train or fine-tune public foundation models by default. | 高 | SE004, SE005 |
| CE012 | Harvey differentiates from horizontal AI tools by combining primary legal database grounding, native DMS connectors, firm template playbooks, and strict confidentiality controls. | 高 | SE004, SE005 |
| CE013 | Harvey operates as an enterprise-only software service requiring demo-based sales qualification, with no self-serve onboarding, unspecified database update intervals, and a 100,000-document ceiling per vault. | 中 | SE005, SE006 |
| CE014 | By 2026, Harvey's platform scaled to support more than 100,000 legal professionals and 25,000 active custom agents across 1,300 organizations globally. | 高 | SE001, SE007 |
| CU001 | Harvey segments its enterprise customer base primarily into global law firms, mid-sized regional firms, corporate in-house legal departments, and multi-disciplinary professional services firms. | 高 | SU003, SU007 |
| CU002 | Harvey reported customer scale expanding from 1,500+ legal organizations in mid-2026 to more than 2,400 organizations by September 2026 across more than 70 countries. | 高 | SU003, SU007 |
| CU003 | Harvey has penetrated a majority of the AmLaw 100, reaching more than 75 of the top 100 US law firms as active platform customers by September 2026. | 高 | SU003, SU005, SU007 |
| CU004 | Harvey's homepage displayed more than 200,000 professionals on the September 25, 2026 access snapshot; the reviewed page does not define the activity threshold or whether every professional is an active user. | 中 | SU007 |
| CU005 | Harvey's official use-case article says GSK Stockmann co-designed an end-to-end due-diligence workflow and describes PwC's deal team in live M&A engagements, without independently verifying rollout scope or quantified outcomes. | 中 | SU003 |
| CU006 | Harvey's official use-case article says HubSpot's Legal Operations team selected Harvey for a first-pass contract-review pattern, while Repsol reported 96% adoption across its legal department; contract status, rollout detail, and business impact are not independently verified. | 中 | SU003 |
| CU007 | In-house legal teams report saving an average of more than 25 hours per month by deploying Harvey for first-pass contract analysis and regulatory tracking. | 高 | SU003, SU007 |
| CU008 | In a Glean-hosted interview, Curtis Salinas says Harvey builds confidence in AI through trusted context, thoughtful governance, hands-on experimentation, and cross-functional collaboration across legal, IT, security, and business teams. | 中 | SU001 |
| CU009 | Customers have built and deployed more than 25,000 custom Workflow Agents across over 500 legal practice groups to codify firm-specific drafting and diligence standards. | 中 | SU003, SU005 |
| CU010 | Harvey maintains an inverted enterprise sales motion where adoption at flagship elite law firms serves as reference validation to drive subsequent in-house corporate legal sales. | 中 | SU006, SU009 |
| CU011 | The March 2026 $11 billion financing price divided by the January 2026 $190 million ARR estimate equaled approximately 57.9x; this was a historical valuation-to-ARR ratio, not recognized revenue and not a current trading multiple. | 中 | SU006 |
| CU012 | Harvey depends on upstream frontier base models from third-party providers, creating commoditization risk, while the reviewed evidence does not establish the scope of Harvey's distinct post-training or application-level intellectual property. | 中 | SU006, SU008 |
| CU013 | Harvey reported a 92% monthly adoption rate across its enterprise customer base for active legal practice workflows. | 中 | SU003 |
| CU014 | Harvey's customer journey expands from single-matter transactional diligence pilots into multi-practice agent orchestration and firm-wide annual licensing. | 中 | SU003, SU004 |
| CU015 | Separate source snapshots report 100,000+ lawyers and 142,000+ lawyers, but the reviewed excerpts do not establish comparable observation dates or show that those populations match the homepage's broader professionals definition. | 中 | SU003, SU006 |
| CU016 | Clinkbill estimates that Harvey reached approximately $350 million ARR by July 2026, citing The Information and Sacra; the figure is third-party reported and unaudited. | 中 | SU009 |
| CR001 | Harvey's document processing ingestion scaled 26-fold year-over-year from 0.94 million to 24.8 million documents per week, handling 56 terabytes of customer data. | 高 | SR003, SR006 |
| CR002 | Judicial precedents including Mata v. Avianca establish that practicing attorneys remain strictly and personally accountable for AI hallucinations and non-existent citations. | 高 | SR005, SR007 |
| CR003 | Professional conduct rules ABA Model Rules 1.1, 1.6, and 5.1 impose non-delegable duties of technological competence, client confidentiality, and supervisory oversight on AI adoption. | 高 | SR005, SR007 |
| CR004 | Dated Grow Law and Fast.io reviews describe Microsoft Azure and OpenAI in Harvey's stack, while Value Add VC discusses dependence on frontier models from OpenAI and Anthropic; the reviewed sources do not establish current allocation, exclusivity, or vendor shares. | 中 | SR001, SR002, SR009 |
| CR005 | Fast.io, a competitor review that says Harvey does not publish pricing, reports a 20-seat minimum and commonly cited $1,200 monthly seat price; $288,000 is the review's illustrative 20-by-$1,200-by-12 calculation, not a verified universal annual floor. | 中 | SR002 |
| CR006 | Diligence triggers should monitor unverified model citations, downstream vector-database rate limits, and unverified third-party reports of subscription price uplifts. | 中 | SR002, SR003 |
| CR007 | Harvey redesigned its document pipeline on an internal Job Framework that isolates extraction failures from chunking and vector indexing tasks. | 中 | SR003 |
| CR008 | Grow Law warns that Harvey may produce incorrect citations or fabricated case law without fact-checking; no independently evaluated Harvey-specific accuracy rate was established by the reviewed evidence. | 中 | SR001 |
| CR009 | Grow Law and Fast.io publish unverified per-seat estimates that suggest affordability pressure for smaller firms, but Harvey's official pricing is undisclosed and the reviewed evidence does not establish that all solo or small firms are excluded. | 中 | SR001, SR002 |
| CR010 | Harvey migrated its primary vector database via dual-writing and shadow reads while adopting Arrow IPC serialization to alleviate memory pressure and backpressure. | 中 | SR003 |
| CR011 | Harvey platform adoption claims exceed 200,000 professionals across more than 2,400 legal organizations and 75 AmLaw 100 law firms. | 中 | SR006 |
| CR012 | Direct competitor Thomson Reuters CoCounsel surpassed 1 million users by embedding AI directly into Westlaw primary legal database subscriptions. | 中 | SR002 |
| CR013 | Value Add VC argues that improvements in general-purpose frontier models could narrow Harvey's differentiation and expose the company to pricing pressure from better-capable legal AI substitutes. | 中 | SR009 |
| CR014 | Harvey Academy provides structured training modules for legal engineering and vendor breach liability analysis to improve firm-level AI literacy. | 中 | SR008 |
| CV001 | In September 2026, Harvey announced a $550 million funding round at a $15.5 billion valuation co-led by Diffusion and Lightspeed Venture Partners. | 高 | SV003, SV005 |
| CV002 | Prior to the September 2026 round, Harvey closed a $200 million growth round in March 2026 co-led by GIC and Sequoia Capital that valued the company at $11 billion. | 中 | SV001, SV007 |
| CV003 | Harvey experienced four successive repricings across five reviewed valuation marks—$3 billion, $5 billion, $8 billion, $11 billion, and $15.5 billion—during 2025 and 2026. | 中 | SV001, SV003, SV004, SV008 |
| CV004 | Harvey achieved approximately $190 million in annual recurring revenue as of January 2026, expanding from $100 million reported in August 2025. | 中 | SV001, SV007 |
| CV005 | The March 2026 $11 billion valuation was approximately 57.9x the January 2026 ARR estimate of $190 million; pairing the September 2026 $15.5 billion mark with that same older denominator yields approximately 81.6x, not a current revenue multiple. | 中 | SV001, SV003, SV005, SV007 |
| CV006 | The core investment thesis rests on category leadership in legal AI, rapid expansion from law firm partners into corporate legal departments, and workflow standardization. | 中 | SV003, SV008 |
| CV007 | Harvey's current homepage reports more than 2,400 legal organizations, more than 200,000 professionals, and 75+ AmLaw 100 firms using the platform. | 中 | SV006 |
| CV008 | Adverse commentary's approximately 57.9x and 80x ratios refer to the March 2026 $11 billion mark with January 2026 ARR and the late-2025 $8 billion mark with then-reported revenue, respectively; neither is Harvey's current entry multiple. | 中 | SV004, SV007 |
| CV009 | Adverse analysis suggests potential commoditization risk if frontier foundation model providers enhance native legal reasoning capabilities and disintermediate specialized wrapper applications. | 中 | SV004, SV007 |
| CV010 | Under a base-case scenario, Harvey expands ARR to $450 million over three years with a normalized 22x exit multiple, yielding a fair-value valuation of roughly $9.9 billion. | 中 | SV001, SV007 |
| CV011 | The recommended stance is track with high risk and stretched valuation stance because disclosed adoption is strong, but prevailing private pricing still assumes flawless multi-year execution without sufficient margin visibility. | 中 | SV001, SV007 |
| CV012 | The September 2026 financing announcement does not disclose liquidation preferences, ratchets, debt, or secondary mechanics, so their effect on common and earlier preferred equity cannot be quantified from public evidence. | 中 | SV003, SV005 |
| CV013 | Recommendation logic synthesizes market leadership against extreme private multiples, establishing a structured decision framework that requires financial auditability prior to a buy rating. | 中 | SV001, SV007 |
| CV014 | Harvey has raised over $1.5 billion in cumulative capital according to its September 2026 financing announcement and third-party tracking. | 中 | SV003, SV005 |
| CV015 | Harvey's own guidance says legal AI augments rather than replaces lawyers and cannot take responsibility for legal advice, supporting a valuation downside constraint on full labor-replacement assumptions without establishing quantitative compliance costs or commercial pricing. | 中 | SV010 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Harvey AI Corporation | Harvey – Professional Class AI | Firms and legal departments do their best work with Harvey. Run everything your team and practice needs on one collaborative platform. |
| SO002 | Harvey AI Corporation | Meet the team behind our legal AI company | Backed by world-class investors including Sequoia, Kleiner Perkins, GV, OpenAI Startup Fund, Coatue, Andreessen Horowitz, and EQT, Harvey is used by 2,400+ customers in 70+ countries. |
| SO003 | Harvey AI Corporation | Harvey Raises at $11 Billion Valuation to Scale Agents Across Law Firms and Enterprises | Harvey, the legal infrastructure for law firms and in-house teams, today announced $200 million in new funding co-led by returning investors GIC and Sequoia. The round values Harvey at $11 billion and includes participation from existing investors Andreessen Horowitz, Coatue, Conviction Partners, Elad Gil, Evantic, and Kleiner Perkins. The investment will be used to expand the agents customers run on Harvey and grow the embedded legal engineering teams supporting them globally. |
| SO004 | Value Add VC | How Is Harvey AI Worth $11B? The Legal AI Breakdown | At ~$11B on ~$190M ARR, Harvey trades near 58x revenue — pricing in years of flawless execution. The core risk is vertical-AI defensibility: Harvey doesn't own a frontier model, so as OpenAI and Anthropic ship cheaper, more capable models... the multiple leaves zero room for a stumble. |
| SO005 | Silicon Valley Investclub | Harvey AI — Company profile — Silicon Valley Investclub | Harvey AI is a pioneering legal technology company founded in 2022 by Winston Weinberg and Gabriel Pereyra. The company specializes in domain-specific artificial intelligence for legal and professional services sectors. Harvey provides AI-powered solutions for tasks such as contract analysis, due diligence, legal research, compliance, and litigation support. |
| SO006 | Clinkbill | Harvey AI ARR 2026: $350M Revenue Timeline for Legal AI | Legal was supposed to be the slowest industry to buy AI. Harvey reached roughly $350 million in annual recurring revenue by July 2026, up from $195 million at the end of 2025 — an 80% climb in about six months. Here is the sourced revenue timeline. |
| SO008 | Harvey AI Corporation | How to Use AI as a Lawyer: The Workflows, Risks, and Rules | Between 2023 and 2025, bar associations and oversight bodies in the US, the UK, the EU, and elsewhere issued guidance on responsible AI use, covering competence, confidentiality, and the duty to verify what a machine produces. |
| SO009 | Harvey Academy | Library | See how a Harvey Agent researches liability exposure after a vendor data breach, mapping statutory damages, regulatory fines, and indemnity recovery into a brief for the incident response team. |
| SM001 | Clio | Harvey AI Alternatives: Legal AI for Law Firms - Clio | While Harvey AI ignited interest in generative AI for the legal industry, its enterprise focus often limits accessibility for most firms. |
| SM002 | Harvey | Harvey | AI software for legal and professional services | The world's top law firms and in-house legal teams trust Harvey with their highest-stakes work. |
| SM003 | Harvey Academy | Library | An intermediate course for legal professionals ready to design, execute, and verify large-scale AI document analysis — from structuring extraction prompts to catching errors before they reach a client or filing. |
| SM004 | Silicon Valley Investclub | Harvey AI — Company profile | Its platform is used by global law firms and Fortune 500 enterprises. On March 25, 2026, Harvey confirmed it had closed a $200M Growth Round co-led by GIC and Sequoia at an $11 billion valuation — bringing total funding to over $1.2 billion. The company reached $190 million in ARR in January 2026, serves 1,300+ customers across 60 countries (including a majority of AmLaw 100 firms), and now runs more than 25,000 custom AI agents for clients. |
| SM005 | Value Add VC | Harvey AI 2026: $11B Valuation, $190M ARR | Legal is a near-perfect AI market: enormous billable-hour budgets, document-heavy work, low tolerance for error, and buyers who will pay for tools that are demonstrably accurate and secure. |
| SM006 | Harvey | How to Use AI as a Lawyer: The Workflows, Risks, and Rules | Between 2023 and 2025, bar associations and oversight bodies in the US, the UK, the EU, and elsewhere issued guidance on responsible AI use, covering competence, confidentiality, and the duty to verify what a machine produces. |
| SM007 | Capitaly | How Harvey AI Became a Billion-Dollar Legal Platform - Capitaly | Harvey went after the $600 billion legal services industry by selling to the people controlling the budgets: partners at top law firms and general counsel at Fortune 500 companies. |
| SM008 | Litmas AI | The Real Litigator's Guide to Evaluating Legal AI Tools in 2026: Litmas AI vs. Harvey AI vs. Lexis AI - Litmas AI - Litmas AI | More importantly, Harvey is primarily geared towards BigLaw transactional work and in-house legal departments. Its core strengths lie in contract review, due diligence, regulatory compliance, and document analysis. It integrates with Microsoft 365 and offers workflow automation tools that make sense for high-volume corporate legal departments processing thousands of similar documents. What it doesn’t do is think like a litigator. |
| SP001 | Maryland State Bar Association | An Overview of Harvey AI's Features for Lawyers | Harvey AI does not publish the costs of using its AI legal tool. However, Harvey AI offers a free demonstration. Pricing amounts shared on the internet vary, and wait times for a free demonstration are reported to be long. |
| SP002 | Forbes | Harvey Hits $11 Billion Valuation With $200 Million Fundraise | Harvey is up against fast growing rivals like Swedish startup Legora which reached a $1.8 billion valuation in October. Startups have also cropped up to go after legal niches, like EvenUp's tools for personal injury lawyers, Finch's focus on paralegals and Supio, which is addressing plaintiff law. |
| SP003 | Harvey AI | Harvey – Professional Class AI | 2,400+ Legal Organizations Build on :Harvey: |
| SP004 | Harvey Academy | Library | Harvey Academy | Learn to turn ad hoc prompts into reliable Workflow agents your whole team can run — covering building, grounding, and sharing for consistent results. |
| SP005 | Silicon Valley Investclub | Harvey AI — Company profile — Silicon Valley Investclub | Collaborative legal AI platform described as a direct Harvey challenger for enterprise firms. |
| SP006 | Value Add VC | How Is Harvey AI Worth $11B? The Legal AI Breakdown | At ~$11B on ~$190M ARR, Harvey trades near 58x revenue — pricing in years of flawless execution. The core risk is vertical-AI defensibility: Harvey doesn't own a frontier model, so as OpenAI and Anthropic ship cheaper, more capable models with longer context and better tool use, the gap between "Harvey" and "Claude with a legal prompt library" could narrow. |
| SP007 | Harvey AI | How to Use AI as a Lawyer: The Workflows, Risks, and Rules | General-purpose AI tools... answer questions on almost any subject but know nothing about your jurisdiction, your client, or the duty you owe the court. Dedicated legal AI is built for law firms and in-house teams. |
| SP008 | Capitaly | How Harvey AI Became a Billion-Dollar Legal Platform | Competitive moat: By owning the relationship with law firms and in-house counsel, Harvey had built switching costs. Customers weren't going to rip out Harvey and replace it with a competitor; they were going to deepen integration. |
| SI002 | Harvey AI | Harvey Raises $550M at a $15.5B Valuation to Help Legal Teams Own Their Intelligence | Today, Harvey announced a $550M funding round at a $15.5B valuation, co-led by Diffusion and Lightspeed Venture Partners. The raise follows the introduction of Harvey’s first post-trained open-weight model and the launch of Harvey LAB, the Legal Agent Benchmark. It will accelerate Harvey’s efforts to help law firms, in-house legal teams, and professional services firms build and own their intelligence at scale. |
| SI003 | Purple Law | Harvey AI Review 2025: Strengths & Weaknesses for Law Firms | A deep dive into Harvey AI's capabilities for law firms — what it gets right, where it falls short, and whether it's worth the investment. |
| SI004 | LawNext | Harvey AI To Move Out Of Early Access Phase, Release More Affordable Versions Of Its Custom AI Models | They will be coming out of the early-access phase during the third quarter of the year and launching versions that they say will be more affordable to firms of various sizes, depending on their needs. |
| SI005 | Harvey AI | Harvey – Professional Class AI | 2,400+ Legal Organizations Build on :Harvey: |
| SI006 | Harvey AI | How to Use AI as a Lawyer: The Workflows, Risks, and Rules | Cloud-based legal AI platforms now offer the same core workflows with flexible pricing, so a two-person practice can run the kind of document review that used to need a litigation support team. |
| SI007 | Harvey Academy | Library | Harvey Academy | See how a Harvey Agent researches liability exposure after a vendor data breach, mapping statutory damages, regulatory fines, and indemnity recovery into a brief for the incident response team. |
| SI008 | Harvey AI | The Guide to AI-Powered Due Diligence for M&A Professionals | AI-powered due diligence is the use of artificial intelligence to investigate, analyze, and verify information before a business transaction closes. It covers the same ground that legal and financial teams have always covered. Contracts get reviewed. Risks get flagged. Obligations get tracked. |
| SI009 | Value Add VC | How Is Harvey AI Worth $11B? The Legal AI Breakdown | At ~$11B on ~$190M ARR, Harvey trades near 58x revenue — pricing in years of flawless execution. The core risk is vertical-AI defensibility: Harvey doesn't own a frontier model... the multiple leaves zero room for a stumble. |
| SI010 | ClinkBill | Harvey AI ARR 2026: $350M Revenue Timeline for Legal AI | Legal was supposed to be the slowest industry to buy AI. Harvey reached roughly $350 million in annual recurring revenue by July 2026, up from $195 million at the end of 2025 — an 80% climb in about six months. Here is the sourced revenue timeline. |
| SE001 | Harvey AI Corporation | Harvey – Professional Class AI | Firms and legal departments do their best work with Harvey. Run everything your team and practice needs on one collaborative platform. |
| SE002 | Harvey AI Corporation | Introducing Harvey's Legal Agent Benchmark | The first version of LAB includes more than 1,200 agent tasks across 24 legal practice areas, and is evaluated by over 75,000 expert-written rubric criteria. |
| SE003 | Harvey Academy | Library | Harvey Academy | Watch how Harvey's thread-based creation flow turns a plain-language description into a ready-to-run Review Table. |
| SE004 | Harvey AI Corporation | How to Use AI as a Lawyer: The Workflows, Risks, and Rules | AI speeds up the reading, the searching, and the first-draft writing. It can't take responsibility for legal advice... The judgment stays with you. The tool shortens the distance to it. |
| SE005 | AI Agents List | Harvey AI Review 2026: Legal AI for Law Firms and In-House Teams | A purpose-built AI platform for law firms and enterprise legal departments. Harvey handles legal drafting, document analysis, due diligence, contract review, and workflow automation — grounded in vetted legal databases and firm-specific documents, not general training data. |
| SE006 | Eigent | Harvey AI Alternative (Free & Open Source) | Eigent | Harvey AI Alternative (Free & Open Source) | Eigent |
| SE007 | Silicon Valley Investclub | Harvey AI — Company profile — Silicon Valley Investclub | AI Agents25,000+ custom agents operating autonomously across law firms and legal departments — completing workflows from start to finish. |
| SE008 | Value Add VC | How Is Harvey AI Worth $11B? The Legal AI Breakdown | Harvey sells AI agents and assistants for legal and professional-services work — drafting documents, reviewing contracts, running due diligence, and completing multi-step workflows that used to consume associate hours. |
| SU001 | Glean Technologies | Harvey Customer Story | Glean Work AI | Harvey AI is helping legal teams adopt AI through trusted context, thoughtful governance, and hands-on experimentation. |
| SU003 | Harvey AI | Top Harvey AI Use Cases for Law Firms and In-House Teams | More than 142,000 lawyers across 1,500+ organizations in 60 countries use Harvey to research, draft, review, and collaborate... Harvey maintains a 92% monthly adoption rate across the customer base. |
| SU004 | Harvey Academy | Library | Harvey Academy | See how a Harvey Agent researches liability exposure after a vendor data breach, mapping statutory damages, regulatory fines, and indemnity recovery into a brief for the incident response team. |
| SU005 | Silicon Valley Investclub | Harvey AI — Company profile — Silicon Valley Investclub | Domain-specific AI platform transforming legal and professional services with frontier AI — 100,000+ lawyers, 1,300+ organizations, 60 countries |
| SU006 | Value Add VC | How Is Harvey AI Worth $11B? The Legal AI Breakdown | At ~$11B on ~$190M ARR, Harvey trades near 58x revenue — pricing in years of flawless execution. The core risk is vertical-AI defensibility: Harvey doesn't own a frontier model... Add well-funded competitors and pricing pressure, and the multiple leaves zero room for a stumble. |
| SU007 | Harvey AI | Harvey – Professional Class AI | Professionals using Harvey... 200,000+ |
| SU008 | Harvey AI | How to Use AI as a Lawyer: The Workflows, Risks, and Rules | Courts and regulators between 2023 and 2026 have scrutinized algorithmic decision-making closely, with the sharpest attention in criminal justice and immigration. |
| SU009 | Clinkbill ARR Leaderboard | Harvey AI ARR 2026: $350M Revenue Timeline for Legal AI | Harvey reached roughly $350 million in annual recurring revenue by July 2026, up from $195 million at the end of 2025. |
| SR001 | Grow Law | Harvey AI Review [2026]: Is It Right for Your Law Firm? | Harvey may produce incorrect citations or fabricated case law. Without fact-checking, these errors can pose real malpractice risks. |
| SR002 | Fast.io | Harvey AI Review 2026: Features, Pricing, and Alternatives | Harvey does not publish pricing. Every contract is negotiated through enterprise sales... At the commonly cited $1,200/seat/month rate, that works out to $288,000 per year before implementation costs. |
| SR003 | Harvey AI | Scaling Document Processing Across Harvey | A year ago, that platform handled just under one million documents in a busy week. Users rarely think about document processing when it is working. Over the last 12 months, our work has been focused on keeping it that way while weekly volume moved from hundreds of thousands of documents to tens of millions. |
| SR005 | Legal Intaker | Harvey AI for Law Firms| | Legal Intaker | Under ABA Model Rule 1.1, attorneys have a duty of technological competence. Under Rule 1.6, client confidentiality must be protected. Under Rules 5.1 and 5.3, supervisory obligations extend to the AI tools a firm deploys. Before any attorney uses Harvey on a client matter, firm leadership must know where client data is stored, whether it is used for model training, what encryption standards apply, and how outputs are reviewed before reaching clients. |
| SR006 | Harvey AI | Harvey – Professional Class AI | The world's top law firms and in-house legal teams trust Harvey with their highest-stakes work. |
| SR007 | Harvey AI | How to Use AI as a Lawyer: The Workflows, Risks, and Rules | Regulators have noticed. Between 2023 and 2025, bar associations and oversight bodies in the US, the UK, the EU, and elsewhere issued guidance on responsible AI use, covering competence, confidentiality, and the duty to verify what a machine produces. None of it bans the technology. All of it expects you to understand the tools you rely on. |
| SR008 | Harvey Academy | Library | Harvey Academy | See how a Harvey Agent researches liability exposure after a vendor data breach, mapping statutory damages, regulatory fines, and indemnity recovery into a brief for the incident response team. |
| SR009 | Value Add VC | How Is Harvey AI Worth $11B? The Legal AI Breakdown | The core risk is vertical-AI defensibility: Harvey doesn't own a frontier model... Add well-funded competitors and pricing pressure, and the multiple leaves zero room for a stumble. |
| SV001 | Teahose | Harvey AI Valuation (2026): $11B — Four Rounds in Twelve Months | At ~$190M ARR the round prices Harvey near 58x revenue, betting on continued doubling, single-platform standardization, and incumbents not commoditizing the layer. |
| SV003 | Harvey AI | Harvey Raises $550M at a $15.5B Valuation to Help Legal Teams Own Their Intelligence | Today, Harvey announced a $550M funding round at a $15.5B valuation, co-led by Diffusion and Lightspeed Venture Partners. |
| SV004 | Connected Dots (Medium) | Someone is Smoking Crack, or Harvey AI is Just a Proxy to Participate in an OpenAI $1 Trillion IPO? | No, the issue this time is basic freaking MATH and sound financial investing principles. The facts are simple, if Harvey tried to go public at an $8 Billion valuation on $100m in revenue, that would be an 80x multiplier on revenue to valuation. |
| SV005 | Silicon Valley Investclub | Harvey AI — Company profile — Silicon Valley Investclub | Valuation (Sep 2026)$15.5B ... ARR (Jan 2026)$190M ... Total Funding$1.7B ... Customers (Mar 2026)1,300+ ... Countries60+ |
| SV006 | Harvey AI | Harvey – Professional Class AI | 2,400+ Legal Organizations Build on :Harvey: |
| SV007 | ValueAdd VC | How Is Harvey AI Worth $11B? The Legal AI Breakdown | At ~$11B on ~$190M ARR, Harvey trades near 58x revenue — pricing in years of flawless execution. |
| SV008 | Capitaly | How Harvey AI Became a Billion-Dollar Legal Platform | Competitive moat: By owning the relationship with law firms and in-house counsel, Harvey had built switching costs. Customers weren't going to rip out Harvey and replace it with a competitor; they were going to deepen integration. |
| SV010 | Harvey AI | How to Use AI as a Lawyer: The Workflows, Risks, and Rules | It can't take responsibility for legal advice, and it can't stand in for you in front of a client or a court. The judgment stays with you. |