DLA Piper Deploys AI Platform to Automate Legal Contract Work

DLA Piper部署AI平台实现法律合同工作自动化 与利用企业级人工智能扩大利润空间并开辟新客户收入流相比，仅关注单年效率提升是一项失败的策略。 年营收超过40亿美元、拥有4000多名律师的DLA Piper选择积极拥抱生成式AI作为战略优势，而非将其视为对传统按小时计费模式的生存威胁。该律所围绕后台业务运营、律师个人生产力和内部软件创新团队三大核心方向构建企业AI方案，并组建了由约24名工程师、大语言模型专家及数据科学家组成的内部团队开发自主解决方案。目前该律所超过50%的法律事务采用替代性收费协议（AFA），使财务激励与AI驱动的运营效率相统一，同时严格恪守数据隐私界限，绝不使用企业客户的敏感数据训练外部或通用AI模型。 全球大型律所 DLA Piper（年营收超 40 亿美元、拥有 4,000 多名律师）将生成式 AI 转化为核心战略优势，构建了涵盖后台业务、律师生产力工具及 20 多人内部软件开发团队的三轨体系，并依托超 50% 的替代性收费模式（AFA）摆脱传统计时收费的束缚。 (0:21) 在诉讼业务场景中，DLA Piper 的自研 AI 平台在 36 小时内完成海量突击证据审查，排除了不合资格的集体诉讼成员并直接减少逾 200 万美元和解支出；同时在大规模侵权诉讼中替代离岸人工审查团队，帮客户削减了 50% 的数据挖掘成本。 (0:09) 针对复杂商业合同梳理，该 AI 平台在约 90 天内完成了原本预计耗费 10 万律师工时的 50 万份合同族系分析，不仅实现逾 90% 的法律成本缩减，还助力客户提前数年落地每年 1 亿美元的经常性业务收入。 (0:09) DLA Piper 强调单纯追求单年效率是不可持续的策略，企业 AI 的目标在于将重复性劳务转化为高溢价战略洞察、扩大利润率并开辟新收入流，且整个 AI 落地过程中未裁减任何内部或客户岗位。 (1:17)

Stephen Henn, Managing Director of AI Technology Services at DLA Piper, argues that law firms must move beyond basic efficiency metrics to deploy generative AI as a driver of client revenue, margin expansion, and higher-value advisory work. Aligning Enterprise Strategy and Billing Structures (0:21) Rather than treating generative AI as an incremental productivity tool or a threat to billable hours, DLA Piper organized its approach around three operational areas: back-office business operations, individual attorney productivity, and proprietary software development. The firm maintains strict policies ensuring client data is never used to train external models, while rolling out over 1,500 licenses for platforms like Harvey, alongside CoCounsel and Microsoft 365 Copilot. Crucially, over 50% of the firm's matters already utilize alternative fee arrangements rather than standard hourly billing. According to Henn, this fee model aligns the firm's financial outcomes with software-driven speed, allowing DLA Piper to protect its margins while automating repetitive legal tasks across its more than 4,000 attorneys. In-House Engineering to Bridge Legal Tech Gaps (0:07) To counter pressures from Big Law rivals investing hundreds of millions in technology, commercial legal-tech vendors, and enterprise clients like Wells Fargo building in-house tools, DLA Piper established its own internal software unit. Operating with about two dozen software engineers, data scientists, LLM specialists, and legal practitioners, the unit develops and patents custom software to fill capability gaps left by off-the-shelf products. Henn explains that while combining fast-moving software development cultures with traditional legal practice generates friction, building domain-specific, proprietary technology provides the firm with a defensible competitive moat against market peers. Quantifying High-Impact Returns in Complex Matters (0:09) Henn categorizes the firm's AI returns by the scale of financial impact delivered to clients. At the seven-figure level, the firm's platform analyzed a last-minute data dump within a 36-hour court deadline, identifying ineligible class-action participants to cut a settlement payout by over $2 million. In another mass-tort engagement, replacing an offshore manual document-mining team with AI halved client costs while opening a new revenue stream for the firm. Larger scale engagements yielded eight- and nine-figure impacts. For a transactional team reviewing 250-page deal documents, custom AI inferred unstated contractual structures, cutting review times by up to 90%. In a corporate product launch, DLA Piper analyzed 500,000 interlinked contract documents in roughly 90 days—a scope previously estimated by a competitor at 100,000 attorney hours. This analysis reduced legal costs by over 90% and allowed the client to accelerate a $100 million recurring revenue stream years ahead of schedule. Prioritizing Margin and Advisory Value Over Headcount Cuts (1:17) Henn rejects tracking single-year efficiency gains, characterizing them as a 'loser's game' because productivity margins quickly normalize. Instead, DLA Piper evaluates technology investments by their ability to generate enterprise revenue lines, expand margins, and elevate lawyers from processing routine tasks into delivering strategic business insights. Despite the scale of automation across massive document reviews, Henn emphasizes that zero jobs have been cut at DLA Piper or within the examined client organizations, indicating that software has acted as an operational capacity multiplier rather than a replacement for human staffing. Henn frames AI deployment not as a cost-cutting measure, but as a strategic reallocation of legal talent toward complex insight; however, maintaining this model relies heavily on clients continuing to embrace alternative fee structures over traditional billable hours.

DLA Piper人工智能技术服务董事总经理Stephen Henn指出，面对生成式AI对传统按小时计费模式的冲击，律所不应仅局限于追求单年效率提升，而应通过自研专利技术、重构收费机制与释放客户商业价值，将技术创新转化为数百万至数亿美元级别的业务回报。 三轨驱动架构与自研软件创新体系 (0:21) 面对生成式AI对律所高业务量、重复性工时计费带来的冲击，拥有4,000多名律师、年营收逾40亿美元的全球第三大律所DLA Piper确立了涵盖三大运营轨道的企业级应对战略。第一轨道为“法律业务运营”，利用AI优化后勤管理与账单审计，确保严格契合90家主要客户中约80种定制计费安排；第二轨道为“律师个人生产力”，部署了包括1,500多套Harvey许可证以及CoCounsel和Microsoft 365 Copilot在内的商用工具，并配合严格的合规与道德培训，以防止引发声誉风险；第三轨道则是在律所内部设立独立的自研软件开发团队。 该自研团队由约20名软件工程师、大语言模型专家、数据科学家及法律执业人员组成，专门填补现成商用软件与复杂业务需求之间的缺口。为构筑竞争壁垒，DLA Piper开发并获得了专有企业级AI技术栈的专利保护。在数据安全与收费机制方面，律所严格执行不使用任何客户数据训练外部或通用模型的政策，且超过50%的法律业务已采用替代性计费结构（AFA），使机构的盈利动力与AI驱动的高效作业保持一致。 多维竞争格局与跨界技术整合壁垒 (0:07) Henn指出，驱动DLA Piper持续加大技术投入的核心动力来自三个维度的外部竞争压力。首先是大型律师事务所之间的直接博弈，面对同行宣布投入高达5亿美元打造专有AI平台的资本动作，DLA Piper认为这印证了行业方向，并强调自身具备起步更早的先发优势；其次是来自法律科技软件供应商的挑战，律所需要弥合习惯于初创节奏的商业软件工程师与传统律师之间的文化张力，而律所内部的跨界团队成功将技术与深厚法律专业知识深度融合。 第三项关键压力则源于企业客户内部法务部门的技术演进，如富国银行（Wells Fargo）等大型金融机构正加速部署内部AI解决方案。DLA Piper认为，外部律所必须在技术能力上走在客户前列或与之同步，才能持续维持战略级核心合作伙伴的地位，避免因客户自研能力增强而被边缘化。 验证企业ROI：从数百万到数亿美元实证 (0:09) DLA Piper通过多个诉讼与交易案例展示了AI对客户财务与业务的直接赋能。在诉讼业务中，AI展现了七位数级别的财务成效：面对原告律所在庭审前36小时突发提交包含数十万份文件的“数据倾倒”，人工团队无法按时完成审查，而DLA Piper的自研AI平台在36小时窗口内识别出对方试图混入集体诉讼的不合资格人员，直接为客户削减了超过200万美元的和解金赔付；在多地区诉讼（MDL）与大众侵权案件中，律所用自研AI替代了客户原先雇佣的离岸手动数据挖掘团队，在保证更高准确度的同时降低了50%的成本，不仅为客户节省费用，还为律所开辟了全新的收入线。 在更高层级的复杂交易中，AI创造了八位数至九位数的经济价值。针对平均达250页的数千份复杂交易文件，DLA Piper构建的AI工作流能从多章节上下文中准确推断未明确注明的“美式”、“欧式”或“混合”法律架构，将律师审查时间缩减最高达90%；在另一项涉及新产品上线的审查中，AI平台在约90天内完成了传统竞品律所预估需10万工时（约合50人年）的50多万份关联合同审查，不仅实现法律成本降低逾90%，更助推客户提前数年实现了每年1亿美元的可持续经常性收入流。 摆脱单纯效率陷阱与重构人力资本价值 (1:17) Henn在总结中明确反对将AI的应用局限于单一维度的“效率提升”，直言追求单年效率收益是难以年年复现的“失败者游戏”。DLA Piper衡量AI成效的核心标准，在于其能否有效扩大机构的整体利润率、解锁客户新的营收增长，并推动律师角色从从事重复性数据处理的基础操作者，升级为提供高阶商业战略洞察的顾问。律所的目标是将高价值的专业见解货币化，而非出售基础劳动力。 在人力资本结构方面，Henn特别强调，在DLA Piper内部及其客户组织中，上述AI工具的大规模部署并未导致任何工作岗位被削减。律所的核心战略始终聚焦于利用技术驱动营收增长和提升利润率，而非通过裁员缩减成本，从而实现了技术创新与员工队伍稳定的平衡。 虽然DLA Piper展示了AI在大规模合同与证据审查中的显著财务回报，但该模式高度依托于律所雄厚的资本积累、专职自研工程团队以及超过半数的非工时计费业务基础。对于缺乏自研技术能力、受制于传统按小时计费模式或合规预算受限的中小型律所而言，如何复制此类专有技术壁垒并化解工时缩减对短期营收的冲击，仍是法律行业普遍面临的未决挑战。

## Background 该内容由 Ai4 于 2026 年 8 月 17 日发布在 YouTube 平台上，记录了欧华律师事务所（DLA Piper）人工智能技术服务常务董事 Stephen Henn 的主题演讲。视频简介指出，当生成式 AI 被普遍视作传统法律执业的生存威胁时，DLA Piper 另辟蹊径将其转化为战略优势，通过打造企业级 AI 平台与自研技术落地，为律所及客户创造可量化的财务与商业回报，走出单纯提效迈向企业级 ROI。 ## TL;DR Stephen Henn 阐述了 DLA Piper 如何通过后台业务管理、个人生产力赋能与内部孵化软件研发这“三条赛道”，重塑年营收逾 40 亿美元律所的商业模式。依托内部专业的工程师团队与绝不使用客户数据训练基础模型的严格安全红线，律所打造出自研专利平台，并展示了多层级的震撼 ROI：从 36 小时内穿透分析原告突袭文件并削减 200 万美元和解金的七位数案例，到长文本非显式条款智能推演缩减 90% 审阅时间的八位数案例，再到 90 天内穿透 50 万份合同族以护航年化 1 亿美元新业务上线的九位数标杆。最后，他语出惊人地提出“效率是个输家”，呼吁企业跳出边际递减的效率思维，转向追求单笔利润率、顶线营收增长以及始终坚守零裁员底线的高阶商业价值。 ## Atmosphere & style 整场演讲洋溢着一位自称“正在康复中的律师”特有的风趣、犀利与从容气场，兼具顶尖律所合伙人的宏观视野与一线软件团队的极客敏锐。主讲人以幽默化解 AI 带来的生存恐慌，在调侃竞争对手与行业沉疴的同时，步步为营地拆解数百万至数亿美元级的商业硬仗。听感极其务实且毫不悬浮，不仅打破了传统法律行业按小时计费的保守偏见，更用真刀真枪的战术穿透力，为观众带来商业认知重塑与工程落地的双重快感。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=x6drSr11S9c&t=4s) 生成式 AI 正在根本性重塑法律工作流，DLA Piper 选择主动打破传统按小时计费模式的束缚，将颠覆性威胁转化为战略增长契机。 - [01:48](https://www.youtube.com/watch?v=x6drSr11S9c&t=108s) 律所构建了清晰的三大 AI 落地赛道：优化计费合规等后台法律运营、赋能个人生产力（大规模采购 1,500 张 Harvey 许可），以及孵化内生软件工程团队填补市场空白。 - [03:37](https://www.youtube.com/watch?v=x6drSr11S9c&t=217s) 恪守受托责任与数据安全红线，DLA Piper 坚决承诺不使用任何客户数据训练基础模型，并依靠自研专利架构建立技术竞争壁垒。 - [03:37](https://www.youtube.com/watch?v=x6drSr11S9c&t=217s) 跳出单纯出售工时的传统模式，DLA Piper 已有超过 50% 的业务采用另类收费安排（AFA），实现按商业价值定价。 - [07:21](https://www.youtube.com/watch?v=x6drSr11S9c&t=441s) 律师角色正在由纯法律顾问升级为商业合伙人，将法律视为支撑企业财务报表和商业运转底层的“工程执行层”。 - [10:09](https://www.youtube.com/watch?v=x6drSr11S9c&t=609s) 在数百万美元级诉讼突袭中，自研 AI 平台在 36 小时内穿透分析数十万份文档，精准识破原告瑕疵，为客户削减逾 200 万美元和解金。 - [10:09](https://www.youtube.com/watch?v=x6drSr11S9c&t=609s) 通过 AI 自动化以极高准确率和半价成本替代低端离岸外包团队，成功夺回多区诉讼等海量数据挖掘业务，开辟全新增量收入。 - [12:09](https://www.youtube.com/watch?v=x6drSr11S9c&t=729s) 攻克非显式条款的复杂多跳法律推理难题，使资深律师在数百页核心法律文本上的审阅耗时锐减 90%，并与客户透明共享效率收益。 - [13:54](https://www.youtube.com/watch?v=x6drSr11S9c&t=834s) 在 90 天内完成超 50 万份关联合同族的深度穿透（相当于节省 50 个律师工作年），将法务成本降低 90% 以上，全力护航年化 1 亿美元增量新产品的合规上线。 - [16:54](https://www.youtube.com/watch?v=x6drSr11S9c&t=1014s) 提出“效率是个输家”的核心论断，指出效率提升边际效应递减，AI 衡量的终极度量衡应聚焦于单笔利润率、顶线营收增长，以及坚持“零裁员”的人文底线。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=x6drSr11S9c&t=4s) 重新思考商业模式：从“生存威胁”转向积极拥抱 DLA Piper（欧华律师事务所）人工智能创新常务董事 Stephen Henn 首先以轻松幽默的口吻开场。作为一名自称“正在康复中的律师”（recovering lawyer），他调侃必须先展示免责声明，而这份声明甚至是 ChatGPT 起草的。他特别指出左下角那句经典的“过往业绩不保证未来结果”，并反驳道：“不，未来的结果只会更好。” 谈及生成式 AI（Generative AI）刚爆发时的行业震荡，当时法律界充斥着“AI 将对法律执业构成生存危机”的恐慌论调。这种担忧并非空穴来风，因为 AI 从根本上重塑了工作流，极大压缩甚至消除了律师传统上按小时计费（billable hours）的大量基础工作。这意味着整个法律行业必须反思其商业模式。面对这种颠覆性冲击，DLA Piper 决定迎难而上（lean into this），绝不把 AI 仅仅当成一个辅助小工具，而是主动探索如何自保并开辟前所未有的商业空间。 --- ### [01:48](https://www.youtube.com/watch?v=x6drSr11S9c&t=108s) 全球第三大律所的 AI 布局：三大业务赛道 DLA Piper 拥有雄厚的转型资产：作为全球第三大律师事务所，年营收超过 40 亿美元，拥有 4,000 多名律师。Stephen 打趣道，这也意味着他必须向其中的 1,200 位合伙人“老板”汇报工作，而令人欣慰的是，这些合伙人展现出了极高的支持度。基于庞大的组织体量，律所将 AI 的应用划分为清晰的“三条赛道”： ```mermaid flowchart TD accTitle: DLA Piper AI Application Strategy accDescription: DLA Piper's three-track approach to enterprise AI integration. subgraph ThreeLanes ["DLA Piper AI 落地三大赛道"] L1["后台法律业务<br/>(Business of Law)<br/>计费管理与客户合规"] L2["个人生产力赋能<br/>(Individual Productivity)<br/>1500张 Harvey 许可与伦理培训"] L3["内生软件工程创新<br/>(Proprietary Software)<br/>自研专利技术与市场填空"] end ThreeLanes --> Target["提升利润率 & 驱动客户商业价值"] ``` 第一条赛道是“后台法律业务”（Business of Law）。这涵盖案件管理、客户计费等运营后端。律所的计费结构极其繁杂——“如果看 90 个客户，可能就有 80 种不同的计费安排”。通过 AI 实现账单合规与管理，极大消除了摩擦，确保严丝合缝地遵循客户指引。 第二条赛道是“个人生产力赋能”。DLA Piper 大规模部署了 Harvey（采购了 1,500 张许可证），同时配合 CoCounsel、Copilot 等 AI 工具。但这伴随着极严格的培训机制，确保每位律师负责任、合乎伦理地使用技术。正如业内同仁所警告的，如果在《纽约时报》头版因为违规使用 AI 闹出大丑闻，那么再多的技术投资都得不偿失。 --- ### [03:37](https://www.youtube.com/watch?v=x6drSr11S9c&t=217s) 律所内的软件开发团队与数据安全壁垒 第三条赛道也是 DLA Piper 真正与众不同的地方：律所决定打破传统边界，在内部孵化一家专门的软件开发团队。Stephen 所在的技术团队汇集了约二十多位专业人才，包括软件工程师、大语言模型（LLM）工程师、数据科学家以及法律实务专家。该团队的唯一目标就是找出“市场现有工具与客户真实需求之间的空白”，并亲自编码填补鸿沟。 作为顶级律所，他们拥有得天独厚的优势——接触海量高度机密和敏感的客户数据。但这同时也是沉重的受托责任（fiduciary responsibility）。为此，团队建立了绝对严密的网络与数据安全架构： - **坚决不使用客户数据训练任何基础模型**：“我们绝不拿数据去训模型，永远不会。”尽管从纯商业视角看，将所有零售或医药客户的数据打通聚合训练将带来巨大技术优势，但受托责任底线不容践踏，目前必须坚守极其审慎的合规红线。 - **积极参与行业政策与治理**：深度参与联合国“AI for Good”倡议，并荣获钱伯斯（Chambers）等权威机构在 AI 创新领域的顶尖评级。 - **推动计费文化与创新**：行业内不少机构仅有约 20% 的案件采用另类收费安排（AFA），而在 DLA Piper，**超过 50% 的业务已经属于另类收费安排**，早已大幅跳出单纯出卖工时的商业逻辑。 - **专利构筑技术护城河**：团队研发的核心 AI 架构已申请专利并拥有自主知识产权，建立起同业难以复制的竞争壁垒。 --- ### [07:21](https://www.youtube.com/watch?v=x6drSr11S9c&t=441s) 应对三方竞合与“法律即业务执行层”的定位升级 团队在规划技术路线图时，坦诚地剖析了面对的三大竞争力量： 1. **其他大型跨国律所（Big Law）**：例如近期有同级别竞争对手高调宣布斥资 5 亿美元打造自研 AI 平台进军市场，Stephen 戏谑地引用电影台词回应：“欢迎来到派对，伙计（Welcome to the party, pal），祝你好运。” 2. **专业法律科技软件公司**：这类竞争充满了跨界摩擦。习惯于初创公司快节奏的工程师与传统律所里严谨保守的律师在一起工作，文化张力与摩擦不可避免。 3. **客户自身的法务与技术团队**：例如富国银行等大型机构客户完全有能力自主开发部署内部方案。DLA Piper 不希望被客户甩在身后，而是必须走在前方，成为客户不可或缺的共创伙伴。 这种竞争推导出一个核心认知转型：**律师不能只做法律顾问，而必须成为真正的商业顾问**。 > “在一家公司的财务报表中体现出任何数字之前，它首先在法律文件中表现为一项风险、一次机遇或一种法定义务。法律才是让商业飞轮运转起来的工程执行层。” --- ### [10:09](https://www.youtube.com/watch?v=x6drSr11S9c&t=609s) 七位数影响（百万美元级）：诉讼突袭防御与离岸替代 Stephen 通过具体实战案例展示了技术落地的财务影响，首先是数百万美元级的“七位数案例”： 第一个故事发生在某个周四深夜。原告律所突然发起“数据倾倒”战术（data dump），向 DLA Piper 扔过来数十万份文档，并放话“36 小时后法庭见”。面对如此庞大的文档量，即便是调集全所 4,000 名律师进行人工肉眼审查也绝无可能完成。团队立刻将全套卷宗注入自研专利 AI 平台，在 36 小时内完成了深度穿透分析。结果发现，原告律所试图浑水摸鱼，悄悄塞进了大量根本不符合集体诉讼资质的申请人。凭借这一精准打击，团队在开庭前直接**为客户削减了超过 200 万美元的和解金额**，令原告方措手不及。 第二个故事展示了通过 AI 开辟增量业务（new revenue stream）。在涉及多区诉讼（MDL）或大规模侵权（mass torts）的案件中，通常需要海量的数据挖掘工作。客户原本习惯将这类繁重低级的人工工作外包给低成本离岸团队。DLA Piper 主动出击，向客户承诺：“我们可以保证提供远高于人工的准确率，而收费只需原外包成本的一半。”最终客户果断签约。通过 AI 自动化，律所重新拿回了以往因自身高单价而被挤出市场的低端长尾业务，不仅开辟了新利润源，还帮客户**直接节省了 50% 的账单开销**。 --- ### [12:09](https://www.youtube.com/watch?v=x6drSr11S9c&t=729s) 八位数影响（千万美元级）：复杂长文本与非显式条款的智能推演 步入千万美元级的“八位数案例”，团队处理的是极其复杂的交易分析业务。某个大型实务组日常需要审查成千上万份单份长达 250 页的核心法律文本，以评估底层资产与交易是否具备推进价值。 最棘手的技术难题在于法律语境的非显式表达。例如，合同中涉及某项机制到底是采用“美式方法”（American approach）、“欧式方法”还是“混合方法”（hybrid approach）。但在实际合同条款中，“美式”或“欧式”这类关键词根本不会直接出现。模型必须通读多处不同条款的逻辑交互，进行多跳法律推理才能判定其属性分类。 团队构建的 AI 解决方案攻克了该难题，**将资深律师的审阅耗时大幅缩减了高达 90%**。在商业回馈上，律所将这部分效率红利透明地与客户共享。客户非常理解律所合理的商业盈利诉求，愿意按实际业务价值进行溢价激励，双方共同享受时间和资金成本的大幅释放。 --- ### [13:54](https://www.youtube.com/watch?v=x6drSr11S9c&t=834s) 九位数影响（亿美元级）：50 万份合同穿透与年化 1 亿新产品上线 最震撼的是涉及数亿美元级影响的“九位数项目”。某大型企业客户计划上线一款重大新产品，但必须事先厘清其现存的庞大合同库中是否存在展业授权或权利限制冲突。审查面临的实际挑战极其严峻： - **审查总量庞大**：涉及超过 50 万份有效合同。 - **关联脉络错综复杂**：并非一户一约，每个业务单元平均关联约 7 份衍生文件形成的“合同族”（family），包括原始协议、第一至第四修正案、多语言翻译件以及抽屉式补充协议（side agreements）。 - **人工成本高昂且耗时过长**：竞品律所测算，纯人工完成该审查**需要 10 万个律师工时（约合 50 个律师工作年）**。即便部分工作转交薪酬较低的律师助理，乘以常规计费费率后依然是无法承受的天文数字。 - **商业延误代价沉重**：该新产品一旦顺利推向市场，将带来**每年 1 亿美元的净增量收入**。因此，快速上线的战略意义远超单纯节省法务开销。 团队通过自研平台介入后，在**约 90 天内完成了全量分析，将法务审查成本降低了 90% 以上**。更关键的是，如果不借助这一自动化架构，该客户原本的策略是完全放弃前期律师全面审查，仅做微量的事后抽检，承受着极大的潜在暴雷风险。AI 使客户能够以极高的确定性和底气，提前将年入 1 亿美元的业务稳健推向市场。 --- ### [16:54](https://www.youtube.com/watch?v=x6drSr11S9c&t=1014s) 终极度量衡：“效率是个输家”，聚焦利润与零裁员 在总结 ROI 衡量标准时，Stephen 语出惊人地提出了一个核心反思观点：“我讨厌‘效率’，效率是个输家（Efficiency is a loser）。” ```mermaid flowchart LR accTitle: AI Impact Hierarchy accDescription: Hierarchy of AI business outcomes from low-value efficiency to high-value revenue and retention. E["效率提升 (Efficiency)<br/>边际效应递减 / 一次性收益"] -->|升级| M["利润率 (Margin)<br/>单笔交易盈利能力跃迁"] M -->|跃迁| R["业务营收 (Revenue)<br/>创造亿美元级增量与战略伙伴关系"] R -->|底线| J["人文与团队 (Zero Job Loss)<br/>不以削减岗位为目的"] ``` 他深入阐述道，效率提升往往是“一次性”的（one and done）。即便第一年通过工具挤出了 20% 的效率冗余，未来十年也不可能持续挤压下去，绝大多数效率红利在最初 12 个月内便消耗殆尽。因此，企业绝不能仅仅追逐效率指标，而应该将 AI 的度量衡锚定在更高层级的维度： - **利润率（Margin）**：不论是律所还是企业客户，能否通过 AI 提升每一次商业交付的单位盈利能力。 - **顶线收入（Revenue）**：与客户并肩作战，通过释放新业务、加速上市流程直接驱动收入增长，从而让外部律师真正晋级为核心“商业合伙人”。 - **变现核心洞见**：重塑律所价值主张，确保体现在 42 亿美元营收报表中的，是高含金量的核心“法律洞见”（legal insight），而非低技术含量的流水线初级劳务。 - **人文导向与零裁员理念**：Stephen 自豪地强调，无论是 DLA Piper 还是其客户项目中，**因实施 AI 导致被裁撤的岗位数量是：零（Zero）**。 > “在与客户的所有探讨中，我们的核心共识是绝不把‘将员工优化出局’作为目标。我们真正专注要工程化实现的，是收入、是利润，然后顺带收获一些效率。” ## Notable quotes > [00:04](https://www.youtube.com/watch?v=x6drSr11S9c&t=4s) 不，未来的结果只会更好。 — Stephen Henn > [01:48](https://www.youtube.com/watch?v=x6drSr11S9c&t=108s) 如果看 90 个客户，可能就有 80 种不同的计费安排 — Stephen Henn > [03:37](https://www.youtube.com/watch?v=x6drSr11S9c&t=217s) 我们绝不拿数据去训模型，永远不会。 — Stephen Henn > [07:21](https://www.youtube.com/watch?v=x6drSr11S9c&t=441s) 欢迎来到派对，伙计，祝你好运。 — Stephen Henn > [07:21](https://www.youtube.com/watch?v=x6drSr11S9c&t=441s) 在一家公司的财务报表中体现出任何数字之前，它首先在法律文件中表现为一项风险、一次机遇或一种法定义务。法律才是让商业飞轮运转起来的工程执行层。 — Stephen Henn > [10:09](https://www.youtube.com/watch?v=x6drSr11S9c&t=609s) 我们可以保证提供远高于人工的准确率，而收费只需原外包成本的一半。 — Stephen Henn > [16:54](https://www.youtube.com/watch?v=x6drSr11S9c&t=1014s) 我讨厌‘效率’，效率是个输家 — Stephen Henn > [16:54](https://www.youtube.com/watch?v=x6drSr11S9c&t=1014s) 在与客户的所有探讨中，我们的核心共识是绝不把‘将员工优化出局’作为目标。我们真正专注要工程化实现的，是收入、是利润，然后顺带收获一些效率。 — Stephen Henn ## Mentioned Stephen Henn, DLA Piper, ChatGPT, Harvey, CoCounsel, Copilot, 钱伯斯 (Chambers), 联合国 AI for Good 倡议, 富国银行 (Wells Fargo) ## Video description Stephen Henn - Managing Director, AI Technology Services, DLA Piper LLP When generative AI emerged, many law firms viewed it as an existential threat to the traditional practice of law. DLA Piper chose a different path. This session tells the story of how the firm transformed AI from a disruptive force into a strategic advantage—building an enterprise AI platform that delivers measurable business outcomes for both the firm and its clients. Rather than focusing on hype or future possibilities, this presentation shares real-world examples of AI deployments that have generated measurable financial and business impact through improved legal operations, accelerated client outcomes, and new service models. Attendees will leave with practical insights into AI strategy, governance, organizational change, and how to move beyond productivity gains to enterprise-scale ROI. [Source](https://www.youtube.com/watch?v=x6drSr11S9c)

[0:04] Stephen Henn: 不，未来的结果只会更好。 [1:48] Stephen Henn: 如果看 90 个客户，可能就有 80 种不同的计费安排 [3:37] Stephen Henn: 我们绝不拿数据去训模型，永远不会。 [7:21] Stephen Henn: 欢迎来到派对，伙计，祝你好运。 [7:21] Stephen Henn: 在一家公司的财务报表中体现出任何数字之前，它首先在法律文件中表现为一项风险、一次机遇或一种法定义务。法律才是让商业飞轮运转起来的工程执行层。 [10:09] Stephen Henn: 我们可以保证提供远高于人工的准确率，而收费只需原外包成本的一半。 [16:54] Stephen Henn: 我讨厌‘效率’，效率是个输家 [16:54] Stephen Henn: 在与客户的所有探讨中，我们的核心共识是绝不把‘将员工优化出局’作为目标。我们真正专注要工程化实现的，是收入、是利润，然后顺带收获一些效率。

[0:04] Stephen Henn: past results do not guarantee future outcomes [0:21] Stephen Henn: above the fold in the New York Times [0:07] Stephen Henn: Welcome to the party, pal

[0:04] Um again I am Steve Hen. I am managing director  of AI innovation at DA Piper and I am here to [0:11] talk to you and build on the presentations that  you may have seen earlier as well too. So here [0:16] I'd really like to talk about impact ROI and some  of the success stories that we've seen in terms of [0:22] our use of AI both within the firm and within  our clients. But first, since I am a lawyer, [0:30] recovering lawyer, I got to do the disclaimers.  Now, I will say that this is Chat GPT's version [0:35] of disclaimer. So, I really, really like the  one on the lower left that says past results [0:41] do not guarantee future outcomes. My comment  to that is no, future outcomes will be better. [0:47] So um just to recap a little bit to to build up to  what we're talking about here is that when Genda [0:54] AI came out, one of the big things about um its  application to the practice of law was that there [1:01] were a lot of stories about how this is going to  be an existential threat to the practice of law. [1:06] And you know there are a lot of very good reasons  for that, right? Um it fundamentally changes [1:12] the workflow. um fundamentally potentially  gets rid of a lot of things that lawyers do [1:18] that they've traditionally build out. So it  involves us to rethink the business model [1:23] um and to take a look at other ways and other and  how to how to holistically look at the application [1:31] of AI to the benefit of the firm of its clients.  So DA to its credit to the firm's credit decided [1:37] they're going to lean into this. We're not simply  going to use it as a tool. We're going to figure [1:41] out how to protect ourselves and potentially  take a look at other things that can be done. [1:48] So, a little bit about DA. DA Piper is  the third largest law firm in the world. [1:53] We have over 4,000 attorneys. We're over four  billion dollars a year in revenue, right? So, we [1:58] have a substantial amount of assets to potentially  use. Now, what you don't see in that is um so [2:06] I have as part of this 1,200 bosses of those 4,000  lawyers that I have to report to. Um and again, [2:16] thankfully to their credit, the 1,200 bosses  have been very very supportive of this. [2:21] So, let's talk about our response. Some of these  responses that you've already seen. For example, [2:26] we decided that there were three lanes we were  going to look at with regard to AI application [2:33] of AI within the firm. The first one is the what  I call the business of law, back office functions. [2:40] Want to make sure the matter management, client  billing, things like that are done. I can tell you [2:44] that if I had if I if I looked at 90 clients, we  probably have 80 different billing arrangements. [2:51] So the use of AI in that regard makes it very very  efficient and helps our helps us and our clients [2:56] with the frustration of making sure that we are  con we are um adhering to client guidelines. [3:02] That's very important. The second and a lot of  what you've heard about today is this idea of [3:07] individual productivity. We're a big Harvey user.  We have 1500 licenses with that involves a lot of [3:14] training to make sure that the individual lawyer  uses the tools Harvey co-consel co-pilot [snorts] [3:21] other AI enabled software within the firm  responsibly and ethically. If that's not done [3:28] I think Greg said that if you know if you appear  above the fold in the New York Times about a big [3:32] problem with what you've done you know it's really  not worth it. It's not worth the investment. [3:37] And then finally a firm also did something  different which is what I'm part of. We have [3:42] decided firm decided to innovate by creating a  software development company within the law firm. [3:51] So I am part of about two dozen software  engineers, large language modu model engineers, [3:59] data scientists, legal practitioners whose sole  job it is is to take a look at what might be a gap [4:08] between what the market offers and what  we're hearing our firm our clients need [4:13] and to address that gap. Now to address that  gap, we have some really good advantages as [4:20] the third largest law firm in the world, right?  We have a massive data access. We have access [4:25] to some of the most confidential and sensitive  information that our our clients have. That's also [4:31] uh an a a challenge because as a law firm we have  a fiduciary responsibility to make sure we keep [4:38] everything in highly confidential and absolutely  develop the data security structures around the [4:45] use of that. So that means no training of models  at all ever. Right? We don't do that. Now, [4:53] there's a part of me, there's a there's a  business part of me that says that's kind of [4:57] a shame in the sense that if imagine I can take  all of my retail customers or farmer customers [5:02] and be able to pull all that data, that might  be an advantage. But we're not there yet. [5:05] We're certainly not there yet. And right now,  we've taken a very very conservative approach [5:12] to client data with regard to that.  Second step, policy leadership. [5:18] Right? We have a large governance uh element  to our AI practice as well too. We are um [5:25] we participate in UN AI for good. Um we are  a chambers recognized firm for our use in AI. [5:32] So the idea of driving the conversation  is really important for us as well too [5:37] and being part of that enables us as a software  company to think about things differently as well [5:43] too. cultural innovation and this is an important  part and it's been addressed in a number of the [5:48] different panels. This idea that the lawyers  have got to want it not only want it but think [5:54] about creative ways to use it both in the um  in in helping uh the firm and in helping our [6:00] clients. Now, one of the things I probably should  note here is that, you know, I think there was a [6:06] comment of one of the firms saying that they have  about 20% of their um matters are are alternative [6:13] fee arrangement. At DA Piper, over 50% of our  matters are alternative fee arrangements. So, [6:19] we're already moving away from the bill billable  model. And this is just another mechanism for us [6:25] to take a look at how do we drive efficiency  and make that work better for our clients. [6:29] The last thing is is competitive mode.  And this is what we've invested in. [6:35] The the the the system, the solution  that I'm about to talk to you about, um, [6:40] excuse me, that we use is proprietary and  patented. So, we're actually looking to [6:47] invest in a technology that is unique to the  industry. So, that's important as well, too. [6:56] Uh the last thing I'll say about that is I think  that um as you start to think about you know your [7:02] interaction whether you're with corporate  whether you're with a law firm as well too [7:06] there are all sorts of ways that you can invest  in your individual uniqueness utilizing AI [7:12] to give you a competitive advantage. What it takes  is creativity and the commitment to do so. Um, [7:21] one last thing I'll say here before I I go on is  that one of the calculus, one of the calculuses, [7:28] calculi calcul one one of the part of the one  part of the calculus here is that we looked [7:33] at the competitive landscape as well too and  here I need to be a little honest with you. We [7:37] identified three big competitive areas  that we needed to figure out how to address [7:44] as part of this discussion early on. First,  as you can imagine, is other big law, [7:49] right? Um there was a uh an announcement um  about a month ago where one of our big law [7:56] competitors is going to invest $500 million  in building their own AI platform and going [8:02] to the market. Our response is, "Welcome to the  party, pal. Glad to have you here. Um good luck." [8:09] The other part of this is that um the other the  other competitor is software companies. Now, [8:16] as you can imagine, thinking like a lawyer and  thinking like a software engineer is different. [8:21] Marrying those two can be a little bit tricky.  Um, I'm not going to say that either side is [8:28] prickly or not, but you can imagine that there's  cultural tension um, occasionally between being, [8:34] you know, for for engineers used to being in  a startup and uh, lawyers in a legal practice. [8:41] But the one that really drove us forward was  the third competition and that was our clients. [8:47] Our clients have the ability as we heard  from Wells Fargo to also implement solutions [8:53] and we wanted to be with them on that journey.  We didn't want them to be ahead of us. [8:58] In fact, we wanted to be perceived as an important  partner in there as well too. So this wraps up [9:04] the sort of circle that we took a look at. One is  we wanted to look at a strategic evolution here. [9:10] Lawyers as business adviserss, not simply legal  adviserss. Because as soon as you start delving [9:16] into legal documents, you start becoming a  business adviser. Before anything appears [9:24] in the financials of a company, it starts  out as a risk, opportunity, or obligation [9:31] in a legal document. So legal is the engineer  engineering part that gets business rolling. [9:37] They're the execution layer of a business.  Being able to tap that is incredibly important. [9:43] Operational excellence. We want to obviously  reduce course costs. Speed is important. Get to [9:48] that in an example. AI [clears throat] for good.  We want to make sure that whatever we're doing, [9:55] we're doing it in a proper manner. And then  client impact. As much as there's a lot of inward [10:02] um interest in driving our margins, which we  have, client impact is extremely important. So, [10:09] let me talk to you about a few success.  Let's start with a sevenf figureure impact. [10:13] Couple of successes there. Two stories related to  this. The first one on the left, client impact. [10:19] Thursday night, plaintiff's law firm comes on in,  does a data dump on us. Hundreds of thousands of [10:25] documents, says, "We'll see you in 36 hours in  front of the judge." Well, even with, you know, [10:33] uh, 4,000 lawyers, we were there's no way we were  going to be able to get through that humanly. [10:38] So we put it in our platform. We were ready in  36 hours and were able to in fact save because [10:47] not that this ever happens. The planer firm tried  to sneak in a whole bunch of of individuals who [10:52] were actually not part of the class who weren't  eligible to be part of the class. We were able [10:57] to reduce the settlement by over $2 million for  the client. They were quite surprised at that. [11:04] Uh the plaintist firm client was quite happy.  So that's our first seven figure impact. The [11:11] second one is a different additional uh different  revenue stream was for us again large litigation. [11:19] We do a lot of data mining in large litigations  right MDLs or mass tors. So, um, we had a client [11:26] that does this as a routine and we said to  them, "Look, we could do this utilizing AI." [11:32] And they said, "No, we've got a we've got a manual  process outside of the firm that we offshore that [11:37] we're perfectly happy to we're happy to continue  using." And I said, "Well, we can guarantee you [11:43] much better accuracy at half the cost." They  said, "Sign us up." Again reflecting something [11:50] um that Greg said earlier is this is business that  because of our size we have traditionally been [11:55] priced out of yet AI enables us to add this as a  line and start offering it to clients and there's [12:01] a huge benefit to them as well too. They're saving  50% on the bill to do what they need to do here. [12:09] So that is our seven figure impact. Let's try  eight figure impact. Now we're in the tens of [12:17] millions of dollars. [sighs] So we have um this  is a a share here. We have a um large practice [12:25] that crawls through documents that are about  250 pages um a piece and there thousands of [12:33] them. That's a lot of data and they need to  pull out information to be able to assess [12:39] um I'm trying to be a little vague here to  protect it but to assess uh effectively the [12:46] the you know that whether whether that document  is representing a deal that is worth pursuing. [12:52] That's probably a good way to say it. Um  there are certain very interesting things in [12:57] it. And it has a clause in it that asks about  whether an element is an American approach, [13:02] a European approach or a hybrid approach to this  particular question. Well, guess what? American, [13:09] a European or hybrid doesn't appear in  the document. That has to be ascertained [13:14] by the by other areas in the document. So, you  have to construct a way to be able to do that. [13:19] So we've been able to do that um and  reduce the lawyer time up to 90%. [13:28] Now what do we do with that?  We share that with the client [13:31] and the clients have generally been receptive,  right? We don't we you know I it's maybe a credit [13:36] to our clients. They understand that we need to  make money as well too. They want to reward us for [13:41] our efficiency. But we work with the clients and  we make sure that whatever we do, we want to make [13:47] sure that we extend both the time savings uh and  the and the money savings to our clients as well. [13:54] Can we go one more? Our nine figure impact a  case I'm current this is an issue I'm currently [13:59] working on right now. Last one was a large complex  complex contracts. This is different. This is a [14:07] client who is looking to launch a new product.  But to do that, they need to understand in their [14:12] corpus of documents whether they have the right to  be able to do that, whether they have the ability [14:17] to do that. And sometimes they do, sometimes they  don't, sometimes it's not clear. [snorts] How do [14:22] you do that over a half a million contracts?  And it wor and it's worse. Not every um every [14:30] every document doesn't necessarily u  represent a different um vendor or client. [14:38] Each particular element, each particular business  unit has about seven contracts related to it. So [14:44] you have the original agreement, you have  an amendment, you have a second amendment, [14:49] a third amendment, a fourth amendment, you have  translation agreements, you have side agreements [14:54] average of seven documents related to what we  call a family that relates to the business impact [15:00] of this. How do you do that? the estimate if you  uh based on what they did with a competitive firm, [15:07] the estimate that it would take about a 100,000  lawyer hours of work to do this manually. [15:16] Now, you might be able to put a bunch of it onto  parallegals or legal assistants at a cheaper cost, [15:22] but those are still not necessarily  um inexpensive. start doing the math [15:28] of what your average billable rate is for your law  firm and figure out how much that's going to cost. [15:35] Um, and then by the way, I think that translates  to something like 50 lawyer years, too. So, this [15:40] is a lot of lawyers working a lot of time. The  impact here is if they could get this to market, [15:46] it's a hundred million dollars a year  revenue stream. So it's not simply, you know, [15:52] it it is not simply a a co in this particular  case, this is not simply a cost savings. [15:57] It's be you're able to onboard a revenue stream  much quicker at Infinite $100 million a year. [16:05] We were able to do that in approximately 90 days.  That's the that's what we're going to do it in [16:10] at a 90 over I'm sorry, it's a well  over 90% reduction in our in cost. [16:18] Now, here's the other thing about this project. [16:22] The alternative for the company was not to  throw lawyers at it. This again was a situation [16:29] where whether it was us or a different firm, there  were no lawyers that were going to be involved. [16:33] They might do some spot-checking, but it always  is going to be after the fact. here. They can walk [16:39] into the into this new business with a high degree  of confidence that what they have is accurate [16:46] and um um and actionable in  terms of running the business. [16:54] So, I'm going to wrap up here by saying what's the  impact at a glance? When we look at this, we look [17:03] at client impact, right? And here's the thing I'm  going to tell you. Efficiency. I hate efficiency. [17:08] Efficiency is a loser. Efficiency is one and done.  If I can pull 20% out of something for a year. [17:14] I'm not doing that for the next 10 years. Usually,  you're you're done with the vast majority of your [17:20] efficiency impact the first 12 months. So, what  we look for is we want to try to impact margin, [17:27] the the ability to to drive profit on each  individual sale, whether it's our clients or or [17:33] ourselves. And then [clears throat] excuse me  revenue. How can we work with our client to [17:40] drive revenue impact because that starts getting  us not simply beyond not simply as a legal adviser [17:48] but as a business partner which is what we're  really trying to to accomplish. again firm impact. [17:54] Now we are trying to fundamentally transform  our abilities and again to the to a prior point [18:01] working on make sure that what we monetize what is  reflected in our revenue line that $4.2 2 billion [18:09] is really legal insight, not the commodity  work that lawyers are often billing for. [18:16] And then the final thing I think it's really  important. How many jobs have been lost [18:22] at the firm or our clients because of this? Zero.  No jobs have been lost. That's the other thing [18:29] that we're thinking that that we're very conscious  about. Part of part of our discussion with our [18:36] with our clients is making sure that we  don't engineer people out as the focus. [18:42] What we're trying to do is engineer revenue,  profit, and then maybe some efficiency. [18:52] And that's the end.