CB Insights Deploys AI Agents to Replace Legacy Market Research

CB Insights 部署 AI Agent 取代传统市场研究 数据提供商必须将原始数据库转化为嵌入工作流程的自动化预测智能，否则企业客户将陷入长达数周的人工分析停滞。 Jason Saltzman 概述了 CB Insights 的组织转型，旨在将长达数周的研究与商业智能（BI）周期大幅缩短为实时的规范性市场情报。该平台以追踪超过 1000 万家公司及数万亿数据点的统一商业图谱为核心，利用大语言模型（LLM）将数据接入周期从数月压缩至数周，并实现了并购尽调和 SWOT 分析等复杂工作流的自动化。Saltzman 将传统数据销售比作提供未提炼的金矿石，指出数据提供商如今必须直接为用户完成数据提取、上下文映射和分析综合。为实现这一目标，CB Insights 依靠四层架构，将结构化情报无缝嵌入到包括 Microsoft Copilot、Salesforce Agentforce 和 Snowflake Cortex 在内的日常企业工作流中。 CB Insights 完成从传统数据服务商向 AI 预测情报平台的转型，旨在打破需要数周时间的手动清洗与静态看板分析瘫痪，实现即时战略洞察与决策建议。 (0:04) 平台构建了涵盖超 1000 万家公司及数万亿数据点的统一商业图谱，摆脱仅出售“原始矿石”的淘金制图师模式，直接为客户自动提炼并输出高价值成果。 (3:21) 通过引入大语言模型（LLM）实现结构化与非结构化数据的自动化摄取，CB Insights 将新数据集的接入周期从 6–12 个月压缩至短短数周，并赋能并购筛选与自动化 SWOT 分析等核心工作流。 (6:02) 该体系涵盖数据获取、AI 映射、动态洞察与嵌入式分发四层架构，并将预测情报深度集成至 Microsoft Copilot、Salesforce Agentforce 和 Snowflake Cortex 等第三方工作流中，避免上下文切换。 (10:15)

According to Jason Saltzman of CB Insights, enterprise data vendors must evolve from selling raw, fragmented databases to delivering automated, proactive intelligence. By integrating large language models across a unified business graph, the platform aims to collapse multi-week research cycles into on-demand strategic recommendations embedded directly in client workflows. Replacing Raw Data Extraction with Automated Synthesis (0:04) Traditionally, market intelligence platforms required clients to endure multi-week cycles across data engineering, analyst synthesis, and business intelligence dashboard creation, resulting in analysis paralysis. Saltzman likens legacy data vendors to gold mining cartographers who merely sold access to riverbeds of raw dirt and mud, forcing customers to hire crews and sift through rock for nuggets. By anchoring its infrastructure to a unified business graph tracking over 10 million companies and trillions of data points, CB Insights uses AI to automate the sifting process. Rather than delivering static snapshots or expecting users to build custom analytics pipelines, the platform synthesizes data against specific business contexts. The stated objective is to build market intelligence that continuously monitors developments, explains why they matter, and prescribes concrete actions in response to natural language inquiries. Accelerating Ingestion Velocity and Autonomous Due Diligence (6:02) To build an active business ontology, CB Insights deployed large language models to automate the extraction and structuring of public web announcements as well as proprietary qualitative inputs, such as founder interviews, analyst survey protocols, and investor networks. This automated pipeline compressed the timeline required to onboard and link new external datasets from 6 to 12 months down to a few weeks. This structured foundation supports automated, high-stakes diligence workflows. For corporate development teams assessing acquisition targets in emerging sectors like voice AI or text-to-video, autonomous agents traverse company relationship graphs to immediately generate prioritized candidate lists, competitive stack rankings, and automated SWOT analyses, bypassing months of manual landscape mapping. Embedding Four-Layer Intelligence into Enterprise Workflows (10:15) Saltzman outlines a four-layer architecture required to make predictive intelligence actionable: automated data acquisition, AI-driven entity and taxonomy mapping, a dynamic insights layer that eliminates static publication bottlenecks, and a multi-channel distribution layer. Rather than trapping intelligence inside a standalone terminal, the strategy embeds outputs directly into third-party enterprise tools, including Microsoft Copilot, Salesforce Agentforce, and Snowflake Cortex. Delivering context directly into existing platforms is intended to prevent context-switching and provide validated intelligence where decision-makers already operate. Implementation Strategy, Quality Controls, and Market Expansion (14:26) For data providers seeking to become intelligence partners, Saltzman recommends auditing existing workflows to locate data silos, decision latency, and repetitive manual post-processing. Following a "think big, start small" philosophy, organizations should isolate and automate high-value, recurring analytical tasks using specialized models before launching enterprise-wide deployments. Simplifying data access has expanded CB Insights' addressable audience beyond traditional private equity investors and corporate strategy units to include go-to-market teams and management consultants. However, Saltzman emphasizes that this model depends on strict data quality controls: because AI systems operate on a "garbage in, garbage out" basis, validated, clean proprietary data remains the indispensable guardrail for trustworthy predictive output. Saltzman concludes that data enterprises must deploy AI across the entire stack—from ingestion to workflow distribution—to bridge the gap between information and execution. However, the ultimate reliability of autonomous market intelligence remains strictly constrained by the quality and continuous validation of the underlying proprietary data assets.

CB Insights 的 Jason Saltzman 指出，随着企业分析工具激增引发决策瘫痪，传统数据服务商必须从被动交付海量原始数据转向提供端到端预测性智能。通过构建覆盖逾千万家企业的商业知识图谱并全栈引入大语言模型与智能体，企业能够将耗时数月的调研管线压缩至实时交付，将提炼后的洞察直接嵌入客户现有工作流。 传统分析管线瓶颈与即时决策需求 (0:04) 传统商业智能的数据供应链高度割裂且消耗资源。在过往流程中，数据工程师通常需要数天时间查找与清洗分散的数据集，行业分析师耗费数周在商业智能工具中解读并生成报表，管理层最终仅能拿到过时的静态快照。随着分析工具爆发式增长，繁杂的数据并未提升决策效率，反而导致严重的“分析瘫痪”（analysis paralysis）。在市场格局瞬息万变的背景下，企业难以承受长达数周的人工分析周期。Saltzman 强调，面对这种变化速度，“必须依靠 AI 才能跟上 AI 的发展步伐”。 为了消除原始数据与业务行动之间的脱节，CB Insights 在过去两年开展结构性转型，引入新技术团队并在业务全流程整合生成式 AI。其目标不再是单纯向企业客户提供庞大的专有数据库和定期发布的分析师报告，而是通过大语言模型与智能体工作流直接赋能底层数据，彻底瓦解传统中间环节，直接为决策者提供具备上下文背景的即时市场洞察与行动建议。 商业图谱构建与“淘金炼金”的比喻 (3:21) 为消除分析链路中的摩擦，CB Insights 构建了围绕统一商业图谱的基础设施，追踪超过 1000 万家科技与新兴市场企业以及数万亿个数据点。该图谱将新兴技术、企业创新动态、投资人网络、竞争格局和合作伙伴生态紧密互联，允许用户结合特定业务情境进行自然语言交互。其产品负责人将这一愿景概括为：打造一套“在用户睡眠时持续运转、如影随形提示发生何事、阐释其重要性并指明应对举措”的市场智能体系。 Saltzman 用早期淘金业的地图绘制来类比传统数据行业的缺陷：过去的数据提供商仅向客户售卖地图以及充满泥沙杂石的河道开采权，客户必须自备筛子、雇佣采金队在矿渣中艰难寻觅金块。而融入专用 AI 模型后，平台颠覆了这一协作模式——由自动化系统负责定位高价值矿藏并完成沉重的洗沙筛选，直接向客户交付提炼纯净的“金块”乃至定制尺寸的“现成戒指”。 自主数据摄取与自动化尽职调查 (6:02) 两年前，CB Insights 对人工智能的参与仅限于分析师层面的行业研究，内部尚未部署相关技术。随后，公司通过招募具备 AI 客服和聊天机器人架构背景的技术主管，全面改造底层数据基础设施。团队运用大语言模型自动化摄取与结构化两类数据源：一类是新闻资讯与合作公告等公开信息，另一类则是分析师调研、专有创始人访谈和投资人交流等非公开质性数据。 这种以自动化 AI 管线取代人工录入的做法，将新外部数据集的接入与结构化周期从 6 至 12 个月压缩至数周。这种结构化本体大幅赋能了高门槛的战略决策场景。例如，当 Meta 等企业寻求语音 AI 或文本生成视频领域的并购标的时，企业发展团队无需再耗费数月在分散工具间拼凑全景图，AI 智能体即可遍历企业档案、分类与关系网络，瞬间生成具有优先级排序的候选名单、竞品排名以及全自动的 SWOT 分析。 四层价值架构与无缝嵌入工作流 (10:15) CB Insights 提出预测性智能的落地依托于四个核心运行层级：第一层是数据获取层，通过大语言模型将数月的人工工程周期缩短为数周；第二层是 AI 增强的数据映射层，将分散的公司实体、投资关系与分类法构建为高度互联的商业本体；第三层是动态洞察层，打破静态研究报告的发布排期限制，支持用户以自然语言对垂直利基市场进行深度提问；第四层则是分发层，直接提供成品级智能与规范性决策，摆脱对原始数据导出的依赖。 为了让预测性情报真正发挥效用，Saltzman 指出智能必须嵌入专业人员日常工作的核心系统，而非将其束缚在孤立的独立门户中。除了提供统一的集中式终端外，CB Insights 推进嵌入式分发，通过合作将情报直接集成至 Microsoft Copilot、Salesforce Agentforce 和 Snowflake Cortex 等主流企业工作流与智能体平台，使用户无需在割裂的应用间频繁切换上下文即可获取经过验证的商业洞察。 拓展服务客群与数据质量红线 (14:26) 从原始数据提供商转变为主动型智能伙伴，从根本上拓宽了数据企业的潜在市场空间（TAM）。CB Insights 以往主要服务于私募股权投资机构与企业战略部门，但在降低数据质询门槛后，其服务对象已延伸至市场拓展（GTM）团队、被投企业价值创造团队和管理咨询顾问。关于落地方法，Saltzman 提出了“大处着眼、小处着手”（think big, start small）的原则，建议企业先审查自身价值链，找出决策延迟最严重、手动后处理最繁琐的高价值重复分析流程，优先通过专用模型和智能体实施自动化改造。 不过，全栈 AI 的有效性高度依赖严苛的数据质量控制。Saltzman 警示，人工智能系统始终严格遵循“垃圾进，垃圾出”（garbage in, garbage out）的规律；唯有将经过严密清洗与高度验证的专有数据作为底层基石与安全护栏，预测性智能才能具备可靠的决策参考价值，真正帮助客户消除信息获取与战略执行之间的鸿沟。 Saltzman 总结认为，现代数据企业必须在数据提取、映射、高级分析及工作流交付的全栈各个环节深度部署 AI，承担起端到端的“精炼”责任，紧密贴近客户最终要解决的业务问题。然而，预测性智能与自主智能体的成效完全受制于底层专有数据的真实性与质量；在缺乏严密验证的数据基底之上，自动化分析不仅无法消除决策延迟，反而可能放大潜在风险。

## Background 本视频由 Ai4 频道于 2025 年 8 月 27 日发布在 YouTube 平台，主讲人为 CB Insights 洞察业务负责人 Jason Saltzman。简介指出，演讲聚焦于 CB Insights 在过去 18 个月内如何全面践行 AI 转型，从传统的私有市场数据集与人工研究报告提供商，蜕变为 AI 驱动的预测性情报领军者，并系统分享了利用专有数据和 AI Agent 将企业战略决策速度提升 10 倍的核心框架。 ## TL;DR Jason Saltzman 深刻剖析了传统数据价值链漫长、割裂且导致决策瘫痪的行业痛点，提出了从“出售淘金许可证”到由 AI 直接交付“提炼黄金与定制首饰”的范式转变。CB Insights 依托覆盖全球超 1000 万家企业的万亿级商业知识图谱，利用 LLM 将新数据集摄入周期从 6 至 12 个月压缩至数周，并实现了并购标的筛选、SWOT 分析等高阶交付物的全自动生成。演讲系统梳理了数据获取、关系映射、动态洞察与智能分发的四层 AI 演进架构，并提出“大处着眼，小处着手”的落地方法论，强调将情报无缝嵌入 Microsoft Copilot、Salesforce Agentforce 等日常办公与决策工作流中。 ## Atmosphere & style 整场演讲呈现出务实、干练且极具前瞻性的科技峰会氛围。Jason 凭借深入浅出的“淘金与打金”生动比喻，将复杂的企业知识图谱与智能体架构阐释得清晰透彻，展现出对数据行业范式颠覆的笃定与自信。在严谨的战略框架与方法论输出中，主讲人偶尔因翻过幻灯片而自然打趣的松弛台风，也为硬核的商业技术分享增添了亲和力与现场感染力。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=4s) CB Insights 在过去两年中全面融入 AI 能力，成功从传统的数据与研究销售平台转型为提供即时洞察和战略行动建议的预测性情报公司。 - [00:04](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=4s) 传统数据价值链存在分散孤岛与工具链蔓延等严重痛点，容易导致数周至数月的决策延迟和“分析瘫痪”，迫使企业必须利用 AI 压缩从数据到行动的链路。 - [03:21](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=201s) 数据服务模式正经历从“出售河流/矿山采矿权（原始数据）”到“代客筛金（即时洞察）”，再到“直接交付定制首饰（端到端行动建议）”的价值跃迁。 - [03:21](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=201s) 底层构建了覆盖全球超过 1000 万家公司、整合数万亿数据点的商业知识图谱（Business Graph），能够 24 小时主动为用户监测市场变化并提供行动建议。 - [06:02](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=362s) 引入 LLM 结构化与理解能力后，新数据集的挖掘、摄入与图谱整合周期从传统的 6-12 个月大幅缩短至数周，加速了数据飞轮的运转。 - [08:22](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=502s) AI 赋能下的 M&A 标的筛选场景支持自然语言交互，Agent 可自动遍历海量图谱并即时生成候选标的推荐、竞品列表与主动式 SWOT 分析。 - [10:13](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=613s) AI 重塑数据业务遵循四层演进架构：自动化数据获取层、关系映射层、动态洞察层以及深度融入原生终端与第三方生态的智能分发层。 - [10:13](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=613s) 战略情报突破了独立看板的限制，通过无缝集成至 Microsoft Copilot、Salesforce Agentforce 和 Snowflake Cortex 等生态，直接嵌入用户日常工作流。 - [14:23](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=863s) 传统数据资产智能化转型应遵循“大处着眼，小处着手”原则，优先聚焦客户高价值、高 ROI 的单一切入点（如并购初筛）打通端到端闭环。 - [14:23](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=863s) 随着自然语言降低数据交互门槛，市场情报的受众群体已从投资机构与战略部横向扩展至 GTM 业务团队、投后管理和咨询顾问等更广泛的人群。 - [14:23](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=863s) AI 落地过程中必须严守高质量数据治理与安全护栏，防止“垃圾进，垃圾出”，确保输出预测性战略洞察的绝对准确性与可靠性。 - [16:20](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=980s) 智能体（Agents）的部署起点应紧扣客户核心痛点，推动业务从“提供数据”向“赋予情境（Context）”与“交付行动与智能”的深层范式转移。 - [17:06](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=1026s) 实现全栈 AI 转型的核心在于将自身重构为客户的深度协作情报伙伴，贯穿数据抓取、清洗映射、预测洞察到工作流嵌入的全流程。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=4s) 破除传统数据价值链的困局与 CB Insights 的转型 CB Insights 洞察业务负责人 Jason Saltzman 开篇指出，传统上 CB Insights 是一家数据与研究公司，主要向客户出售庞大数据平台的访问权限，并在其上叠加人工撰写的行业研究报告。然而在过去两年多时间里，公司引入了具备早期 AI 专业知识的团队，全面构建 AI 能力，完成了从“提供海量数据的公司”向“提供洞察、情报与具体行动建议的公司”的重大转型。当下整个商业世界的数据比以往任何时候都要庞大，数据固然是 AI、洞察和决策的基石燃料，但现实中大多数企业却深陷于分散、孤岛化且极度消耗资源的传统价值链之中。 在创新与技术日新月异的今天，市场变化速度要求组织内的每个人都具备即时决策的战略情报能力。然而传统的决策流程充斥着繁复的痛点：业务人员不得不耗费数周时间在混乱的数据集里筛选淘金，依赖大量分析师的工时支持，并在极其臃肿的分析与 BI 工具链（Tool stack/Tech stack sprawl）中艰难地拼凑关联数据。其结果往往是长达数周甚至数月的决策延迟，或者直接陷入“分析瘫痪”（Analysis paralysis）。如果数据公司依然只是将海量未经提炼的原始数据打包丢给客户，无异于直接给客户推卸了巨大的工作负担。Jason 强调，“现在我们已经到了必须用 AI 来跟上 AI 发展步伐的临界点”，数据企业必须借助大语言模型（LLM）与 AI Agent，压缩从数据获取到最终行动的链路，直接将客户带到价值链的终点。 ```mermaid flowchart LR accTitle: Traditional vs AI-Powered Market Intelligence Flow subgraph Traditional["传统模式（周期长、资源密集）"] direction TB A1["分散的原始数据/数据孤岛"] --> A2["数据清洗与提取"] A2 --> A3["人工分析师解读建模"] A3 --> A4["静态看板 / BI / 报告"] A4 --> A5["数周/数月后的滞后决策"] end subgraph AIPowered["CB Insights AI 模式（即时、主动）"] direction TB B1["全网非结构化/专有数据"] --> B2["AI 商业知识图谱 (10M+ 企业)"] B2 --> B3["自然语言交互 / 专用 AI Agent"] B3 --> B4["即时生成洞察、SWOT 分析与决策建议"] B4 --> B5["深度嵌入工作流 (Copilot / CRM / 终端)"] end ``` ### [03:21](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=201s) 从“淘金向导”到“成品定制”：智能压缩价值链 Jason 提出了一个生动的淘金比喻来阐释这种商业模式的根本颠覆。在传统时代，数据公司就像是淘金热初期的“地图测绘师”，仅仅向客户出售通往河流或矿山的许可证，给客户呈现的是一堆混杂着泥沙、碎石的庞大数据山，客户必须自己掏钱购买滤网和挖掘工具、雇佣矿工，耗时费力地去淘洗那几克金子。而 CB Insights 当前的愿景，则是利用 AI 压缩整个数据价值链，不仅帮助客户在更多泥沙中发现黄金，更由 AI 系统自动完成所有的筛洗提炼工作，直接把纯金送到客户手上；甚至进一步演进到客户只需提出需求（例如“我需要一枚这种尺寸和款式的金戒指”），系统便能端到端交付最终业务成果。 为了支撑这一愿景，CB Insights 依托 AI 构建了涵盖超过 1000 万家公司、数万亿个数据点的海量商业知识图谱（Business Graph）。该图谱将全球科技企业的技术底座、创新成果以及复杂的企业间商业关系全面打通。CB Insights 产品负责人曾有一句极具代表性的概括：“我们的目标是打造‘在你睡觉时也能持续工作’的市场情报——它就坐在你的肩膀上，时刻主动告诉你发生了什么、为什么重要，以及你接下来应该采取什么行动。”用户不再需要面对复杂晦涩的数据库，只需根据自身业务上下文直接提出战略问题，AI 即可端到端输出行动建议。 - **核心数据规模**：底层商业知识图谱覆盖全球超过 1000 万家科技与新兴企业，整合数万亿（Trillions）级数据点与商业关联网络。 - **价值链演进阶段**： - *第一阶段（原始数据）*：提供混杂泥沙的河流访问权（原始数据平台）； - *第二阶段（提炼洞察）*：AI 自动化淘洗筛选，直接呈现高纯度金块（即时分析洞察）； - *第三阶段（定制成果）*：根据个性化业务诉求，直接交付定制化首饰/成果（可落地的战略行动指南）。 ### [06:02](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=362s) 重塑数据获取与建网飞轮：从 6-12 个月缩短至数周 两年前，CB Insights 对 AI 的涉足还仅仅停留在“将其作为一个研究行业赛道进行跟踪报道”，自身并未在底层业务和客户服务中充分落地。转型开始时，公司从客服与聊天机器人领域引入了具备早期 AI 技术背景的专业人才，并果断决定**从数据本身切入变革**。过去，无论是抓取公开资讯、整合非公开与专有数据，还是通过分析师调研问卷（Analyst survey protocols）以及直接对话企业创始人与投资人所收集的大量定性数据，其接入、清洗、打标签和结构化的过程都极度依赖手工流程。 在引入 LLM 之后，CB Insights 打造了一个能够自主加速扩张的数据飞轮系统： - **数据接入周期质的飞跃**：过去引入并上线一个全新的数据集，需要制定冗长的新操作规范（SOP）并培训大量人员，通常耗时 6 到 12 个月；如今借助大模型的结构化与理解能力，新数据集的挖掘、摄入与图谱整合缩短至**数周**即可完成。 - **非结构化定性数据价值释放**：借助大模型解析海量非公开调研与行业交流记录，深入提取企业背后的战略意图、行业趋势与隐性伙伴关系。 - **知识图谱的自动关联**：AI 能够自动将分散的实体、投资者、竞争格局、技术分类（Taxonomy）精准映射进底层商业图谱，构建出不可替代的高壁垒数据底座。 ### [08:22](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=502s) AI Agent 驱动的核心场景：以并购（M&A）标的筛选为例 拥有了万亿级数据点的商业知识图谱之后，关键在于如何赋予业务人员高效的开采工具。Jason 以企业战略并购（M&A Target Analysis）这一经典且高价值的业务场景为例：假设 Meta 的企业发展与并购团队希望在 Voice AI（语音 AI）或 Text-to-Video（文生视频）领域物色下一个收购标的，面对市场上每周大量涌现的初创公司，传统的尽职调查与扫描需要投入庞大的人力、跨多个独立数据库交叉比对，耗时漫长。 而在当前 CB Insights 的 AI 驱动平台下： 1. **自然语言直接发起意图**：M&A 团队只需在系统提示框中输入“我们正在寻找某某垂直领域的并购标的”。 2. **全图谱智能遍历与匹配**：系统底层 Agent 会自动遍历全部 1000 万家公司的业务描述、分类标签、技术图谱与合作生态。 3. **自主生成战略交付物**：系统不仅实时推荐最契合的候选标的与其竞品列表，还能主动为各候选企业生成定制化的 SWOT 分析与市场地位综合排名，极大压缩了前期尽调与筛选周期。 ```mermaid flowchart TD accTitle: M&A Analysis Workflow Powered by CB Insights subgraph Input["用户战略意图输入"] A["M&A 团队提出自然语言需求<br/>（如：寻找 Voice AI 领域潜在收购标的）"] end subgraph Engine["CB Insights 底层图谱与 Agent 引擎"] B["检索 1000M+ 实体商业知识图谱"] C["实体标签/技术分类/上下游生态自动匹配"] B --> C end subgraph Output["自动化高阶洞察交付"] D1["核心候选标的 & 竞品梯队推荐"] D2["即时生成主动式 SWOT 分析"] D3["市场地位排名 & 战略协同价值评估"] C --> D1 C --> D2 C --> D3 end A --> B ``` ### [10:13](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=613s) 四层赋能架构与无缝分发：嵌入核心工作流 Jason 总结了 AI 彻底重塑数据业务的四层完整演进架构，并指出数据公司必须摒弃把数据孤立在独立看板或 BI 工具中的做法，而是将情报主动输送至用户执行深度工作的场景中： | 架构层级 | 变革前（传统模式） | 变革后（AI 赋能模式） | |:--- |:--- |:--- | | **1. 数据获取层 (Data Acquisition)** | 依赖人工规范与繁杂流程，耗时 6-12 个月 | LLM 驱动自动化接入与非结构化定性挖掘，缩短至数周 | | **2. 关系映射层 (Data Mapping)** | 数据孤岛割裂，人工打标与建立分类 | 基于本体（Ontology）和 AI 自动打通投资人、赛道、伙伴与竞品关系网 | | **3. 动态洞察层 (Dynamic Insights)** | 依赖分析师排期调研，“给什么看什么”的被动覆盖 | 允许任何人在良好映射的图谱上针对任意企业/行业自由提问 | | **4. 智能分发层 (Distribution)** | 导出 CSV、静态报表、独立 BI 看板 | 深度嵌入原生终端及第三方生态（Microsoft Copilot、Agentforce、Snowflake Cortex） | 在分发层面，CB Insights 既打造了统一的一体化终端（Terminal），又积极开展生态集成，与 Microsoft Copilot 合作，将数据深度嵌入日常办公软件，并接入 Salesforce Agentforce、Snowflake Cortex Knowledge Engine 等企业级智能体平台。这使得战略情报能够在用户办公的第一现场即时呈现。 ### [14:23](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=863s) 从数据到智能企业的落地路径：“大处着眼，小处着手” 对于希望将自身数据资产转型为智能业务的企业，Jason 给出了务实的落地方法论。他强调首先需要对现有状态进行全面评估：识别内部或客户当前面临的数据孤岛在哪里、哪些业务决策耗时过长、数据交付后还需要多少繁琐的后期人工处理。 在执行策略上，CB Insights 信奉 **“大处着眼，小处着手”（Think big, start small）**： - **锁定高价值单一切入点**：不要试图一次性重构所有业务，而是先聚焦客户最核心、ROI 最高的单一工作流（如并购初筛、竞品雷达等），利用 LLM、智能体或自动化工具打通端到端闭环。 - **拓展全新客群边界**：随着数据交互门槛由编写复杂 SQL 或筛选繁琐报表降低为自然语言对话，CB Insights 的客群已从原先单一的投资机构与企业战略部，成功横向扩展至 GTM 业务团队（市场拓展/销售）、投后价值创造团队以及管理咨询顾问等广泛群体。 - **坚持高质量数据治理原则**：在构建 AI 解决方案时，必须时刻警惕“垃圾进，垃圾出”（Garbage in, garbage out）的铁律。CB Insights 投入了极大的精力与严密机制，在利用 AI 扩大数据覆盖面的同时，依托权威验证数据与上下文环境构建安全护栏，确保输出洞察的绝对准确性与战略可靠性。 ### [16:20](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=980s) 部署智能体与将情报融入核心工作流 在探讨如何落地智能体（Agents）时，杰森·萨尔茨曼（Jason Saltzman）指出，部署智能体的起点应当紧扣客户当下最核心的情报工作流。企业必须从“客户目前正试图用你们的数据独立解决什么痛点”这一根本问题切入。一旦明确了这些核心需求，接下来的演进路径就是将提炼出的高阶智能直接接入并嵌入客户现有的 AI 工具体系，乃至整个通用的企业工具链中，深入到专业人员进行最深度思考与决策的核心工作环境里。 舞台大屏幕上展示了一个二维码，萨尔茨曼提到，观众可以通过它深入了解 CB Insights 对智能体未来演进路线的研判——这不仅关乎他们自身的产品方向，也反映了整个市场的大趋势。整个行业正在经历一场深刻的范式转移：企业不再仅仅停留在交付原始数据的层面，而是通过层层提炼，实现从“提供数据”到“赋予情境（Context）”，最终跨多项服务转化为“行动与智能（Actions and Intelligence）”。 ```mermaid flowchart LR accTitle: 数据向行动智能的演进路径 subgraph S1["数据底座"] A["原始数据与痛点挖掘"] end subgraph S2["理解与提炼"] B["情境构建 (Context)"] end subgraph S3["成果交付"] C["行动与智能预测"] end subgraph S4["深度嵌入"] D["客户现有 AI 工具栈"] end A --> B B --> C C --> D ``` --- ### [17:06](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=1026s) 从“金矿与河流”看数据公司的全栈 AI 转型 萨尔茨曼在翻页时出现了一点小插曲，笑着打趣自己“又翻过头了好几页幻灯片（I went too many slides again）”，随后将话题拉回到 CB Insights 完成从传统“数据提供商”跨越为“情报公司”的核心心得。他强调，利用 AI 的关键不仅在于技术赋能，更在于将自身重构为一个深度协作的“情报伙伴（Intelligence Partner）”。传统模式最大的弊端在于直接把一堆枯燥的数据丢给客户，然后“寄希望于他们能自己搞定并交出最好的工作成果”；而真正的转型要求企业借助 AI 极大缩短客户从获取数据到完成最终工作目标之间的距离。 为了阐释这种转变，萨尔茨曼重申了他贯穿全场的生动比喻：数据就像是一座金矿，或者一条流淌着砂金的河流。企业不能只充当把客户带到河边或矿坑入口的引路人，而是应该尽可能深入整个开采与冶炼的全过程，直接将客户梦寐以求的金子——也就是高纯度的行业洞察与预测性情报——交付到他们手中。这种变革的底层逻辑在于将 AI 贯穿至数据公司的全技术栈： - **数据抓取与提取（Data Extraction & Capture）**：利用 AI 实现多源异构非结构化数据的高效捕获； - **数据映射与清洗（Data Mapping）**：通过自动化管道完成实体对齐与深层逻辑关联； - **核心情报生成（Ultimate Intelligence）**：将分散数据提炼为具备预测能力的深度行业洞察； - **嵌入式工作流（Embedding into Workflows）**：将生成的情报无缝注入客户日常所依赖的业务生态，实现战略决策的十倍速跃升。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=4s) 现在我们已经到了必须用 AI 来跟上 AI 发展步伐的临界点 — Jason Saltzman > [03:21](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=201s) 我们的目标是打造‘在你睡觉时也能持续工作’的市场情报——它就坐在你的肩膀上，时刻主动告诉你发生了什么、为什么重要，以及你接下来应该采取什么行动。 — CB Insights 产品负责人 > [03:21](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=201s) 我需要一枚这种尺寸和款式的金戒指 — Jason Saltzman > [06:02](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=362s) 从数据本身切入变革 — Jason Saltzman > [08:22](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=502s) 我们正在寻找某某垂直领域的并购标的 — Jason Saltzman > [14:23](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=863s) 大处着眼，小处着手 — Jason Saltzman > [14:23](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=863s) 垃圾进，垃圾出 — Jason Saltzman > [16:20](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=980s) 客户目前正试图用你们的数据独立解决什么痛点 — Jason Saltzman > [17:06](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=1026s) 又翻过头了好几页幻灯片 — Jason Saltzman > [17:06](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=1026s) 寄希望于他们能自己搞定并交出最好的工作成果 — Jason Saltzman > [17:06](https://www.youtube.com/watch?v=QvAmnX8xs8o&t=1026s) 数据就像是一座金矿，或者一条流淌着砂金的河流。企业不能只充当把客户带到河边或矿坑入口的引路人，而是应该尽可能深入整个开采与冶炼的全过程，直接将客户梦寐以求的金子——也就是高纯度的行业洞察与预测性情报——交付到他们手中。 — Jason Saltzman ## Mentioned Jason Saltzman, CB Insights, Ai4, Meta, Microsoft Copilot, Salesforce Agentforce, Snowflake Cortex, 商业知识图谱 (Business Graph) ## Video description Jason Saltzman, Head of Insights, CB Insights Abstracts What if you could have always on intelligence that predicts which companies and technologies will reshape entire industries? Two years ago, CB Insights' relationship to AI was to cover the space through our research. Over the last 18 months, we have implemented AI across every part of the business to evolve from a company known for datasets and research to the leader in AI-powered predictive intelligence; turning months of analysis into AI tools and agents that leverage proprietary data to deliver instant insights on the world's companies. How did we go from data and manual research to an AI-first company? Attendees will discover how a leading data and research company is leveraging specialized AI models trained on proprietary datasets to generate insights and deliverables that 10x strategic decision-making speed and accuracy, moving beyond reactive research to proactive market intelligence. You'll learn the frameworks behind building AI systems that don't just process data, but predict the future of private markets and emerging technologies; leveraging data as the fuel and guardrails for effective AI implementation Join this session to see how the convergence of data, context, and AI tools & agents is creating the next evolution of market intelligence; and how you can apply these principles to transform your own data and AI strategy. [Source](https://www.youtube.com/watch?v=QvAmnX8xs8o)

[0:04] Jason Saltzman: 现在我们已经到了必须用 AI 来跟上 AI 发展步伐的临界点 [3:21] CB Insights 产品负责人: 我们的目标是打造‘在你睡觉时也能持续工作’的市场情报——它就坐在你的肩膀上，时刻主动告诉你发生了什么、为什么重要，以及你接下来应该采取什么行动。 [3:21] Jason Saltzman: 我需要一枚这种尺寸和款式的金戒指 [6:02] Jason Saltzman: 从数据本身切入变革 [8:22] Jason Saltzman: 我们正在寻找某某垂直领域的并购标的 [14:23] Jason Saltzman: 大处着眼，小处着手 [14:23] Jason Saltzman: 垃圾进，垃圾出 [16:20] Jason Saltzman: 客户目前正试图用你们的数据独立解决什么痛点 [17:06] Jason Saltzman: 又翻过头了好几页幻灯片 [17:06] Jason Saltzman: 寄希望于他们能自己搞定并交出最好的工作成果 [17:06] Jason Saltzman: 数据就像是一座金矿，或者一条流淌着砂金的河流。企业不能只充当把客户带到河边或矿坑入口的引路人，而是应该尽可能深入整个开采与冶炼的全过程，直接将客户梦寐以求的金子——也就是高纯度的行业洞察与预测性情报——交付到他们手中。

Jason Saltzman: What if you could have always on intelligence that predicts which companies and technologies will reshape entire industries? Jason Saltzman: turning months of analysis into AI tools and agents that leverage proprietary data to deliver instant insights on the world's companies. [0:04] Jason Saltzman: you need AI to keep up with AI. [3:21] Jason Saltzman: works while you sleep that sits on your shoulder and tells you what happened, why it matters, and what you should do about it. [14:26] Jason Saltzman: think big, start small [14:26] Jason Saltzman: garbage in, garbage out [16:20] Jason Saltzman: data into context into actions and intelligence [17:03] Jason Saltzman: handing them off data hoping that they can go and do their best work alone

[0:04] Great to meet you all. Uh I'm Jason [0:05] Salswin. I'm the head of insights at CB [0:07] Insights. We're a market intelligence [0:10] company. Now, you know, traditionally we [0:12] were a data and research company selling [0:14] access to sort of a broad database data [0:17] platform. Uh and then layering our own [0:19] research on top of that. in the last two [0:21] years and change we've sort of you know [0:24] brought on a suite of new team members [0:26] and new AI capabilities that have [0:28] enabled us to go from a you know here [0:32] here's a bunch of data company to a [0:34] here's a bunch of intelligence here are [0:36] a bunch of insights and here's what you [0:37] can do about it company. Um, right. This [0:39] talk is hopefully going to give you an [0:41] example of sort of what we've done and [0:43] provide some context on what most data [0:45] companies can now tackle to get their [0:47] customers and clients closer to the [0:49] insights and actions that they're [0:50] looking to get out of their data. Um, so [0:53] maybe to where where we've started, [0:55] right? We all have more data than ever. [0:57] We all know data is the fuel for AI. We [0:59] all know data is the fuel for insights. [1:01] Data is the fuel for intelligence and [1:03] data is the fuel for decision-m. But [1:06] right where we all start is a place with [1:08] sort of scattered siloed resource [1:10] intensive value chain between data [1:12] intelligence and action and with [1:14] companies technologies and innovation [1:16] moving faster than ever. Every person in [1:18] your organization now needs to have the [1:21] sort of strategic intelligence to make [1:23] decisions. Not after weeks of sifting [1:25] through data, not after countless [1:26] analyst resources, and not after sort of [1:30] peering through tons of different tools [1:31] to try to connect different elements of [1:35] data together uh to get to decisions [1:38] weeks or months from now when [1:41] technologies change every day. We're [1:43] sort of at the point where you need AI [1:45] to keep up with AI uh and everything [1:47] that is happening. Um you know some [1:49] interesting numbers here right we look [1:51] at the amount of data that has sort of [1:53] proliferated over the last decade. We [1:55] look at the sort of tool stack tech [1:57] stack sprawl for insights and analytics [2:00] tools and we look at the amount of time [2:03] that it generally takes companies to get [2:04] to this insight. And you know we're [2:07] looking at a place where you know [2:08] there's almost a paralysis that can [2:10] happen and especially if you're trying [2:13] to as a data company provide more value [2:15] to your customers instead of just [2:17] handing them almost a bunch of work to [2:20] go and do to extract that insight and [2:22] extract that action. AI now will allow [2:24] us all to get our customers and clients [2:30] uh closer to the end of this value chain [2:32] which is what they're looking for really [2:34] the insights and decisions and actions [2:36] that they want to act on. where we all [2:38] sit is sort of in a place where data [2:40] teams have to go and search for, look [2:42] for, find, clean, synthesize data into a [2:45] point where it an analyst can then go [2:47] and do the work on top of it to [2:49] interpret it, to analyze it, to try to [2:51] get to those insights, put them into BI [2:53] tools, put them into a dashboard, hand [2:55] them off to whoever, you know, within an [2:58] executive team, within a uh [3:00] decision-making team will try to get [3:02] those insights in that packaged form and [3:04] then make that decision. Right? Data [3:06] companies now have the ability with AI, [3:07] with LLM, with tool, AI tools, AI agents [3:11] now to get your clients closer and [3:13] closer to that final decision-making [3:14] step. Even often acting proactively for [3:17] them to make recommendations on what [3:19] they should do next. Right? The analogy [3:21] that is going to become I'm going to get [3:23] to as we come through here uh maybe one [3:27] more slide uh is that you know we had [3:30] this vision for basically condensing [3:32] that value chain all into one, right? We [3:34] want to be a place where we can use AI [3:36] to get the most comprehensive data [3:38] possible. We want to use AI to map all [3:42] of that data together across 10 million [3:44] plus companies, trillions of data [3:46] points. Create this business graph, this [3:48] knowledge graph of, you know, the [3:50] world's tech companies, their [3:52] technologies, their innovation, and how [3:54] it all maps together and what the [3:56] business relationships are between all [3:57] of them, right? All of that again starts [4:00] with a massive or several massive piles [4:03] of data that turn into a mountain and [4:05] then often the old way was handing that [4:08] pile of data off hoping that someone [4:10] would comb through it, find their golden [4:11] nugget, figure out what to do with it. [4:13] Right? We've taken the idea of what if [4:16] we can just, you know, let you shout out [4:18] a question, let you tell us based on [4:20] your business context, based on the [4:21] personalized things that matter most to [4:23] you and your company and your role, [4:25] right? What if we just gave you [4:26] intelligence that just happened? it just [4:29] worked. You told us what mattered most [4:30] to you and we gave you the insights and [4:32] the actions to go from that question to [4:37] you know the ultimate deliverable and [4:39] then what you will do about it. Um, so [4:41] we take a look at that and right this is [4:43] a quote that I stole from our head of [4:45] product, right? Our goal is to make [4:47] market intelligence that works while you [4:48] sleep that sits on your shoulder and [4:50] tells you what happened, why it matters, [4:52] and what you should do about it, right? [4:54] Again, from one massive pile of data [4:57] made up of other several smaller piles [4:59] of data all mapped together in a way [5:01] that's proactive and matter, you know, [5:03] sort of communicates that value to you [5:05] and your business. [5:07] So uh the analogy that I'm going to use [5:10] for all of this starting with that [5:12] mountain of data is that you know [5:13] traditionally data companies you know [5:16] are the very start of a gold mining [5:18] operation. We're telling we're the [5:19] ctographers. We're selling you access to [5:22] the river or to the mountain, to this [5:24] pile of dirt and gold and mud and [5:26] whatever else might be there and giving [5:28] you the option to, you know, buy the [5:31] tools, buy the saves, buy the mining [5:33] materials, hire the miners, and then [5:36] hopefully get to that piece of gold, [5:37] that nugget of insight, that piece of [5:39] action that you can go and take, right? [5:41] We now can use AI, use LLMs, and to a [5:46] find more places where that gold is [5:48] hiding, more piles of dirt, and then do [5:50] as much of that sifting, that mining for [5:52] you in a way that gets your customers [5:55] that nugget of gold, your clients that [5:56] nugget of gold faster and faster, and [5:59] gives them something to do about it. [6:02] So, where did we start? Right? Two years [6:04] ago, our relationship to AI was just to [6:06] cover it as a sector. We could talk the [6:08] talk. We knew what was happening in the [6:09] space, but we weren't putting it to use [6:11] for ourselves or for our customers. We [6:13] brought on a new team with early AI [6:15] expertise uh from the customer service [6:18] segment and the chatbot segment and we [6:21] decided that the place to start was with [6:23] our data itself, right? We, you know, [6:27] had this ability to sort of start from a [6:31] point of millions of companies and use [6:33] the process. You know, think about the [6:35] manual processes that we were using to [6:37] collect that data, to source that data, [6:39] to structure that data, whether it was [6:42] publicly available data or non-public [6:44] data or proprietary data, whether it was [6:47] from news sources, from conversations [6:49] with customers and their relationships [6:52] with other vendors, their partners. uh [6:54] we had the opportunity to create you [6:56] know basically use AI to expand the [6:59] processes we were using to start right [7:01] go and source data more efficiently more [7:03] effectively using LLM and then also to [7:06] go and expand this nonpublic proprietary [7:10] data sourcing through our analyst survey [7:13] protocols through some of the other [7:15] programs that we have that you know talk [7:17] directly to these companies and to uh [7:20] the investors in these companies to [7:22] understand what is happening in their [7:23] space and in their sector sector, right? [7:25] Creating a ton of qualitative data that [7:27] then again once an LLM sits on top of [7:29] it, right? You're able to extract more [7:31] meaning, more strategy, more company [7:33] trends from. Uh so where we started uh [7:37] right, we had something no one else did. [7:39] Uh we had structured, sourced, and [7:41] mapped intelligence and this became the [7:42] foundation for everything that we built. [7:44] Next, we started to build a machine that [7:46] sort of compounds its own ability to [7:48] acquire data and new data sets using [7:50] LLMs where you would have previously had [7:52] to train people to come up with a whole [7:54] set of processes and operating [7:56] procedures that would have taken, you [7:58] know, generally 6 to 12 months for us to [8:00] ingest any new data set to go and source [8:02] it to go and find it and to incorporate [8:04] it into this business graph. Um you know [8:07] we now have the ability to launch data [8:10] sets and source new data sets in months [8:13] uh or sorry in weeks rather than months [8:15] and incorporate it into this trillion of [8:18] trillions of data points on 10 million [8:19] plus companies. [8:22] What happens once you have all of that [8:24] data, right? That's simply the mountain, [8:26] that's the river, that's the place where [8:27] all the gold hides, right? Now, you have [8:30] to give people the data, the tools and [8:32] the miners to make use of this data to [8:35] go from, you know, okay, we want to find [8:38] one of the best use cases for us is M&A [8:40] targeted analysis, right? 10 million [8:42] plus companies, you know, the companies [8:44] that, you know, take the news last week, [8:47] uh, or we'll take Meta as a great [8:49] example, uh, given their recent [8:50] acquisition spree, right? You know, what [8:52] who should they go acquire next in the [8:54] voice AI space, in the text to video [8:56] space, wherever it might be, right? You [8:58] know, there are new companies popping up [9:00] every day, every week. I'm sure there [9:02] are dozens of them here that play in [9:04] those spaces, right? If I'm Meta's [9:06] corporate uh, M&A team, right? How do I [9:09] do the initial diligence on those [9:10] companies to understand where their [9:12] partnerships lie, who they're working [9:14] with, and what uh matters most in terms [9:17] of where our business is trying to [9:18] generate value from this M&A project. Uh [9:22] the old way starts with a mountain of [9:24] data probably disparate data sources and [9:26] silos and forces people to use tool or [9:29] you know to use intensive resources [9:30] across tooling data and people to get to [9:33] the ultimate insight of either a long [9:35] list or short list of companies that we [9:37] actually want to go and talk to. Now [9:39] with our, you know, this expansive [9:41] business graph and the agents and [9:43] tooling we've built on top of it, you [9:46] know, M&A teams can sort of shout into [9:47] the void, we're looking to acquire a [9:48] company and insert market here. And our [9:52] platform will go look at all of the [9:54] company descriptions, look at all the [9:55] appropriately sourced, tagged, [9:57] taxonomied companies across the entirety [10:00] of the business graph and suggest [10:02] companies competitors and run proactive [10:05] SWAT analyses uh market analyses on that [10:08] company in where they rank and stack [10:10] accordingly. [10:13] Um [10:15] AI has sort of become this clearing has [10:18] cleared the path for data companies to [10:20] put context and intelligence everywhere [10:21] that people are doing their best work. [10:23] uh thinking about you know all of the [10:25] places that you go currently whether [10:27] it's you know we have to use some [10:30] examples we have our own terminal right [10:31] a lot of this is or almost all of this [10:34] is available there right we wanted to [10:36] make one centralized station one single [10:37] pane of glass and ways for people to [10:40] sort of integrate that intelligence into [10:41] their daily workflows we have a great [10:44] partnership with Microsoft copilot where [10:46] we can embed into the tools that you use [10:48] in your day-to-day tasks so that you're [10:50] not forced to go into another tool that [10:52] will only expand And as we continue to [10:54] do more partnerships, putting our data [10:56] into other agent platforms, other [10:58] intelligence platforms, other places [11:00] where people are already doing their [11:01] deepest work, right, we see that with [11:03] Salesforce and Agent Force and Snowflake [11:05] and their new Cortex knowledge uh [11:08] engine. And right, we're trying to put [11:10] these insights and intelligence rather [11:12] than just raw data into these places [11:14] where companies are doing and [11:16] individuals are doing their best work. [11:18] Um but you know we sort of backing out [11:22] from that entire process from business [11:24] graph to insight and how AI has impacted [11:27] that layer right we've looked at it as [11:29] what can we do at every layer of how we [11:31] deliver value you know both to ourselves [11:33] as a company and to our clients uh and [11:36] their deepest work right we've looked at [11:38] this across data acquisition uh which [11:42] took months now we can do in weeks we [11:44] looked at this by marrying these [11:46] disparate data sets uh and trying to [11:48] enhance our using AI enhanced data [11:50] mapping to create all of these [11:52] relationships between the companies that [11:54] we cover and their various relationships [11:57] whether it's investors uh competitors [12:00] the markets they belong to um or [12:03] potential targets and relationships [12:05] there right we have the ability to then [12:08] go deeper into the [12:12] um [12:14] sort of dynamic insights layer right you [12:16] know we used to have or we still have an [12:19] incredible research team, but we sort of [12:21] used to force our clients and our [12:22] customers to either we would cover the [12:26] space they cared most about or great, [12:28] come ask us for it. We can see if we can [12:30] fit it into the research program. Now, [12:33] layering LLM and AI over the top of a [12:36] well-mapped data ontology allows us to [12:40] let anyone ask any question of the [12:42] markets, companies they care most about. [12:45] And then from the final layer, right, [12:47] it's the distribution layer. Where do [12:49] you put that intelligence? Right? You [12:51] know, instead of putting data into a [12:53] dashboard or into a BI tool, right? Now, [12:56] where do you put the final intelligence [12:58] where people do their deepest work so [13:00] that it's there ready when they need it? [13:04] Right? We've gone, you know, again, to [13:06] use this gold mining analogy, we went [13:08] from a river, access to mud and rocks [13:11] and some gold inside of it to a full, [13:15] you know, here's the piece of gold. [13:16] Here's the thing you can do with it. And [13:18] we're getting even closer to to blow my [13:20] analogy out. Giving someone access to [13:22] the ability to say, I want a ring in [13:25] this size that looks like this. Please [13:27] help me get to there faster and faster. [13:29] We are going from data to decisions to [13:32] outcomes for all of the core [13:34] functionality and core teams that we [13:35] serve uh across a number of different [13:37] functions. Right? The an LLM allows you [13:40] to and AI agents allow you to either [13:44] provide general intelligence, the [13:45] ability to ask any question or the [13:47] ability to get specific insights, [13:49] outcomes and deliverables for the [13:52] clients and the way that you understand [13:54] them. Right? You know, I think a great [13:55] place to start with this is to think [13:57] about, you know, how do you instead of [13:59] selling access to this river, start [14:01] producing and getting, you know, you we [14:05] all know our customers well enough to [14:06] know the things that they do with our [14:07] data to know the things that they are [14:09] trying to accomplish, right? How do we [14:10] go from here's a bunch of data to here's [14:14] the thing that you are trying to do with [14:15] this data. We are now an intelligence [14:17] partner to you rather than a data and [14:20] then go do your own work partner. [14:23] Um, [14:26] did I just go past two? Ah, all right. [14:29] Uh, right. When we think about the ways [14:31] to go from data to intelligence, right, [14:34] a lot of it starts with assessing your [14:35] current state, what are you offering to [14:37] either internal stakeholders or your [14:39] clients, uh, in terms of where the data [14:41] silos are, which decisions are taking [14:43] too long, how much post-processing needs [14:45] to happen for your data, and what's [14:47] available to sort of blow out your data [14:49] collection and mapping capabilities, [14:52] right? The best place is always to start [14:54] with one use case. Start, you know, we [14:56] like to say at CB Insights, think big, [14:58] start small. Um, thinking about the ways [15:00] in which you can, okay, we want to get [15:02] to this broad intelligence partner uh [15:06] starting point or uh companion to our [15:09] clients. What are the things that they [15:11] are, you know, sort of the highest value [15:13] use cases, highest value workflows that [15:14] we can start to automate, whether it's [15:16] through LLMs or through agents um or [15:19] even just through intelligent [15:20] automation. uh and then you can sort of [15:23] continue along that to scale [15:25] intelligence across the entire [15:26] organization. Right? We've gone from you [15:28] know traditionally selling to investors [15:30] and corporate strategy teams to now [15:33] serving go to market teams um value [15:36] creation teams, consultants, etc. Right? [15:39] simply by making this data easier to [15:42] interact with and easier to interrogate [15:44] and easier to get the ultimate uh nugget [15:47] of insights or intelligence or actions [15:49] out of. Um [15:52] so how can you transform your data [15:55] business into an intelligence business? [15:57] Right? The first one is to fuel your own [15:58] internal and external um AI initiatives [16:02] with validated data and context. Right? [16:04] You know, as with anything, data and AI [16:07] serves as a garbage in, garbage out [16:09] exercise, right? We've spent an [16:11] tremendous amount of time and care [16:13] thinking about the ways in which we can [16:14] use AI to increase our coverage, but [16:17] also increase the quality of our data, [16:19] right? We you can think about starting [16:20] to deploy agents for your most important [16:22] intelligence workflows. Again, starting [16:24] with the problems that your customers [16:26] are trying to solve with their with your [16:28] data on their own. And then you can [16:30] start to integrate intelligence into [16:32] your existing AI tools, into their [16:34] existing AI tools or their general stack [16:36] of tools and where they do their deepest [16:38] work. Um right if we want to I guess we [16:42] put the QR code on here uh to talk about [16:44] you know what we see as sort of the [16:45] future of agents uh both for ourselves [16:48] and for the market as a whole as people [16:51] transition from you know sort of turning [16:53] data into context into actions and [16:56] intelligence for their customers and [16:58] clients across a number of services. And [17:01] then finally, [17:03] I went too many slides again. Uh, right, [17:06] you know, I think the key takeaways in [17:08] the transition that we've made from a [17:10] data provider to an intelligence company [17:12] are really thinking about how we can use [17:14] AI to enable uh ourselves to become an [17:17] intelligence company, to become an [17:18] intelligence partner to get our clients [17:20] closer and closer to the work that [17:21] they're trying to do with our data [17:23] rather than just handing them off data [17:25] hoping that they can go and do their [17:26] best work alone. uh right to use my [17:30] analogy again data is a gold mine or [17:32] data is a river right you want to get as [17:35] close to that full mining process to get [17:37] people the insights and the intelligence [17:38] that they are looking for most and right [17:41] the big thing that we've seen is that AI [17:44] can be applied sort of across the entire [17:45] stack of a data company uh from [17:48] everything from sort of data extraction [17:49] data capture to data mapping to [17:52] providing that ultimate intelligence and [17:54] then to embedding it into existing [17:56] workflows. Thank you very much.