Invisible Technologies
具备真实收入的后期企业 AI 工作流平台,但估值已提前反映强执行预期
Invisible Technologies 已经拿到真实的企业 AI 采用和可信的产品宽度,但最新 >$2 billion 估值已经预设了比公开证据能证明的更强前瞻增长,以及更接近软件的经济模型。
封面要素
公司概况
Invisible Technologies 是一家私有企业 AI 公司,总部位于旧金山,创立于 2015 年,目前由 CEO Matthew Fitzpatrick 领导;他此前负责 McKinsey 的 QuantumBlack Labs。公司已从技术赋能的外包运营, 演进为模块化 AI 软件平台,覆盖数据基础设施、工作流映射、专家市场、评测和智能体编排。公开证据支持其规模真实——2024 年收入 $134 million、2025 年 9 月 $100 million 增长轮将已披露累计资本推至 $144 million, 客户证据横跨 Microsoft、AWS、Cohere、Nasdaq、Headway、保险公司和公共部门项目——但支撑当前 >$2 billion 估值的公开证据,对 2025/2026 年收入、利润率结构、集中度和融资条款仍然有限。
- 成立时间
- 2015-01-01
- 创始人
- Francis Pedraza
- 创立地点
- San Francisco, California, United States
- 总部
- San Francisco, California, United States
- 产品
- Invisible 销售模块化企业 AI 平台,由数据基础设施、工作流映射、专家型人在回路能力、标注和评测工具、智能体编排组成, 通常配合前线部署工程师和结果导向服务落地。
- 客户
- 大型企业、AI 模型厂商、受监管行业和公共部门组织,需要工作流自动化、AI 训练、评测和运营落地支持。
- 商业模式
- 混合软件加服务模式,组合平台模块、前线部署工程、专家劳动力和托管式 AI 运营;公开案例研究暗示其企业合同更偏结果或工作流计费, 而非透明自助定价。
- 阶段
- Series B / Growth stage
- 融资情况
- 2025 年 9 月由 Vanara Capital 领投的 $100 million 增长轮,据报道估值超过 $2 billion, 已披露累计融资达到 $144 million。
执行摘要
主要优势
- 2024 年公开收入达到 $134 million,并有 2025 年 $100 million 融资佐证,企业部署证据也较扎实。
- 产品范围不再止于标注,已经延伸到数据基础设施、工作流梳理、评估、专家劳动力和智能体编排。
- 客户案例在医疗理赔、Nasdaq 入职流程、保险运营和 AI 模型改进工作流中,都给出了可量化 ROI。
- 管理层已补入具备企业 AI 实战经验的操盘手,包含 McKinsey/QuantumBlack 背景,技术班底也更厚。
主要风险
- 最新 >$2 billion 估值对仍有劳动力辅助的模式,已经隐含偏高的历史收入倍数。
- 公开披露对 2025/2026 年收入、毛利率、烧钱速度、客户集中度、续约质量和融资条款仍很薄。
- 劳动力模式审查、用工实践诉讼敞口和治理监督,都会抬高执行与声誉风险。
- 合成数据、自建工具,以及更大的工作流或 BPO 平台,都可能挤压 Invisible 历史上的 AI 训练切入口。
未决问题
- 2025 年实际业绩、2026 年收入运行率,以及软件与服务毛利率的桥接。
- 主要产品族的客户集中度、NRR/GRR、续约队列和合同期限。
- 2025 年融资完整条款清单、优先股堆叠、董事会权利和老股流动性历史。
- 证明企业工作流收入,而非传统标注或 RLHF 劳动力,如今已成为主增长引擎的证据。
目录
01公司概况
1.1 身份定位与运营模式
Invisible Technologies 现在把自己定位为企业 AI 软件平台,但本次核查记录显示,它更像混合模式,而不是纯 SaaS 厂商。官网首页、关于页面、how-we-work 页面和隐私政策都描述了同一个系统:整理混乱的企业数据,部署智能体工作流, 并在单靠软件不够时加入领域专家或智能体。运营上,Invisible 称前线部署工程师会把客户系统接入其模型无关平台, 同时客户数据留在客户环境中。这套架构与其 AI 训练材料相吻合:材料强调多语言评测、强化学习环境、红队测试和专家动员, 而不是打包销售的单点软件。 独立来源也支持这一点:公司 2015 年起步时是虚拟助理 / 外包服务,之后沿价值链上移到 RLHF、AI 训练和企业 AI 基础设施。这段演进很重要,因为它同时解释了公司为什么能赢得模型厂商信任,也解释了它为什么仍依赖分布式人力。 它还说明 invisible.ai 是一个实质性的研究陷阱:该域名属于另一家制造业计算机视觉公司,不是本章的目标公司。[CO001, CO002, CO003, CO004, CO005, CO006]
运营模型把企业数据、Invisible 软件和托管专家层连在一起,而不是纯自助工作流。
[CO001, CO002, CO003, CO004, CO028, CO042]1.2 领导层与治理
围绕 2024-2025 年交接,领导层证据最扎实。Ben Plummer 在 2024 年 1 月公共部门业务发布和 2024 年 11 月 Deloitte 公告中仍以 CEO 身份被引用,但 2025-01-21 Invisible 宣布 Matthew Fitzpatrick 出任 CEO。Fitzpatrick 此前负责 McKinsey 的 QuantumBlack Labs,说明公司想要一位能把模型厂商经验转化为企业部署的运营者。 Francis Pedraza 仍是记录中贯穿始终的基础治理人物:Sacra 认定他是创始人,现代奴役声明显示他以创始人、总裁兼董事长身份签署, 2025 年融资公告也列其为董事长。 公开治理披露仍不完整,但 2025 年 9 月融资材料列出了具体董事会阵容——Pedraza、Charlie Songhurst、Doug Clinton、John Lee、Robyn Scott,以及新加入的 Vanara 合伙人 Hayden Lekacz。Wes Green 被任命为首位全球公共部门 SVP 也值得注意,因为这是最清晰的岗位级信号之一,表明 Invisible 将政府业务视为持久扩张方向。 因此,主要关键人依赖是:Pedraza 支撑创始人叙事和治理连续性,Fitzpatrick 负责企业软件再定位。[CO007, CO009, CO010, CO011, CO012, CO013]
| 人物 | 职务 | 有证据支撑的背景 | 覆盖范围 | 关键人物依赖 |
|---|---|---|---|---|
| Francis Pedraza | 创始人兼董事长 | 从 2015 年起打造 Invisible,至今仍以创始人 / 总裁 / 董事长身份签署治理声明 | 创始人叙事与治理连续性 | 高 |
| Matthew Fitzpatrick | CEO | 曾任 McKinsey QuantumBlack Labs 全球负责人,2025 年 1 月获任 CEO | 企业 AI 商业化与运营节奏 | 高 |
| Ben Plummer | 2024 年公开材料中的前 CEO | 在 2024 年 1 月和 11 月公司公告中以 CEO 身份被引用 | 领导层交接背景清楚,但当前角色不明 | 中 |
| Wes Green | 全球公共部门 SVP | 前空军军官和行业老将,受聘开拓政府垂直领域 | 公共部门扩张执行 | 中 |
| Hayden Lekacz | 通过 2025 年轮次进入董事会 | Vanara 管理合伙人,该笔投资附带董事会席位 | 资本市场连接与投资者监督 | 中 |
仅为部分公开名单;Invisible 未在已审阅来源中公布完整高管组织图或董事会委员会结构。
[CO009, CO010, CO011, CO012, CO013, CO014]| 利益相关方 | 角色 | 控制权或经济重要性 | 当前证据 | 尽调问题 |
|---|---|---|---|---|
| Vanara Capital | 2025 年领投方 | 领投 $100M 增长轮并获得董事会席位 | 新资本叠加治理影响力 | 要求提供完整投资者权利和董事会观察员条款 |
| Francis Pedraza | 创始人兼董事长 | 在董事会和合规材料中仍被点名为治理锚点 | 创始人连续性可见,但持股未披露 | 要求提供当前股权结构表和创始人投票权 |
| Acrew Capital、Greycroft、Backed VC 与 BY Ventures | 跟投老股东 | 既有支持方在 2025 年轮次继续加码 | 显示内部支持,但不显示经济条款 | 要求提供轮次分配和按比例参与细节 |
| Princeville、HOF、Freestyle、Rocketeer 与 Tallwoods | 新参与投资方 | 新资金随 Vanara 一同进入增长轮 | 资本基础更多元,但条款不透明 | 要求提供工具类型及董事会 / 同意权 |
| Doug Clinton 与 Deepwater Asset Management | 董事会关联既有投资者 | 融资披露同时提到 Deepwater 参与和 Clinton 董事席位 | 经济和治理角色可能超过被动出资 | 要求提供持股比例和委员会角色 |
| Charlie Songhurst | 独立董事 | 新闻材料突出其另任 Meta 董事 | 增加 AI 网络触达,但委员会任命未披露 | 要求提供董事会职责和冲突政策 |
| John Lee 与 Jazz Venture Partners | 董事 | 公开点名的董事,带有风投代表性 | 暗示前序董事会连续性,但时间未披露 | 要求提供初始任命日期和保护性条款 |
| Robyn Scott / Apolitical | 董事 | 公开点名的董事,背景契合政策和公共部门熟悉度 | 可能影响受监管市场扩张 | 要求提供委员会席位和决策权 |
映射公开点名的投资者和董事,不包括持股比例、清算优先权或投票控制权。
[CO011, CO015, CO016, CO017]1.3 资本、规模与商业验证
资本和商业验证指向同一结论:Invisible 似乎已经从小众 AI 运营厂商,跨入资本更充足的企业 AI 基础设施玩家。 官方和独立来源一致显示,公司在 2025 年 9 月融资 $100M,使累计融资达到 $144M,并引入新老投资者组合。 最可信的外部估值读数来自 SiliconANGLE 和 Sacra 的“超过 $2B”,而 2025 年 1 月 CEO 公告锚定的上一台阶, 是 2024 年初 $500M。运营动能看起来真实,不只是叙事驱动。CEO 交接帖称收入从 2023 年到 2024 年翻倍以上,达到 $134M;后续融资材料重复了同一数字,并称 2020 年至 2023 年增长 24x。Sacra 还认为公司在这一规模下已经盈利,估计 EBITDA 约 $15M。 商业验证对一家私人公司来说也异常具体:官方案例研究声称 Headway 理赔处理提速 8x、某配送平台入驻提速 233%、Nasdaq 节省 10,000 开发者工时;WEF、AWS marketplace 和 FeaturedCustomers 都重复了 Invisible 曾服务 80% 以上领先 AI 模型提供商的说法,其中包括 AWS、Microsoft 和 Cohere。不过,确切客户数量仍未披露。[CO015, CO016, CO017, CO018, CO019, CO020]
| 指标 | 数值或状态 | 截至 | 置信度 | 缺口或备注 |
|---|---|---|---|---|
| 公司身份 | 企业 AI 平台,配套托管专家运营 | 2026-06-04 | 高 | 混合软件 + 人工模式,而非纯自助式 SaaS |
| 成立 | 2015 | 历史 | 中 | 可访问官方页面未显示准确注册成立日期 |
| 运营基地 | 旧金山;在加州登记的特拉华州公司 | 2026-06-04 | 高 | 城市 / 总部推断来自官方发稿地和诉状,而不是专门的总部页面 |
| 最新 CEO | Matthew Fitzpatrick | 2025-01-21 | 高 | Ben Plummer 直到 2024 年末仍主导公开沟通 |
| 累计融资 | $144M | 2025-09-16 | 高 | 早期逐轮经济条款仍不完整 |
| 隐含估值 | >$2B | 2025-09-16 | 中 | 有外部来源印证,但完整投后估值和任何老股交易占比未披露 |
| 2024 年收入 | $134M | FY2024 | 高 | 公司反复提及,并由 Sacra 印证 |
| 盈利能力 | 盈利 5+ 年;Sacra 估计 EBITDA 约 $15M | FY2024-FY2025 叙述 | 中 | EBITDA 数字是外部估计,并非经审计披露 |
| 人力足迹 | 35+ 个国家 / 地区的 3,000+ 名代理人员,外加约 350 名全职团队成员 | 2025 年快照 | 中 | 当前 2026 年准确员工数仍未核实 |
| 客户证据 | 领先 AI 模型提供商中 >80%;点名 AWS、Microsoft、Cohere | 2026-06-04 | 中 | 准确客户数和集中度未披露 |
| 主要公开下行风险 | 加州劳动集体诉讼 | 2023-11-17 | 中 | 结果和任何补救措施在可访问来源中尚未解决 |
混合了官方说法、独立估计和可访问公共记录;准确客户数、债务条款和当前 2026 年员工数仍未披露。
[CO001, CO005, CO007, CO009, CO015, CO016]这组 KPI 视角把规模信号和主要未决尽调标记放在一起,而不是逐行重复快照表。
劳动力数字是 Sacra 估计;估值来自外部佐证,而不是披露的官方投后条款清单。
[CO001, CO016, CO018, CO020, CO027, CO037]1.4 时间线与未决风险
公开里程碑显示,公司形态变化很快,但尽调信号并非全是正面。时间线从 2015 年创立起点和 2020 年疫情期间的扩张故事开始,后者说明在企业 AI 成为热门概念之前,Invisible 为什么先以运营执行力出名。2024 年,公司既启动公共部门业务,也凭 Deloitte Fast 500 排名获得外部增长认可。2025 年 1 月,公司更换 CEO; 到 2025 年 9 月,公司完成大额增长轮、扩大已披露董事会成员,并开始把自己定义为企业 AI 基础设施,而不只是 AI 训练运营。2026 年 3 月收购 WeCP 进一步强化这条轨迹,因为该交易补上了与高精度验证工作流直接相关的专家评测工具和面试数据。 最清晰的公开反向事项是 2023 年 11 月在加州提起的集体诉讼,指控公司存在加班、休息、工资单、费用报销和带薪病假违规。 诉讼本身不能证明责任成立,但意味着劳动力实践尽调不能只依赖公司叙事。风险还被放大:本轮核查中 Indeed 评论页和 BBB 投诉页都被验证墙拦住,Crunchbase/PitchBook 也没有提供可用的结构化数据。因此,与融资标题、收入或领导层事实相比, 当前员工情绪、确切员工数和细粒度融资条款的验证程度更低。[CO012, CO028, CO029, CO035, CO036, CO037]
| 日期 | 事件 | 类型 | 金额、估值或状态 | 参与方 | 含义 |
|---|---|---|---|---|---|
| 2015 | Invisible 成立,起步为外包 / 助理式运营模式 | 创立 | 已成立 | Francis Pedraza / 早期团队 | 确立服务较重的起点,之后演进为 AI 基础设施 |
| Mar 2020 | 疫情需求冲击期间,配送平台入职计划扩容 | 合作 | 入职速度 +233%;每月 1.5M 个数据点 | 未具名配送平台 / Invisible 运营团队 | 证明后来 AI 平台叙事之前的大规模运营执行力 |
| 2023-11-17 | Jordan Crowley 在 San Francisco Superior Court 提起集体诉讼 | 反向 | Case CGC-23-610522 | Jordan Crowley 诉 Invisible Technologies Inc. | 使劳动力实践尽调成为仍在发生的风险项 |
| Jan 2024 | 全球公共部门业务启动,Wes Green 获任 | 扩张 | 新垂直领域启动 | Invisible / Wes Green | 显示公司扩张到私营企业和模型构建方之外 |
| 2024-11-21 | 入选 Deloitte Technology Fast 500 第 61 名 | 扩张 | 排名期内增长 2,342% | Deloitte / Invisible | 外部信号显示近期增长速度 |
| 2025-01-21 | Matthew Fitzpatrick 获任 CEO | 治理 | CEO 交接 | Invisible / Fitzpatrick / Pedraza | 领导层转向企业 AI 商业化 |
| 2025-08-05 | 现代奴隶制声明获董事会批准并由 Pedraza 签署 | 监管 | 合规声明发布 | 董事会 / Francis Pedraza | 劳动力治理的公开合规材料 |
| 2025-09-16 | 增长轮公布 | 融资 | 融资 $100M;累计 $144M | Vanara 领投财团 | 重定价公司,并为下一阶段平台建设提供资金 |
| 2025-09-16 | 董事会成员随 Vanara 新席位一同披露 | 治理 | 董事会扩充 / 披露 | Pedraza、Lekacz、Songhurst、Clinton、Lee 与 Scott | 显示融资后公开掌握治理影响力的人 |
| 2026-03-10 | WeCP 收购协议公布 | 产品 | 18,000+ 个评估框架;2M+ 条面试记录 | Invisible / WeCP | 加深专家验证和评估工具栈 |
本时间线穷尽了截至 2026-06-04 在已审阅公开来源集中浮现的有日期里程碑;没有公开日期的项目留在表外。
[CO005, CO012, CO013, CO015, CO020, CO029]有日期锚点的里程碑显示,Invisible 从运营供应商根基转向资本加持的企业 AI 平台,但劳工风险披露仍然存在。
[CO005, CO012, CO013, CO015, CO016, CO029]1.5 证据项
02市场分析
2.1 市场边界与现状替代方案
Invisible Technologies 应放在企业 AI 运营、AI 训练和评测、工作流自动化预算里比较——而不是与 invisible.ai 服务的工厂车间视觉市场,或完整生成式 AI 基础设施栈比较。公司自有材料始终把产品定义为围绕数据、智能体、人在回路和评测系统的模块化部署, 并绑定真实运营结果;RL 环境产品也明确围绕会计、银行、法律、合规等企业任务组织。因此,最接近的替代品是一组混合对手: Appen、Labelbox 这类标注优先厂商,UiPath 这样的编排软件,以及 TaskUs 这类外包数字服务提供商。 这也意味着,通用服务器支出、超大规模云厂商资本开支和无关计算机视觉部署,都应排除在市场边界之外。关键分析结论是: Invisible 争夺的是工作流预算,在这些场景里,正确性、可审计性和人工升级比直接模型访问或商品化标注吞吐更重要。[CM001, CM002, CM003, CM004, CM005, CM006]
| 细分 / 类别 | 纳入支出 | 排除支出 | 买方 / 付款方 | 关联度 |
|---|---|---|---|---|
| 企业 AI 运营 | 与运营 KPI 绑定的工作流自动化、数据摄取、异常处理、人工复核和受监控输出 | 通用云计算、无关 SaaS 模块和非工作流 AI 试验 | COO、共享服务负责人、运营 VP、业务单元负责人 | Invisible 切入后台、入职、理赔和支持工作流的核心楔子 |
| AI 训练与 RLHF 服务 | 专家数据生成、多语言训练、多模态标注、红队测试和训练后评估 | 商品化点击任务、仅未差异化的合成数据或通用模型托管 | CTO、首席 AI 官、模型 / 产品负责人 | Invisible 切入前沿实验室和企业模型团队的核心楔子 |
| 企业 RL 环境 | 面向智能体训练的工作流模拟、可验证奖励、评分器、轨迹和可回放运行 | 与 Invisible 点名工作流无关的消费者聊天机器人、通用基准和工厂车间视觉试点 | 模型研究负责人、应用 AI 负责人、创新预算负责人 | 新兴但对 Invisible 具战略重要性的类别 |
| 工具优先的数据与评估平台 | 标注工具、评估 UI、托管审核员、模型辅助功能 | 端到端工作流重设计、深度遗留系统集成 | ML 平台团队、研究运营、采购 | 简单或更早期项目中的替代方案 |
| 现状替代方案 | BPO、内部运营团队、传统自动化套件和人工专家复核 | 不替代既有工作流成本中心的净新增 AI 专项预算 | 运营、客户体验和 IT 预算负责人 | Invisible 往往靠替代这类既有支出来销售,而不是创造全新预算线 |
边界有意排除 invisible.ai 的制造业视觉类别和更广泛的 AI 基础设施资本开支;重点是面向 Invisible Technologies 的企业工作流、训练后和受治理部署支出。
[CM001, CM002, CM003, CM004, CM005, CM006]2.2 规模测算视角与证据约束下的市场估计
针对 Invisible 精准细分市场的公开市场规模数据很少,因此本章不用单一 TAM 标题,而是采用多重视角。最干净的公开下限是受治理自动化支出: UiPath 一家公司就在 2026 年报告了 $1.901 billion ARR,说明大型企业已经为编排后的人加软件工作流投入大规模预算。 第二个视角来自 AI 数据和评测基础设施:Appen 报告 50M+ 平台小时、20K+ AI 项目和 100M LLM 数据元素,而 Invisible 称其服务超过全球顶尖 AI 公司 80%,并指向前沿实验室评测和 RL 环境需求。第三个视角是工作流层面的 ROI:金融、医疗和保险案例显示的节省或提速足以支撑经常性预算。合在一起,这些信号支持一个更窄的 2026 年 SAM 估计,大约 $2.0 billion 到 $6.0 billion;基准情景为 $3.8 billion,覆盖专家人在回路的企业 AI 运营、评测和 RL 环境工作。相对广义生成式 AI 叙事,这一区间刻意保守,因为 Invisible 的产品交付密集、集成重, 并受专家供给约束。[CM009, CM010, CM011, CM012, CM013, CM014]
| 发布方 | 年份 | 地理范围 | 数值 | CAGR | 方法 | 置信度 | 限制 |
|---|---|---|---|---|---|---|---|
| UiPath IR | 2026 | 全球企业自动化买方 | 1.901 | n/a | 业务编排和自动化需求的公开 ARR 下限 | 高 | 单一厂商收入是下限,不是总市场规模 |
| Appen 平台 | 2026 | 全球 AI 构建者 | 50M+ 平台小时;20K+ 项目;100M LLM 元素 | n/a | AI 训练和评估工作的运营规模视角 | 中 | 不是收入数字,也不特指 Invisible 的精确细分市场 |
| Labelbox 定价 | 2026 | 全球工具优先评估买方 | 免费层最多 30 名用户;订阅层加付费服务 | n/a | 包装视角显示训练后和评估工作的工具优先预算入口 | 中 | 未披露公开 GMV 或收入 |
| Invisible 客户证据集 | 2024-2026 | 金融、医疗、保险、企业 AI | 已记录 ROI:速度 8x、入职时间 -63%、成本 -37% 至 -57%、节省 10k+ 小时 | n/a | 来自具名客户结果的工作流级 ROI 视角 | 中 | 案例研究由公司撰写,不等同于市场规模数据 |
| 作者综合 SAM 估算 | 2026 | 全球受监管且数据密集的企业 AI 工作流 | 2.0-6.0 ($B),基准 3.8 | n/a | UiPath 公开支出下限,加上 Appen、Labelbox、AWS Marketplace 和 Invisible 工作负载组合所证明的训练后 / 评估需求上修 | 低 | 作者推导值,因为没有独立分析师单独划出 Invisible 的精确市场边界 |
本章采用受证据约束的规模测算,而不是一个宽泛的生成式 AI TAM。主要数字要么是上市公司运营指标,要么是作者基于这些公开视角明确推导出的估计。
[CM009, CM010, CM011, CM012, CM013, CM014]广义企业 AI 工作流支出大于当前公开自动化收入,但 Invisible 更近的市场会大幅收窄到专家在环、重集成工作。
三层都是作者推导,应理解为受证据约束的规模测算层,而非分析师发布的 TAM/SAM/SOM 数字。数值锚定 UiPath 公开 ARR 底线、Appen/Labelbox 的后训练需求指标,以及 Invisible 自身跨垂直客户证明。
[CM013, CM014, CM015, CM016, CM020]三组从公开数据到作者推导的视角说明,Invisible 面向的市场最好看作一个数十亿美元级区间,而不是单一精确数字。
区间值是情景估计,不是第三方市场报告。统一口径是年度市场支出,单位为十亿美元。
[CM009, CM011, CM015, CM032]2.3 买方、预算与采用路径
Invisible 的买方图谱异常跨职能。训练、评测和 RL 环境项目通常由 CTO、首席 AI 官、模型或产品负责人发起, 他们关心基准缺口、领域质量和部署准备度。工作流自动化项目则通常由 COO、共享服务、理赔、入驻、支持或合规负责人证明合理性, 这些人负责吞吐、错误率和劳动力成本 KPI。从 Invisible 材料和案例研究里浮现的采用路径,不是“先买模型,再扩规模”。 它从一个高摩擦工作流开始,进入与遗留系统的严格限定集成,用历史数据验证,之后才扩展到受监控的生产环境。这也是为什么工具优先平台和通用外包商在简单项目中仍是可行替代品; 但更高价值的机会在于,买方需要一个能把领域专业能力、工作流设计和可量化运营指标结合起来的供应商。横跨资产管理、医疗、保险、金融入驻、 多语言模型评测和 RAG 调优的客户证据表明,公司近期最可信市场是受监管、数据密集的企业 AI 切片,而不是广义 SMB 自助需求。[CM021, CM022, CM023, CM024, CM025, CM026]
| 细分 | 买方 | 用户 | 付款方 | 工作流 | 预算负责人 | 采用触发因素 |
|---|---|---|---|---|---|---|
| 前沿实验室和基础模型团队 | 首席 AI 官 / 模型负责人 | 研究员、评估团队、训练员 | R&D / 模型预算 | 训练后、RLHF、红队测试、多语言评估 | CTO / 首席 AI 官 | 基准饱和、领域扩张或智能体质量缺口 |
| 受监管企业运营 | COO / 运营负责人 | 运营分析师、审核员、理赔员、案件团队 | 运营或共享服务预算 | 理赔、入职、对账、文档工作流 | COO / 运营副总裁 | 积压、SLA 违约或劳动力成本压力 |
| 企业产品与平台团队 | 产品副总裁 / 工程副总裁 | 应用 AI、搜索、信任与安全团队 | 产品 / 工程预算 | RAG 排序、对话审核、提示词和回复质量 | 产品副总裁 / 工程副总裁 | 模型相关性差、幻觉或质量漂移 |
| 合规敏感职能 | 首席风险官 / 合规负责人 | 合规分析师和审核员 | 风险 / 合规预算 | 审计证据、工作流日志、人工监督、受控部署 | 首席风险官 / 法务运营 | 监管期限、审计发现或 AI 治理要求 |
| 沿用外包现状的买家 | 客户体验或共享服务负责人 | 坐席、主管、BPO 经理 | 现有外包预算 | 重复性客服、理赔或后台处理 | COO / CX 负责人 | 需要替换或改进低成本人力型服务交付 |
买家图谱综合了 Invisible 的交付模式、工作流案例研究,以及 TaskUs、Appen、UiPath、Labelbox 等相邻替代品。本图谱展示预算归属和工作流切入点,而不是穷尽市场份额。
[CM021, CM022, CM023, CM025, CM027]Invisible 短期最强的细分市场,是预算耐久度高、又强烈需要领域专长、可审计性和工作流集成的场景。
[CM021, CM024, CM026, CM030]Invisible 适用的采用路径通常会从宽泛工作流痛点,收窄到有治理、有指标、有监督的生产项目。
指数值是示意性阶段权重,不是实测转化率;它们展示从宽泛工作流兴趣到耐久生产项目的相对收窄。
[CM018, CM019, CM022, CM023, CM024, CM026]2.4 增长驱动、采用约束与估值相关性
对 Invisible 来说,三项结构性驱动最重要。第一,企业和前沿实验室正从预训练转向定制评测、后训练和智能体工作流, 推高对专家数据、评分员和 RL 环境的需求。第二,真实买方价值越来越多产生在混乱的运营系统内部,而不是孤立演示里, 这有利于能把部署工程和人在回路结合的厂商。第三,监管正在把治理变成商业门槛:EU AI Act 抬高了对日志、文档、 人类监督和透明度的预期;美国州级规则也继续围绕就业、训练数据披露和高风险用途增多。同样的力量也构成主要逆风。 定制评测和 RL 环境受专家供给限制;定义糟糕的奖励函数和评分员会摧毁 ROI;买方还要面对不断扩张的合规、供应商监督和 AI 洗白担忧。映射到估值,机会真实且很可能达到数十亿美元级,但收入耐久性取决于 Invisible 能否把定制项目变成可复制、 受治理的项目,而不是停留在服务密集型单点解决方案厂商。[CM031, CM032, CM033, CM034, CM035, CM036]
| 驱动因素 / 约束 | 方向 | 时间 | 影响 | 尽调问题 |
|---|---|---|---|---|
| 从预训练转向后训练和定制评估 | 顺风 | 现在-2028 | 推高对专家数据、评分员、RL 环境和评估服务的需求 | 询问管理层:新销售管线中评估 / 后训练相对基础数据工作的占比是多少 |
| 企业需要可衡量的工作流 ROI | 顺风 | 现在-2028 | 利好能把模型接入遗留系统和运营 KPI 的供应商,而不是只卖通用试点 | 要求提供首个工作流 ROI 及向第二、第三个工作流扩张的队列数据 |
| EU AI Act 和美国州法收紧监管 | 治理能力强的供应商顺风 / 买家逆风 | 2026-2028 | 抬高买家对日志、监督、披露和供应商治理的需求,但也拉长销售周期 | 要求提供活跃部署中使用政策、审计和文档工具的证据 |
| RL 环境和重领域任务中的专家供给瓶颈 | 逆风 | 现在-2028 | 即便需求扩张,也会限制 Invisible 或同业高质量交付的扩张速度 | 按领域、语言和独占模式验证专家网络深度 |
| 奖励劫持、评分失灵和仿真到现实错配 | 逆风 | 立即 | 管线设计差会吃掉 ROI,让试点在投产前失败 | 要求说明验证器校准方法、对抗测试和回滚指标 |
| 工具、BPO、自动化套件和定制集成商割裂市场 | 混合 | 立即 | 给 Invisible 的混合定位留出空间,但也加大品类教育和销售叙事难度 | 要求提供相对 Appen、Labelbox、UiPath、TaskUs 和内部自建的赢单 / 输单数据 |
方向从 Invisible 视角判断。顺风表示需求应会扩张;逆风表示部署更难或更贵;混合表示同时带来需求和摩擦。
[CM031, CM032, CM033, CM034, CM035, CM036]03竞争格局
3.1 格局:直接同行、替代品和竞争外溢
评估 Invisible Technologies 时,应把它看作 invisibletech.ai 这一企业 AI 工作流和训练公司,而不是狭义数据标注单点产品。 证据显示其竞争集确实很宽。Sacra 将 Scale AI 和 Surge AI 放在直接 AI 训练赛道,同时也点名 Appen 这样的标注专家和 TaskUs、Teleperformance 这样的 BPO 替代者。Invisible 自己的对比指南进一步拉宽框架, 把市场拆成工具优先平台、托管标注服务、开源栈和端到端合作伙伴。CB Insights 又加入一层外溢,把 Mimica、SuperAnnotate 和 Hypatos 列为替代方案,意味着买方可以用流程智能、标注或文档自动化产品解决同一问题的相邻部分。实际结论是, Invisible 不是只要打败一个显而易见的对手就能赢。它必须证明:为什么买方应该选择全栈运营模式,而不是更便宜的工具、 规模化服务商或自建路径。[CP022, CP023, CP024, CP025, CP037]
| 竞争对手 | 类别 | 规模 / 融资 | 目标客群 | 差异化 | 限制 |
|---|---|---|---|---|---|
| Invisible Technologies | 端到端 AI 合作伙伴 | $134M 2024 收入;累计融资 $144M;2025 年估值 >$2B;团队 350 人 | 企业 AI 团队和前沿模型构建方 | 一套栈覆盖数据、工作流、专家、评估和智能体自动化 | 公开定价、续约和赢单 / 输单数据仍然不足 |
| Scale AI | 直接数据引擎同业 | Sacra 对比称 ARR 约 $1.5B、估值约 $25B | 需要大规模训练数据的企业 AI 实验室和团队 | 适合标注、API、RLHF、评估和 GenAI 工作流 | 已审阅证据仍主要把它定位为数据引擎,而不是对完整工作流负责 |
| Labelbox | 工具优先的标注平台 | 私有公司;公开定价页显示免费层、付费订阅和附加项 | 自建数据工厂或评估工作流的团队 | 自助入口摩擦低,并带多模态评估功能 | 公开证据在工具层面最强,端到端运营所有权证据较弱 |
| Appen | 托管标注服务老牌厂商 | 1M+ 贡献者;50M+ 人时;20K+ AI 项目;处理 10B 单元 | 需要全球化、多语言标注和评估的大型企业 | 托管劳动力、平台和企业合规能力一起交付 | 仍围绕标注和评估,定价未披露 |
| TaskUs | BPO / CX 替代品 | 规模化上市公司,提供外包数字服务 | 已采购外包数字运营或 CX 支持的企业 | 采购熟悉度和服务交付宽度 | 未被包装成专用 AI 训练或数据基础设施栈 |
| UiPath | 自动化套件替代品 | $1.901B ARR;2,624 家客户 ARR >$100K;374 家客户 ARR >$1M | 规模化自动化工作流的受监管企业 | 存量客户基础带来可信度,并具备受治理编排和企业控制 | 比 Invisible 更少聚焦专家数据运营、RLHF 或人工训练闭环 |
概况使用公开定位和已披露规模信号;缺失融资或定价细节标为未知,而不是猜测。
[CP016, CP017, CP020, CP022, CP023, CP024]序数定位基于工作流拥有范围和人工专家服务强度,而不是经审计市场份额数据。
坐标轴是基于公开定位、包装和披露规模信号综合出的证据支撑序数评分;不是收入份额或 NPS 衡量。
[CP022, CP023, CP025, CP026, CP028, CP031]3.2 能力宽度与打包差异
能力宽度是 Invisible 最清晰的公开差异点。官方产品页覆盖领域专家 AI 训练、RL 环境、多模态数据工作、联络中心 QA、 计算机视觉 QA 和后台自动化。这个范围比工具优先或托管标注竞争对手公开可见的定位更宽。Labelbox 清楚展示了一个带免费层和付费企业功能的标注与评测平台, 因此买方如果想先用工具、再上服务,它是最低摩擦的比较点。Appen 展示的是相反取舍:贡献者基础很大、模态支持广、 企业合规姿态完整,但包装由销售驱动而不是透明定价。Invisible 自己的 Scale-AI 对比指南强化了市场分裂:一类买方要数据集, 另一类买方要生产工作流加领域专业能力。这个差异重要,因为产品打包本身就是竞争武器。最容易试用的厂商,不一定最适合端到端接管一个混乱、 受监管或高度依赖判断的工作流。[CP001, CP005, CP007, CP008, CP009, CP010]
| 采购标准 | Invisible | Scale AI | Labelbox | Appen | TaskUs / UiPath / 内部自建 |
|---|---|---|---|---|---|
| 主要任务 | 带专家运营的生产级 AI 工作流 | 大规模训练数据和评估 | 工具优先的标注和数据工厂工作流 | 全球规模的托管标注和评估 | 外包运营、自动化或自建栈 |
| 人工专家深度 | 深度专家网络加人工在环 | 托管劳动力和审核 | 专家可作为服务附加项使用 | 全球众包加内部专家 | TaskUs 服务深;UiPath 和内部自建需要另配人员 |
| 工作流自动化所有权 | 强:流程映射、智能体自动化、后台和联络中心流程 | 部分:API 和模型工作流,但已审阅证据集中在数据引擎任务 | 部分:平台工作流和评估工具 | 部分:可配置的数据生产工作流 | UiPath 自动化强;TaskUs 由服务牵引;内部自建取决于工程能力 |
| 多模态数据和评估 | 是:多模态数据、RL 环境、评估、QA | 已审阅对比来源显示为是 | 是:标注、模型辅助标注、多模态聊天编辑器 | 是:文本、音频、图像、3D、4D 和评估 | 混合,且往往碎片化 |
| 信任 / 合规姿态 | 声称工作流能满足合规要求,并配有专门治理材料 | 已审阅材料中,公开证据弱于功能证据 | 付费层提供企业控制 | 明确的安全 / 合规资质和云集成 | UiPath 在有治理约束的企业控制上最强;TaskUs 在外包采购熟悉度上最强 |
| 包装可见度 | 定制化,公开材料不透明 | 定制化,已审阅对比来源不透明 | 已审阅集合中公开可见度最高 | 报价驱动 / 已审阅集合中未公开定价 | 通常由捆绑或合同驱动;内部自建把成本转移到工程和运营 |
单元格总结公开来源明确展示的内容;公开证据不完整时,单元格标明限制,而不是推断同等能力。
[CP001, CP007, CP008, CP009, CP010, CP011]| 供应商 | 公开包装 / 定价信号 | 包含能力 | 未知项 | 影响 |
|---|---|---|---|---|
| Invisible Technologies | 定制化、结果导向的企业销售动作;已审阅来源中没有公开费率表 | 模块化平台、专家市场、工作流自动化、评估、智能体编排 | 实际成交价、折扣、最低承诺和利润率未披露 | 外部更难对标;适合顾问式企业销售 |
| Labelbox | 免费层加订阅层和附加项 | 标注平台、Monitor、SSO、定制向量嵌入、多模态模型评估工具,以及作为附加项的专家服务 | LBU 经济性、企业折扣和服务组合未公开 | 已审阅直接同业中评估路径摩擦最低 |
| Appen | 报价驱动 / 已审阅来源中未披露 | ADAP 平台、托管众包、多阶段 QA、工作流定制、API/AWS/Azure 集成 | 已审阅材料没有公开标价、最低用量或单位经济性 | 竞争点是全球规模和托管服务,而不是价格透明 |
| DataAnnotation | 任务型承包者市场,向贡献者支付溢价薪酬 | 专家审核、提示词工作、排序、标注和回复检查 | 公开页面看不到企业套餐、治理 SLA 或采购结构 | 可替代特定任务的专家劳动力,但把编排负担推回买家 |
| UiPath / TaskUs | 企业销售或合同驱动的销售动作 | 规模化受治理自动化,或外包数字服务交付 | 已审阅集合中没有可直接横向比较的价格点 | 既有厂商可凭契合现有预算、采购或自动化路线图赢单 |
本表比较包装姿态,不比较总体拥有成本;公开费率表缺失处保持未知项显式呈现。
[CP021, CP028, CP029, CP031, CP034, CP035]专家劳动力和工作流所有权必须紧耦合时,Invisible 最强;Labelbox 在透明自助包装上最强。
这些标签表示在已审阅来源集中目前可见的公开证据;不应误认为供应商认证的基准分数。
[CP025, CP027, CP028, CP029, CP030, CP031]3.3 切换成本、多供应商并用与买方适配
Invisible 的公开叙事表明,竞争耐久性将来自嵌入式运营,而不是模型锁定。公司称客户数据留在客户系统中,技术栈也不绑定模型; 这对谨慎企业有吸引力,但也意味着护城河必须在别处挣出来。可能来源是工作流设计、历史验证数据、领域专家和变革管理。 有真实证据表明,这些要素在部署后会产生价值:Nasdaq 案例研究提到入驻时间缩短 63%,节省超过 10,000 开发者工时。 但部署前竞争仍然激烈。Labelbox 可以吸引想要更快自助评测的团队。Appen 可以吸引优先考虑规模和全球覆盖的买方。 TaskUs 可以匹配既有外包预算。UiPath 可以借企业自动化路线图和装机基础可信度进入。因此,工作流需要专家判断加持续运营负责时, Invisible 最强;买方主要想要熟悉采购通道、按席位计费工具或狭义标注工厂时,Invisible 最弱。[CP003, CP004, CP013, CP014, CP028, CP031]
3.4 护城河耐久性与反向证据
正向案例足够扎实,值得认真对待。官方和第三方来源围绕有意义的规模相互印证:2024 年收入 $134 million、累计融资 $144 million、2025 年估值超过 $2 billion,客户或合作伙伴证据横跨 Microsoft、AWS、Cohere、Nasdaq、 Swiss Gear、SAIC 和 Charlotte Hornets。这不是一家小众标注店的足迹。但反向案例也很强。Sacra 认为模型厂商正转向合成数据,这会削弱单纯训练数据切入口的防御性,也有助解释 Invisible 的企业转向。同一来源提示劳动力模式审查风险, Alvarez & Marsal 则记录了 AI 治理、披露纪律和第三方合规门槛上升。这些压力重要,因为 Invisible 的公开定价仍不透明,一些竞争对手也提供更容易自助或由采购牵头进入的入口。因此,证据支持一个平衡判断:Invisible 有差异化切入口, 但其耐久性仍需要续约、定价兑现和对更简单替代方案的竞争胜率来证明。[CP015, CP016, CP017, CP018, CP020, CP021]
| 护城河主张 | 威胁 | 严重性 | 缓释措施 / 尽调问题 |
|---|---|---|---|
| 覆盖数据、工作流、专家、评估和智能体自动化的宽度 | 工具优先供应商和自动化套件可拆分整套栈,让买家组合更便宜的点解决方案 | 高 | 要求提供模块附加率、赢单 / 输单原因,以及买家只落地一两个模块的频率 |
| 企业交付可信度 | 免费增值 / 自助工具和 BPO 替代品在深度部署前可能看起来更易试用或更易采购 | 中高 | 要求提供试点转化率、部署周期,以及试点扩张或停滞的原因 |
| 模型无关架构 | 模型锁定效应较低,也意味着如果工作流嵌入不深,专有切换成本较低 | 中高 | 要求按工作负载提供第一年后的留存,并证明历史验证数据能提高续约概率 |
| 信任与治理姿态 | 监管审查、借 AI 包装的质疑和供应商合规审查会拖慢销售或损害可信度 | 高 | 要求提供审计材料、合规事件和客户安全审查结果 |
| AI 实验室传承 | 合成数据采用率提升会削弱历史训练数据切入点的粘性 | 高 | 要求提供当前收入在企业运营与模型构建方工作之间的构成,并证明企业侧多元化可持续 |
严重性反映持久性风险,而非当前产品质量;关键未知在于宽度能否转化为留存和定价权。
[CP038, CP039, CP040, CP042, CP043, CP044]Invisible 的公开叙事在广度和规模上最强,但在定价透明度和外部证明的护城河耐久性上最弱。
这组 KPI 是锚定公开证据的定性综合面板,不是内部评分卡或客户留存数据。
[CP016, CP018, CP038, CP040, CP042, CP043]3.5 证据项
04财务情况
4.1 收入模式与收入流
Invisible 的公开材料支持的是混合收入模式,而不是干净的 SaaS 或纯服务原型。官方页面营销一个端到端平台,组合数据基础设施、 工作流设计、评测工具和人类专家;Sacra 则把商业包装描述为围绕明确工作流和结果销售的运营即服务(operations-as-a-service)。 这种组合对投资判断很重要:平台软件可以随着时间提高交付杠杆,但目前证据仍指向一门业务,其变现绑定具体客户流程、标注量或月度保留费, 而不是公开席位价格。 因此,最有证据支撑的收入流拆分,是(1)面向模型厂商的 AI 训练、RLHF、评测和专家验证工作;(2)面向希望把 AI 嵌入遗留流程的大型组织的企业工作流自动化和定制解决方案。官方案例研究覆盖金融研究、医疗理赔、保险运营、零售招聘和企业数据入驻, 说明公司同时在变现模型改进工作和业务流程转型。缺失的仍是最重要的投资判断层:公开材料没有披露有多少收入来自经常性平台订阅, 多少来自劳动支撑的项目或托管服务;也没有展示客户集中度或续约条款。[CI001, CI002, CI003, CI004, CI005, CI006]
| 收入流 | 机制 | 单位 | 当前价值 / 状态 | 收入质量 | 尽调问题 |
|---|---|---|---|---|---|
| AI 训练、RLHF 和评估工作 | 模型构建方项目使用专家反馈、标注、验证和衡量 | 按工作流、标注量或保留专家团队计费 | 官方 AI 训练页面和 Cohere/Boosted 案例研究清楚显示该业务活跃;确切构成未披露 | 中——需求存在,但业务构成可能暴露于合成数据替代和项目波动 | 要求提供模型构建方工作收入拆分、续约率,以及非经常性项目收入占比 |
| 企业工作流自动化 / 定制解决方案 | Invisible 接入客户系统,自动化或增强运营工作流 | 可能按月保留费或按工作流定价 | Nasdaq、Headway、保险公司、零售商和 Swiss Gear 一类案例提供支撑;无公开合同金额 | 如果嵌入核心工作流,质量高;但经常性特征尚无公开证据 | 要求提供与生产部署而非试点绑定的 ARR 或托管服务收入 |
| 专家市场 / 人工在环验证 | 通过 Invisible 平台接触领域专家和分布式操作员 | 专家任务、项目批次或经质量验证的产出 | 嵌入平台叙事和案例研究;确切独立变现方式不清楚 | 中——可以差异化,但人力密度可能压制利润率 | 要求提供专家市场对软件模块的附加率,以及专家支持工作对应毛利率 |
| 公共部门和受监管行业项目 | 国防、政府、保险和其他受监管工作流中的企业账户 | 定制企业工作说明书(SOW) | 官方新闻稿提到 SAIC / U.S. Navy 活动以及公共部门企业负责人 | 中——合同可能更长期,但采购周期和认证要求不透明 | 要求提供公共部门项目的管线转化率、合同期限和预算来源 |
| 传统高管支持 / 助理工作流 | 历史上的礼宾式或委派任务支持 | 据 Sacra 历史定价,最低 $2,000/月 | 历史变现路径;不是当前定位核心 | 低——传统收入流似乎已被战略性弱化 | 确认该收入流是否仍存在,以及是否仍有重大传统合同 |
本表区分有证据支持的收入流和公开未知项。官方材料证明宽泛工作流类别存在,但产品线收入构成和续约质量仍为私有信息。
[CI001, CI004, CI006, CI038]| 产品 / 单位 | 公开价格 / 单位 | 合同模式 | 标价 vs. 实际成交价 | 折扣 / 未知项 | 来源 |
|---|---|---|---|---|---|
| 历史高管支持 | $2,000/月最低 | 按月服务关系 | 唯一识别到的历史公开价格点;不是当前企业费率表 | 当前可用性、范围和客群未知 | Sacra |
| AI 训练 / 标注工作流 | 未公开列示 | 可能按每 1,000 条标注、批次或托管保留费计价 | 实际成交价未公开;公开的只有代理性单位描述 | 折扣表、质量奖金和最低量未知 | Sacra + 官方 AI 训练定位 |
| 企业工作流自动化 | 未公开列示 | 可能采用定制保留费或工作说明书定价 | 官方页面销售可衡量成果,而不是固定标价 | 无公开标准期限、折扣或实施费表 | 我们如何工作 + 定制解决方案 |
| 案例研究 ROI 套餐 | 未公开列示 | 定制企业合作 | 公开证据展示客户节省成本和缩短周期,而不是合同金额 | 没有合同,无法把节省主张换算成从合同总额到净收入的口径 | Headway / Nasdaq / 保险公司 / 零售商 / Boosted.ai |
| 平台模块(Neuron / Atomic / Synapse / Axon / Expert Marketplace) | 未公开列示 | 可能采用模块化或捆绑式企业定价 | 官方材料证明模块存在,但未展示独立定价 | 捆绑结构、附加率,以及软件是否不带专家服务单独销售均未知 | 融资新闻稿 + 首页 |
Invisible 在保留的公开页面上没有发布当前价格手册。因此,公开变现证据只停留在代理层面,不应被误认为实际成交价。
[CI004, CI005, CI006, CI039]Invisible 如何把工作流问题转成软件加服务混合收入。
定价单位标签基于 Sacra 的历史描述和官方工作流营销材料。Invisible 不公布当前价目表或收入结构百分比。
[CI001, CI003, CI004, CI006, CI038]4.2 GTM 动作与公开结果经济性
Invisible 看起来走的是实施牵引、ROI 优先的销售动作。公司 how we work 页面强调,前线部署工程师连接客户系统, 用历史数据验证,然后在生产中衡量吞吐、错误率、资源效率和单笔交易成本。这和透明自助 SaaS 漏斗很不一样: 它意味着顾问式问题选择、嵌入式工作流重设计,以及绑定可量化运营改善的扩张。 公开案例研究持续强化这个框架。Headway、Nasdaq、未具名全国保险公司、Boosted.ai 和零售招聘案例都发布了节省或提速主张, 但没有披露实际合同价值、年度承诺规模或标准折扣。从尽调角度看,这些案例仍有用。它们表明 Invisible 能在多个垂直行业推销有形 ROI,比泛泛营销文案更能支撑收入质量。它们也暗示,一旦工作流上线,销售效率可能提升,因为客户购买的是可量化流程结果, 而不是实验性试点。代价是不透明:没有合同金额、队列留存或客户数量,就无法有把握地把这些结果收益转化为 CAC、回本周期或 NRR。[CI007, CI008, CI009, CI010, CI011, CI012]
| 客户 / 项目 | 公开成果 | 为何影响 GTM | 财务含义 | 来源 |
|---|---|---|---|---|
| Headway | 理赔处理速度提升 8x;成本较内部团队 -37%;较原 BPO -57% | 医疗健康工作流有强前后对比 ROI 叙事 | 如果 Invisible 能分享部分劳动力节省,就支撑定价权 | Invisible 案例研究 |
| Boosted.ai | AI 投资助理数据工作节省 90% 成本,并获得实时洞察 | 证明 Invisible 能支撑高价值、特定领域 AI 项目 | 当专家标注数据事关关键任务时,可能具备溢价定价 | Invisible 案例研究 |
| Nasdaq | 入职 / 接入时间减少 63%;节省 10,000+ 小时开发者工时 | 证明其在企业数据与金融服务接入场景里能创造价值 | 显示在大型企业账户内具备先落地、再扩张的经济潜力 | Invisible 案例研究 |
| 全国性保险公司 | 节省 $450k;节省 16,000 小时;审批提速 50%;准确率从 75% 到 98% | 证明保险后台能实打实降本、提质 | 隐含 ROI 可能支撑托管服务或按价值定价,但合同金额未披露 | Invisible 案例研究 |
| 零售商招聘工作流 | 每周审阅 500 名候选人;Invisible 预筛 65%;节省 38% 时间 | 显示高量招聘工作流里的运营杠杆 | 表明经济模型是可复制的人力套利叠加工作流软件,而不是纯咨询 | Invisible 案例研究 |
| Cohere 评估项目 | 所引示例中,ADI2 领先 GPT-4o 和 DeepSeek-V3 9 分 | 表明 Invisible 能向前沿模型客户出售质量敏感的评估工作 | 支撑高溢价专家工作定价,但合同经济性仍是私有信息 | Invisible 案例研究 |
这些是客户结果代理指标,并非 Invisible 收入披露。它们有助于推断销售叙事和价值捕获,但不能替代合同层面的经济性。
[CI007, CI008, CI009, CI010, CI011, CI012]从工作流 ROI 到推断生产率和利润率问题的公开证据链。
Invisible 不披露实际定价、CAC、NRR 和毛利率,因此该桥接图在这些环节是定性的。生产率节点是作者用公开收入和团队规模做出的简单计算。
[CI023, CI025, CI034, CI035, CI040]4.3 单位经济代理指标与成本结构
公开印证最强的收入信号是 2024 年收入达到 $134 million,Sacra 另行估计在这一基数上 EBITDA 为 $15 million。如果这些数字方向正确,Invisible 已经摆脱了重度烧钱 AI 服务初创公司最早期的画像。核心未决问题不是有没有收入, 而是利润率路径会向软件式经济性收敛,还是继续受劳动密集度限制。 公开证据两边都有。正面看,公司营销自研软件、连续评测工具和工作流自动化,这些都应随着时间提高吞吐并改善毛利率。限制面是, Sacra 描述的交付引擎覆盖 35+ 个国家的 3,000+ 名代理人员,加上 350 人内部团队,意味着可变人工成分不小。 案例研究也强调速度和劳动力节省,而不是软件席位扩张,再次指向定制交付模式。用已披露的 2025 年 350 人团队规模计算, 最好的公开运营生产率代理指标是 2024 年每名当前团队成员约 $383k 收入,但这只是粗略方向性指标。UiPath 和 Appen 等公开可比公司披露的 ARR、留存和吞吐信息要详细得多,凸显 Invisible 的单位经济研究仍有大量内容在私域。[CI014, CI015, CI018, CI020, CI025, CI026]
| 指标 | 公开数值 / 状态 | 置信度 | 为何重要 | 尽调要求 |
|---|---|---|---|---|
| 2024 年收入 | $134M | 高 | 证明公司已经具备真实规模,不再只是试点阶段的 AI 供应商 | 将 2024 年数字勾稽到已审计月度收入与收入确认政策 |
| 2024 年 EBITDA | ~$15M(Sacra 估计) | 中 | 唯一公开盈利能力代理指标;决定增长是靠现金消耗买来,还是由经营供血 | 要求管理层提供 EBITDA 桥表,以及从 EBITDA 到经营现金流的现金转换 |
| EBITDA 利润率 | ~11%(Sacra 估计) | 中 | 如果方向正确,说明经济性好过重度亏损的服务业务 | 要求提供 GAAP 毛利率和 EBITDA 调节表 |
| 当前团队人均收入 | 按 $134M / 350 人计算约 $383k | 中 | 混合软件 / 服务模型的方向性生产率代理指标 | 用同期平均员工数重算,并按软件与交付人工拆分收入 |
| 交付基础 | 35+ 个国家的 3,000+ 名代理人员,另有 350 名全职员工 | 中 | 说明即便自动化提升吞吐,人力在经济模型里仍然重要 | 要求拆分承包商、员工和软件自动化各自承担的交付占比 |
| 上市可比披露代理 — UiPath | $1.901B ARR;109% DBNRR;2,624 个 $100k+ ARR 客户;374 个 $1M+ ARR 客户 | 高 | 展示上市工作流自动化同行的披露标准 | 要求管理层提供同口径指标,即便公司保持私有 |
| 上市可比披露代理 — Appen | 50M+ 人工小时;20K+ AI 项目;100M LLM 数据元素;处理 10B 个单元 | 中 | 说明 AI 数据同行即便利润率不同于 Invisible,也会披露运营规模指标 | 要求 Invisible 提供对应的吞吐量、项目数和专家规模指标 |
| 上市可比披露代理 — TaskUs/BPO | 存在上市外包数字服务代理公司,但抓取到的概览没有内嵌披露利润率数据 | 中 | 用于测试下行情景利润率时,可作较低利润率服务参照 | 调取完整监管文件,对比毛利率、EBITDA 利润率和人力强度 |
| 毛利率 | 未公开披露 | none | 判断业务到底按软件扩张还是按托管服务扩张的核心问题 | 要求按 AI 训练 / 评估与企业工作流自动化拆分毛利率 |
| CAC / 回本周期 / NRR | 未公开披露 | none | 判断 GTM 效率和先落地、再扩张经济性耐久度所必需 | 要求按分部提供队列留存、销售与营销支出、新增 ARR 和回本周期 |
公开单位经济性证据混合了已印证的收入顶部线、第三方盈利估计、作者的简单计算和上市可比披露标准。所有未披露字段都应视为真实尽调阻断项。
[CI014, CI018, CI025, CI026, CI027, CI028]4.4 资本充足性与融资依赖
Invisible 已披露资本状况方向上有支撑,但对投资判断仍不完整。公司 2025 年 9 月宣布 $100 million 增长轮, 使已披露累计融资达到 $144 million。管理层称,募集资金将投向核心 AI 软件平台和同一新闻稿中描述的领导层扩张。 再结合创始人称公司多年来以盈利方式建设,这降低了即时流动性危机的概率。 话虽如此,公开记录没有给出投资者真正需要用来衡量融资依赖的指标。没有披露账面现金、烧钱速度、月度或季度现金流, 也没有现金跑道指引。保留的公开来源中没有看到债务或项目融资负担,这比发现风险债或契约条款风险要好, 但“未披露”不等于尽调项已关闭。下一轮融资触发条件同样未披露。实际结论是,标题层面的资本充足性看起来可以接受, 因为公司已有融资且显然没有陷入困境;但如果还有下一次融资窗口,没有管理账就无法建模。[CI016, CI017, CI019, CI020, CI030, CI031]
| 项目 | 数值 / 状态 | 置信度 | 含义 | 尽调要求 |
|---|---|---|---|---|
| 账面现金 | 未公开披露 | none | 留存公开记录无法判断当前流动性 | 要求最新不受限现金余额和月度现金桥表 |
| 月度烧钱速度 | 未公开披露 | none | 现金跑道和融资依赖无法建模 | 要求过去 12 个月预算与实际烧钱对比 |
| 现金跑道(月) | 未公开披露 | none | 不存在可辩护的公开现金跑道估计 | 要求管理层提供基准与下行情景下的现金跑道模型 |
| 累计融资 | 披露累计融资 $144M | 高 | 相较极早期 AI 服务同行,公司资金更充足 | 将累计融资与当前股权结构表及任何老股交易勾稽 |
| 最新一轮融资 | 2025 年 9 月完成 $100M 增长融资 | 高 | 新资金缓解近端压力,但不能替代经营披露 | 要求投后所有权、清算堆叠和投资人权利 |
| 计划资金用途 | 进一步投入核心 AI 软件平台,并支持领导层 / 一线扩张 | 高 | 指向产品建设,而非应急融资 | 要求按平台、招聘、GTM 和地域拆分经董事会批准的资金分配 |
| 下一轮触发条件 | 未公开披露 | none | 不清楚下一轮融资取决于收入里程碑、产品里程碑,还是整体市场窗口 | 要求管理层说明融资计划、目标里程碑和下行情景应对动作 |
| 债务 / 项目融资义务 | 留存公开材料未披露 | 低 | 表面上利好,但没有披露不能替代尽调 | 要求债务明细、银行授信、最低支出承诺和契约包 |
资本充足性只能通过融资和管理层评论做方向性判断。公开来源没有披露真实承销所需的现金、烧钱速度或现金跑道输入项。
[CI016, CI017, CI019, CI030, CI031, CI040]该矩阵把财务承销中已披露、已估计和仍不可得的信息分开。
中间一行不是经审计的公司披露。它们要么来自第三方估计,要么是作者基于保留公开来源做出的推断。
[CI016, CI020, CI029, CI040]4.5 财务缺口与投资判断结论
Invisible 公开财务画像最重要的特征,不是收入存在,而是缺少判断收入质量和耐久性所需的指标。公开证据足以得出结论: 公司有真实规模、真实客户价值,也有足够融资继续投入。但这些证据不足以验证经常性收入质量、毛利率扩张、CAC 效率或集中度风险。 缺失项不是装饰。它们决定 Invisible 到底是在成为一个软件牵引的企业平台,把劳动力作为赋能层;还是一家高端技术赋能服务公司, 利润率会被结构性封顶。 反向案例也真实存在。Sacra 明确指出,模型实验室正在转向合成数据生成,这威胁到历史上的 RLHF 和数据标注收入流。 这让企业工作流自动化转向更加重要,但公开材料没有披露这次结构迁移已经走到哪一步。合并来看,财务结论是:对标题层面的牵引和近期资本化谨慎正面, 对尽调完整性明确负面。公司有足够需求证据和近期融资,仍处于可投资范围;但严肃投资判断仍需要合同级定价、收入结构桥、毛利率披露、 集中度数据和当前现金 / 烧钱模型,之后信心才可能越过“继续研究”。[CI029, CI032, CI038, CI039, CI040]
| 缺失的私有指标 | 对承销的影响 | 公开来源为何不足 | 精确尽调路径 |
|---|---|---|---|
| 按产品 / 工作流 / 客户类型拆分的收入结构 | 无法判断收入中有多少是经常性软件、托管服务或项目制工作 | 官网页面证明有多条变现路径,但不披露各自占比 | 要求按 AI 训练、评估、专家市场和企业工作流自动化提供季度收入桥表 |
| 客户集中度和垂直行业结构 | 无法检验增长是否分散,还是依赖少数旗舰账户 | 案例研究展示客户标识和用例,不展示收入集中度 | 要求前十大客户集中度、续约时间表,以及按行业 / 地域拆分的收入 |
| 毛利率和 COGS 构成 | 无法判断 Invisible 应该套用软件式还是服务式估值逻辑 | 只有 Sacra 提供 EBITDA 估计;公开渠道没有毛利率披露 | 要求毛利率桥表、人工成本分摊,以及按工作流类型拆分的自动化节省 |
| CAC、回本周期、NRR 和队列留存 | 无法验证销售效率说法,也无法判断先落地、再扩张增长的耐久度 | 公开材料强调 ROI 轶事,而不是漏斗或队列数据 | 要求季度队列、销售与营销支出、转化率,以及按分部拆分的 NRR / GRR |
| 现金、烧钱速度和现金跑道 | 无法建模融资依赖或下行韧性 | 融资标题没有披露当前流动性或现金消耗 | 要求最新现金报表、月度烧钱速度、现金跑道模型和债务授信 |
| 实际成交价格和折扣结构 | 无法把工作流 ROI 转换成收入质量或利润率质量判断 | 留存来源没有公开价目表或合同条款表 | 审阅现行价格手册、折扣矩阵和已签 SOW 样本 |
这些缺口,就是有说服力的公开叙事与可投资财务档案之间的差距。毛利率、集中度和现金 / 烧钱速度仍是最重要的阻断项。
[CI028, CI029, CI039, CI040]05产品与技术
5.1 产品定义与模块地图
Invisible Technologies 应被理解为工作流 AI 运营商,而不是狭义标注厂商或通用模型封装器。公司各解决方案页面描述的是一个模块化栈: 摄取非结构化企业数据,映射业务逻辑,部署智能体,在置信度低时插入人工复核,并围绕运营 KPI 持续评测输出。 公开产品覆盖后台自动化、联络中心质量和路由、预测、计算机视觉、AI 训练和强化学习环境。这个宽度重要,因为它显示 Invisible 围绕具体客户任务打包可复用工作流模块,而不只是出售定制咨询工时。 最清晰的模块地图来自 2025 年融资材料和 2026 年 WeCP 收购公告。这些来源描述的核心平台围绕数据基础设施、流程映射、 专家市场或 Meridial 能力、评测和编排组织。实践中,解决方案页面可以整齐映射回这些层:后台自动化强调摄取、路由、 证据呈现和人工升级;联络中心强调每次互动中的受治理上下文和政策级评测;预测强调数据统一加定制模型;计算机视觉强调安全部署、 标注、QA 和洞察交付;AI 训练加 RL 环境强调专家判断、评分员、评分细则和可回放任务运行。因此,产品故事是一组相互协作的模块和工作流模板, 而不是单一 SKU。[CE001, CE002, CE003, CE004, CE005, CE006]
| 模块 / 资产 | 主要用户 | 状态 / 成熟度 | 差异化 | 尽调缺口 |
|---|---|---|---|---|
| Neuron 数据基础设施 | 前置部署工程师和企业 IT 团队 | 公开披露的核心模块;成熟度足以支撑企业定位 | 整合并转换结构化与非结构化数据,供下游工作流使用 | 留存资料中没有公开架构图或连接器目录 |
| Atomic 工作流映射器 | 运营负责人和交付团队 | 公开披露的核心模块;定位为工作流设计层 | 用可视化流程映射固化业务逻辑,而不是强推模板优先的自动化 | 没有公开截图、变更日志或规则编写文档浮出 |
| Meridial / 专家市场 | AI 训练团队、领域专家、评估人员 | 已公开披露,并通过收购 WeCP 扩展 | 把专家寻源、RLHF、验证和评估基础设施组合起来 | 专家筛选和留存的公开质量指标仍是私有信息 |
| Synapse 评估层 | 模型团队和 QA 负责人 | 公开披露的核心模块;技术文档集提供强支撑 | 衡量性能,支持标注、微调和持续改进 | 没有公开基准测试仪表盘或模型评估 API 参考浮出 |
| Axon 编排层 | 运营团队和智能体负责人 | 公开披露的核心模块 | 跨系统编排任务和决策,而不是困在单一聊天界面内 | 没有公开支持 SLA 或运行时治理文档浮出 |
| 解决方案封装(后台、联络中心、预测、视觉、AI 训练、RL 环境) | 业务单元负责人和 AI 运营者 | 已有多个面向客户的解决方案页面和案例研究 | 把核心技术栈打包成面向特定工作流的产品,并用可衡量 KPI 讲清价值 | 包装广度是公开的,但独立模块定价和附加销售率未披露 |
各行区分平台层与工作流封装。状态指公开材料对该层的证据明确程度,不代表内部路线图置信度。
[CE001, CE005, CE007, CE008, CE011, CE012]| 用户任务 | 当前工作流问题 | Invisible 方案 | 可衡量收益 | 限制 |
|---|---|---|---|---|
| 后台文档处理 | 扫描文档、邮件、发票和例外事项拖慢合规要求高的运营 | 抽取、标准化,按置信度路由,呈现证据,并把不确定决策升级给人工复核 | 可用于合规的数据和更低手工负担,是公开承诺的核心 | 没有公开的受支持系统清单或 SLA 目标 |
| 联络中心质量和分诊 | 抽样会漏掉政策违规,分散的渠道数据也会掩盖趋势 | 受治理的跨渠道视图、100% 互动评估、情绪 / 风险浮出和人工可控交接 | 政策级 QA 覆盖和更快路由是明确公开主张 | 没有公开证据证明实时客户数或支持可用时间 |
| 需求预测 | 规划团队受困于分散的 ERP、POS、人力和外部数据 | 统一数据底座,训练定制模型,并交付仪表盘和建议 | 可直接支撑决策的预测和价值链可见性,是明确公开承诺 | 未披露公开准确率基准测试或刷新节奏 |
| 计算机视觉运营 | 原始视频难以运营化,模型在复杂环境里会退化 | 标注、QA、安全部署、边缘 / 本地部署选项,以及持续再训练闭环 | 结构化事件流、更好的漂移管理和客户数据控制,是卖点的核心 | 公开证据最强的是叙事文档,而不是可访问技术规格 |
| AI 训练和 RL 环境 | 通用基准测试和众包评分抓不住企业判断 | 专家评审员、可验证奖励、可回放运行和定制评估框架 | 公开证据包括 You.com 的 20k 次评估,以及为 Cohere 提供可信人工评估 | 留存资料中没有 RL 环境的公开 API 或定价界面 |
收益只涵盖公开主张和案例研究结果。缺失指标反映公开披露缺位,并不代表负面产品证据。
[CE002, CE003, CE004, CE005, CE006, CE007]Invisible 的公开产品叙事把数据基础设施、工作流逻辑、专家复核、评估和编排叠在客户工作流周围。
[CE011, CE012, CE013, CE014, CE015, CE016]5.2 架构、运营模式与部署机制
Invisible 的公开技术叙事对部署方式说得异常具体。how-we-work 页面称,前线部署工程师从客户工作流出发,将遗留系统和运营数据库连接到 Invisible 平台,让客户数据留在客户自己的环境中,用历史数据验证,然后进入受监控的生产状态。Forward Deployed Engineering 手册也强化了这一运营模式,把交付定义为嵌入式执行,而不是战略咨询。架构刻意保持模型无关:客户工作流和业务逻辑是稳定层, 模型、专家、智能体和评估器坐在这条集成主干之上。 技术文档层面也解释了 Invisible 为什么如此倚重评测、验证器和人工监督。其 AI 评测报告认为,标准排行榜不足以支撑企业部署, 定制评测框架应围绕业务特定错误类型、治理模型和多轮互动搭建。RL 环境和评分员问题材料更进一步:它们描述了可回放运行、奖励函数、 人工标注参考轨迹,以及包含结构测试、对抗性模型攻击和人类专家复核的三阶段验证器流程。对于计算机视觉和多模态系统,Invisible 自己的文章强调事件流输出、接入 ERP/WMS/CRM 系统的 API 桥、边缘或本地部署、再训练循环,以及按模态设计故障处理。 架构故事是连贯的:先集成,再评测,只有工作流可衡量之后才自动化。[CE019, CE020, CE021, CE022, CE023, CE024]
| 层 / 流程 | 角色 | 公开依赖 | 风险 |
|---|---|---|---|
| 遗留系统连接器和数据管道 | 把结构化和非结构化企业数据接入工作流技术栈 | 客户系统、运营数据库、数据仓库、ERP/WMS/CRM 目标系统 | 连接器广度和变更管理负担没有公开文档 |
| 工作流映射和业务逻辑设计 | 把混乱的真实工作转成明确路由、约束和升级路径 | 前置部署工程师加 Atomic 式流程映射 | 如果工作流定义不清,智能体输出可能会优化错误目标 |
| 模型层 | 选择最适合任务的模型,同时保持模型无关 | 第三方模型和客户环境约束 | 模型漂移和供应商依赖仍是持续风险 |
| 人工专家层 | 提供领域判断、标签、轨迹、评审和例外处理 | Meridial / 专家市场,加上收购来的 WeCP 评估基础设施 | 质量、吞吐和劳动治理指标尚未完全公开 |
| 评估和评分器层 | 衡量输出质量、校准奖励,并在上线前抓住失败模式 | Synapse 加定制评估框架、评分标准、对抗测试和人工复核 | 公开验证器指标偏薄,买方仍需尽调假阳性 / 假阴性率 |
| 监控和编排层 | 运行实时工作流、记录动作、比较版本并追踪运营 KPI | Axon 编排、可回放 RL 运行、受监控生产状态 | 留存资料没有公开可用时间页面、事故历史或支持 SLA 界面 |
本表把架构视作运营模型。风险聚焦公开材料尚未为尽调量化的内容。
[CE019, CE020, CE021, CE024, CE025, CE026]公开交付动作从选择工作流、连接系统开始,经过历史验证、上线运行,再进入持续复核。
[CE019, CE020, CE021, CE022, CE030, CE033]Invisible 的交付模式不仅靠模型可用性,还依赖系统访问、专家供给、验证器质量以及信任 / 治理证明。
该依赖图综合了公开材料对交付依赖的描述;它不是内部系统蓝图。
[CE019, CE023, CE025, CE026, CE017, CE018]5.3 部署证据、可靠性闭环与成熟度信号
公开案例研究显示,Invisible 的产品主张绑定可量化工作流结果,而不是抽象功能清单。Nasdaq 集成项目聚焦异构数据平台之间的互操作性, 将入驻时间缩短 63%,并节省 10,000 开发者工时。Headway 的理赔验证工作流使用批处理、并行处理和熟练的全球团队, 把处理速度提高 8x,同时成本低于内部团队和 BPO 替代方案。保险案例在更广流程尺度上展示了同样模式:Invisible 将自动化应用于发票核对、W9 处理、理赔信函和合规文档工作,并报告了准确率、周转时间和节省人工小时的提升。 这些仍是公司自撰证据,但它们展示的是围绕具体运营指标的实施。 产品成熟度在 Invisible 能指向可重复、评测密集型工作流的地方最强。You.com 项目在结构化相关性系统内使用了 20,000 次评测;Cohere 案例强调可信数据、连续可观测性,以及多语言或推理导向的人工评测。2025 年融资材料也通过点名平台领导层招聘和工程组织翻倍, 显示技术成熟度上升。限制是,Invisible 没有公开发布说明、正常运行时间报告、认证细节或正式支持 SLA 界面, 而这些通常是企业买方对成熟软件供应商的期待。因此,成熟度在工作流交付和评测引擎层面可信,但在软件治理层面只部分可见。[CE036, CE037, CE038, CE039, CE040, CE041]
| 日期 / 阶段 | 功能或里程碑 | 状态 | 含义 | 来源 |
|---|---|---|---|---|
| 2025-09-16 | 五层平台在融资材料中披露 | 公开宣布 | 模块地图被明确摊开,叙事也比早期只讲工作流更产品化 | 融资新闻稿 + Business Wire |
| 2025-09-16 | 工程组织翻倍;新增平台 CTO 和现场 CTO | 公开宣布 | 指向更重的软件和部署投入,但公开发布治理仍然偏薄 | 融资新闻稿 + Business Wire |
| 2026-03-10 | 收购 WeCP,补入评估库和面试记录 | 协议已宣布 | 强化专家验证基础设施和 RL 模拟资产 | WeCP 收购公告 |
| 当前公开界面 | 案例研究显示,接入、理赔、搜索和评估场景反复出现部署模式 | 公开证实 | 说明工作流在实施层面已有成熟度 | Nasdaq / Headway / 保险 / You.com / Cohere 案例 |
| 当前公开界面 | 信任门户和公开治理材料仍然较难获取 | 部分有证据 | 认证深度、事件透明度、支持成熟度仍缺少公开材料 | 信任门户 + 隐私政策 |
本表记录有日期的里程碑和有证据支撑的当前状态信号。它不是完整的软件发布日志,因为证据集中没有保留公开更新日志。
[CE017, CE018, CE036, CE037, CE040, CE041]公开证据对工作流交付和重评估用例最强,但对信任控制、正常运行时间披露等软件治理材料较弱。
成熟度评分是基于证据深度和部署证明得出的分析标签,不是公司提供的评级。
[CE036, CE040, CE041, CE042, CE049, CE050]5.4 差异化、专家网络与信任 / 合规控制
Invisible 最清晰的差异化,是把专家劳动力、工作流设计和评测基础设施绑进同一个交付模式。公司自己的 AI 训练和 RL 环境材料反复主张,领域专家、高质量人类轨迹和对抗性验证,比通用基准分数更重要。WeCP 收购通过把技术测评库和面试记录直接加入 Meridial 层,强化了这一投资逻辑。这让 Invisible 对单点解决方案厂商拥有一个可信切入口:它不是只卖模型端点, 也不是只卖工作流工具,而是把数据塑形、流程逻辑、专家验证和持续测量打包成企业 AI 采用的一个操作系统。 信任和治理故事则更混合。正面看,Invisible 称客户数据留在客户系统,发布了宽泛隐私政策,并链接到信任门户。隐私政策还罕见地具体披露了 代理人员监控、录制、客户访问工作信息和用户隐私权。负面看,本轮保留的信任中心抓取结果没有暴露详细控制映射或可访问认证证据, 准备好的证据集中也没有公开正常运行时间或事件披露入口。这很重要,因为外部法律和安全来源显示,围绕隐私、透明度、同意、训练数据披露和风险管理的环境正在变严。 相较 Appen 和 Cohere 的公开可比披露,Invisible 可访问的信任文档比其产品叙事更薄。[CE045, CE046, CE047, CE048, CE049, CE050]
| 控制或要求 | 状态 | 范围 | 缺口 |
|---|---|---|---|
| 客户数据留在客户系统内 | 明确主张 | 前置部署企业实施和安全视觉部署 | 留存资料中没有公开审计材料证明每条产品线都按此执行 |
| 用户隐私权(访问、可携带、更正、限制、删除) | 明确披露 | 隐私政策覆盖的网站和服务用户 | 政策层权利已公开,但产品特定保留时间表没有公开 |
| 代理人员工作监控和记录披露 | 明确披露 | 代理人员软件、线上会议和客户账户 | 敏感度高,因为文中提到键盘输入、屏幕截图和摄像头图像 |
| 信任门户 / 认证披露界面 | 门户存在,但留存抓取无法访问细节 | 安全与合规证明界面 | 本轮没有暴露可访问的控制映射或认证证据 |
| 红队测试、政策知情评估和专家验证 | 明确主张 | AI 训练、多模态和 RL 环境产品 | 公开方法以叙事为主;买方仍需要实证缺陷 / 升级指标 |
| 外部法律和安全预期 | 上升 | 隐私、透明度、同意、训练数据披露和 AI 风险管理 | Invisible 的公开信任界面比法律和伙伴比较基准暗示的要求更薄 |
本表区分政策披露和运营控制证明。最重要的缺口是详细信任材料能否访问,而不是门户链接是否存在。
[CE034, CE045, CE046, CE047, CE048, CE049]06客户情况
6.1 客户分层与广度
Invisible 的客户证据足够宽,可以判断公司不是单一小众厂商。保留的证据集覆盖前沿模型厂商、搜索和问答产品、金融数据与投资工作流、 健康和保险运营、零售目录工作、配送入驻、招聘平台运营、太阳能和居家服务工作流,以及早期公共部门和体育叙事。各细分里的买方和付款方不同: 模型提供商似乎购买评测和专家反馈,而企业运营负责人购买特定工作流中的吞吐、准确率和周期时间改善。 这种广度重要,因为它同时支撑两种不同增长动作。第一,Invisible 可以服务 Cohere 以及 WEF 和 AWS Marketplace 提到的更广模型提供商群体等技术要求高的 AI 构建者。第二,它可以向企业运营者销售工作流自动化;这些人关心入驻、理赔、 目录丰富或客户支持吞吐,而不是基准分数。来源组合仍有缺口:没有公开客户数量,头部客户名单未披露,分细分市场收入贡献不透明。 但可见证据足以说明,公司有真实的多垂直采用,而不是集中在单一用例。[CU001, CU002, CU003, CU004, CU005, CU006]
| 细分市场 | 买方 / 用户 / 付款方 | 用例 | 规模 / 证明 | 收入 / 战略价值 | 缺口 |
|---|---|---|---|---|---|
| 前沿模型提供商 | 模型评估负责人 / 标注员 / 模型构建方预算负责人 | 企业任务评估、RLHF、多语言和代码基准 | Cohere 具名证明,以及 >80% 头部提供商队列说法 | 有标杆客户标识的战略锚定客群 | 具体客户数和收入占比未披露 |
| 搜索和答案引擎 | 产品和搜索相关性团队 / 评分员 / 产品预算负责人 | RAG 相关性评分和搜索质量评估 | You.com 具名证明,以及上下文对话创业公司案例 | 一旦嵌入,可支撑周期性评估工作流 | 合同期限和部署范围未披露 |
| 金融数据和投资平台 | 产品、工程和研究负责人 / 最终用户是分析师和客户导入团队 / 企业软件预算 | 数据互操作和 AI 投资助手训练 | Nasdaq 与 Boosted.ai 具名证明 | 受监管信息工作流里的高信号客户标识 | 未披露合同金额 |
| 医疗科技运营 | 理赔 / 收入周期经理 / 理赔操作人员 / 运营预算 | 理赔处理吞吐和保险验证 | Headway 具名证明 | 证明 Invisible 能处理合规敏感型后台工作,信号有用 | 续约和量级增长未披露 |
| 保险后台 | 自动化和合规负责人 / 财务运营人员 / 运营预算 | 发票对账、W9 处理、理赔审批 | 全国性保险公司案例,有量化节省 | 显示可复制的降本和合规价值 | 客户未具名 |
| 零售与电商商品运营 | 市场平台或目录负责人 / 商品运营人员 / 收入运营预算 | SKU 信息丰富化和搜索可发现性 | 四大零售商案例,覆盖 50,000 个 SKU,ROI 为 9x | 可切入大批量目录经济 | 客户未具名 |
| 市场平台与配送入驻 | 入驻和供给增长负责人 / 入驻运营 / 运营预算 | 餐厅 / 菜单入驻和 OCR 驱动的数据抽取 | 配送平台案例,每月 1.5M 个数据点 | 指向大规模托管运营能力 | 客户未具名 |
| 招聘与人才平台 | 运营 / 数据质量负责人 / 质检操作员 / 运营预算 | 每日招聘帖质检和地点数据补全 | Getro 具名证明,节奏持续 | 有用的重复使用和满意度代理指标 | 未公开合同期限 |
| 太阳能与家居服务运营商 | 销售运营和财务团队 / 支持运营人员 / 获客预算 | 方案生成、融资合同支持和监控 | 太阳能供应商案例,峰值 180 份合同 / 日 | 若能持续,就是强落地扩张模式 | 客户未具名 |
| 公共部门与体育扩张 | 政府项目负责人或体育分析团队 / 分析师 / 项目或部门预算 | 模拟支持、模型评估和球探分析 | 公共部门和体育页面,加上 Hornets 叙事 | 可能具备战略价值的新垂直 | 证据质量较低,采购摩擦较高 |
本表只归类公开可见的细分客群,不代表收入占比、客户数量或完整市场覆盖。
[CU001, CU002, CU003, CU005, CU006, CU007]| 指标 | 数值 | 日期 / 新鲜度 | 来源 | 置信度 | 含义 | 缺失分母 |
|---|---|---|---|---|---|---|
| 具名客户证明广度 | 留存来源中有 6 个具名客户,外加 4 个量化匿名部署 | 当前页面抓取于 2026-06-04 | 案例研究 | 中 | 多个垂直里的采用是真实的 | 无公开客户数 |
| 第三方推荐证明广度 | 7 条评价和 16 个案例研究 / 客户故事 | 当前页面抓取于 2026-06-04 | FeaturedCustomers | 中 | 推荐证明不止 Invisible 自有页面 | 与相同底层客户标识的重叠未知 |
| 头部 AI 提供商队列说法 | >80% 的领先 AI 模型提供商,包括 Microsoft、AWS 和 Cohere | 最近的资料页面在 2026 年仍在线 | WEF、AWS Marketplace 与 CaseStudies.com | 中 | 指向强模型提供商定位 | 未披露具名名单或收入占比 |
| Nasdaq 入驻改善 | -63% 入驻时间;节省 10,000+ 开发者小时 | 当前案例研究 | Invisible | 高 | 具备生产特征的企业部署,并有量化 ROI | 合同金额未披露 |
| Headway 运营改善 | 快 8x;较内部团队 -37%;较此前 BPO -57% | 当前案例研究 | Invisible | 中 | 显示其替代了此前交付模式 | 理赔量和合同规模未披露 |
| You.com 搜索质量项目 | 20,000 次评估;相关性 +70% | 当前案例研究 | Invisible | 中 | 指向活跃且可衡量的评估工作流 | 时间窗口和基线未披露 |
| Boosted.ai 赋能 | 节省 90% 成本;第三批数据被称为解锁团队 | 当前案例研究 | Invisible | 中 | 暗示该账户内学习曲线快速改善 | 无持续运行率或续约期限 |
| 四大零售商目录项目 | 50,000 个 SKU;ROI 9x;30 天搭建后执行 16 天 | 当前案例研究 | Invisible | 中 | 显示能很快扩至高量级零售工作流 | 无稳态收入或复购数据 |
| 配送平台入驻爬坡 | 速度 +233%;成本 -50%;30 天组建 200 人团队;每月 1.5M 个数据点 | 当前案例研究 | Invisible | 中 | 大规模运营采用的强信号 | 量级是否持续未知 |
| 太阳能供应商扩张 | 方案支持扩至融资合同;峰值 180 份合同 / 日 | 当前案例研究 | Invisible | 中 | 明确的相邻工作流扩张信号 | 未披露合同期限或客户标识 |
| 全国性保险公司自动化 | 节省 $450k;节省 16,000 小时;审批速度提升 50%;准确率从 75% 到 98% | 当前案例研究 | Invisible | 中 | 受监管后台工作里的强 ROI 证明 | 未披露客户身份或完整流程数量 |
| Getro 持续节奏 | 每日批次,100% 质检记录,每两周通话 | 当前案例研究 | Invisible | 中 | 显示重复使用和服务管理节奏 | 无续约日期或年度支出 |
本表记录公开采用和 ROI 代理指标,不是已披露客户数或队列指标。缺失分母列标出公开证据还不足以支撑投资判断的位置。
[CU001, CU004, CU011, CU012, CU013, CU014]基于公开证据集,映射 Invisible 通常如何从一个工作流问题走向嵌入式交付和账户扩张。
阶段来自案例研究叙事的综合,而不是管理层披露的漏斗指标。
[CU020, CU021, CU022, CU035, CU040]6.2 具名证据与采用质量
最强的客户证据集中在少数具名案例研究,它们超越客户标识,提供了部署特定结果。Nasdaq、Headway、Cohere、Boosted.ai、 You.com 和 Getro 都具名出现。其中,Nasdaq、Headway、You.com 和 Boosted.ai 给出了最清晰的量化前后对比结果; Getro 则给出了最清晰的持续服务节奏和引用式满意度代理。Cohere 的证据更多关于评测质量,而非经济影响,但仍然重要, 因为它把 Invisible 放进了高要求的企业 AI 工作流,而不是通用标注任务。 不过,证据质量并不均匀。一些最好的数值结果来自保险、零售、配送和太阳能领域的匿名企业案例,意味着运营收益可见, 但客户身份不可见。第三方目录改善了广度,却大多只是聚合或概述公司叙事,而不是独立证明续约或支出。Hornets 案例最清楚地显示, 一个高知名度客户标识带来的叙事热度高于证据深度:反向文章表明,这一主张是通过 Invisible 托管的营销语言传播, 而不是官方球队发布。合并来看,本章支持真实采用,但信心水平仍应按证据质量分层,而不是把每个可见客户标识都视为同等。[CU011, CU012, CU013, CU014, CU024, CU025]
| 客户 | 细分市场 | 部署 / 用例 | 生产 / 试点 | 结果 / 证明 | 局限 |
|---|---|---|---|---|---|
| Cohere | 前沿模型提供商 | Command A 的企业任务评估和质量控制 | 接近生产的模型改进工作流 | 引述的质量门槛、盲测人类评估、企业任务焦点 | 未披露合同经济性或期限 |
| Nasdaq | 金融数据平台 | 新产品的数据互操作和客户导入 | 生产工作流 | 入驻速度提升 63%,节省 10,000+ 开发者小时 | 未披露支出或续约数据 |
| Headway | 医疗科技运营 | 带批处理和并行处理的理赔处理工作流 | 生产工作流 | 处理速度快 8x,成本低于内部团队和此前 BPO | 未披露量级分母或期限 |
| Boosted.ai | 投资研究平台 | 围绕 SLM 构建的 AI 投资助手的数据生产 | 支撑生产的工作流 | 节省 90% 成本,客户称第三批数据解锁了产品迭代 | 未披露上线后续约指标 |
| You.com | 搜索 / 答案引擎 | 结构化评分系统和搜索相关性评估 | 生产评估工作流 | 20,000 次评估,相关性提升 70% | 未披露收入金额或合同期限 |
| Getro | 招聘 / 人才平台 | 每日招聘帖地点处理,配质检和账户管理 | 生产托管服务节奏 | 每日处理、100% 质检记录、正向满意度引述 | 未披露年度支出或期限 |
| Charlotte Hornets | 体育分析 | AI 辅助选秀验证和计算机视觉分析叙事 | 不清晰 / 有争议的营销证明 | AWS Marketplace 和 WEF 提到该用例;反向文章称公开证明大多由 Invisible 托管 | 未找到 Hornets 官方发布或合同细节 |
本表是不完整的具名证明枚举,聚焦于有足够公开细节可描述部署的客户或品牌标识。匿名但有量化数据的案例研究不列入本表,而是在其他位置处理。
[CU001, CU024, CU027, CU031, CU032, CU033]| 证明载体 | 具名客户 | 可观察新鲜度 | 第二来源佐证 | 部署可信度 | 保留项 |
|---|---|---|---|---|---|
| 官方案例 | Nasdaq | 2026-06-04 仍在线 | 公共部门页面 + Nasdaq 首页上下文 | 高 | 仍是单边营销材料,未披露合同金额 |
| 官方案例 | Cohere | 2026-06-04 仍在线 | 体育 / 公共部门垂直页面 + Cohere 首页上下文 | 中高 | 未保留客户方确认 |
| 官方案例 | Headway | 2026-06-04 仍在线 | 未保留客户方页面 | 中 | 指标强,但证据仅由公司撰写 |
| 官方案例 | You.com | 2026-06-04 仍在线 | You.com 首页上下文 | 中高 | 未披露合同期限或 ACV |
| 官方案例 | Getro | 2026-06-04 仍在线 | 未保留客户方页面 | 中 | 节奏和引语不错,经济性可见度弱 |
| 第三方目录资料 | 混合引用 | 2026-06-04 仍在线 | FeaturedCustomers 与 CaseStudies.com | 中 | 覆盖面信号可能混有复用营销文案 |
| 市场 / 文章引用 | Charlotte Hornets / 顶级 AI 提供商 | WEF 和 AWS 资料在 2026 年仍在线;文章日期为 2026-02-24 | WEF + AWS Marketplace + 负面 OpenCourt 文章 | 中低 | 叙事曝光高,但佐证不对称 |
本表评估证据质量和新鲜度,而不是客户价值;目的在于把扎实的具名生产证据与较弱的叙事型引用区分开。
[CU024, CU025, CU031, CU032, CU033, CU039]按结果具体性、持续使用可见度和佐证强度比较具名客户证明。
置信度单元格反映来源质量和佐证情况,而不是客户价值或收入重要性。
[CU024, CU031, CU032, CU033, CU039, CU041]6.3 留存与满意度代理指标
公开留存数据是 Invisible 客户档案中最弱的一部分。没有可见的 NRR、GRR、流失率、续约率、合同期限或队列披露, 因此本章无法直接验证客户耐久性。这一缺失不是装饰;它是客户章节止步于高信心耐久性结论的主要原因。如果 Invisible 的收入基底由项目工作主导,或由少数超大型模型实验室客户主导,经济性可能明显弱于案例研究叙事。 不过,来源集也给出了一些次优耐久性代理。Getro 描述了每日协作、100% QC 记录和双周客户经理电话。配送平台案例称, Invisible 在 90 天内完全接入客户内部系统,之后每月处理 1.5 million 个唯一数据点。太阳能供应商案例称, 客户在初始工作流之外要求下游支持,这是典型先落地再扩张信号。这些都是账户深度和运营粘性的有意义指标, 但不能替代队列指标。因此,正确尽调姿态是把耐久性视为可信但文档不足,仍需要正式留存、合同期限和集中度数据。[CU019, CU020, CU021, CU022, CU023, CU030]
| 指标 / 代理 | 数值 | 细分市场 | 置信度 | 尽调要求 |
|---|---|---|---|---|
| 净收入留存率(NRR) | 公司整体 | none | 索取 AI 实验室与企业运营队列的 NRR | |
| 总留存 / 流失 | 公司整体 | none | 索取过去 24 个月的客户标识留存和流失数量 | |
| 平均合同期限 | 公司整体 | none | 审阅 MSA / SOW 样本和续约日历 | |
| 第三方推荐证明广度 | 7 条评价和 16 个案例研究 / 客户故事 | 公开推荐基础 | 中 | 确认多少推荐证明对应当前付费客户 |
| 日常运营节奏代理 | Getro 每日批次,100% 质检记录 | 招聘 / 人才平台 | 中 | 确认该节奏是否已持续 12+ 个月 |
| 嵌入系统代理 | 配送平台在 90 天内与内部系统完全集成 | 市场平台 / 配送 | 中 | 索取当前月度量级和续约条款 |
| 扩张请求代理 | 太阳能供应商请求下游融资合同和监控支持 | 太阳能 / 家居服务 | 中 | 索取范围变更历史和增量 ACV |
| 引述满意度代理 | Getro 称赞每日文档和每两周一次的客户经理通话 | 招聘 / 人才平台 | 中 | 索取近期 NPS、CSAT 或客户推荐电话 |
空值表示该指标未公开披露。非空行是耐久性或满意度代理,不可替代正式留存指标。
[CU019, CU021, CU022, CU023, CU024, CU030]展示有证据支撑的路径:从初始客户证明走向重复使用和扩张。
该图为定性图:它映射证据推进,不是已披露的数值销售漏斗。
[CU019, CU020, CU021, CU022, CU029, CU035]6.4 扩张与集中度风险
Invisible 的公开客户故事展示了可信的先落地再扩张模式。太阳能账户从方案生成扩展到融资合同支持和监控; 配送平台部署从流程重设计推进到集成后的月度处理;Getro 似乎按每日节奏经常性运行。这些都说明,一旦 Invisible 证明 ROI,成功工作流可以拓宽。公共部门和体育页面显示,管理层也在尝试把既有行业证据转化为相邻市场进入动作。 风险面同样重要。服务 80% 以上顶尖 AI 公司这一叙事具有战略吸引力,但也可能意味着少数超大 AI 实验室或超大规模买方对收入影响过大。 同时,公司没有公开客户数量、头部客户集中度数据或细分市场收入结构。由此无法判断公开证据集代表的是广泛付费账户基础, 还是几个旗舰客户标识之上的营销层。最安全的解释是,扩张潜力真实存在;但在管理层提供客户数量、ACV 和 top-10 敞口数据之前, 集中度仍是活跃的投资判断风险。[CU010, CU022, CU034, CU035, CU036, CU037]
| 扩张驱动因素 | 集中度风险 | 影响 | 尽调路径 |
|---|---|---|---|
| 账户内相邻工作流扩张 | 若只有标杆账户扩张,可能局限于少数大客户标识 | 支撑 ACV 增长,但可能遮蔽集中度 | 索取账户级扩张历史和 ACV 桥接 |
| 系统集成和持续节奏 | 嵌入式部署会提高切换成本,但只适用于完成集成的那部分客户 | 若大范围铺开,可能改善留存 | 索取装机基础按试点和已集成生产的分层 |
| 前沿模型提供商定位 | 少数超大规模 AI 实验室可能主导收入 | 若粘性强,上行空间大;若单一实验室流失,下行也大 | 索取头部 AI 实验室账户的收入结构 |
| 公共部门与体育扩张 | 采购周期长、工作流定制化,可能拖慢转化 | 有战略选择权,但现金兑现更慢 | 索取管线阶段、采购负责人和转化时间 |
| 匿名企业案例研究 | 匿名赢单难以尽调,可能夸大覆盖面 | 削弱集中度分析置信度 | 索取匿名化收入集中度表和客户标识授权 |
| 目录 / 评价广度 | 目录可能重复计算公开故事或过期客户标识 | 广度信号不错,收入信号弱 | 将目录引用映射到活跃账户和时间新鲜度 |
| Hornets 营销叙事 | 证据质量弱于核心案例研究,可能夸大体育牵引力 | 如果引用过猛,有声誉下行风险 | 索取官方客户推荐,或降低该说法优先级 |
| 无公开客户数 | 无法对标账户集中度或销售效率 | 尽调结论仍稳稳落在“继续研究” | 索取客户数、前 10 大集中度和队列数量 |
本表把可见的落地扩张机制,与仍阻碍集中度判断的缺口配对。
[CU022, CU035, CU036, CU037, CU038, CU039]6.5 客户判断
截至 2026-06-04,Invisible 跨过了最重要的第一道客户门槛:具名且量化的公开证据足够多,可以排除需求只是概念性的说法。 公司展示了跨多个垂直行业的真实采用,至少能指向数个具名客户,并反复用运营结果而不是空泛转型语言来定义客户价值。 这很重要,因为它说明产品或服务正在落进真实工作流,并产生可衡量后果。 限制因素是耐久性透明度,而不是采用可见度。公开来源没有披露客户数量、集中度、合同条款、NRR、GRR 或队列结果。 因此,本章支持一个正面但不完整的客户观点:采用是真实的,账户内扩张看起来可信,集中度 / 留存仍是最大未解问题。 投资者可以对客户牵引持建设性态度,但没有内部账户级数据,就不能诚实地称客户基础已被完全验证。[CU001, CU024, CU030, CU036, CU040]
6.6 证据项
07风险
7.1 监管与法律暴露
Invisible 的法律风险敞口,首先来自公司自己公布的控制边界。2026 年 3 月的隐私政策写明,Invisible 会把业务流程外包给代理人员,会处理客户和代理人员的个人信息,可能使用自动化决策或画像技术,也可能向服务提供商、商业伙伴、API 或 SDK、关联方和交易对手共享数据。对工作流自动化平台来说,这并不反常;关键在于,Invisible 的产品页和案例研究把公司放进了理赔审批、W9 处理、保险验证、财务入职和政府现代化项目,而不是低风险沙盒场景。换句话说,公司卖的不是抽象模型工具;它已经贴近真实工作流运营,隐私、公平性和告知义务都在那里变得重要。 2026 年的外部法律环境让风险更尖锐。欧盟《AI 法案》要求高风险系统配备人工监督、后市场监测和事件报告,其透明度义务也将在 2026 年 8 月生效。Baker Botts 描述了美国州法层面的同步碎片化局面,覆盖加州、德州、伊利诺伊州和科罗拉多州;多篇法律评论也提示职场 AI 歧视、隐私和监控风险。Alvarez & Marsal 又补上一条相邻执法线索:AI 包装和披露审查。Invisible 的年度现代奴隶制和供应商风险披露有缓释价值,但也确认了一个分布式劳动力与供应商模式,需要持续尽调,不能盲目信任。[CR007, CR008, CR009, CR010, CR011, CR014]
| 风险领域 | 规则 / 触发点 | 风险敞口成因 | 发生概率 | 严重程度 | 缓释成熟度 | 剩余敞口 | 尽调路径 |
|---|---|---|---|---|---|---|---|
| 自动化决策、隐私与就业 AI 法律 | EU AI Act 监督和透明度义务,以及 2026 年州级 AI 与职场规则 | Invisible 在保险、医疗、金融和类 HR 企业任务中发布自动化决策表述和案例 | 中高 | 极高 | 中等 | 高 | 审阅 DPA、通知、影响评估和客户工作流地图 |
| 跨境数据传输和供应商治理 | 全球数据传输、服务提供商、API/SDK 和交易共享义务 | 隐私政策覆盖代理和客户数据、国际传输以及多条第三方共享路径 | 高 | 高 | 中等 | 高 | 按产品线索取子处理方清单、SCC 和数据流图 |
| 公共部门采购和安全合规 | 政府采购、安全审查和关键任务可靠性预期 | Invisible 启动公共部门打法,并引用联邦机构工作,但公开授权证据较薄 | 中 | 高 | 中低 | 高 | 获取采购工具、安全问卷和仍在服务的参考客户 |
| 劳工与供应链合规 | Modern Slavery Act 及更广泛的劳工 / 供应商监督 | Invisible 依赖代理、供应商和全球分布式运营,同时承认采购类别风险更高 | 中 | 中高 | 中等 | 中 | 检查入职控制、审计节奏和升级记录 |
| AI 包装与信息披露风险 | 监管会审查夸大的 AI、控制或合规声明 | 成长期 AI 营销、估值信号和模型供应商覆盖声明,可能招致披露审查 | 中 | 中高 | 中低 | 中高 | 用客户合同、指标和控制证据核对营销表述 |
严重程度排序基于公司政策表述、受监管用例证据与 2026 年外部法律进展的组合;客户特定法律状态不公开,因此覆盖不完整。
[CR007, CR008, CR009, CR010, CR011, CR014]Invisible 六个主要风险集群的发生概率、影响、缓释成熟度和剩余严重性。
评分是作者基于公开证据库的判断;缓释成熟度只反映保留来源中可见的控制。
[CR028, CR032, CR037, CR041, CR046, CR048]7.2 运营、安全与质量风险
Invisible 确实有运营证据,但这些证据是双刃剑。公司称,前置部署工程师会把客户的遗留系统和运营数据库接入其平台,生产部署会跟踪吞吐量、错误率、资源效率和单笔交易成本,并保留可审计文档。这套表述指向运营责任,而不是轻量软件辅助。案例研究进一步证明这一点:Invisible 正在触及保险理赔审批、医疗验证、财务入职、专家财务 QA 和企业模型评估。公司离真实运营流这么近,主要风险就不再是 AI 能否创造价值,而是工作流跨更多行业和交易对手扩张后,质量保证、异常处理和控制证据能否继续撑住。 公开缓释因素也存在。Invisible 对外强调持续评估、红队测试、专家网络和人工监督,Boosted.ai 也明确称,没有人工复核,这套工作流跑不起来。这些控制重要。剩余问题在于透明度。留存的公开来源没有点名第三方安全审计机构,没有披露事件日志,也没有确认公共部门工作的授权范围。《International AI Safety Report》正好框定了这个问题:智能体和通用系统需要持续监测治理,而不是口头声明治理。因此,投资者看到的运营风险严重程度中高;设计意图不错,但控制有效性的外部证据仍不完整。[CR005, CR006, CR012, CR013, CR021, CR022]
| 失效模式 | 发生概率 | 严重程度 | 缓释成熟度 | 剩余敞口 | 未解决缺口 |
|---|---|---|---|---|---|
| 理赔、入职或审批等受监管运营中的工作流错误 | 中 | 高 | 中等 | 高 | 未公开按客户工作流划分的错误率历史或异常率披露 |
| 服务提供商、API 或跨境传输引发的数据治理或隐私故障 | 中 | 高 | 中等 | 高 | 未披露具名审计方、事件日志或详细控制范围证据 |
| 专家网络在不同领域和 80+ 种语言中的质量不一致 | 中 | 中高 | 中高 | 中高 | 公开来源未披露审阅员校准、缺陷漏出或返工率 |
| 代理式或人机协同工作流中的模型上线迁移失败 | 中 | 中高 | 中等 | 中高 | 未公开生产可靠性看板或模型回滚指标 |
| 分布式运营中的供应商或劳动力监督失灵 | 中 | 中 | 中等 | 中 | 公开信息仅有年度现代奴隶制审查和入职项目 |
发生概率和剩余风险评分是作者基于公开运营模型作出的判断;事故或认证信息未公开披露,应视为尽调事项,而不是没有风险的证据。
[CR005, CR006, CR012, CR013, CR021, CR022]Invisible 的核心运营和模型风险如何传导到客户、利润率、融资和投资逻辑结果。
[CR005, CR022, CR032, CR037, CR041, CR042]7.3 合作伙伴、平台与客户依赖风险
Invisible 的生态杠杆同时也是一层依赖栈。管理层称,公司为全球领先模型提供商中的 80% 以上训练过基础模型,并点名 Cohere、Microsoft 和 AWS;AWS Marketplace 也确认了一条可见的伙伴上市渠道。这些关系增强可信度,但也意味着公司的增长叙事部分依赖外部生态,而这些生态可以很快改变定价、平台规则或自研 / 采购取向。同样逻辑也适用于 WeCP:收购扩展了专家验证和 RL-gym 能力,但产品和团队完全整合前,它既是缓释因素,也是执行依赖。 客户和渠道依赖同样真实,尽管公开集中度数据缺失。最强的公开证据点集中在金融、医疗、保险和公共部门邻近工作流,这些行业价值高,但推进慢,采购和治理预期也更苛刻。Invisible 自己的 Scale-AI 对比页承认,受监管买方需要比通用标注平台更强的控制、访问管理和决策文档。与此同时,Appen 仍在营销全球贡献者网络,UiPath 则用公开 ARR 和大客户指标来营销受治理的自动化。结果是,Invisible 的依赖画像要求它同时稳住多个生态:超大规模云厂商、专家供给、收购来的验证资产,以及对可信度高度敏感的企业客户。[CR014, CR015, CR016, CR027, CR034, CR035]
| 依赖项 | 交易对手 | 作用 | 集中度 | 失效情景 | 严重程度 | 缓释措施 | 剩余敞口 |
|---|---|---|---|---|---|---|---|
| 模型厂商和超大云厂商生态 | Microsoft / AWS / Cohere 及其他头部模型供应商 | 需求信号、参考可信度和平台关系 | 高 | 支出减少、内部自建或价格压力削弱 Invisible 的训练和平台叙事 | 高 | 转向企业工作流和更广的软件模块 | 高 |
| 市场和平台销售路径 | AWS Marketplace / AWS 生态 | 分销和伙伴可发现性 | 中高 | 市场准入、费用或战略一致性变化削弱可见企业渠道 | 中高 | 直销叠加更广伙伴网络 | 中高 |
| 收购获得的验证能力 | WeCP 团队和评估库 | 专家验证、RL 训练场和招聘信号基础设施 | 中 | 整合延误或人才流失导致预期质量或速度提升落空 | 中高 | Meridial 整合计划和保留的产品重点 | 中高 |
| 公共部门渠道和采购打法 | 联邦部门、机构和公共部门买家 | 新增长向量和受监管工作可信度 | 中 | 销售周期慢、采购失败或安全审查摩擦延迟收入转化 | 中高 | 专职公共部门领导和行业化信息传递 | 中高 |
| 大型企业参考客户 | 保险、医疗、金融和企业 AI 客户 | 价值证明和工作流嵌入 | Unknown | 少数参考账户或行业可能贡献过高证据权重和收入集中度 | 中高 | 跨行业多个用例,但集中度仍未披露 | 中高 |
集中度根据公开叙事评分,而非已披露收入占比,因为客户数量和合作伙伴收入数据不公开。
[CR014, CR015, CR016, CR027, CR034, CR035]对 Invisible 剩余风险画像最重要的交易对手和运营接触面。
[CR014, CR015, CR016, CR034, CR039, CR040]7.4 团队与执行风险
Invisible 正在同时推进数个艰难转型。短时间内,公司更换 CEO、工程团队翻倍、办公室扩展到 New York、San Francisco、Washington, D.C. 和 London,启动专门的公共部门打法,还增加了一项必须整合进核心平台的收购。这些动作本身都不负面。事实上,Fitzpatrick 的企业 AI 背景是文件里更清晰的缓释因素之一。风险来自并发推进。一家 350 人公司叠加分布式代理人员模式,需要扩大的不仅是人才招聘,还有管理系统;当工作流横跨受监管行业和政府买方时,控制负担还会进一步上升。 公开记录显示 Invisible 明白严谨性的必要,但还没证明稳定的运营节奏。支撑乐观判断的同一批来源,也暗示了执行拉伸:资本正在投向软件模块和领导层扩张,现代奴隶制监督依赖年度审查和供应商入职,WeCP 整合还需要产品、文化和销售落地对齐。公共部门扩张再次抬高门槛,因为安全审查和采购周期比私人试点更不容错。这不是危机画像,但公司仍需要把快速组织变化转化为可重复的控制证据。[CR003, CR004, CR010, CR014, CR016, CR039]
| 角色 / 职能 | 依赖或缺口 | 发生概率 | 严重程度 | 缓释措施 | 尽调路径 |
|---|---|---|---|---|---|
| CEO 与高管梯队 | Fitzpatrick 是近期上任的 CEO,必须证明组织变化后仍能维持稳定运营节奏 | 中 | 高 | 深厚企业 AI 背景,加上创始人在董事长层面保持连续性 | 访谈核实现任高管梯队、继任规划和董事会运转节奏 |
| 工程和产品扩张 | 2025 年工程团队翻倍,同时公司在扩展软件模块、客户工作流和地域 | 中 | 高 | 新资金和可见技术领导招聘 | 索取组织架构、发版节奏、事件复盘流程和平台可靠性 KPI |
| 公共部门和地域扩张 | 华盛顿 D.C. 和伦敦扩张增加采购与执行复杂度 | 中 | 中高 | 专职公共部门和 EMEA 领导 | 审阅分部层面的管线质量、成交率和合规人员配置 |
| 收购整合 | WeCP 整合增加产品、人员和商业化协同工作 | 中 | 中高 | 围绕专家验证的聚焦整合逻辑 | 审阅交割后里程碑、留任包和产品路线图整合 |
| 劳动力和供应商监督 | 分布式代理、供应商和国际传输带来超出简单软件公司的控制开销 | 中 | 中高 | 已有年度风险审查和入职控制 | 检查审计节奏、升级数据和质量治理人员配置 |
严重程度按领导层更替、工程扩张、地域增长和收购整合同时发生的压力判断,而不是基于任何已披露组织失灵。
[CR003, CR004, CR010, CR014, CR016, CR039]7.5 财务与商业模式风险
Invisible 的财务风险画像好于多数私有 AI 基础设施故事的中位数,但支撑它的证据还不完整。公司有真实收入规模、意义明确的 2025 年增长轮,以及第三方盈利信号。这降低了即时融资压力。更难的问题是商业模式耐久性。Sacra 分析称,实验室越来越转向合成数据生成,并明确把 Invisible 的回应描述为转向企业部署。公开案例研究支持这个转向,但没有告诉投资者,如今收入有多少来自软件模块、有多少来自劳动力支撑的服务,客户群集中度多高,或不同交付模式下毛利率如何。 这种不确定性重要,因为 Invisible 现在已经按规模化 AI 平台故事定价。SiliconANGLE 报道其估值超过 $2 billion,而 UiPath 等公开自动化可比公司会披露 ARR、大客户数量和留存背景,Invisible 没有。风险不在于 Invisible 缺需求;公开证据恰恰相反。风险在于,如果企业端结构迁移放慢,或劳动力强度在经济上比投资者预期更久地占主导,那么偏高估值、不完整披露和潜在合成数据替代压力会一起放大。[CR001, CR002, CR017, CR018, CR019, CR020]
7.6 缓释框架与否决条件
Invisible 看起来不像粗放运营者。政策、产品和案例研究材料里,公司反复回到人工专业能力、评估闭环、工作流指标、可审计文档,以及在客户系统内受治理地落地。这些主题,正是把 AI 推进受监管运营界面时需要抓住的主题。现代奴隶制声明和供应商风险计划也说明,管理层至少试图把供应链和劳动力监督当成真正的治理任务。这是正面逻辑。 因此,决定性标准不是通用创业指标,而是证据门槛。投资者应要求公司证明:隐私和自动化决策控制在真实部署中撑得住;安全尽调和事件报告不止停留在营销语言;公共部门工作拥有真实授权,而不是只有愿景;企业软件收入占比确实上升得足够快,可以抵消旧 AI 训练收入线中的合成数据替代。如果公司拿不出这些证明,或者质量指标和关键关系恶化、估值预期仍居高不下,投资逻辑就应从风险可控的上行,转为回避或重新定价。[CR009, CR010, CR012, CR027, CR032, CR041]
| 风险 | 可监控触发点 | 阈值 / 事件 | 行动含义 |
|---|---|---|---|
| 隐私和自动化决策合规 | 监管或客户法律升级 | 正式调查、DPIA 失败,或无法为上线工作流展示自动化决策通知和 DPA | 先暂停判断,直到法律控制包完成审阅;若主动执法将 Invisible 或其工作流列为根因,投资假设破裂 |
| 受监管工作流中的运营质量 | 工作流质量恶化 | 理赔或入职错误重复出现、缺少异常看板,或无法提供事件复盘数据 | 假设留存韧性更低、流失风险更高;若拿不出质量指标则回避 |
| 安全和控制成熟度 | 独立控制证据缺口 | 尽调要求后仍无具名安全审计方或事件历史材料包 | 对管理层说法打折,在证据到位前把控制成熟度视为未经证明 |
| 合作伙伴和生态依赖 | 关键关系或整合承压 | 重要模型供应商或市场关系流失,或 WeCP 整合里程碑延误超过 12 个月 | 下调增长和缓释假设,重新承销平台杠杆 |
| 人才和执行 | 领导层或扩张断裂 | 意外高管流失、工程交付持续放缓,或公共部门扩张缺乏合规人员配置 | 转入观察名单,或在运营节奏稳定前回避 |
| 财务和模式组合 | 企业业务占比提升未能抵消合成数据压力 | 下一轮融资平轮或降价、无收入组合桥,或证据显示劳动支撑交付仍是主要利润率驱动 | 回避或大幅重新定价,因为估值支撑不再有证据基础 |
这些阈值是锚定公开风险栈的投资者监控启发式标准,而不是公司披露的内部护栏。
[CR009, CR012, CR027, CR032, CR034, CR041]08估值
8.1 融资背景与入场纪律
Invisible 进入估值讨论时,公开证据强于多数后期私有 AI 公司:公司、Business Wire、SiliconANGLE、TechNews180 和 Intelligence360 均印证了 2025 年 9 月的 $100 million 增长轮,该轮把已披露融资推至 $144 million;官方和分析师来源也把 2024 年收入锚定在 $134 million。这些证据足以否定随口一句「纸面独角兽」。更难的问题是价格纪律。即便用保守的 $2.0 billion 作为最新标记下限,公司按 2024 年收入计算的历史收入倍数也超过 14.9x。这个水平低于最激进的前沿模型式私有 AI 可比公司,但这家公司公开证据显示 EBITDA 估计为 11%、交付引擎依赖 3,000 多名代理人员,且未披露毛利率、烧钱速度或股权结构细节;这样的价格仍然偏贵。因此,正确入场姿态不是不相信业务,而是不要在缺少 2025/2026 年收入、软件 vs 服务毛利率以及下行保护条款的私下桥接证据时,为名义估值付钱。[CV001, CV002, CV003, CV004, CV005, CV006]
| 维度 | 评估 | 证据基础 |
|---|---|---|
| 投资建议 | 继续研究 — 只有在私有尽调能降低价格风险或揭示更低有效入场价时才保持跟进 | 公开证据证明规模和客户价值,但当前财务细节不足,无法承销最新估值 |
| 置信度 | 中 | 融资轮、收入和客户证据佐证充分,但股权结构、留存和毛利率可见度仍缺失 |
| 风险评级 | 高 | 公司是真实业务,但估值倍数压缩风险、劳动密集度、治理审查和条款缺失叠在一起,容错空间很小 |
| 估值立场 | 按公开证据看偏高 | >14.9x 的历史收入倍数只有在当前收入、利润率组合和企业客户韧性较上一个公开收入锚点大幅改善后才站得住 |
| 回报 / 持有视角 | 按当前估值,在公开假设下,基准情形总 MOIC 在 3-5 年持有期内约为 0.9x-1.4x | 除非私有尽调证明前瞻经济性强得多,否则低于晚期风险投资式入场通常目标 |
| 决策含义 | 进入观察名单以外的尽调前,必须拿到收入桥、利润率瀑布、集中度数据和融资条款 | 没有这些文件,名义估值对业务方向可以大体正确,但对新增资金仍可能缺乏吸引力 |
本建议刻意对价格敏感。它评估当前公开证据是否支撑最新估值,而不是抽象判断 Invisible 是否是一家优秀公司。
[CV004, CV007, CV010, CV020, CV035, CV041]基于公开的 2024 年收入 $134 million 锚点,计算不同收入倍数下的企业价值。
本图刻意把收入保持在最近公开的 2024 年基数不变,让读者看清争议有多少来自分母风险,又有多少来自倍数选择。
[CV008, CV010, CV029, CV030, CV033, CV041]8.2 投资逻辑、反向逻辑与可比公司组
核心乐观逻辑有证据支撑。Invisible 有真实产品宽度,不只是劳动力市场:材料描述了面向数据基础设施、工作流映射、专家工作、评估和编排的模块化软件层,客户案例也显示,在医疗、保险、金融数据入职和 AI 模型改进中产生了可衡量部署 ROI。WEF 画像和官方发布也支持这样一家企业:已经盈利、广泛部署,并越来越围绕企业工作流所有权展开,而不是只守着狭窄的 RLHF 切口。反向逻辑同样真实。Sacra 对收入和利润率的框定,仍然更像技术赋能服务,而不是纯软件;同一来源还警告,合成数据采用会挤压传统标注和 RLHF 需求。行业基准让这种张力更尖锐。应用 AI 和数据智能业务仍可获得溢价定价,但 Finro 和 Aventis 都认为,后期投资者越来越会把前沿模型基础设施,与运营强度更高的应用模型区分开。这意味着,Invisible 需要企业耐久性和软件式利润率扩张,才能继续捍卫远高于公开软件常态的倍数。[CV011, CV012, CV013, CV014, CV015, CV016]
| 论点 | 证据 | 改变判断的条件 |
|---|---|---|
| 正方:企业工作流证据真实 | Headway、Nasdaq、保险客户和 Boosted.ai 都展示了可衡量的运营或成本结果,而不只是愿景式试点 | 如果客户集中度实际很高,或已计量部署无法续约,论点会削弱 |
| 正方:平台不只是劳动力市场 | Invisible 描述五个模块层,并支持模型无关部署;官方产品页展示数据、工作流、评估和编排能力 | 更强结论需要证明软件模块能不依赖专家劳动力推动毛利率扩张 |
| 正方:当前估值不如头部私有 AI 异常值极端 | 按保留倍数锚点筛选,Invisible 低于 Mercor,也略低于 Scale AI | 只有新的收入披露显示实际当前倍数已低于 2024 年历史口径,判断才会改善 |
| 反方:经济性仍像劳动辅助模式 | Sacra 的 3,000+ 名代理足迹和约 11% EBITDA 估计,说明交付引擎尚未变成纯软件 | 如果管理层能按产品线证明软件驱动毛利率和经营杠杆,担忧会缓解 |
| 反方:传统 RLHF 和标注需求可能商品化 | Sacra 警告模型实验室正转向合成数据,这会削弱纯训练数据楔子的价值 | 如果企业工作流收入已经明确成为主导增长引擎,担忧会缓解 |
| 反方:公开分母陈旧,融资条款清单隐藏 | 尽管估值台阶很大,公开信息没有披露 ARR、NRR、集中度或优先权栈 | 当前董事会材料和干净融资文件会最快推动建议变化 |
本表把公司质量论点和价格质量论点拆开。业务可以很强,但当前入场价仍可能过不了新增资金回报测试。
[CV011, CV012, CV015, CV016, CV017, CV018]| 可比对象 | 指标 | 倍数 / 估值 / 状态 | 可比意义 | 局限 |
|---|---|---|---|---|
| Invisible 当前估值标记 | 2024 年收入 $134M;2025 年估值 >$2B | 隐含 >14.9x 往绩收入 | 本次承销判断的直接对象 | 使用过时的公开分母,优先股堆叠未知 |
| Scale AI | Sacra 称 ARR 约 $1.5B,估值 $25B | ~16.7x 收入 | 最接近的存续私有参照,代表高溢价 AI 数据 / 对齐业务 | 比 Invisible 更暴露于前沿实验室,稀缺性叙事也更强 |
| Mercor | Sacra 称收入运行率约 $50M,估值 $2B | ~40x 收入 | 说明市场可以把具备软件型增长和人才网络杠杆的 AI 平台定到多高 | 运营模型轻得多、规模也更早期,只能作为上限,不是真正同业 |
| 大型交易 AI 中位数 | Aventis 大型 AI 融资与 M&A 样本 | 24.2x 收入倍数中位数 | 可作为后期 AI 融资的上限市场基准 | 样本偏向更大的赢家和融资估值,不适合直接当作现实新钱入场纪律 |
| 应用 AI / 上市软件基准 | Aventis + Finro 对 2025 年末基准的看法 | AI 融资约 25x-30x,但上市 SaaS 约 6x,应用型细分会向软件倍数回归 | 用来校验 Invisible 在变得难以自圆其说前还能拉伸多远 | 行业基准,不是直接公司可比 |
| UiPath / Appen 披露基准 | UiPath 披露 ARR $1.901B、DBNRR 109%;Appen 披露 1M+ 贡献者、10B 单元处理量 | 上市公司或规模化披露可比对象,KPI 维度比 Invisible 更丰富 | 有助于判断退出准备度,以及成熟买家已经能拿什么来比 | 所用来源没有在同一时点给出两家公司干净的 EV/revenue 标记 |
这组可比混合了直接私有参照和适配模型的基准,因为企业 AI 工作流运营商很少披露足够公开数据,无法拼出干净的同口径表格。
[CV010, CV024, CV025, CV026, CV028, CV029]以投委会视角给 Invisible 当前的市场位置、验证强度、经济性、治理就绪度和估值纪律打分。
分数只是方向性判断工具,不是机械模型;较低分值主要反映证据缺口和价格风险,并不否认业务质量。
[CV015, CV029, CV031, CV035, CV036, CV037]8.3 情景区间与回报逻辑
由于最后一个干净的公开收入锚点是 2024 年,情景测算必须明确哪些是已观察事实、哪些是假设。已观察锚点很直接:2024 年收入 $134 million;按最新一轮计算,隐含历史收入倍数超过 14.9x;Sacra 给出 11% EBITDA 估计;运营足迹偏劳动力密集;市场证据显示,应用 AI 现在的定价更接近软件基准,而不是前沿模型的极端水平。这些事实指向当前入场价下不对称的回报测算。悲观情景里,增长停滞、市场倍数压缩到 6x-9x 收入,最新估值会有明确下行。基准情景里,Invisible 继续切入企业工作流并维持十倍出头的倍数,大致能守住名义估值,但很难明显产生风投式回报。只有乐观情景——2025/2026 年收入显著跑赢滞后的公开基数,且软件驱动利润率改善——才会创造清晰且有吸引力的上行。换句话说,当前价格也许扛得住,但不宽容。[CV006, CV007, CV008, CV009, CV010, CV020]
| 情景 | 收入假设 | 倍数逻辑 | 指示性价值 / 总 MOIC | 概率信号 | 主要下行 / 上行触发因素 |
|---|---|---|---|---|---|
| 悲观 | $150M-$170M 收入基数 | 6x-9x 收入;如果增长回归常态、业务组合仍偏劳动力密集,更接近应用软件和科技赋能服务的结果 | $0.90B-$1.53B / 相对 >$2B 入场价约 0.4x-0.8x | ~30%:如果企业客户组合停滞,或市场压缩到接近软件型倍数,这一情景说得通 | 合成数据替代、利润率扩张乏力,或客户组合出现意外 |
| 基准 | $180M-$210M 收入基数 | 10x-14x 收入;假设企业工作流继续复合增长,公司相对上市软件保持溢价,但尚未证明前沿模型公司级稀缺性 | 价值 $1.80B-$2.94B;总 MOIC ~0.9x-1.4x | ~50%:最贴合当前公开证据 | 需要一条清晰的 2025/2026 桥接,软件贡献毛利尚可,且没有难看的优先股负担 |
| 乐观 | $220M-$260M 收入基数 | 14x-18x 收入;假设软件占比更高、企业留存稳固,且数据智能使能方继续拿到溢价定价 | 价值 $3.08B-$4.68B;总 MOIC ~1.5x-2.3x | ~20%:需要执行表现兑现,而当前公开记录还没证明 | 企业规模化、利润率拐点和干净融资条款共同支撑高溢价退出路径 |
区间只是指示性,不是精确预测。它们有意展示:如果缺少当前收入桥接,也没有比公开记录已证明的更好利润率组合,新钱能看到的上行空间很有限。
[CV010, CV029, CV030, CV031, CV032, CV041]以最后公开的收入基数为锚,按明确的收入和倍数假设,拼出悲观、基准、乐观三种估值区间。
区间受证据约束,但假设成分仍然很重;公开来源没有披露 2025 或 2026 年实际收入、产品组合或融资条款。
[CV010, CV030, CV041, CV045, CV046, CV048]8.4 建议、触发条件与最终尽调问题
建议是继续研究,置信度中等、风险高、估值立场偏高。这个判断刻意对价格敏感。公司已有足够公开证据,值得留在尽调清单:真实收入、真实企业成果、可信的 2025 年融资,以及超越简单标注的产品故事。但同一批证据也显示,尽调支撑缺口仍大。Invisible 没有公开给出 2025/2026 年收入桥、软件 vs 服务的毛利率组合、集中度和续约数据,也没有披露名义估值背后的优先股堆叠。公开 AI 和自动化可比公司维持着投资者关系页面和文件披露节奏,Invisible 还没跟上,因此近期 IPO 视角为时过早。实际含义很简单:只有尽调能验证企业工作流增长具备耐久性、利润率扩张真实、2025 年轮次条款干净,这个名字才值得保持跟踪。如果这些检查失败,当前估值标记应被视为脆弱,而不是鼓舞人心。[CV035, CV036, CV037, CV038, CV041, CV042]
| 触发因素 | 阈值 | 对投资逻辑的传导 | 行动含义 |
|---|---|---|---|
| 前瞻收入桥接不及预期 | 2025 年实际值或 2026 年收入运行率,相比公开的 2024 年 $134M 基数没有明显台阶 | 估值会不再像高溢价数据智能标记,而更像过时的后期轮价格 | 将情景重置到悲观价值区间,不再把最新轮作为可辩护锚点 |
| 毛利率组合仍偏劳动力密集 | 管理层无法证明软件带动毛利率扩张,或贡献毛利随规模改善 | 这家公司仍主要是劳动力辅助服务模式的反向逻辑会占上风 | 将倍数框架下调到上市软件或科技赋能服务区间 |
| 企业转型跑不赢 RLHF 商品化 | 模型开发方工作仍是主引擎,而合成数据替代侵蚀定价权 | 支撑乐观情景的核心战略转向就没有得到证明 | 把公司视为结构性暴露于传统数据标注压缩压力 |
| 治理或供应商控制缺口浮现 | 尽调发现 AI 治理、供应商监督或文档相对企业买家预期偏弱 | 公司会失去高溢价企业定位最强的论据之一 | 暂停承销,直到控制措施被修复并验证 |
| 退出路径仍不透明 | 管理层无法展示可信的战略出售、财务赞助方或最终上市准备路径,并配有可衡量里程碑 | 即使业务站得住,以当前价格仍可能交出差回报 | 将项目保留在继续研究状态,而不是推进到可提交 IC 的承销判断 |
这些触发因素聚焦会快速改变价格纪律的可衡量阈值。设计目的不是用通用风险装饰备忘录,而是阻止自我合理化。
[CV035, CV036, CV037, CV040, CV043, CV044]| 主题 | 缺失证据 | 为何重要 | 负责人或尽调路径 |
|---|---|---|---|
| 当前财务桥接 | 董事会批准的 2025 年实际值、2026 年收入运行率,以及从公开 2024 年基数出发的增长桥接 | 没有当前分母,最新估值就无法换算成真实入场倍数或回报情景 | CEO、CFO 和月度董事会材料 |
| 利润率架构 | 按软件模块、专家工作和托管服务交付拆分的毛利率瀑布 | 核心估值争论是 Invisible 是否在变成软件主导,还是结构上仍靠劳动力辅助 | FP&A、产品财务和分部盈利能力拆分 |
| 需求质量 | Top-10 客户集中度、续约、NRR 或队列留存,以及 AI 实验室与企业工作流的组合 | 如果增长更多元、粘性更强,而不是依赖波动的标注项目或少数大客户,它的价值更高 | 收入运营和客户成功分析 |
| 股权结构表和下行保护 | 股权类别、清算优先权、参与权、认股权证,以及 2025 年融资中的任何附函 | 如果普通股经济性明显差于投后估值暗示,光看名义估值不够 | 主办律师和财务运营 |
| 治理就绪度 | AI 治理框架、供应商监督控制,以及面向受监管企业买家的文档 | 治理越来越像销售护城河的一部分,也是拿到可信高溢价倍数的前提 | 首席法务官、合规负责人和审计材料包 |
| 退出地图 | 战略买家地图、财务赞助方兴趣,以及任何最终上市公司就绪的里程碑 | 入场价偏高仍可能跑通,前提是退出可选性真实且有时间边界 | CEO、董事会材料和投行参考 |
每项请求都直接绑定会改变建议的变量。没有一项是装饰性要求;这些是最可能改变价格纪律或击穿投资逻辑的缺失文件。
[CV035, CV037, CV042, CV043, CV044, CV045]真实运营证据、人工辅助经济模型、披露缺口和估值倍数纪律如何合力导向继续研究结论。
这是决策逻辑,不是确定性模型。它显示在当前价格下,哪些证据会推动建议变化。
[CV004, CV010, CV037, CV041, CV044, CV045]免责声明
本报告依据 Invisible Technologies 在 invisibletech.ai/invisible.co 的公开证据;invisible.ai 看起来是另一家公司,本报告将其作为身份冲突数据点,而不是研究对象。
证据索引
| 编号 | 陈述 | 可信度 | 来源 |
|---|---|---|---|
| CO001 | Invisible positions itself as an enterprise AI platform that structures messy data, deploys agentic workflows, and adds human experts where needed. | 高 | SO001, SO002, SO003 |
| CO002 | Invisible says forward-deployed engineers connect customer legacy systems to a model-agnostic platform while customer data stays in customer environments. | 中 | SO003 |
| CO003 | Invisible’s AI training offering spans domain-expert training, multilingual evaluation across 80+ languages, reinforcement-learning environments, multimodal data generation, and red-teaming. | 中 | SO004 |
| CO004 | Invisible’s privacy policy states that the company delivers digital work by outsourcing business processes to human agents. | 中 | SO007 |
| CO005 | Sacra says Invisible was founded in 2015 and evolved from a virtual-assistant service into an outsourcing and automation platform for AI training and enterprise workflows. | 中 | SO020, SO021 |
| CO006 | The target company is Invisible Technologies at invisibletech.ai rather than Invisible AI at invisible.ai, whose site describes a separate manufacturing visual-intelligence product. | 高 | SO001, SO030 |
| CO007 | Official announcements and the California complaint both place Invisible Technologies in San Francisco, supporting San Francisco as the publicly evidenced operating base. | 高 | SO009, SO013, SO025 |
| CO008 | The California labor complaint describes Invisible Technologies Inc. as a Delaware corporation registered to do business in California with its principal place of business in San Francisco. | 中 | SO025 |
| CO009 | Matthew Fitzpatrick became CEO of Invisible on 2025-01-21. | 高 | SO009, SO013, SO018 |
| CO010 | Before joining Invisible, Fitzpatrick led QuantumBlack Labs at McKinsey and oversaw roughly 1,000 engineers and product leaders. | 高 | SO009, SO013, SO018 |
| CO011 | Francis Pedraza is the founder and chair or executive chairman in the reviewed governance materials. | 高 | SO008, SO013, SO018 |
| CO012 | Ben Plummer was still quoted as CEO in 2024 public company materials, implying a leadership transition between 2024 and Fitzpatrick’s January 2025 appointment. | 高 | SO010, SO011, SO009 |
| CO013 | Wes Green was appointed as Invisible’s first SVP of Global Public Sector to lead federal and government expansion. | 高 | SO011, SO005 |
| CO014 | Invisible’s modern slavery statement says the board approved the 2024 fiscal-year statement and that Pedraza signed it on 2025-08-05 as founder, president, and chair. | 中 | SO008 |
| CO015 | Invisible announced a $100M growth round on 2025-09-16 led by Vanara Capital. | 高 | SO013, SO018, SO019 |
| CO016 | After the 2025 round, Invisible’s total capital raised reached $144M. | 高 | SO013, SO018, SO020 |
| CO017 | Participants in the 2025 round included Princeville, HOF, Freestyle, Rocketeer, Tallwoods, Acrew, Greycroft, Backed, BY Ventures, and Deepwater. | 高 | SO013, SO018 |
| CO018 | SiliconANGLE and Sacra both pegged Invisible’s 2025 round valuation at more than $2B. | 中 | SO019, SO020 |
| CO019 | Invisible’s January 2025 CEO announcement said the company had achieved a $500M valuation in early 2024. | 中 | SO009 |
| CO020 | Official 2025 announcements and Sacra all say Invisible’s 2024 revenue more than doubled from 2023 to reach $134M. | 高 | SO009, SO013, SO018, SO020, SO021 |
| CO021 | Invisible’s Deloitte Fast 500 post says the company grew 2,342% across Deloitte’s ranking period. | 中 | SO010 |
| CO022 | Invisible’s 2025 announcements also described a 24x increase between 2020 and 2023 before the 2024 revenue step-up. | 高 | SO009, SO013, SO018 |
| CO023 | WEF and Sacra characterize Invisible as profitable for more than five years or half a decade. | 中 | SO020, SO022 |
| CO024 | Sacra estimates Invisible’s 2024 EBITDA at about $15M, or roughly an 11% margin on $134M revenue. | 中 | SO020 |
| CO025 | Sacra says Invisible’s operating model used 3,000+ agents in 35+ countries plus a 350-person full-time team during its 2025 scale-up. | 中 | SO020, SO021 |
| CO026 | Invisible’s join-us page confirms a global-hub, remote-friendly workforce with equity, flexible PTO, parental leave, health coverage, and a 401(k) program. | 中 | SO006 |
| CO027 | FeaturedCustomers, WEF, and AWS Marketplace all say Invisible has worked with more than 80% of the world’s leading AI model providers, including AWS, Microsoft, and Cohere. | 中 | SO024, SO022, SO023 |
| CO028 | The public-sector launch and industry page show that Invisible deliberately expanded from private-enterprise work into federal and public-sector programs in 2024. | 高 | SO011, SO005 |
| CO029 | Invisible’s WeCP acquisition added 18,000+ assessment frameworks and 2M+ interview records to strengthen expert validation and reinforcement-learning environments. | 中 | SO012 |
| CO030 | WEF and AWS Marketplace profiles show Invisible marketing cross-industry references such as Swiss Gear, SAIC, and the Charlotte Hornets beyond AI labs. | 中 | SO022, SO023 |
| CO031 | Invisible’s Headway case study says the company made claims processing 8x faster while cutting cost 37% versus an internal team and 57% versus the prior BPO. | 中 | SO014 |
| CO032 | Invisible’s delivery-platform case study says the company boosted onboarding speed 233%, reduced onboarding cost 50%, and structured 1.5M data points monthly. | 中 | SO015 |
| CO033 | Invisible’s Nasdaq case study says the company cut onboarding times 63% and saved more than 10,000 developer hours. | 中 | SO016 |
| CO034 | Invisible’s Cohere case study says Cohere previously used Invisible for Command R hallucination reduction and later for enterprise-agent evaluation on Command A. | 中 | SO017 |
| CO035 | Invisible ranked 61st on Deloitte’s 2024 Technology Fast 500. | 中 | SO010 |
| CO036 | Invisible’s 2025 fundraise materials described the company as the No. 2 fastest-growing AI company on the 2024 Inc. 5000 list. | 高 | SO013, SO018 |
| CO037 | The California class-action complaint alleges unpaid overtime, unpaid meal and rest premiums, unpaid minimum wages, inaccurate wage statements, unreimbursed expenses, and paid-sick-leave failures. | 中 | SO025 |
| CO038 | Because the complaint was filed on 2023-11-17 and remained a live public document in this run, labor-law exposure is a material diligence item rather than a resolved historical footnote. | 中 | SO025 |
| CO039 | The Indeed review URL was reachable only through a Cloudflare verification interstitial during this run, so current employee sentiment could not be independently verified from page content. | 中 | SO026 |
| CO040 | The BBB complaints URL likewise returned a verification or 403 barrier in this run, limiting direct review of complaint details. | 中 | SO027 |
| CO041 | Crunchbase returned a Cloudflare challenge and PitchBook produced no usable page content in this run, leaving database-style headcount and financing fields unverified. | 中 | SO028, SO029 |
| CO042 | The privacy policy and how-we-work page together indicate that Invisible remains a hybrid software-plus-human-operations business rather than a pure self-serve SaaS vendor. | 高 | SO007, SO003 |
| CO043 | The move from a roughly $500M valuation in early 2024 to more than $2B by September 2025 implies a step-change in market perception over about 20 months, even though the secondary mix is undisclosed. | 中 | SO009, SO019, SO020 |
| CO044 | No accessible source reviewed in this run disclosed an exact current customer count for Invisible. | 中 | SO001, SO013, SO024 |
| CO045 | The identity collision between invisibletech.ai and invisible.ai creates a practical research risk because the latter’s manufacturing-computer-vision pages are easy to confuse with the target company’s AI-operations narrative. | 高 | SO001, SO030 |
| CO046 | Sacra and the official site together show Invisible’s arc from outsourcing and virtual-assistant roots to RLHF or model-builder work and then to enterprise AI software and infrastructure. | 高 | SO021, SO002, SO004, SO013 |
| CO047 | The public-sector launch, WeCP acquisition, and 2025 growth fundraise together show a 2024-2026 push toward government work, expert-validation tooling, and enterprise platform positioning rather than only AI training services. | 高 | SO011, SO012, SO013, SO018 |
| CM001 | Invisible defines its product around embedding AI into core workflows with data, agents, humans-in-the-loop, and evaluations rather than selling generic model access alone. | 高 | SM003, SM004, SM007 |
| CM002 | Invisible’s AI training offer includes domain experts, multilingual training, multimodal labeling, red-teaming, and RL environments, so the relevant market includes post-training and evaluation services as well as automation. | 高 | SM001, SM002, SM005 |
| CM003 | Invisible’s RL-environment offer focuses on auditable enterprise tasks in coding, accounting, banking, legal, and compliance, which places it in enterprise agent-training infrastructure rather than factory-floor computer vision. | 高 | SM002, SM015, SM018 |
| CM004 | Status-quo substitutes for Invisible include annotation platforms, managed labeling vendors, BPO operators, and automation-orchestration suites, not only other model-training specialists. | 中 | SM019, SM022, SM023, SM024 |
| CM005 | Labelbox packages complex post-training and evaluation work as a tooling-first offer with curated expert networks and multimodal evaluation features, illustrating the self-serve end of the substitute set. | 中 | SM024 |
| CM006 | Appen positions itself as a leader across data sourcing, preparation, and real-world model evaluation, illustrating the managed-data-services substitute that competes on scaled workforce supply. | 中 | SM019, SM020 |
| CM007 | UiPath frames the adjacent budget pool as governed business orchestration where AI agents, robots, and people are combined inside regulated workflows. | 中 | SM022 |
| CM008 | Because Invisible integrates legacy systems, workflow metrics, and human review, its nearer market excludes generic infrastructure capex and pure model-hosting spend. | 高 | SM003, SM004, SM006 |
| CM009 | UiPath reported $1.901 billion of ARR and 2,624 customers above $100,000 ARR as of April 30, 2026, providing a public floor for enterprise willingness to pay for governed automation and orchestration. | 中 | SM022 |
| CM010 | Appen reports 50M+ people hours on platform, 20K+ AI projects, 100M LLM data elements, and 10B units of data processed, showing that AI training and evaluation work is already production-scale. | 中 | SM019 |
| CM011 | Invisible and its AWS marketplace profile both claim work with over 80% of the world’s top AI companies, implying strong penetration into frontier-lab and top-tier model-builder demand. | 高 | SM001, SM021 |
| CM012 | Invisible’s cited customer set spans asset management, financial data onboarding, insurance, healthcare, multilingual model evaluation, and enterprise RAG, indicating a cross-vertical serviceable market rather than a single-industry wedge. | 中 | SM008, SM009, SM010, SM011, SM012, SM013, SM014 |
| CM013 | Invisible’s broad relevant TAM is the combined budget pool for enterprise AI operations, post-training and evaluations, and governed workflow automation, which is larger than current monetized automation spend alone but narrower than headline generative-AI TAMs. | 低 | SM009, SM010, SM019, SM022, SM024 |
| CM014 | Invisible’s nearer SAM is enterprises and frontier labs buying expert-in-the-loop workflow automation, custom evaluation, or RL environments for regulated or high-value tasks rather than commodity annotation or generic RPA. | 中 | SM001, SM002, SM003, SM005, SM019, SM022, SM024 |
| CM015 | An evidence-constrained 2026 SAM range for Invisible-relevant spend is roughly $2.0 billion to $6.0 billion with a $3.8 billion base case, anchored by UiPath’s public automation floor and uplift for post-training and evaluation services evidenced by Appen, Labelbox, and Invisible’s frontier-lab workload. | 低 | SM019, SM020, SM021, SM022, SM024 |
| CM016 | A realistic three-to-five-year SOM for Invisible is measured in hundreds of millions rather than tens of billions because winning requires custom delivery, domain experts, and workflow integration that constrain throughput even when demand is broad. | 低 | SM004, SM006, SM018 |
| CM017 | Boosted.ai needed ten times more data throughput and more advanced ground-truth data before its small-language-model assistant could meet enterprise standards, showing buyers pay when generic fine-tuning is insufficient. | 中 | SM008 |
| CM018 | Nasdaq cut onboarding times by 63% and saved more than 10,000 developer hours, showing workflow-automation ROI can justify spend well before full autonomy. | 中 | SM009 |
| CM019 | The national insurer case shows AI workflow automation can improve W9 accuracy, claim-response time, document throughput, and labor cost in adjacent processes after an initial process win. | 中 | SM010 |
| CM020 | Headway’s case shows Invisible competes directly for spend already allocated to internal operations teams and prior BPO providers, not only for new AI line items. | 中 | SM011, SM023 |
| CM021 | Budget ownership splits by workflow: CAIO, CTO, and ML leaders sponsor training and evaluation work, while COO, shared-services, and compliance leaders sponsor workflow automation and governed deployment. | 中 | SM001, SM003, SM004, SM005, SM007, SM010, SM011 |
| CM022 | Invisible’s back-office automation offer is explicitly built for operations buyers who need evidence-backed outputs, confidence routing, and human review instead of unattended black-box autonomy. | 高 | SM004, SM007 |
| CM023 | Invisible’s documented delivery motion starts with operations leads, connects legacy systems, validates on historical data, and scales only after monitoring throughput, error, and cost metrics. | 高 | SM004, SM006 |
| CM024 | Tool-first evaluation platforms can be the entry point for smaller teams, but Invisible is better positioned once buyers need services, subject-matter experts, or workflow integration into live systems. | 中 | SM003, SM024 |
| CM025 | TaskUs represents the status-quo outsourcing substitute for many customer-experience and digital-operations budgets that Invisible aims to displace with AI-native workflows. | 中 | SM023 |
| CM026 | Appen and Labelbox show buyers can start cheaply with tooling or labeling services, but Invisible’s case studies show demand shifts toward higher-judgment work once production quality and workflow nuance matter. | 中 | SM012, SM013, SM014, SM019, SM024 |
| CM027 | Invisible’s visible proofs cluster in regulated and data-heavy verticals including asset management, insurance, healthcare, and enterprise information workflows, which narrows the most credible near-term buyer set. | 中 | SM008, SM009, SM010, SM011, SM013, SM014 |
| CM028 | Cohere used Invisible for multilingual, coding, reasoning, and enterprise-task evaluations, showing frontier labs and enterprise-model providers treat human evaluation as a performance lever rather than just QA overhead. | 中 | SM012 |
| CM029 | Invisible’s RAG customers used ranking, rating, and conversation review workflows to improve trust, relevance, and response quality, showing enterprise adoption depends on post-deployment tuning loops. | 中 | SM013, SM014 |
| CM030 | Invisible’s market is inherently multi-stakeholder because ML teams need expert data, operations teams need workflow automation, and governance teams need evaluation and observability. | 高 | SM001, SM003, SM004, SM005, SM007 |
| CM031 | Benchmark saturation and data exhaustion make reinforcement learning and custom evaluation more important for further model capability gains than simply extending pre-training. | 中 | SM005, SM015 |
| CM032 | RL environments are becoming an explicit budget item because reliable agent training depends on auditable tasks, verified rewards, human trajectories, and domain-specific graders. | 高 | SM002, SM015, SM017, SM018 |
| CM033 | Domain expertise is the binding constraint in enterprise RL environments because banking, legal, compliance, and operations workflows require experts to define what correct looks like. | 高 | SM001, SM002, SM017, SM018 |
| CM034 | Poor reward functions, simulation-to-real mismatch, stale trajectories, and grader gaming are material adoption constraints that raise deployment cost and slow productionization. | 中 | SM016, SM017 |
| CM035 | Invisible’s evaluation framework argues standard benchmarks miss enterprise-specific error types, governance requirements, and multi-turn interactions, increasing demand for custom evaluation spend. | 中 | SM005 |
| CM036 | The EU AI Act makes logging, documentation, human oversight, dataset quality, and transparency mandatory for many high-risk AI uses, raising the value of governed deployment partners. | 高 | SM025, SM028 |
| CM037 | U.S. AI regulation remains fragmented, with multiple state laws effective in 2026 covering employment, training-data transparency, disclosures, and high-risk systems. | 高 | SM026, SM027, SM029 |
| CM038 | AI washing, disclosure quality, and third-party vendor compliance have become board-level governance concerns for AI buyers and investors. | 高 | SM027, SM028 |
| CM039 | Employment and workplace use cases face special risk because discrimination, notice, audit, and bias-assessment duties attach to hiring and worker-management systems. | 高 | SM025, SM026, SM027, SM029 |
| CM040 | These regulatory burdens favor vendors that can provide human oversight, audit-ready documentation, and controlled deployment, all capabilities Invisible emphasizes across its workflow and RL materials. | 高 | SM002, SM004, SM007, SM025, SM028 |
| CM041 | The market is still fragmented because pure labeling, BPO, orchestration software, and custom AI integrators solve different slices of the workflow, leaving room for Invisible’s hybrid position. | 中 | SM003, SM019, SM022, SM023, SM024 |
| CP001 | Invisible positions itself as a modular enterprise AI platform spanning data, agents, humans-in-the-loop, and evaluations. | 中 | SP001 |
| CP002 | Invisible says AI value comes from embedding models into core workflows, operational data, and measurable business metrics rather than treating AI as a pilot. | 中 | SP001 |
| CP003 | Invisible says its forward deployed engineers connect legacy systems while customer data remains in the customer’s own systems. | 中 | SP003 |
| CP004 | Invisible says its platform is model-agnostic rather than locked to a single model vendor. | 中 | SP003 |
| CP005 | Invisible says it can mobilize specialized domain experts for AI training across highly technical disciplines. | 中 | SP004 |
| CP006 | Invisible says it can train and evaluate models in more than 80 languages. | 中 | SP004 |
| CP007 | Invisible says it offers reinforcement-learning environments and step-based agentic training workflows. | 中 | SP004 |
| CP008 | Invisible says it offers multimodal data generation and labeling plus frontier-grade red-teaming and compliance evaluations. | 中 | SP004 |
| CP009 | Invisible’s contact-center solution claims it can evaluate 100% of interactions against policy and quality standards. | 中 | SP005 |
| CP010 | Invisible’s contact-center solution claims humans remain in control through triage, reply suggestions, and handoffs. | 中 | SP005 |
| CP011 | Invisible’s computer-vision solution combines domain-expert training, annotation, QA, secure deployment, and operational insights. | 中 | SP006 |
| CP012 | Invisible’s back-office automation solution combines data ingestion, adaptive process mapping, custom AI agents, and human verification for compliance-ready outputs. | 中 | SP008 |
| CP013 | Invisible’s Nasdaq case study says onboarding times fell by 63% after Invisible streamlined data integration. | 中 | SP009 |
| CP014 | Invisible’s Nasdaq case study says the project saved more than 10,000 developer hours. | 中 | SP009 |
| CP015 | Invisible says it has trained foundation models for more than 80% of the world’s leading AI model providers, including Cohere, Microsoft, and AWS. | 高 | SP011, SP016, SP017 |
| CP016 | Official 2025 funding materials say Invisible reached $134 million of revenue in 2024. | 高 | SP011, SP013 |
| CP017 | Official 2025 funding materials say Invisible raised $100 million in 2025 and total capital raised reached $144 million. | 高 | SP011, SP013 |
| CP018 | Official 2025 funding materials say Invisible had a team of 350 and doubled the size of its engineering organization during 2025. | 高 | SP011, SP013 |
| CP019 | Sacra says Invisible was founded in 2015 as a virtual-assistant service and later evolved into a human-plus-automation platform. | 中 | SP015 |
| CP020 | Sacra estimates Invisible grew from $60 million of 2023 revenue to $134 million of 2024 revenue, up 123% year over year, with roughly 11% EBITDA margin. | 中 | SP014, SP015 |
| CP021 | Sacra describes Invisible’s monetization as outcome- or process-based rather than a transparent software seat price. | 中 | SP014, SP015 |
| CP022 | Sacra identifies Scale AI and Surge AI as direct AI-training competitors to Invisible. | 中 | SP014 |
| CP023 | Sacra says Invisible also competes with BPOs such as Accenture, TaskUs, and Teleperformance for outsourced enterprise workflows. | 中 | SP014 |
| CP024 | Sacra says Invisible also competes with annotation specialists including Appen, iMerit, Toloka, and Prolific. | 中 | SP014 |
| CP025 | Invisible’s Scale-AI comparison guide groups rivals into tool-first platforms, managed labeling services, open-source tools, and end-to-end AI partners. | 中 | SP019 |
| CP026 | Invisible’s comparison guide describes Scale AI as strongest in high-volume annotation, APIs, generative-AI workflows, RLHF, and evaluation. | 中 | SP019 |
| CP027 | Invisible’s comparison guide says mature buyers increasingly want domain expertise, operational integration, and enterprise controls beyond labeling-only support. | 中 | SP019 |
| CP028 | Labelbox’s public pricing page shows a free tier with up to 30 users, up to 50 projects, and one workspace. | 中 | SP020 |
| CP029 | Labelbox reserves SSO, custom embeddings, monitoring, multimodal chat evaluation, and extra services for paid subscription or add-on tiers. | 中 | SP020 |
| CP030 | Appen says ADAP supports text, audio, image, 3D point-cloud, and 4D annotation with configurable workflows. | 中 | SP021 |
| CP031 | Appen says ADAP integrates internal experts with Appen’s global crowd and offers API, AWS, and Azure integrations plus enterprise compliance credentials. | 中 | SP021 |
| CP032 | Appen’s investor materials say the company has a global crowd of over one million skilled contributors across the AI lifecycle. | 中 | SP022 |
| CP033 | Appen’s platform page cites 50M+ people hours, 20K+ AI projects, 100M LLM data elements, and 10B units of data processed. | 中 | SP021 |
| CP034 | DataAnnotation presents itself as a flexible contractor marketplace where experts rate outputs, refine prompts, label data, and get paid per project. | 中 | SP023 |
| CP035 | TaskUs positions itself as an outsourced digital-services and next-generation customer-experience provider serving multiple enterprise sectors. | 中 | SP024 |
| CP036 | UiPath positions itself as a regulated-enterprise automation vendor with $1.901 billion of ARR, 2,624 customers over $100K ARR, and 374 customers over $1M ARR. | 中 | SP025 |
| CP037 | CB Insights lists Mimica, SuperAnnotate, and Hypatos as additional alternatives to Invisible, signaling spillover competition from process intelligence, annotation, and document automation. | 中 | SP018 |
| CP038 | Invisible’s stated differentiation is breadth: data infrastructure, workflow mapping, expert marketplace, annotation and evaluation, and agentic automation in one stack. | 中 | SP011, SP019 |
| CP039 | Any switching cost Invisible earns is likely to come from embedded workflows, validation data, and expert loops rather than model lock-in because the company says data stays in customer systems and models are interchangeable. | 中 | SP003, SP004 |
| CP040 | Pricing transparency is weak across most reviewed direct peers because Invisible and Appen expose no public rate card in the reviewed set while Labelbox exposes a freemium self-serve entry point. | 中 | SP014, SP020, SP021, SP022 |
| CP041 | BusinessWire, FeaturedCustomers, and AWS partner materials show Invisible already references customers or partners such as Microsoft, AWS, Cohere, Nasdaq, Swiss Gear, SAIC, and the Charlotte Hornets. | 中 | SP013, SP016, SP017 |
| CP042 | Sacra says AI labs are increasingly moving toward synthetic data, which is one reason Invisible is pivoting harder toward enterprise clients. | 中 | SP015 |
| CP043 | Sacra flags labor-model scrutiny as a risk because Invisible’s economics benefit from global wage differences. | 中 | SP014 |
| CP044 | Alvarez & Marsal says AI vendors face rising scrutiny on AI disclosures, governance controls, and third-party vendor compliance. | 中 | SP026 |
| CP045 | Invisible’s modern-slavery statement shows workforce and supply-chain governance are explicit board-level risk topics for the company. | 中 | SP012 |
| CP046 | BusinessWire and SiliconANGLE describe Invisible’s 2025 financing as valuing the company at more than $2 billion. | 高 | SP013, SP027 |
| CP047 | Invisible argues enterprises often need custom computer-vision models and workflow-specific tuning rather than generic off-the-shelf tools. | 中 | SP028 |
| CP048 | Invisible argues frontier labs outsource RL environments when domain coverage, expert judgment, and evaluation design matter more than raw labeling throughput. | 中 | SP029 |
| CP049 | Invisible argues RL pipeline failures often happen before model training because task design, grading logic, and evaluation structure are weak, which supports demand for managed evaluation workflows. | 中 | SP030 |
| CP050 | GetLatka classifies Invisible Technologies inside AI and machine-learning operationalization software while mapping a wider long-tail of adjacent alternatives. | 中 | SP031 |
| CP051 | NVIDIA frames manufacturing AI as a market built around digital twins, robotics, quality inspection, and predictive maintenance, underscoring how large platform vendors can substitute for some computer-vision-led enterprise workflows. | 中 | SP032 |
| CI001 | Invisible sells a modular AI platform that combines data infrastructure, workflow software, human experts, evaluation tooling, and agentic automation. | 高 | SI001, SI011 |
| CI002 | Invisible's 2025 financing materials describe five platform components: Neuron, Atomic, Expert Marketplace, Synapse, and Axon. | 高 | SI001, SI013 |
| CI003 | Invisible says customers hand work into its platform and can track status, throughput, error rates, resource efficiency, and cost per transaction. | 高 | SI002, SI011 |
| CI004 | Sacra describes Invisible as an operations-as-a-service model with usage-based and outcome-oriented monetization rather than pure hourly staffing. | 中 | SI015, SI016 |
| CI005 | Sacra says Invisible historically set a $2,000 per month minimum spend for individual executive-support customers. | 中 | SI015 |
| CI006 | Sacra says corporate engagements are structured around defined process units such as per-1,000 annotation rates or monthly retainers for onboarding workflows. | 中 | SI015 |
| CI007 | Invisible tells prospects to measure time saved, cost reduced, revenue gain, or quality improvement on monthly or quarterly reviews, signaling ROI-led selling. | 中 | SI011 |
| CI008 | Invisible's Headway case study reports 8x faster claims processing, 37% lower cost than internal teams, and 57% lower cost than a prior BPO. | 中 | SI007 |
| CI009 | Invisible's Boosted.ai case study reports 90% cost savings for a financial-analysis assistant data program. | 中 | SI006 |
| CI010 | Invisible's Nasdaq case study reports 63% lower onboarding time and more than 10,000 developer hours saved. | 中 | SI008 |
| CI011 | Invisible's retailer recruiting case study reports 500 candidates reviewed weekly, 65% pre-screened by Invisible, and 38% recruiter time savings after six months. | 中 | SI010 |
| CI012 | Invisible's national-insurer case study reports $450k cost savings, 50% faster claim approvals, 16,000 labor hours saved, 6,000 W9 requests handled annually, and accuracy improvement from 75% to 98%. | 中 | SI009 |
| CI013 | Invisible's Cohere case study says Invisible-supported human evaluation helped Command A achieve a 9-point lead over GPT-4o and DeepSeek-V3 on Arabic-dialect ADI2 scoring. | 中 | SI005 |
| CI014 | Invisible's September 2025 funding materials say 2024 revenue more than doubled versus 2023 to reach $134 million. | 高 | SI001, SI013, SI017 |
| CI015 | Sacra independently estimates Invisible generated $134 million of 2024 revenue, up 123% from $60 million in 2023. | 中 | SI015, SI016 |
| CI016 | The September 2025 growth round added $100 million and brought disclosed lifetime capital raised to $144 million. | 高 | SI001, SI013, SI017 |
| CI017 | SiliconANGLE, citing Bloomberg, reports the 2025 financing valued Invisible at more than $2 billion. | 中 | SI014, SI015 |
| CI018 | Sacra estimates Invisible produced about $15 million of EBITDA in 2024, or roughly an 11% EBITDA margin. | 中 | SI015 |
| CI019 | Invisible's chairman said the company had been built quietly and profitably for years before the 2025 round. | 中 | SI013 |
| CI020 | The funding release says Invisible had a team of 350 in 2025, doubled its engineering organization that year, and opened offices in New York, San Francisco, Washington, D.C., and London. | 高 | SI001, SI013 |
| CI021 | Invisible's November 2024 Deloitte Fast 500 announcement says the company grew 2,342% from 2020 to 2023. | 中 | SI012 |
| CI022 | Official messaging centers a forward-deployed engineering model that plugs into customer systems and starts with operations leads rather than a pure API self-serve motion. | 高 | SI002, SI011 |
| CI023 | Invisible emphasizes live production monitoring around throughput, error rates, resource efficiency, and cost per transaction, suggesting implementation-led selling tied to operating KPIs rather than seat growth alone. | 中 | SI002, SI011 |
| CI024 | The 2025 release adds a public-sector enterprise accounts leader and cites SAIC and U.S. Navy work, supporting a push into regulated and government workflows that typically have longer procurement cycles. | 高 | SI001, SI013 |
| CI025 | Using $134 million of 2024 revenue and a 350-person 2025 team implies roughly $383k of revenue per current team member as a rough public productivity proxy. | 中 | SI001, SI013 |
| CI026 | Appen's ADAP page discloses 50M+ people-hours, 20K+ AI projects, 100M LLM data elements, and 10B units processed, showing that public AI-data peers expose large-scale operating metrics that Invisible does not. | 中 | SI018, SI019 |
| CI027 | UiPath's investor site discloses $1.901B ARR, 109% dollar-based net retention, 2,624 customers with $100K+ ARR, and 374 with $1M+ ARR as of April 30, 2026. | 高 | SI021, SI024 |
| CI028 | UiPath's investor page explicitly anchors those KPIs to its Annual Report on Form 10-K filed with the SEC on March 25, 2026, providing a filing-backed benchmark disclosure standard absent at Invisible. | 高 | SI021, SI024 |
| CI029 | Across Invisible's official website, funding announcement, and third-party research, there is no public disclosure of ARR, NRR, gross margin, CAC, burn rate, cash on hand, or customer concentration. | 中 | SI001, SI002, SI011, SI015, SI016 |
| CI030 | Management said the 2025 proceeds would be invested into Invisible's core AI software platform across data infrastructure, workflow mapping, expert marketplace, evaluation, and agentic automation. | 高 | SI001, SI013 |
| CI031 | No debt, venture debt, or project-finance obligation is disclosed in the retained 2025 financing materials or other retained public sources. | 低 | SI001, SI013, SI014 |
| CI032 | Sacra argues Invisible is pivoting away from AI-lab RLHF because leading labs are increasingly using synthetic data generation, a genuine demand-side risk to one historical revenue stream. | 中 | SI016 |
| CI033 | TaskUs describes itself as an outsourced digital services provider, making it a public BPO-style proxy for lower-margin labor-heavy delivery economics rather than software-only margins. | 中 | SI020, SI025 |
| CI034 | Sacra says Invisible operates with 3,000+ agents in 35+ countries plus 350 full-time staff, indicating a delivery model with substantial variable labor cost even as automation improves throughput. | 中 | SI015, SI016 |
| CI035 | Case-study economics repeatedly emphasize labor and cycle-time savings rather than published list prices, implying Invisible sells bespoke ROI packages whose realized price and margin are negotiated privately. | 中 | SI006, SI007, SI008, SI009, SI010 |
| CI036 | The official release says Invisible has trained foundation models for more than 80% of the world's leading AI model providers. | 高 | SI001, SI013 |
| CI037 | Official references to Microsoft, Swiss Gear, SAIC, the Hornets, Cohere, Headway, Nasdaq, insurers, asset managers, and banks imply end-market diversification across model builders, enterprise software, regulated industries, and operations workflows. | 中 | SI001, SI005, SI007, SI008, SI009, SI010, SI026, SI027 |
| CI038 | Combining official platform pages with Sacra's description supports a two-part revenue mix: AI training and evaluation work alongside enterprise workflow automation and custom solutions. | 中 | SI001, SI003, SI004, SI015, SI016 |
| CI039 | Public pricing visibility is limited to historical or proxy-style descriptions; official Invisible pages market outcomes and workflows but do not publish current rate cards, standard discounts, or contract durations. | 中 | SI002, SI003, SI004, SI011, SI015 |
| CI040 | Public evidence suggests capital adequacy is better than an immediate cash crunch because the company claims recent profitability and raised $100 million, yet underwriting still depends on management disclosing cash, burn, runway, and concentration data. | 中 | SI001, SI013, SI015, SI016 |
| CI041 | C3.ai and Palantir maintain dedicated investor-relations surfaces, confirming there is a public enterprise-AI peer set even though the fetched overview pages here do not expose detailed inline unit-economics data. | 低 | SI022, SI023 |
| CE001 | Invisible presents its offer as a modular platform that lets customers use combinations of data, agents, humans-in-the-loop, and evaluations rather than one monolithic product. | 中 | SE005, SE007 |
| CE002 | Invisible’s back-office solution is designed to turn unstructured inputs into compliance-ready data. | 中 | SE001 |
| CE003 | Invisible says back-office agents draft outputs, surface source evidence for verification, and escalate uncertain decisions for human review. | 中 | SE001 |
| CE004 | Invisible’s contact-center product claims a governed cross-channel view and evaluation of 100% of interactions against policy and quality standards. | 中 | SE002 |
| CE005 | Invisible’s forecasting offer unifies ERP, POS, e-commerce, labor, operations, and external signals into a single demand-data foundation. | 中 | SE003 |
| CE006 | Invisible’s forecasting offer includes custom forecast models and decision-ready dashboards rather than generic packaged forecasts. | 中 | SE003 |
| CE007 | Invisible’s computer-vision offer bundles human training, an annotation platform, end-to-end evaluations and QA, secure deployment, and recommendation outputs. | 中 | SE004 |
| CE008 | Invisible’s AI-training offer spans domain-expert training, agentic workflows, 80-plus languages, multimodal data generation, and red-teaming or compliance-oriented evaluation. | 中 | SE005 |
| CE009 | Invisible’s RL-environment product focuses on real work in coding, accounting, banking, legal, and compliance instead of generic benchmarks alone. | 中 | SE006 |
| CE010 | Invisible’s public case-study hub shows the company packaging workflow solutions across enterprise operations rather than around a single narrow product niche. | 中 | SE016 |
| CE011 | Invisible’s 2025 financing materials disclose a five-layer core platform organized around data infrastructure, workflow mapping, a human-expert engine, evaluation, and orchestration. | 高 | SE008, SE009 |
| CE012 | Neuron is the layer Invisible says integrates and transforms structured and unstructured data. | 高 | SE008, SE009 |
| CE013 | Atomic is the layer Invisible says codifies workflows and business logic through visual process mapping and building. | 高 | SE008, SE009 |
| CE014 | Invisible’s public materials use Meridial or Expert Marketplace language for the expert-workforce layer that supplies RLHF, training, validation, and specialist judgment. | 高 | SE005, SE008, SE010 |
| CE015 | Synapse is the layer Invisible says measures performance, enables annotation, supports fine-tuning, and drives continuous improvement. | 高 | SE008, SE009 |
| CE016 | Axon is the layer Invisible says orchestrates tasks and decisions across systems. | 高 | SE008, SE009 |
| CE017 | The WeCP acquisition adds more than 18,000 scope-specific technical assessments and over two million interview records to Invisible’s stack. | 中 | SE010 |
| CE018 | Invisible says WeCP will be integrated into Meridial to improve expert validation and reinforcement-learning workflow precision. | 中 | SE010 |
| CE019 | Invisible says forward-deployed engineers connect legacy systems, operational databases, and warehouses to the platform while customer data stays in customer systems. | 中 | SE007 |
| CE020 | Invisible says deployments are validated against historical data before production. | 中 | SE007 |
| CE021 | Invisible says live deployments are monitored using throughput, error rates, resource efficiency, and cost-per-transaction metrics. | 中 | SE007 |
| CE022 | Invisible’s FDE playbook frames forward deployment as the mechanism that turns AI from slideware into operational reality. | 中 | SE023 |
| CE023 | Invisible’s AI-evaluation report says standard benchmarks miss enterprise-specific business value and that enterprises need custom evaluation frameworks tied to unique use cases and objectives. | 中 | SE022 |
| CE024 | Invisible’s RL-environment technical writing argues that post-training environments, rather than larger pre-training runs alone, are now the key capability lever for agentic systems. | 中 | SE020 |
| CE025 | Invisible’s grader-problem article says RL environment quality depends on a verifier that aligns with expert judgment and is stress-tested against reward hacking. | 中 | SE021 |
| CE026 | Invisible’s grader-problem article describes a three-tier verification process of automated structural checks, adversarial LLM attacks, and human expert review. | 中 | SE021 |
| CE027 | Invisible’s multimodal systems guide says teams should decompose systems, engineer pipelines explicitly, and treat metrics as diagnostics rather than leaderboard endpoints. | 中 | SE024 |
| CE028 | Invisible’s enterprise multimodal playbook says multimodal deployment raises higher demands on governance, data, infrastructure, and change management than text-only AI. | 中 | SE025 |
| CE029 | Invisible’s computer-vision technical doc says edge processing converts video into lightweight structured metadata instead of shipping raw footage upstream. | 中 | SE018 |
| CE030 | Invisible’s computer-vision technical doc says useful deployment requires an API bridge that maps visual events into ERP, WMS, or CRM business logic. | 中 | SE018 |
| CE031 | Invisible’s computer-vision comparison doc says custom models are the better path for high-volume, high-stakes workflows, while commodity tasks can use prebuilt tools. | 中 | SE017 |
| CE032 | Invisible’s computer-vision comparison doc says ownership of model weights and on-prem or edge deployment improves control and long-run economics. | 中 | SE017 |
| CE033 | Invisible’s computer-vision degradation doc says continuous feedback loops and automated retraining pipelines are needed to prevent silent model drift. | 中 | SE019 |
| CE034 | Invisible’s computer-vision solution page says deployments can be local for secure or remote environments with customer-controlled data. | 中 | SE004 |
| CE035 | Invisible says RL-environment runs are logged, replayable, auditable, and scored with built-in rewards, rubrics, and automated checks. | 中 | SE006 |
| CE036 | Invisible says its Nasdaq engagement cut onboarding time by 63% and saved more than 10,000 developer hours through interoperability work. | 中 | SE012 |
| CE037 | Invisible says its Headway deployment used batching and parallel processing to achieve 8x faster claims processing with lower cost than internal or BPO alternatives. | 中 | SE011 |
| CE038 | Invisible says its insurance automation program improved W9 accuracy from 75% to 98% and reduced claim-response time by 50%. | 中 | SE013 |
| CE039 | Invisible says the same insurance program raised compliance document throughput from 40 documents per week to 350 and saved managers over 16,000 hours. | 中 | SE013 |
| CE040 | Invisible says its You.com engagement used 20,000 evaluations and a structured relevance, freshness, and diversity rating system to lift result relevance by 70%. | 中 | SE015 |
| CE041 | Cohere says Invisible maintained a high bar for talent and continuous observability that made its evaluation data trustworthy. | 中 | SE014 |
| CE042 | Invisible’s 2025 financing disclosures say the company doubled its engineering organization and added a platform CTO plus multiple field CTOs during 2025. | 高 | SE008, SE009 |
| CE043 | Invisible’s public careers page emphasizes hubs, in-person meetups, equity, and benefits rather than exposing technical artifacts such as docs, repos, or package surfaces. | 中 | SE027 |
| CE044 | DataAnnotation’s public recruiting page is more explicit than Invisible’s own careers page about assessment-gated expert review work, highlighting that Invisible’s practitioner-facing public signal is comparatively thin. | 中 | SE027, SE036 |
| CE045 | Invisible’s privacy policy says clients may record online meetings with agents and store those recordings in their accounts, subject to notice. | 中 | SE026 |
| CE046 | Invisible’s privacy policy says installed agent software may collect keystrokes, mouse clicks, screenshots, and webcam pictures as work information. | 中 | SE026 |
| CE047 | Invisible’s privacy policy says client organizations can access usage data and the contents of communications and files associated with accounts. | 中 | SE026 |
| CE048 | Invisible’s privacy policy offers access, portability, correction, restriction, consent-withdrawal, and erasure rights subject to applicable law. | 中 | SE026 |
| CE049 | Invisible links to a public trust portal, but the retained trust-page fetch did not expose detailed control mappings or certification evidence in this run. | 中 | SE028 |
| CE050 | Appen’s public ADAP page discloses GDPR, SOC, HIPAA, and ISO 27001 claims plus API and cloud integrations, a publicly visible trust benchmark Invisible does not match in accessible materials. | 中 | SE028, SE031 |
| CE051 | Cohere’s homepage publicly commits to VPC, on-prem, or dedicated deployment with customer data control, a partner benchmark consistent with Invisible’s own data-stays-in-your-systems positioning. | 中 | SE007, SE032 |
| CE052 | The American Bar Association’s 2025 AI review says current litigation and legislation themes center on privacy, fairness, transparency, consent, and training-data disclosure. | 中 | SE034 |
| CE053 | The same ABA review notes California enacted 2024 laws on training-data transparency and AI transparency, which are directly relevant to AI-training and evaluation vendors. | 中 | SE034, SE005 |
| CE054 | The International AI Safety Report 2026 describes itself as a rigorous assessment of AI risk management built with over 100 experts and input from more than 30 countries and organizations. | 中 | SE033 |
| CE055 | FeaturedCustomers lists seven reviews or testimonials and sixteen case studies for Invisible, giving some third-party visibility into customer-proof volume even though the underlying evidence is still shallow. | 中 | SE029 |
| CE056 | The World Economic Forum organization page repeats Invisible’s positioning as a platform that structures data, builds workflows, deploys agentic solutions, evaluates impact, and mobilizes human experts. | 中 | SE030 |
| CE057 | Invisible.ai is a separate manufacturing-vision company offering on-prem factory-floor visual intelligence and should not be confused with Invisible Technologies. | 中 | SE035 |
| CU001 | Invisible's public customer proof spans at least six named customers—Cohere, Nasdaq, Headway, Boosted.ai, You.com, and Getro—plus several quantified but unnamed deployments in insurance, retail, delivery, and solar workflows. | 中 | SU002, SU005, SU006, SU007, SU011, SU012, SU008, SU009, SU010, SU013 |
| CU002 | The visible customer base spans both frontier-model builders and enterprise operations buyers rather than a single narrow customer archetype. | 中 | SU002, SU005, SU006, SU007, SU011, SU016, SU017 |
| CU003 | Buyer, user, and payer roles vary by segment: model-evaluation teams buy expert feedback, while operations leaders buy workflow redesign and downstream business users consume faster or more accurate outputs. | 中 | SU002, SU006, SU007, SU008, SU011, SU013 |
| CU004 | Third-party profile pages repeat Invisible's claim that it has improved models for more than 80% of the world's top AI companies, including Microsoft, AWS, and Cohere. | 高 | SU016, SU017, SU019 |
| CU005 | Financial-data and investment workflows are proven customer segments via the Nasdaq and Boosted.ai case studies. | 中 | SU005, SU006 |
| CU006 | Healthcare and insurance operations are proven customer segments via Headway and the national insurer case study. | 中 | SU007, SU009 |
| CU007 | Search and answer-quality workloads are proven customer segments via You.com and the unnamed contextual-conversation startup case study. | 中 | SU004, SU011 |
| CU008 | Retail and marketplace operations are proven customer segments via the big-4 retailer and delivery-platform case studies. | 中 | SU010, SU013 |
| CU009 | Renewable-energy customer support and financing workflows are evidenced by the solar-provider case study. | 中 | SU008 |
| CU010 | Public-sector and sports pages show Invisible actively packaging customer proof into government-adjacent and sports-focused vertical-entry narratives by June 2026. | 中 | SU014, SU015 |
| CU011 | Nasdaq is more than a logo in Invisible's public materials: the company attributes a 63% reduction in onboarding time and 10,000+ developer hours saved to that deployment. | 高 | SU006, SU014 |
| CU012 | Headway's case study claims 8x faster claims processing with 37% lower cost than an internal team and 57% lower cost than a prior BPO provider. | 中 | SU007 |
| CU013 | Boosted.ai's case study claims 90% cost savings and says the customer felt unlocked by the third batch of training data. | 中 | SU005 |
| CU014 | You.com's case study says Invisible supported 20,000 evaluations and a 70% increase in relevance. | 中 | SU011 |
| CU015 | The big-4 retailer case study says Invisible enriched 50,000 dormant SKUs, delivered a reported 9x ROI, and generated nearly $1 million of revenue from 3,100+ revived items within 30 days. | 中 | SU010 |
| CU016 | The delivery-platform case study says onboarding speed improved 233%, onboarding costs fell 50%, and the deployment later processed 1.5 million unique data points monthly. | 中 | SU013 |
| CU017 | The solar-provider case says Invisible expanded from proposal generation into financing-contract support and post-installation monitoring, reaching 180 contracts per day at peak. | 中 | SU008 |
| CU018 | The national insurer case claims $450,000 of savings, 16,000 labor hours saved, 50% faster claim-approval response times, and W9 accuracy improvement from 75% to 98%. | 中 | SU009 |
| CU019 | Getro's case study shows a repeat-service cadence with daily batches, 100% QC logging, and biweekly account-manager calls. | 中 | SU012 |
| CU020 | Several case studies describe Invisible becoming embedded in customer systems or adjacent workflows instead of remaining one-off pilots. | 中 | SU008, SU012, SU013 |
| CU021 | The delivery-platform case explicitly says Invisible was fully integrated with the client's internal technology systems within 90 days. | 中 | SU013 |
| CU022 | The solar-provider case explicitly says the customer requested follow-on downstream support after the initial proposal-generation workflow. | 中 | SU008 |
| CU023 | Getro's testimonial praising daily documentation and biweekly calls is a positive customer-satisfaction proxy. | 中 | SU012 |
| CU024 | FeaturedCustomers lists 7 reviews and testimonials plus 16 case studies or customer stories for Invisible, implying a broader public reference base than the chapter's named case list alone. | 中 | SU018 |
| CU025 | CaseStudies.com also presents Invisible as a customer-success vendor profile, corroborating that third-party directories see a non-trivial body of customer references. | 中 | SU019 |
| CU026 | Across company-owned case studies, directory profiles, and customer homepages, Invisible's proof set points to enterprise-scale counterparties rather than SMB or self-serve buyers. | 高 | SU016, SU017, SU018, SU021, SU022, SU023, SU024, SU025 |
| CU027 | Cohere's case study frames Invisible as a provider of enterprise-task evaluation and quality control rather than generic commodity labeling. | 中 | SU002, SU025 |
| CU028 | Cohere's quote that Invisible maintained a high bar for talent and challenged the model with complex questions is a quality signal for demanding enterprise AI customers. | 中 | SU002 |
| CU029 | The public proof base skews toward production-style workflow outcomes, but exact contract values and denominator metrics are almost never disclosed. | 中 | SU005, SU006, SU007, SU008, SU009, SU010, SU011, SU012, SU013 |
| CU030 | Public sources do not disclose customer count, revenue concentration, average contract length, NRR, or GRR for Invisible. | 中 | SU001, SU018, SU019 |
| CU031 | Reference quality varies materially: Nasdaq, Headway, Cohere, Boosted.ai, You.com, and Getro are named proofs, while insurer, retailer, delivery, and solar outcomes are quantified but anonymous. | 中 | SU002, SU005, SU006, SU007, SU008, SU009, SU010, SU011, SU012, SU013 |
| CU032 | The Charlotte Hornets example extends Invisible's customer narrative into sports, but public support is materially weaker than for the main case studies. | 中 | SU017, SU020 |
| CU033 | The adverse basketball article does not deny that Invisible worked with the Hornets; it questions whether the “gifted us Kon Knueppel” phrasing is marketing overreach rather than independently verified customer proof. | 中 | SU020 |
| CU034 | The public-sector and sports pages reuse Nasdaq, Cohere, insurance, and model-evaluation outcomes as reusable proof points for adjacent-market selling. | 中 | SU014, SU015, SU006, SU002, SU009 |
| CU035 | Land-and-expand is visible in several cases: solar moved into financing and monitoring, delivery moved into fully integrated monthly processing, and Getro operates on a recurring daily batch. | 中 | SU008, SU012, SU013 |
| CU036 | Customer concentration risk remains material because the company advertises relationships with major model providers and large enterprises while withholding revenue mix and top-customer exposure. | 中 | SU016, SU017, SU018, SU019 |
| CU037 | The over-80%-of-top-AI-companies claim implies strategic relevance but may also imply dependence on a relatively small set of very large model-lab buyers. | 中 | SU016, SU017, SU019 |
| CU038 | Procurement friction is likely higher in public-sector and regulated workflows because the visible proof set emphasizes custom operations, compliance-sensitive tasks, and enterprise integrations rather than self-serve adoption. | 中 | SU014, SU008, SU009, SU006 |
| CU039 | Review and directory evidence improves breadth, but it does not prove renewal economics or production scale for every logo mentioned on profile pages. | 中 | SU018, SU019 |
| CU040 | The customer chapter supports real adoption and meaningful expansion potential, but it does not support a fully verified retention or concentration model. | 中 | SU006, SU007, SU008, SU012, SU016, SU018, SU020 |
| CU041 | WEF and AWS Marketplace both cite Microsoft, AWS, and Cohere in Invisible's top-AI-provider cohort narrative, showing that large-enterprise references are central to the company's customer story. | 高 | SU016, SU017 |
| CU042 | The retained source set touches at least ten public customer domains or verticals: frontier AI labs, cloud, finance, health, insurance, search, delivery, retail, solar, public sector, and sports. | 中 | SU002, SU005, SU006, SU007, SU008, SU009, SU010, SU011, SU014, SU015, SU016, SU017 |
| CU043 | The freshest customer-narrative surfaces are the 2026 public-sector, sports, and Hornets materials, while many classic workflow case studies remain live but undated. | 中 | SU014, SU015, SU020, SU001 |
| CR001 | Invisible announced a $100 million growth round in September 2025, bringing disclosed lifetime capital raised to $144 million. | 高 | SR007, SR016 |
| CR002 | Invisible said 2024 revenue reached $134 million after more than doubling from 2023. | 高 | SR007, SR016 |
| CR003 | Invisible said it had a team of 350 and had doubled the size of its engineering organization in 2025. | 高 | SR007, SR016 |
| CR004 | Matthew Fitzpatrick became CEO in January 2025 after leading McKinsey's QuantumBlack Labs and overseeing 1,000 engineers and product leaders. | 高 | SR006, SR016 |
| CR005 | Invisible's implementation model connects customer legacy systems, operational databases, and data warehouses to its platform while leaving data in customer systems. | 中 | SR002 |
| CR006 | Invisible says production deployments track throughput, error rates, resource efficiency, and cost per transaction and produce audit-ready documentation. | 中 | SR002 |
| CR007 | Invisible's privacy policy, updated March 25 2026, says the company delivers digital work by outsourcing business processes to human agents. | 中 | SR003 |
| CR008 | Invisible's privacy policy says it processes personal information from both clients and agents and may share that data with service providers, business partners, affiliates, APIs or SDKs, and transaction counterparties. | 中 | SR003 |
| CR009 | Invisible's privacy policy says it may use automated decision-making or profiling technology but will not use it for decisions that significantly affect people unless contract, consent, or law permits. | 中 | SR003 |
| CR010 | Invisible's board-approved modern slavery statement says the company reviews modern slavery risks annually and operates a third-party risk program during vendor onboarding. | 中 | SR004 |
| CR011 | Invisible's modern slavery statement says the business is lower risk than manufacturing but still flags electronics procurement and certain service categories as higher-risk areas that require controls. | 中 | SR004 |
| CR012 | Invisible markets red-teaming, policy-informed evaluations, and continuous model evaluation as built-in controls for AI training and deployment. | 中 | SR005 |
| CR013 | Invisible says synthetic-data approaches still need humans and that its expert network spans complex domains and more than 80 languages. | 中 | SR005 |
| CR014 | Invisible publicly launched global public-sector operations and said it would work with federal departments and government agencies on data modernization and efficiency programs. | 中 | SR008 |
| CR015 | Invisible says it has trained foundation models for more than 80% of the world's leading AI model providers, including Cohere, Microsoft, and AWS. | 高 | SR007, SR016, SR020 |
| CR016 | Invisible agreed to acquire WeCP and integrate its assessment library and team into Invisible's Meridial AI training platform. | 中 | SR009 |
| CR017 | Sacra says Invisible evolved from a concierge-style virtual assistant service into an AI training, RLHF, and automation platform using specialists in 35 or more countries. | 中 | SR018, SR019 |
| CR018 | Sacra estimates Invisible reached about $134 million of 2024 revenue with roughly an 11% EBITDA margin. | 中 | SR018, SR019 |
| CR019 | Sacra says frontier labs are moving toward synthetic data generation and that Invisible is strategically pivoting toward large enterprise clients in response. | 中 | SR018, SR019 |
| CR020 | SiliconANGLE reported that Invisible's 2025 financing valued the company at more than $2 billion. | 中 | SR017, SR016 |
| CR021 | Invisible's custom-solutions page says it automates invoice reconciliation, W9 processing, claim approval letters, and compliance support. | 中 | SR001 |
| CR022 | Invisible's insurance case study says it automated claim approval, W9 processing, and compliance workflows, improved W9 accuracy from 75% to 98%, and cut claim response times by 50 percent. | 中 | SR010 |
| CR023 | Invisible's Headway case study says it delivered 8x faster insurance validation and 37% lower cost than Headway's internal team. | 中 | SR011 |
| CR024 | Invisible's Nasdaq case study says an onboarding integration program saved more than 10,000 developer hours and emphasized accuracy and reliability controls. | 中 | SR012 |
| CR025 | Invisible's Boosted.ai case study says the work required expert-level QA and that the customer considered the workflow impossible without human oversight. | 中 | SR013 |
| CR026 | Invisible's Cohere case study says human evaluations on enterprise and HR-style tasks produced a nine-point Arabic-dialect lead over GPT-4o and DeepSeek-V3 in the cited example. | 中 | SR014 |
| CR027 | Invisible's own competitive framing says regulated or high-risk environments need stronger controls for data access, work review, and decision documentation than annotation-first platforms typically provide. | 中 | SR015 |
| CR028 | The EU AI Act page says deployers must ensure human oversight and monitoring while providers must operate post-market monitoring and incident reporting for high-risk systems. | 高 | SR021, SR022 |
| CR029 | The EU AI Act page says transparency rules for many AI systems take effect in August 2026 and that high-risk employment-related rules apply on a longer timeline ending in December 2027 for certain areas. | 高 | SR021, SR022 |
| CR030 | Baker Botts says California, Texas, and Illinois AI laws took effect at the start of 2026 and Colorado's comprehensive AI Act becomes effective on June 30 2026. | 中 | SR022 |
| CR031 | Legal commentary from Cooley, Fisher Phillips, the American Bar Association, Harvard Journal on Legislation, and National Law Review describes rising discrimination, privacy, surveillance, and labor-law risk around workplace AI tools. | 中 | SR023, SR024, SR025, SR028, SR029 |
| CR032 | Alvarez & Marsal says regulators and plaintiffs are pursuing AI-washing, disclosure, discrimination, and vendor-compliance theories, especially for multinationals and boards. | 中 | SR026 |
| CR033 | The International AI Safety Report 2026 treats capable general-purpose and agentic systems as a governance problem that requires reliability monitoring, misuse mitigation, and human oversight. | 中 | SR027 |
| CR034 | Invisible sells through AWS Marketplace and publicly names AWS among its leading model-provider relationships. | 中 | SR030, SR016 |
| CR035 | Appen still positions a global contributor network and human evaluation as core to the AI lifecycle, showing that labor-backed AI data work remains a contested and competitive market. | 中 | SR031 |
| CR036 | UiPath emphasizes governed automation for regulated industries and publicly discloses ARR and large-customer metrics that Invisible does not publish. | 中 | SR032 |
| CR037 | Because Invisible connects models to legacy systems and measured business workflows, deployment failures can transmit directly into customer operations rather than remaining isolated model-quality problems. | 中 | SR002, SR027 |
| CR038 | Invisible's privacy policy puts international data transfers and service-provider governance at the center of its compliance perimeter. | 中 | SR003 |
| CR039 | Public-sector expansion raises procurement, security, and mission-critical reliability requirements beyond a typical commercial workflow deployment. | 中 | SR008, SR032 |
| CR040 | WeCP integration expands Invisible's expert-validation capability but adds integration, retention, and execution risk until product and team assimilation are proven. | 中 | SR009 |
| CR041 | Synthetic-data substitution pressure makes the company's shift from model-builder training work toward enterprise software and workflow revenue strategically important. | 中 | SR005, SR018 |
| CR042 | Invisible said it is deploying new capital into software modules and leadership expansion rather than simply preserving cash, which raises execution expectations for the next 12 to 24 months. | 中 | SR007, SR016 |
| CR043 | Case studies and public-sector materials place Invisible in insurance, healthcare, finance, and government-adjacent workflows where errors could carry regulated or monetary consequences. | 中 | SR008, SR010, SR011, SR012 |
| CR044 | Invisible's public mitigation stack relies on human expertise, continuous evaluation, red-teaming, workflow metrics, and documented processes rather than fully autonomous deployment. | 中 | SR002, SR005, SR013, SR014 |
| CR045 | Sacra and the customer case studies together suggest expert-backed service delivery still remains material to Invisible's economics even as software modules expand. | 中 | SR010, SR013, SR018 |
| CR046 | BusinessWire and the World Economic Forum both present Invisible as having been profitable for years, reducing immediate liquidity risk without resolving disclosure gaps on mix, margin, and concentration. | 中 | SR016, SR020 |
| CR047 | Invisible competes simultaneously against annotation-first vendors, AI-data specialists, and governed automation platforms, which increases pricing pressure and raises buyer expectations for control maturity. | 中 | SR015, SR031, SR032 |
| CR048 | The combination of a 350-person team, a distributed agent model, vendor onboarding controls, and international data transfers makes workforce and supplier oversight a first-order operating discipline rather than a back-office function. | 中 | SR003, SR004, SR007 |
| CR049 | No retained public source names a third-party security auditor, discloses an incident log, or confirms public-sector security authorizations for Invisible as of the run date. | 中 | SR003, SR005, SR008 |
| CR050 | Invisible maintains a privacy portal that centralizes privacy-policy access and request handling, implying ongoing operational work around data-subject rights. | 中 | SR033 |
| CR051 | Invisible markets healthcare as a dedicated industry vertical, extending its exposure to sensitive-data and regulated workflow environments beyond isolated case studies. | 中 | SR034 |
| CR052 | Invisible markets insurance as a dedicated industry vertical, reinforcing that insurer workflows are a strategic go-to-market lane rather than a one-off deployment. | 中 | SR035 |
| CR053 | Invisible markets life sciences as a dedicated industry vertical, widening the company's potential exposure to regulated processes and compliance-heavy customers. | 中 | SR036 |
| CR054 | Invisible markets private equity as a dedicated industry vertical, showing continued push into high-stakes financial workflows where accuracy and auditability matter. | 中 | SR037 |
| CR055 | Invisible markets energy and oil-and-gas operations as a dedicated industry vertical, adding critical-industry execution risk to its expanding sector footprint. | 中 | SR038 |
| CR056 | Invisible markets consumer-industry workflows as a dedicated vertical, indicating that the platform is broadening sector coverage faster than public control evidence is broadening. | 中 | SR039 |
| CV001 | Invisible announced a $100 million growth funding round on 2025-09-16. | 高 | SV006, SV007, SV008 |
| CV002 | The 2025 financing brought Invisible's total disclosed capital raised to $144 million. | 高 | SV006, SV007, SV014, SV015 |
| CV003 | The 2025 financing materials said the new capital would be invested in Invisible's core AI software platform. | 中 | SV006, SV007, SV014 |
| CV004 | Invisible reported $134 million of revenue for 2024. | 高 | SV006, SV007, SV009, SV016 |
| CV005 | Official and analyst sources say Invisible's revenue more than doubled from 2023 to 2024. | 高 | SV006, SV007, SV009, SV010 |
| CV006 | Sacra estimated Invisible grew from $60 million of 2023 revenue to $134 million in 2024, or about 123% year over year. | 中 | SV009, SV010 |
| CV007 | Sacra estimated Invisible generated roughly $15 million of EBITDA in 2024, implying about an 11% EBITDA margin. | 中 | SV009, SV010 |
| CV008 | Sacra said Invisible was valued at about $500 million in early 2024, equal to roughly 8.3x its then-$60 million revenue base. | 中 | SV009, SV010 |
| CV009 | SiliconANGLE reported that the September 2025 financing valued Invisible at more than $2 billion. | 中 | SV008, SV007 |
| CV010 | A $2.0 billion valuation against $134 million of 2024 revenue implies a trailing revenue multiple above 14.9x. | 中 | SV006, SV007, SV009 |
| CV011 | Invisible's 2025 financing materials described five product layers: Neuron, Atomic, Expert Marketplace, Synapse, and Axon. | 中 | SV006, SV007 |
| CV012 | Invisible describes its offering as a modular platform combining data, agents, humans-in-the-loop, and evaluations rather than a single point product. | 中 | SV001, SV005 |
| CV013 | Invisible says forward-deployed engineers connect customer legacy systems to a model-agnostic platform while customer data remains in customer systems. | 中 | SV003, SV002 |
| CV014 | Invisible tells customers to track throughput, error rates, resource efficiency, and cost per transaction after deployment. | 中 | SV003 |
| CV015 | Invisible says it has trained foundation models for more than 80% of the world's leading AI model providers, including Microsoft, AWS, and Cohere. | 高 | SV004, SV006, SV007, SV016 |
| CV016 | The Headway case study reports 8x faster claims processing, 37% lower cost than internal teams, and 57% lower cost than the prior BPO. | 中 | SV017 |
| CV017 | The Nasdaq case study reports 63% lower onboarding time and more than 10,000 developer hours saved. | 中 | SV019 |
| CV018 | The national-insurer case study reports $450,000 of cost savings, 16,000 labor hours saved, and 50% faster claim approvals. | 中 | SV020 |
| CV019 | The Boosted.ai case study reports 90% cost savings and describes Invisible as critical for expert-ground-truth data on an AI investment assistant. | 中 | SV018 |
| CV020 | The World Economic Forum profile says Invisible has been profitable for over half a decade. | 中 | SV016 |
| CV021 | Sacra says Invisible operates with 3,000+ agents in 35+ countries alongside a 350-person full-time team. | 中 | SV009, SV010 |
| CV022 | The 2025 funding release says Invisible had 350 employees and doubled the size of its engineering organization in 2025. | 中 | SV006, SV007, SV014, SV015 |
| CV023 | The 2025 release says customers include Microsoft, Swiss Gear, SAIC, and the Charlotte Hornets. | 中 | SV006, SV007, SV014, SV015 |
| CV024 | Appen says it has a global crowd of more than 1 million skilled contributors and positions itself as a global leader in high-quality AI datasets. | 中 | SV023, SV024 |
| CV025 | Appen says its platform has processed 50M+ people hours, completed 20K+ AI projects, processed 10B units of data, and completed 100M LLM data elements. | 中 | SV023, SV024 |
| CV026 | UiPath discloses $1.901 billion of ARR, 109% dollar-based net retention, 2,624 customers above $100K ARR, and 374 above $1M ARR as of April 30, 2026. | 高 | SV026, SV030 |
| CV027 | TaskUs describes itself as a leading provider of outsourced digital services and next-generation customer experience for innovative companies. | 中 | SV025, SV029 |
| CV028 | C3 AI and Palantir both maintain dedicated investor-relations surfaces and SEC filing histories, highlighting the disclosure stack expected of public AI software comps. | 中 | SV027, SV028, SV031, SV032 |
| CV029 | Aventis says the median revenue multiple for AI companies in its large-transaction sample was 24.2x. | 中 | SV011 |
| CV030 | Aventis says 2025 AI fundraising medians sit around 25-30x EV/revenue while public SaaS trades closer to about 6x EV/revenue. | 中 | SV011, SV012 |
| CV031 | Finro says valuation premiums remain highest for model builders and rails, while applied AI categories track closer to familiar software benchmarks. | 中 | SV012, SV011 |
| CV032 | Finro says data intelligence still commands strong pricing relative to many applied AI niches. | 中 | SV012 |
| CV033 | Sacra says Scale AI was running at about $1.5 billion of ARR and a $25 billion valuation, implying roughly 16.7x revenue. | 中 | SV010 |
| CV034 | Sacra says Mercor was at roughly $50 million of revenue run rate and a $2 billion valuation, implying about 40x revenue. | 中 | SV010 |
| CV035 | Public sources reviewed still do not disclose Invisible's ARR, NRR, gross margin, cash balance, burn rate, or customer concentration. | 中 | SV001, SV003, SV006, SV009, SV010 |
| CV036 | Sacra says leading AI labs are moving toward synthetic data generation, which pressures pure RLHF and labeling demand. | 中 | SV010 |
| CV037 | Alvarez & Marsal says regulators and investors are scrutinizing AI-washing, governance controls, and third-party vendor oversight more aggressively. | 中 | SV022 |
| CV038 | Invisible's modern slavery statement says the board approved an annual review of labor and supplier risks for the 2024 financial year. | 中 | SV021 |
| CV039 | The official product pages, customer cases, and World Economic Forum profile support a thesis that Invisible is pivoting from model-builder services toward enterprise AI workflow ownership. | 中 | SV001, SV005, SV006, SV016, SV017, SV019, SV020 |
| CV040 | Sacra's 11% EBITDA estimate and 3,000+ agent footprint imply that Invisible's economics still look more labor-assisted than software-pure today. | 中 | SV009, SV010 |
| CV041 | Invisible's >14.9x trailing multiple is below elite frontier-style private AI comps such as Scale's 16.7x and Mercor's 40x, but well above the ~6x public SaaS benchmark cited by Aventis. | 中 | SV006, SV007, SV010, SV011 |
| CV042 | Because the company has no public cap-table or preference disclosures, the headline valuation above $2 billion may overstate common-equity value. | 中 | SV006, SV007, SV008 |
| CV043 | Compared with public AI and automation comps that maintain investor-relations surfaces and SEC filing histories, Invisible is not yet disclosure-ready for a near-term IPO standard. | 中 | SV001, SV006, SV025, SV026, SV027, SV028, SV029, SV030, SV031, SV032 |
| CV044 | The current evidence supports a strategic sale or another late-stage private financing more credibly than a near-term public listing. | 中 | SV006, SV025, SV026, SV027, SV028, SV029, SV030, SV031, SV032 |
| CV045 | At an entry price above $2 billion, base-case underwriting only works if 2025-2026 revenue and gross-profit progression materially exceeded the last disclosed 2024 base. | 中 | SV006, SV007, SV009, SV011, SV012 |
| CV046 | If Invisible cannot prove software-led gross-margin expansion and resilient enterprise demand beyond RLHF, the valuation is vulnerable to applied-AI multiple compression. | 中 | SV010, SV011, SV012, SV022 |
| CV047 | The customer outcome evidence across healthcare, financial data, insurance, and onboarding reduces the chance that the 2025 valuation is pure narrative. | 中 | SV017, SV018, SV019, SV020 |
| CV048 | TechNews180 and Intelligence360 independently corroborate the $100 million round, the $144 million total funding figure, and the enterprise AI infrastructure positioning. | 中 | SV014, SV015, SV006 |
| CV049 | GetLatka places Invisible in a broader MLOps software competitor set, reinforcing that buyers can compare it against workflow software vendors rather than only outsourcing peers. | 低 | SV033 |
| CV050 | CaseStudies.com restates Invisible as an end-to-end AI software platform that structures data, deploys agentic solutions, and mobilizes relevant human experts. | 低 | SV034 |
| CV051 | Multiples.vc markets a public-comps and M&A multiples database spanning 238 granular sectors, illustrating why valuation work for Invisible should use model-appropriate comp buckets rather than a single generic AI average. | 低 | SV013 |
| CV052 | Salestools published a brief item labeling Invisible's $100 million financing as a growth raise. | 低 | SV035 |
| CV053 | TELUS Digital presents itself as a customer-experience and digital-solutions provider, supporting use of hybrid digital-operations comps alongside pure AI software names in Invisible's public benchmark set. | 中 | SV036 |
| 编号 | 出版方 | 标题 | 引文 |
|---|---|---|---|
| SO001 | Invisible Technologies | AI Software for Labs and Enterprise | Invisible Technologies | With Invisible’s modular platform, you plug in only the pieces you need (data, agents, humans-in-the-loop, evaluations). |
| SO002 | Invisible Technologies | About Invisible Technologies | We Make Enterprise AI Work | |
| SO003 | Invisible Technologies | How we work | Your data stays in your systems. |
| SO004 | Invisible Technologies | AI Training & RLHF Services | Invisible Technologies | |
| SO005 | Invisible Technologies | Invisible for Public Sector | Invisible Technologies | |
| SO006 | Invisible Technologies | Careers in AI & Operations | Join Invisible Technologies | |
| SO007 | Invisible Technologies | Privacy policy | Invisible Technologies | Invisible Technologies Inc. delivers digital work by outsourcing business processes to human agents. |
| SO008 | Invisible Technologies | Modern Slavery Statement | Invisible Technologies | It has been approved by Invisible Technologies’ board of directors and signed by Francis Pedraza (Founder, President and Chair of the board). |
| SO009 | Invisible Technologies | Invisible Appoints McKinsey AI Leader Matt Fitzpatrick CEO | Invisible achieved a $500 million valuation in early 2024. |
| SO010 | Invisible Technologies | Invisible 61st on the 2024 Deloitte Technology Fast 500™ | Invisible Technologies grew 2,342% during this period. |
| SO011 | Invisible Technologies | Invisible launches global public sector operations | Invisible Technologies has appointed Wes Green as the company’s first senior vice president, Global Public Sector. |
| SO012 | Invisible Technologies | Invisible Technologies acquires WeCP | Press Release | WeCP brings a library of more than 18,000 scope-specific technical assessments and over two million real-world interview records. |
| SO013 | Invisible Technologies | $100M Fundraise to Power the Next Gen of AI Infrastructure | This investment brings Invisible’s total capital raised to $144 million. |
| SO014 | Invisible Technologies | 8x Faster Claims for Headway | Invisible Technologies | Invisible achieved an 8x faster claims processing speed than previous vendors. |
| SO015 | Invisible Technologies | +233% Restaurant Onboarding Speed | Invisible Technologies | Now, streamlined Invisible processes structure 1.5M unique data points monthly, and have reduced onboarding costs by 50%. |
| SO016 | Invisible Technologies | 10k Dev Hours Saved for Nasdaq | Invisible Technologies | Nasdaq reduced onboarding times by 63%, saving over 10,000+ hours of engineers' time. |
| SO017 | Invisible Technologies | Agentic AI for Complex Enterprise Tasks with Cohere | We had partnered with Invisible previously, to train our Command R model for hallucination reduction. |
| SO018 | Business Wire | Invisible Technologies Raises $100 Million to Power the Next Generation of AI Infrastructure for the Enterprise | Revenue more than doubled from 2023 to 2024, reaching $134 million. |
| SO019 | SiliconANGLE | AI data provider Invisible raises $100M at $2B+ valuation | Bloomberg reported today that the deal values the company at more than $2 billion. |
| SO020 | Sacra | Invisible revenue, valuation & funding | With profitability of approximately $15M EBITDA (11% margin) on its 2024 revenue, Invisible has established itself as a financially sustainable player. |
| SO021 | Sacra | Invisible at $134M in revenue | Invisible Technologies (founded 2015) started as a virtual assistant service and has since grown into a platform. |
| SO022 | World Economic Forum | Invisible Technologies | Invisible, profitable for over half a decade, reached $134M in revenue and ranked |
| SO023 | AWS Marketplace | Invisible Technologies makes AI work | Our modular platform adapts models to your business and adds human expertise when needed, the same approach used to improve models for over 80% of the world's top AI companies. |
| SO024 | FeaturedCustomers | 23 Invisible Technologies Customer Reviews & References | Invisible has trained foundation models for more than 80% of the world’s leading AI model providers, including Cohere, Microsoft, and AWS. |
| SO025 | Superior Court of California filing hosted on S3 | Crowley v. Invisible Technologies Inc. class action complaint | Violation of Cal. Labor Code §§ 510 and 1198 (Unpaid Overtime). |
| SO026 | Indeed | Working at Invisible Technologies: reviews | Additional Verification Required. |
| SO027 | Better Business Bureau | Invisible Technologies complaints | Better Business Bureau | Verification successful. Waiting for www.bbb.org to respond. |
| SO028 | Crunchbase | Invisible Technologies | Crunchbase | Attention Required! | Cloudflare. |
| SO029 | PitchBook | Invisible Technologies 2026 company profile | PitchBook | |
| SO030 | Invisible AI | Invisible AI | Visual Intelligence for Manufacturing | Vision Execution System Scale Your Impact Across Every Line, Every Shift. |
| SM001 | Invisible Technologies | AI Training & RLHF Services | Invisible Technologies | Train and evaluate models in 80+ languages, ensuring cultural precision and linguistic accuracy for global deployment. |
| SM002 | Invisible Technologies | Reinforcement learning environments | Invisible Technologies | Tasks are drawn from work that creates real value like coding, accounting, banking, legal and compliance. |
| SM003 | Invisible Technologies | Custom AI solutions for enterprise | Invisible Technologies | With Invisible’s modular platform, you plug in only the pieces you need (data, agents, humans-in-the-loop, evaluations), and drive outcomes you can measure, fast. |
| SM004 | Invisible Technologies | How we work | Track operational metrics: throughput, error rates, resource efficiency, cost per transaction. |
| SM005 | Invisible Technologies | From benchmarks to business value: How to evaluate AI | Enterprises need to adopt custom evaluation frameworks specifically tailored to their unique use cases and business objectives. |
| SM006 | Invisible Technologies | Forward Deployed Engineering | Invisible Technologies | Forward deployment turns AI from a slideware promise into operational reality. |
| SM007 | Invisible Technologies | Enterprise back office solutions | Invisible Technologies | Agents draft outputs, surface source evidence for verification, and flag uncertain decisions for human review. |
| SM008 | Invisible Technologies | AI Investment Assistant for Boosted.ai | Invisible | Actionable Insights & Data Updates: Real-Time Cost Savings: 90% |
| SM009 | Invisible Technologies | 10k Dev Hours Saved for Nasdaq | Invisible Technologies | Onboarding Times: -63% Developer Hours Saved: 10,000+ |
| SM010 | Invisible Technologies | Slashing Costs with AI Automation for National Insurer | Cost savings: $450k Reduction in claim approval response times: 50% Hours in labor savings: 16,000 hours |
| SM011 | Invisible Technologies | 8x Faster Claims for Headway | Invisible Technologies | Cost vs. internal team: -37% Cost vs. BPO: -57% Increased claim processing speed: 8x |
| SM012 | Invisible Technologies | Agentic AI for Complex Enterprise Tasks with Cohere | For example, take Arabic dialects–its ADI2 score (a human evaluation metric) achieved a 9-point lead over GPT-4o and DeepSeek-V3. |
| SM013 | Invisible Technologies | Ranking & Optimizing RAG AI Models for Enterprise Platform | Rating/ranking tasks performed: 1,100 |
| SM014 | Invisible Technologies | 300 RAG Chats Evaluated Weekly | Invisible Technologies | Conversations Evaluated per Week: 300 |
| SM015 | Invisible Technologies | Why Pre-Training Is No Longer Enough: RL Environments | Pre-training gave large language models language. RL environments are giving them judgment. |
| SM016 | Invisible Technologies | RL Pipeline Bottlenecks: What Goes Wrong Before Training | The three primary failure points are: poorly specified reward functions, simulation-to-real mismatch, and off-policy data drift. |
| SM017 | Invisible Technologies | The Grader Problem: Why Most RL Environments Fail Early | Reward hacking occurs when an RL agent finds a path to a high reward signal without actually completing the underlying task. |
| SM018 | Invisible Technologies | Why Frontier Labs Outsource RL Environments | Invisible | Frontier labs outsource RL environments because domain coverage — not compute or algorithms — is the binding constraint in post-training. |
| SM019 | Appen | AI Data Platform (ADAP) | Appen | 50M+ People hours on platform; 20K+ AI projects completed; 100M LLM data elements completed. |
| SM020 | Appen | Investors Relations | Appen | Appen is the global leader in the development of high-quality datasets that are used to build and continuously improve artificial intelligence systems. |
| SM021 | Amazon Web Services | AWS Marketplace seller profile for Invisible Technologies | Invisible Technologies makes AI work... the same approach used to improve models for over 80% of the world's top AI companies. |
| SM022 | UiPath | Investors | UiPath is a leader in business orchestration and automation... $1.901B ARR growing 12% year over year. |
| SM023 | TaskUs | Investor Relations | TaskUs | TaskUs is a leading provider of outsourced digital services and next-generation customer experience to the world’s most innovative companies. |
| SM024 | Labelbox | Plans & Pricing | Labelbox | Highly-skilled AI trainers curated from our Alignerr Network for complex post-training and eval projects. |
| SM025 | European Commission | AI Act | High-risk AI systems are subject to strict obligations before they can be put on the market: logging, documentation, human oversight, and robustness. |
| SM026 | Fisher Phillips | Comprehensive Review of AI Workplace Law and Litigation as We Enter 2025 | Over 30 states have formed AI committees or taskforces that have begun issuing reports and recommendations. |
| SM027 | Baker Botts | U.S. Artificial Intelligence Law Update: Navigating the Evolving State and Federal Regulatory Landscape | Thought Leadership | January 2026 | Baker Botts | Over 1,000 AI-related bills were introduced across states in 2025 alone. |
| SM028 | Alvarez & Marsal | AI Litigation, Enforcement, and Compliance Risk: Q4 2025 Regulatory Update | In the past six months, multiple cases were brought by DOJ, the SEC, and the FTC related to AI washing and AI fraud. |
| SM029 | Harvard Journal on Legislation | The Sound and Fury of Regulating AI in the Workplace – Harvard Journal on Legislation | The use of AI simultaneously presents labor and employment law risks, including introducing or proliferating bias or unlawful discrimination. |
| SP001 | Invisible Technologies | AI Software for Labs and Enterprise | Invisible Technologies | |
| SP002 | Invisible Technologies | About Invisible Technologies | We Make Enterprise AI Work | |
| SP003 | Invisible Technologies | How we work | |
| SP004 | Invisible Technologies | AI Training & RLHF Services | |
| SP005 | Invisible Technologies | Contact Center Solutions | Invisible Technologies | |
| SP006 | Invisible Technologies | Computer Vision Solutions | Invisible Technologies | |
| SP007 | Invisible Technologies | Custom AI solutions for enterprise | Invisible Technologies | |
| SP008 | Invisible Technologies | Enterprise back office solutions | Invisible Technologies | |
| SP009 | Invisible Technologies | Saving 10,000 Hours Through Seamless Interoperability | |
| SP010 | Invisible Technologies | Invisible launches global public sector operations | |
| SP011 | Invisible Technologies | $100M Fundraise to Power the Next Gen of AI Infrastructure | |
| SP012 | Invisible Technologies | Modern Slavery Statement | Invisible Technologies | |
| SP013 | Business Wire | Invisible Technologies Raises $100 Million to Power the Next Generation of AI Infrastructure for the Enterprise | |
| SP014 | Sacra | Invisible revenue, valuation & funding | |
| SP015 | Sacra | Invisible at $134M in revenue | |
| SP016 | FeaturedCustomers | Invisible Technologies Customer Reviews & References | |
| SP017 | Amazon Web Services | AWS Marketplace Seller Profile: Invisible Technologies | |
| SP018 | CB Insights | Top Invisible Alternatives, Competitors | |
| SP019 | Invisible Technologies | Top Scale AI alternatives and competitors for enterprise AI | |
| SP020 | Labelbox | Plans & Pricing | Labelbox | |
| SP021 | Appen | AI Data Platform (ADAP) | Appen | |
| SP022 | Appen | Investor Relations | Appen | |
| SP023 | DataAnnotation | DataAnnotation | Future-Proof Your Career With AI Training Work | |
| SP024 | TaskUs | TaskUs Investor Relations | |
| SP025 | UiPath | UiPath Investor Relations | |
| SP026 | Alvarez & Marsal | AI Litigation, Enforcement and Compliance Risk: Q4 2025 Regulatory Update | |
| SP027 | SiliconANGLE | AI data provider Invisible raises $100M at $2B+ valuation | |
| SP028 | Invisible Technologies | Off-the-Shelf Computer Vision vs. Custom Models: What Enterprises Need | |
| SP029 | Invisible Technologies | Why Frontier Labs Outsource RL Environments: The Domain Coverage Problem | |
| SP030 | Invisible Technologies | RL Pipeline Bottlenecks: What Goes Wrong Before Training | |
| SP031 | Latka | Top Invisible Technologies Alternatives, Competitors & Similar Software | GetLatka | |
| SP032 | NVIDIA | NVIDIA AI in Manufacturing | |
| SI001 | Invisible Technologies | $100M Fundraise to Power the Next Gen of AI Infrastructure | Revenue more than doubled from 2023 to 2024, reaching $134 million. |
| SI002 | Invisible Technologies | How we work | Track operational metrics: throughput, error rates, resource efficiency, cost per transaction. |
| SI003 | Invisible Technologies | Custom AI solutions for enterprise | Invisible Technologies | |
| SI004 | Invisible Technologies | AI Training & RLHF Services | Invisible Technologies | |
| SI005 | Invisible Technologies | Agentic AI for Complex Enterprise Tasks with Cohere | |
| SI006 | Invisible Technologies | AI Investment Assistant for Boosted.ai | Invisible | |
| SI007 | Invisible Technologies | 8x Faster Claims for Headway | Invisible Technologies | |
| SI008 | Invisible Technologies | 10k Dev Hours Saved for Nasdaq | Invisible Technologies | |
| SI009 | Invisible Technologies | Slashing Costs with AI Automation for National Insurer | |
| SI010 | Invisible Technologies | Streamlining Recruitment for Retailer with 700 Locations | |
| SI011 | Invisible Technologies | AI Software for Labs and Enterprise | Invisible Technologies | |
| SI012 | Invisible Technologies | Invisible 61st on the 2024 Deloitte Technology Fast 500™ | Invisible Technologies grew 2,342% during this period. |
| SI013 | Business Wire | Invisible Technologies Raises $100 Million to Power the Next Generation of AI Infrastructure for the Enterprise | This investment brings Invisible’s total capital raised to $144 million. |
| SI014 | SiliconANGLE | AI data provider Invisible raises $100M at $2B+ valuation | Bloomberg reported today that the deal values the company at more than $2 billion. |
| SI015 | Sacra | Invisible revenue, valuation & funding | Sacra estimates that Invisible generated $134M in revenue in 2024, up 123% from $60M in 2023. |
| SI016 | Sacra | Invisible at $134M in revenue | Invisible is strategically pivoting away from serving AI labs with training data (which are increasingly moving toward synthetic data generation) to focus on large-scale enterprise clients. |
| SI017 | Intelligence360 | Invisible Technologies Raises $100 Million to Power the Next Generation of AI Infrastructure for the Enterprise | Revenue more than doubled from 2023 to 2024, reaching $134 million. |
| SI018 | Appen | Investors Relations | Appen | |
| SI019 | Appen | AI Data Platform (ADAP) | Appen | 50M+ people hours on platform. |
| SI020 | TaskUs | Investor Relations | TaskUs | |
| SI021 | UiPath | Investors | $1.901B ARR growing 12% year over year. |
| SI022 | C3.ai | Investor Relations | C3.ai, Inc. | |
| SI023 | Palantir Technologies | Palantir IR | |
| SI024 | Securities and Exchange Commission | Company Information: | |
| SI025 | Securities and Exchange Commission | Company Information: | |
| SI026 | Invisible Technologies | Invisible for Asset Management | Invisible Technologies | |
| SI027 | Invisible Technologies | Invisible for Banking Industry | Invisible Technologies | |
| SE001 | Invisible Technologies | Enterprise back office solutions | Invisible Technologies | Agents draft outputs, surface source evidence for verification, and flag uncertain decisions for human review. |
| SE002 | Invisible Technologies | Contact Center Solutions | Invisible Technologies | Evaluate 100% of interactions against your policies and quality standards, without relying on sampling. |
| SE003 | Invisible Technologies | Demand Forecasting Solutions | Invisible Technologies | Unify ERP, POS, e-comm, labor, ops, and external signals into a single demand foundation. |
| SE004 | Invisible Technologies | Computer Vision Solutions | Invisible Technologies | Local deployment for secure or remote environments. Your data stays in your control. |
| SE005 | Invisible Technologies | AI Training & RLHF Services | Invisible Technologies | Our Meridial Expert Network connects you to vetted trainers who elevate model performance from day one. |
| SE006 | Invisible Technologies | Reinforcement learning environments | Invisible Technologies | Every run is logged and replayable. Debug failures, compare model versions, and show stakeholders exactly what the agent did and why. |
| SE007 | Invisible Technologies | How we work | Your data stays in your systems. |
| SE008 | Invisible Technologies | $100M Fundraise to Power the Next Gen of AI Infrastructure | Invisible will use the funding to invest further in its core AI Software Platform, which consists of five modular components. |
| SE009 | Business Wire | Invisible Technologies Raises $100 Million to Power the Next Generation of AI Infrastructure for the Enterprise | With a team of 350, it has doubled the size of its engineering organization in 2025. |
| SE010 | Invisible Technologies | Invisible acquires WeCP | Invisible will integrate WeCP’s evaluation infrastructure into Meridial, its AI training platform, to support more precise expert validation and reinforcement learning workflows. |
| SE011 | Invisible Technologies | 8x Faster Claims for Headway | Invisible Technologies | Invisible optimized the workflow delivering 8x efficiency through batching and parallel processing. |
| SE012 | Invisible Technologies | 10k Dev Hours Saved for Nasdaq | Invisible Technologies | Invisible implemented a comprehensive integration solution, enabling seamless interoperability between diverse data platforms. |
| SE013 | Invisible Technologies | Slashing Costs with AI Automation for National Insurer | Claim approval processes saw a 50% reduction in response times. |
| SE014 | Invisible Technologies | Agentic AI for Complex Enterprise Tasks with Cohere | They maintain a really high bar for talent, with continuous observability that ensures we can trust the data. |
| SE015 | Invisible Technologies | 20k Evaluations for You.com | Invisible Technologies | Invisible implemented a structured rating system to evaluate article relevance based on the user’s query intent, and evaluated 20,000 model responses. |
| SE016 | Invisible Technologies | AI Case Studies for Enterprise Operations | Invisible | Invisible delivered a faster, more accurate AI investment assistant at 90% lower cost that unlocked real-time insights. |
| SE017 | Invisible Technologies | Off-the-Shelf Computer Vision vs. Custom Models: What Enterprises Need | Custom models can be optimized for edge computing, running directly on the cameras or local servers within your facility. |
| SE018 | Invisible Technologies | What Data Does Computer Vision Produce & How to Use It | Modern computer vision algorithms process the frame at the source and transmit only the essential facts. |
| SE019 | Invisible Technologies | Preventing Computer Vision Model Degradation in Production | Organizations can prevent this degradation by implementing continuous feedback loops, establishing automated retraining pipelines, and maintaining high-quality human-in-the-loop validation. |
| SE020 | Invisible Technologies | Why Pre-Training Is No Longer Enough: RL Environments | Pre-training gave large language models language. RL environments are giving them judgment. |
| SE021 | Invisible Technologies | The Grader Problem: Why Most RL Environments Fail Early | The approach that works operates in three tiers, each designed to catch what the previous tier misses. |
| SE022 | Invisible Technologies | From benchmarks to business value: How to evaluate AI | Enterprises need to adopt custom evaluation frameworks specifically tailored to their unique use cases and business objectives. |
| SE023 | Invisible Technologies | Forward Deployed Engineering | Invisible Technologies | How forward deployment turns AI from a slideware promise into operational reality. |
| SE024 | Invisible Technologies | Designing multimodal systems | Invisible Technologies | Multimodal perception is harder than it looks. Decompose the task, don’t rely on one world model. |
| SE025 | Invisible Technologies | Beyond text: Why multimodal AI demands a different playbook | With the benefits, there are higher demands on data, infrastructure, governance and change management. |
| SE026 | Invisible Technologies | Privacy policy | Invisible Technologies | The software also automatically collects Work Information about the Agent’s work for the Client, such as keystrokes, mouse clicks, screenshots, and webcam pictures of the Agent. |
| SE027 | Invisible Technologies | Careers in AI & Operations | Join Invisible Technologies | Invisible AI has hubs in the cities that matter most, where the people shaping the future of our company come together to push boundaries. |
| SE028 | Invisible Technologies | Trustero | Trustero |
| SE029 | FeaturedCustomers | Invisible Technologies Reviews and Testimonials | Read 7 Invisible Technologies reviews and testimonials from customers, explore 16 case studies and customer success stories. |
| SE030 | World Economic Forum | Invisible Technologies organization page | Invisible Technologies end-to-end AI software platform structures messy data, builds digital workflows, deploys agentic solutions, evaluates/measures impact, and mobilizes relevant human experts. |
| SE031 | Appen | AI Data Platform (ADAP) | Appen | Appen holds GDPR compliance, AICPA SOC certification, HIPAA compliance, and TÜV Rheinland certification to ISO/IEC 27001:2013. |
| SE032 | Cohere | Enterprise AI: Private, Secure, Customizable | Cohere | Secure within your virtual private cloud (VPC), on-premises, or dedicated, Cohere-managed Model Vault. |
| SE033 | International AI Safety Report | International AI Safety Report 2026 | This remains the most rigorous assessment of AI capabilities, risks, and risk management available. |
| SE034 | American Bar Association | Recent Developments in Artificial Intelligence Cases and Legislation 2025 | Emerging themes for both the courts and state and local legislators center around copyright infringement, privacy, fairness/perceived bias, civil rights, transparency and consent. |
| SE035 | Invisible AI | Invisible AI | Visual Intelligence for Manufacturing | Entirely on-premise solution |
| SE036 | DataAnnotation.tech | Open roles and interview process | Complete an assessment aligned with your area of expertise. |
| SU001 | Invisible Technologies | AI Case Studies for Enterprise Operations | Invisible | |
| SU002 | Invisible Technologies | Agentic AI for Complex Enterprise Tasks with Cohere | They maintain a really high bar for talent, with continuous observability that ensures we can trust the data. |
| SU003 | Invisible Technologies | Ranking & Optimizing RAG AI Models for Enterprise Platform | |
| SU004 | Invisible Technologies | 300 RAG Chats Evaluated Weekly | Invisible Technologies | |
| SU005 | Invisible Technologies | AI Investment Assistant for Boosted.ai | Invisible | Cost Savings: 90%. |
| SU006 | Invisible Technologies | 10k Dev Hours Saved for Nasdaq | Invisible Technologies | Nasdaq reduced onboarding times by 63%, saving over 10,000+ hours of engineers' time. |
| SU007 | Invisible Technologies | 8x Faster Claims for Headway | Invisible Technologies | Invisible achieved an 8x faster claims processing speed than previous vendors. |
| SU008 | Invisible Technologies | AI for End-to-End Customer Experience at Solar Provider | At peak, we were generating 180 contracts per day for new installations. |
| SU009 | Invisible Technologies | Slashing Costs with AI Automation for National Insurer | Saved managers over 16,000 hours of manual work, realizing cost savings of over $320,000 and helping avoid the risk of millions in potential penalties. |
| SU010 | Invisible Technologies | 9x ROI for Big 4 Retailer | Invisible Technologies | In 30 days, 3,100+ dead-stock items generated nearly $1M in revenue. |
| SU011 | Invisible Technologies | 20k Evaluations for You.com | Invisible Technologies | Evaluations: 20,000. |
| SU012 | Invisible Technologies | AI Operations Helping Getro Reach Cash-Flow Positivity | We love the daily documentation and service they provide - especially the bi-weekly calls with our account manager that keep us in the loop on work performance and quality! |
| SU013 | Invisible Technologies | +233% Restaurant Onboarding Speed | Invisible Technologies | Now, streamlined Invisible processes structure 1.5M unique data points monthly, and have reduced onboarding costs by 50%. |
| SU014 | Invisible Technologies | Invisible for Public Sector | Invisible Technologies | Invisible helped Nasdaq streamline a data integration process, reducing customer onboarding time and saving 10,000 developer hours. |
| SU015 | Invisible Technologies | Invisible for Sports Industry | Invisible Technologies | Invisible improved Cohere's data quality and scalability, enhancing multilingual, coding, and reasoning capabilities to strengthen its enterprise-ready AI performance. |
| SU016 | World Economic Forum | Invisible Technologies | The same approach used to improve models for over 80% of the world's top AI companies, including Microsoft, AWS, and Cohere. |
| SU017 | Amazon Web Services | AWS Marketplace seller profile for Invisible Technologies | We work across industries - supply chain automation for Swiss Gear, AI-enabled naval simulations with SAIC, and validating NBA draft picks for the Charlotte Hornets. |
| SU018 | FeaturedCustomers | 23 Invisible Technologies Customer Reviews & References | Read 7 Invisible Technologies reviews and testimonials from customers, explore 16 case studies and customer success stories, and watch customer videos. |
| SU019 | CaseStudies.com | Invisible Technologies B2B Case Studies & Customer Successes | Invisible has trained foundation models for more than 80% of the world's leading AI model providers, including Cohere, Microsoft, and AWS. |
| SU020 | OpenCourt Basketball | The Hornets, The Draft, And The Algorithm: Inside The Kon Knueppel AI Story | The “gifted us Kon Knueppel” phrasing appears in Invisible-authored or Invisible-hosted marketing-style content, and the viral spread of the quote has largely been driven by social amplification rather than an official Hornets press release. |
| SU021 | Microsoft | Microsoft – AI, Cloud, Productivity, Computing, Gaming & Apps | |
| SU022 | Amazon Web Services | Cloud Computing Services - Amazon Web Services (AWS) | |
| SU023 | Nasdaq | Nasdaq: Stock Market, Data Updates, Reports & News | |
| SU024 | You.com | The Leading Web Search APIs for AI | Powering web search for leading enterprises. |
| SU025 | Cohere | Enterprise AI: Private, Secure, Customizable | Cohere | Enterprise AI: Private, Secure, Customizable. |
| SR001 | Invisible Technologies | Custom solutions | |
| SR002 | Invisible Technologies | How we work | Daily outputs by location, category, or workflow. Real-time adjustments as conditions shift. Track operational metrics: throughput, error rates, resource efficiency, cost per transaction. |
| SR003 | Invisible Technologies | Privacy policy | Invisible Technologies | We may use technologies to engage in automated decision making or profiling. We will not use these technologies to make automated decisions about you that would significantly affect you, unless such a decision is necessary as part of a contract we have with you, we have your consent, or we are permitted by law to use such technology. |
| SR004 | Invisible Technologies | Modern Slavery Statement | Invisible Technologies | Invisible Technologies maintains a third party risk program that reviews vendor practices during onboarding. |
| SR005 | Invisible Technologies | AI Training & RLHF Services | Invisible Technologies | Red-teaming, fine-tuning, and policy informed evaluations with a dedicated SWAT team to align models with safe and compliant use. |
| SR006 | Invisible Technologies | Invisible Appoints McKinsey AI Leader Matt Fitzpatrick CEO | |
| SR007 | Invisible Technologies | $100M Fundraise to Power the Next Gen of AI Infrastructure | This investment brings Invisible’s total capital raised to $144 million. |
| SR008 | Invisible Technologies | Invisible launches global public sector operations | |
| SR009 | Invisible Technologies | Invisible Technologies acquires WeCP | Press Release | |
| SR010 | Invisible Technologies | Slashing costs with automation for national insurance company | |
| SR011 | Invisible Technologies | Achieving 8x faster claims processing for Headway | |
| SR012 | Invisible Technologies | Saving Nasdaq 10,000 developer hours | |
| SR013 | Invisible Technologies | Launching a better, faster AI investment assistant for Boosted.ai | |
| SR014 | Invisible Technologies | Cohere agentic enterprise tasks | |
| SR015 | Invisible Technologies | Top Scale AI alternatives and competitors for enterprise AI | |
| SR016 | Business Wire | Invisible Technologies Raises $100 Million to Power the Next Generation of AI Infrastructure for the Enterprise | Revenue more than doubled from 2023 to 2024, reaching $134 million. |
| SR017 | SiliconANGLE | AI data provider Invisible raises $100M at $2B+ valuation | |
| SR018 | Sacra | Invisible at $134M in revenue | With the appointment of its ex-McKinsey CEO Matthew Fitzpatrick in 2024, Invisible is strategically pivoting away from serving AI labs with training data (which are increasingly moving toward synthetic data generation) to focus on large-scale enterprise clients. |
| SR019 | Sacra | Invisible revenue, valuation & funding | |
| SR020 | World Economic Forum | Invisible Technologies | |
| SR021 | European Commission | AI Act | Once an AI system is on the market, authorities are in charge of market surveillance, deployers ensure human oversight and monitoring, and providers have a post-market monitoring system in place. |
| SR022 | Baker Botts | U.S. Artificial Intelligence Law Update: Navigating the Evolving State and Federal Regulatory Landscape | Many states, including California, Texas, and Illinois, have enacted significant AI legislation taking effect at the start of 2026, with Colorado’s comprehensive AI Act following on June 30, 2026. |
| SR023 | Cooley | AI in the Workplace: US Legal Developments | |
| SR024 | Fisher Phillips | Comprehensive Review of AI Workplace Law and Litigation as We Enter 2025 | |
| SR025 | American Bar Association | Recent Developments in Artificial Intelligence Cases and Legislation 2025 | |
| SR026 | Alvarez & Marsal | AI Litigation, Enforcement, and Compliance Risk: Q4 2025 Regulatory Update | |
| SR027 | International AI Safety Report | International AI Safety Report 2026 | |
| SR028 | Harvard Journal on Legislation | The Sound and Fury of Regulating AI in the Workplace | |
| SR029 | National Law Review | The Hidden Legal Minefield- Compliance Concerns with AI Smart Glasses, Part 3 –Privacy, Surveillance, and Labor Law Violations | |
| SR030 | AWS Marketplace | AWS Marketplace seller profile | |
| SR031 | Appen | Investors Relations | Appen | |
| SR032 | UiPath | Investors | |
| SR033 | Invisible Technologies | Privacy portal | Invisible Technologies | |
| SR034 | Invisible Technologies | Invisible for Healthcare Industry | Invisible Technologies | |
| SR035 | Invisible Technologies | Invisible for Insurance Industry | Invisible Technologies | |
| SR036 | Invisible Technologies | Invisible for Life sciences | Invisible Technologies | |
| SR037 | Invisible Technologies | Invisible for Private Equity | Invisible Technologies | |
| SR038 | Invisible Technologies | Invisible for Oil & Gas Operations | Invisible Technologies | |
| SR039 | Invisible Technologies | Invisible for Consumer Industry | Invisible Technologies | |
| SV001 | Invisible Technologies | AI Software for Labs and Enterprise | Invisible Technologies | |
| SV002 | Invisible Technologies | About Invisible Technologies | We Make Enterprise AI Work | |
| SV003 | Invisible Technologies | How we work | Track operational metrics: throughput, error rates, resource efficiency, cost per transaction. |
| SV004 | Invisible Technologies | AI Training & RLHF Services | Invisible Technologies | |
| SV005 | Invisible Technologies | Custom AI solutions for enterprise | Invisible Technologies | |
| SV006 | Invisible Technologies | $100M Fundraise to Power the Next Gen of AI Infrastructure | Revenue more than doubled from 2023 to 2024, reaching $134 million. |
| SV007 | Business Wire | Invisible Technologies Raises $100 Million to Power the Next Generation of AI Infrastructure for the Enterprise | This investment brings Invisible’s total capital raised to $144 million. |
| SV008 | SiliconANGLE | AI data provider Invisible raises $100M at $2B+ valuation | Bloomberg reported today that the deal values the company at more than $2 billion. |
| SV009 | Sacra | Invisible revenue, valuation & funding | Sacra estimates that Invisible generated $134M in revenue in 2024, up 123% from $60M in 2023. |
| SV010 | Sacra | Invisible at $134M in revenue | Invisible is strategically pivoting away from serving AI labs with training data (which are increasingly moving toward synthetic data generation) to focus on large-scale enterprise clients. |
| SV011 | Aventis Advisors | AI Valuation Multiples in 2025 | |
| SV012 | Finro | AI Valuation Multiples: Q4 2025 Update | |
| SV013 | Multiples.vc | Valuation Multiples by Industry - Multiples.vc - Public Comps and Valuation Multiples | |
| SV014 | Tech News 180 | Invisible Technologies Just Raised $100M - Here's Why VCs Are Betting Big | |
| SV015 | Intelligence360 | Invisible Technologies Raises $100 Million to Power the Next Generation of AI Infrastructure for the Enterprise | Revenue more than doubled from 2023 to 2024, reaching $134 million. |
| SV016 | World Economic Forum | Invisible Technologies | Invisible, profitable for over half a decade, reached $134M in revenue and ranked |
| SV017 | Invisible Technologies | 8x Faster Claims for Headway | Invisible Technologies | |
| SV018 | Invisible Technologies | AI Investment Assistant for Boosted.ai | Invisible | |
| SV019 | Invisible Technologies | 10k Dev Hours Saved for Nasdaq | Invisible Technologies | |
| SV020 | Invisible Technologies | Slashing Costs with AI Automation for National Insurer | |
| SV021 | Invisible Technologies | Modern Slavery Statement | Invisible Technologies | |
| SV022 | Alvarez & Marsal | AI Litigation, Enforcement and Compliance Risk: Q4 2025 Regulatory Update | |
| SV023 | Appen | Investors Relations | Appen | |
| SV024 | Appen | AI Data Platform (ADAP) | Appen | 50M+ people hours on platform. |
| SV025 | TaskUs | Investor Relations | TaskUs | |
| SV026 | UiPath | Investors | $1.901B ARR growing 12% year over year. |
| SV027 | C3.ai | Investor Relations | C3.ai, Inc. | |
| SV028 | Palantir Technologies | Palantir IR | |
| SV029 | Securities and Exchange Commission | Company Information: | |
| SV030 | Securities and Exchange Commission | Company Information: | |
| SV031 | Securities and Exchange Commission | Company Information: | |
| SV032 | Securities and Exchange Commission | Company Information: | |
| SV033 | GetLatka | Top Invisible Technologies Alternatives, Competitors & Similar Software | GetLatka | |
| SV034 | CaseStudies.com | Invisible Technologies B2B Case Studies & Customer Successes | |
| SV035 | Salestools | Invisible Technologies raises $100M Growth | |
| SV036 | TELUS Digital | TELUS Digital Customer Experience & Digital Solutions |