Consolidate product roadmaps

roadmap/product-roadmap-unified-zh.md

@@ -0,0 +1,721 @@
+# 首席增长官 - 统一产品路线图
+
+_这是一份合并后的首席增长官产品路线图。核心原则是：先有证据，再做扩张。每个阶段只有在证明了本阶段要消除的关键风险后，才进入下一阶段。_
+
+---
+
+## 读者与读后行动
+
+本文面向创始人、产品负责人、增长负责人和投资人。读完后，团队应能明确：下一步该验证什么、构建什么、如何设置阶段门槛、如何设计早期定价，以及什么时候应该拒绝过早的平台化扩张。
+
+---
+
+## 路线图理念
+
+不要一开始就构建完整的"AI首席增长官"。长期愿景可以很大，但第一个产品必须足够窄、足够可信、法律上可持续，并且商业上可验证。
+
+产品应分三层演进：
+
+1. **产品机会雷达** - 从市场信号中发现高质量的产品和传播机会。
+2. **决策智能系统** - 帮助品牌决定要做什么产品、改什么、测什么、说什么。
+3. **增长编排平台** - 协调产品、内容、广告、用户运营和服务工作流。
+
+前 12 个月最重要的问题是：
+
+> 我们能否通过合法、可持续的数据管道和可行的单位经济模型，帮助美妆与个护品牌比自己更早、更准地发现可执行的产品机会？
+
+如果答案是肯定的，更大的"首席增长官"平台才有资格从这个楔子自然生长。如果答案是否定的，增加更多引擎只会叠加风险，而不是形成护城河。
+
+---
+
+## 指导原则
+
+1. **按风险排序，而不是按功能排序。** 真正会杀死公司的风险是数据获取、客户信任、在可行价格下的付费意愿，以及入门产品能否低摩擦接入私有数据。
+2. **把阶段 0 设为硬性数据门。** 不要默认可以获取竞品评论、社交评论、平台信号、知识图谱来源或客户私有数据。必须先证明 PIPL 就绪、数据授权、API 可行性，以及哪些数据源不能碰。
+3. **楔子必须独立成立。** 产品创新引擎本身必须是毛利为正的业务，不能依赖未来多引擎加购来证明经济性。
+4. **信任是逐级赢得的。** 先做决策支持，再逐步走向编排和执行。只有当客户反复采纳系统输出后，才提高自主权。
+5. **深度优先于广度。** 先在一个美妆子品类做深，再加一个相邻引擎。除非有证据表明多引擎使用改善留存、ARPA 或 CAC 回收期，否则不要投入五引擎平台。
+6. **私有数据锁定从入门层开始。** 旗舰问题"我们比竞品 X 弱在哪里？"需要客户自己的数据。入门层必须包含轻量私有数据连接，否则大多数客户买到的核心体验是空的。
+7. **行业情报是可选且高风险的后期业务。** 匿名化或聚合的跨品牌情报必须放在后期，且需要客户明确授权、监管审查和渠道冲突管理。它不是默认资产。
+
+---
+
+## 信任阶梯
+
+信任阶梯贯穿所有阶段。定价权和"首席增长官"定位不是一开始宣称出来的，而是通过逐级爬升赢得的。
+
+| 阶梯 | 产品角色 | AI 做什么 | 主要阶段 |
+| --- | --- | --- | --- |
+| 1 | 洞察 | 呈现信号、排序痛点、标记异常 | 阶段 0-1 |
+| 2 | 推荐 | 给出具体行动建议，并附推理、证据和置信度 | 阶段 1 |
+| 3 | 起草 | 生成可直接使用、供人工审核的材料 | 阶段 2 |
+| 4 | 编排 | 在人工参与下协调跨职能多步骤工作流 | 阶段 2-3 |
+| 5 | 运营 | 在批准的护栏内执行受限工作流，由人工监督 | 阶段 3+ |
+
+每个产品决策都应说明它服务于哪个阶梯。任何暗示高于当前证据水平的功能，都应推迟。
+
+---
+
+## 初始 ICP
+
+初始客户应是中国美妆、护肤、个护或洗护品牌，年 GMV 在人民币 3,000 万至 3 亿元之间，主要通过天猫、抖音、小红书、京东或私域渠道销售。
+
+优先选择满足以下至少三个条件的品牌：
+
+- 频繁推出新品或进行 SKU 迭代。
+- 高度依赖成分、功效、质地或功能性宣称。
+- 创始人或产品负责人直接参与产品决策。
+- 竞品压力大。
+- 明确痛于发现新产品机会。
+- 客户反馈分散在多个平台。
+- 愿意连接轻量私有数据以获得更好的诊断。
+
+---
+
+# 阶段 0：数据门、ICP 与问题验证
+
+## 阶段命题
+
+先赢得构建产品机会雷达的资格。
+
+本阶段把任何"默认能拿到数据"的假设，替换为硬性的可行性门槛。团队应以最少工程投入证明业务成立所依赖的关键假设。
+
+## 时间盒
+
+0-8 周可作为规划估计，但是否进入下一阶段取决于证据，而不是时间。
+
+## 要证明什么
+
+- 竞品评论、电商问答、社交内容和社交评论，能否通过官方 API、授权数据供应商、直接合作、客户授权导出或其他合规方式合法且可持续地获得。
+- 私有数据接入所涉及的 PIPL 与数据处理要求是否清楚，包括客服对话、退款原因、SOV 数据、CRM 标签、销售和流失数据。
+- 能否在一个细分美妆品类中，用可靠来源播种一个窄而深的知识图谱，例如护肤精华、防晒或洗护修复。
+- 目标 ICP 是否存在迫切的产品机会发现痛点，并愿意在完整 SaaS 出现前为专家审核、AI 辅助的输出付费。
+- 战略意图过滤器是否会以客户能感知并重视的方式改变输出。
+
+## 工作流
+
+### 1. 数据获取与合规调查
+
+逐项梳理核心数据源：
+
+- 竞品产品评论。
+- 电商问答。
+- 法律与技术允许范围内的小红书帖子和评论。
+- 法律与技术允许范围内的抖音产品和内容信号。
+- 公开社交提及。
+- 客户提供的客服对话、退款原因、调研、CRM 标签、SKU 销售和区域退货数据。
+
+每个数据源都要记录：
+
+- 法律依据和授权模型。
+- API 或供应商可行性。
+- 限流、稳定性和成本。
+- 爬取是否被禁止或过于脆弱。
+- 数据保留要求。
+- PIPL 义务。
+- 该来源可用于产品、仅可用于礼宾式分析，还是必须排除。
+
+### 2. ICP 与问题验证
+
+访谈 20-30 位创始人、产品负责人和增长负责人，验证：
+
+- 他们当前如何发现产品机会。
+- 哪些产品决策一旦做错成本最高。
+- 他们信任哪些数据。
+- 他们多久看一次竞品评论、社交内容和客服对话。
+- 什么会让他们信任 AI 生成的建议。
+- 他们是否愿意为机会卡、周报或决策工作台付费。
+
+### 3. 礼宾式产品机会雷达
+
+为 5-10 个设计合作伙伴交付 done-with-you 版本，并收取真实费用。AI 可以在后台辅助，但必须保留专家审核。
+
+交付物包括：
+
+- 排序后的痛点。
+- 机会卡。
+- 每周创始人简报。
+- 证据层：代表性反馈、来源、时间窗口、置信度和风险等级。
+- 战略意图变体，例如高端成分导向 vs 性价比导向。
+
+### 4. 知识图谱种子
+
+在一个狭窄品类建立深度，沉淀：
+
+- 成分分类。
+- 功效宣称。
+- 质地词汇。
+- 常见投诉。
+- 监管和宣称风险备注。
+- 竞品 SKU 图谱。
+- 品牌定位维度。
+
+## 继续标准
+
+只有全部满足时，才进入阶段 1：
+
+- 核心雷达有可重复、法律上可防御的数据路径。
+- PIPL 与私有数据处理要求清楚到足以影响产品架构和销售承诺。
+- 至少 5 个付费或半付费设计合作伙伴参与。
+- 至少 70% 受访品牌确认产品机会发现是真实痛点。
+- 至少 3 个品牌在合规流程下提供历史发布、反馈或客服数据用于测试。
+- 至少 5 个品牌的创始人或产品负责人基于输出改变、排序或认真讨论了真实决策。
+- 付费意愿清晰，并且价格能支持预期销售和 onboarding 动作。
+
+## 停止或转向标准
+
+出现以下情况应暂停或转向：
+
+- 核心外部数据无法合法、稳定或可负担地获取。
+- 客户拒绝轻量私有数据连接，导致旗舰诊断过浅。
+- 输出被视为有趣研究，但不影响决策。
+- 付费意愿无法支撑 CAC 回收期和服务成本。
+- 战略意图过滤器不能改善客户感知质量。
+
+---
+
+# 阶段 1：产品机会雷达与产品创新引擎
+
+## 阶段命题
+
+构建一个能自负盈亏的产品。
+
+阶段 1 将经过验证的礼宾式流程产品化为聚焦的 SaaS 加服务楔子。它主要服务于信任阶梯 1-2 级：洞察和推荐。
+
+## 时间盒
+
+第 2-12 个月可作为规划估计。是否进入下一阶段取决于下方证据门槛。
+
+## 定位
+
+第一个可销售包使用"产品机会雷达"，阶段 1 的更完整产品使用"产品创新引擎"。"首席增长官"保留为母愿景，而不是第一天的承诺。
+
+核心承诺：
+
+> 我们帮助美妆品牌从竞品反馈、消费者痛点、市场信号和轻量私有数据中，发现新的产品、质地、成分、功效和传播机会。
+
+## 产品形态
+
+先从混合形态开始，再逐步产品化：
+
+- AI 驱动的数据分析。
+- 专家审核的机会卡。
+- 每周机会简报。
+- 轻量网页工作台。
+- 紧急竞品信号提醒。
+- 入门层包含轻量私有数据连接。
+
+不要从重仪表盘 BI 系统开始。核心产品对象是机会卡。
+
+## 核心模块
+
+### 1. 战略配置中心
+
+客户配置：
+
+- 品类。
+- 价格带。
+- 品牌定位。
+- 核心竞品。
+- 产品线。
+- 英雄成分。
+- 关键宣称。
+- 目标消费者。
+- 战略优先级。
+
+### 2. 市场信号雷达
+
+监测：
+
+- 竞品评论变化。
+- 投诉峰值。
+- 成分趋势。
+- 质地和使用反馈。
+- 社交内容主题。
+- 新兴痛点。
+- 电商问答信号。
+
+### 3. 轻量私有数据连接
+
+这应包含在入门体验中，而不只属于专业版。范围要足够小，以保证快速 onboarding：
+
+- CSV 或平台导出上传。
+- 有限客服对话样本。
+- 退款和退货原因样本。
+- 购后评论样本。
+- 基础本品 SKU 反馈。
+
+它使旗舰比较真正成立：
+
+> 我们比竞品 X 弱在哪里？应该先修什么？
+
+### 4. 竞品评论情报
+
+每个客户跟踪 20-50 个竞品 SKU，识别：
+
+- 正在上升的负面反馈。
+- 重复投诉。
+- 质地问题。
+- 功效质疑。
+- 包装问题。
+- 成分担忧。
+- 价格/价值抱怨。
+- 使用困惑。
+
+### 5. 痛点排序
+
+按以下维度排序痛点：
+
+- 频次。
+- 增长率。
+- 严重程度。
+- 与品牌定位的相关性。
+- 竞争空白。
+- 证据质量。
+
+### 6. 机会卡系统
+
+每张卡应包含：
+
+- 机会名称。
+- 消费者痛点。
+- 代表性证据。
+- 信号强度。
+- 竞品弱点。
+- 在连接私有数据时的本品对比。
+- 品牌适配度。
+- 建议产品方向。
+- 建议宣称或传播角度。
+- 推荐下一步行动。
+- 置信度。
+- 风险等级。
+- 决策状态。
+
+### 7. 决策工作台
+
+团队可以：
+
+- 保存、比较、拒绝、排序和归档机会卡。
+- 分配机会。
+- 添加评论。
+- 投票或打分。
+- 导出内部简报。
+- 跟踪机会是否被采纳。
+
+决策状态包括：
+
+- 新建。
+- 评估中。
+- 测试中。
+- 已采纳。
+- 已拒绝。
+- 已归档。
+
+### 8. AI 分析师与可导出简报
+
+支持自然语言问题：
+
+- "竞品 A 最近弱在哪里？"
+- "防晒品类哪些投诉在上升？"
+- "我们比竞品 X 弱在哪里？"
+- "什么机会最符合我们的敏感肌定位？"
+- "哪个机会最适合下一次上新？"
+
+生成：
+
+- 创始人每周总结。
+- 新品概念简报。
+- 产品改进简报。
+- 内容策略桥接简报。
+- 竞品回应简报。
+
+## 暂不构建什么
+
+不要构建：
+
+- 完整 AI agent 编排。
+- 五引擎平台。
+- 自动广告优化。
+- 广泛品类覆盖。
+- 完全自动化决策。
+- 行业数据产品。
+
+## 包装与定价
+
+定价必须与 CAC 回收期和销售动作经济性共同设计，不能只依赖付费意愿访谈。
+
+### 入门版
+
+面向进入楔子的小型和中型品牌。
+
+包含：
+
+- 竞品监测。
+- 痛点排序。
+- 每周机会简报。
+- 有限机会卡。
+- 轻量私有数据连接。
+- 基础本品 vs 竞品诊断。
+
+入门价格必须足以覆盖 onboarding、数据成本，并支持 12 个月内的目标 CAC 回收期。人民币 19,800-29,800 元/年可以作为起始假设，但只有当真实销售动作在这个价格下成立时才能接受。
+
+### 专业版
+
+面向把产品用于周期性产品复盘的品牌。
+
+包含：
+
+- 更多竞品 SKU。
+- 更深的私有数据连接。
+- 产品决策工作台。
+- 月度策略复盘。
+- 更详细证据层。
+- 从决策到结果的反馈闭环。
+
+人民币 59,800-99,800 元/年只有在毛利和回收期能承受必要服务层时才合理。
+
+### 战略共创版
+
+面向需要更深分析和专家参与的品牌。
+
+包含：
+
+- 自定义分类体系。
+- 更多私有数据。
+- 专家审核。
+- 月度战略工作坊。
+- 定制机会报告。
+
+人民币 150,000-300,000 元/年可作为起始假设，但关键门槛是：该包是否为产品创造可重复学习，而不是退化成定制咨询。
+
+## GTM
+
+- 创始人主导销售。
+- 设计合作伙伴推荐。
+- 围绕创始人的增长方法论做内容营销。
+- 面向美妆创始人社区做教育。
+- 只有当价格支撑时，才采用高触点 onboarding。
+
+必须明确埋点：
+
+- 分渠道 CAC。
+- 销售周期。
+- Onboarding 成本。
+- 数据集成成本。
+- 分层毛利。
+- CAC 回收期。
+- 扩张率。
+- 推荐采纳率。
+
+## 继续标准
+
+只有全部满足时，才进入阶段 2：
+
+- 50-80 个付费客户，或客户数更少但留存和价格证据足以支持聚焦扩张。
+- 净收入留存达到或超过团队目标。
+- 入门产品本身的 CAC 回收期小于 12 个月。
+- 阶段 1 独立毛利为正。
+- 60%+ 月活账户率，或另一个与产品复盘相关的明确活跃基准。
+- 40%+ 客户每月至少导出或分享一份简报。
+- 至少 20 个真实业务决策被产品影响，并有记录。
+- 连接私有数据的客户在留存或活跃上明显优于仅使用公开数据的客户。
+
+## 停止或转向标准
+
+出现以下情况应暂停扩张：
+
+- 产品被当作报告消费，而不是用于决策。
+- 入门层无法包含私有数据且不破坏 onboarding 经济性。
+- 毛利依赖不可规模化的专家工作。
+- CAC 回收期需要不现实的销售动作。
+- 客户在信任机会引擎前就主要索要通用 AI 内容。
+
+---
+
+# 阶段 2：深化楔子，并增加一个相邻引擎
+
+## 阶段命题
+
+先用两个引擎证明协同，再宣称多引擎护城河。
+
+本阶段从推荐走向起草和早期编排。只有在产品创新楔子证明留存和决策影响后，才增加一个相邻引擎。
+
+## 时间盒
+
+第 12-24 个月可作为规划估计。是否继续取决于多引擎价值证据。
+
+## 构建顺序
+
+### 1. 先深化产品创新引擎
+
+在增加广度前：
+
+- 将知识图谱扩展到更多美妆子品类。
+- 改善置信度校准。
+- 强化证据可追溯性。
+- 强化宣称风险意识。
+- 改善私有数据反馈闭环。
+- 跟踪市场信号、品牌决策、执行和业务结果之间的关系。
+
+最强的专有资产不是原始数据，而是历史关系：
+
+> 市场信号 -> 品牌决策 -> 执行 -> 业务结果。
+
+### 2. 将内容激活作为第一个相邻引擎
+
+内容是自然的第二个引擎，因为产品洞察可以直接转化为内容角度。
+
+例如，当机会是"消费者抱怨防晒黏腻厚重"时，系统可以起草：
+
+- 小红书笔记角度。
+- 抖音短视频脚本。
+- 直播卖点。
+- 商品详情页文案。
+- 达人 brief。
+- 创始人解释脚本。
+- 对比宣称。
+- FAQ 和异议处理文案。
+
+### 3. 构建产品到内容工作流
+
+每张机会卡可以转化为：
+
+- 产品概念。
+- 卖点。
+- 内容活动。
+- 达人 brief。
+- 上市传播信息。
+
+跟踪哪些草稿被使用，以及表现如何。这是第一次真正的编排证明：洞察 -> 推荐 -> 起草 -> 执行反馈。
+
+### 4. 增加宣称风险检查
+
+帮助品牌识别高风险、夸大、不受支持或不合规的宣称。这能保护信任，也能让产品区别于通用 AI 内容生成工具。
+
+## 定价与扩张
+
+- 将从产品创新到内容激活的转化作为可衡量的队列变量，而不是假设。
+- 避免 25-40 倍的价格悬崖。
+- 基于实际协同价值、实施成本和回收期影响来定价套餐。
+- 跟踪多引擎账户是否比单引擎账户有更好留存、更高 ARPA 或更短回收期。
+
+## 多引擎证明门
+
+在至少一项被客户数据证明前，不要投入工程资源开发下一个引擎：
+
+- 多引擎客户留存显着好于单引擎客户。
+- 多引擎客户带来明显更高 ARPA，且 CAC 回收期没有恶化。
+- 内容激活提高了产品创新的使用频率或质量。
+- 产品到内容工作流产出的内容被采纳，且区别于通用 AI 内容工具。
+- 扩张销售可以重复，而不是依赖定制咨询。
+
+## 仍然不构建什么
+
+仍不要构建完整广告自动化。广告优化是另一个市场，既有更强的竞争者，也有更高复杂度和更明确的效果责任。
+
+除非证明门显示广度正在改善经济性，否则不要构建用户运营、全链路运营或行业情报。
+
+## 继续标准
+
+只有全部满足时，才进入阶段 3：
+
+- 内容激活 attach rate 达到团队目标，30%+ 可作为初始基准。
+- 50%+ 已激活客户每月导出或使用内容 brief。
+- 多引擎客户在留存或扩张上明显优于可比单引擎客户。
+- 多引擎 ARPA 或 CAC 回收期变好，而不是只带来更高收入和更重服务负担。
+- 跨引擎推荐被客户采纳。
+- 公司在美妆垂直领域拥有可防御的参考客户基础。
+
+## 停止或转向标准
+
+出现以下情况应暂停进一步引擎扩张：
+
+- 内容激活表现得像通用 AI 文案工具。
+- 内容用户没有比楔子用户更高留存或扩张。
+- 第二引擎分散了产品创新留存的注意力。
+- 每次扩张销售都需要定制工作流设计。
+- 宣称风险引发责任或信任问题。
+
+---
+
+# 阶段 3：多引擎增长系统
+
+## 阶段命题
+
+只有当广度改善留存、ARPA 或回收期时，才继续扩张。
+
+阶段 3 走向跨增长职能的编排，但每增加一个引擎都必须通过自己的价值门槛。
+
+## 时间盒
+
+第 24-36 个月及以后可作为规划估计。不能仅因时间到了就进入。
+
+## 进入条件
+
+必须全部成立：
+
+- 产品创新有强留存。
+- 内容激活有有意义的 attach rate 和可衡量协同。
+- 客户把系统用于真实决策，而不仅是读报告。
+- 私有数据集成已经跑通。
+- 公司有足够实施能力。
+- 下一个引擎有明确客户拉力。
+- 多引擎使用已经改善留存、ARPA 或回收期。
+
+## 候选引擎扩张顺序
+
+下一引擎应根据客户拉力和经济证据选择。
+
+### 1. 用户反馈与留存引擎
+
+使用购后、社群、私域和 CRM 反馈识别：
+
+- 复购驱动因素。
+- 不满意点。
+- 流失信号。
+- 产品改进任务。
+- 分人群异议。
+
+### 2. 广告学习引擎
+
+先做学习和诊断，不做自主广告优化：
+
+- 胜出信息分析。
+- 创意角度诊断。
+- 广告评论挖掘。
+- 落地页异议分析。
+- 信息到产品的反馈。
+
+### 3. 全链路运营引擎
+
+只在后期、且客户数据支持时构建：
+
+- 退货原因分析。
+- 客服问题聚类。
+- 物流或区域异常检测。
+- 产品质量反馈闭环。
+- 服务到产品改进任务。
+
+## 编排层
+
+编排器应在同一客户使用多个引擎之后自然出现。它的职责是连接工作流：
+
+- 产品机会变成内容 brief。
+- 内容表现变成产品洞察。
+- 客户投诉变成产品改进任务。
+- 广告异议变成落地页或产品信息改进。
+- 退货原因变成产品或服务修复。
+
+本阶段主要服务于信任阶梯第 4 级：有人参与的编排。
+
+## 继续标准
+
+只有满足以下条件，才走向完整平台：
+
+- 25%+ 客户使用至少两个引擎，或另一个明确阈值达成且经济性更强。
+- 多引擎客户留存显着优于单引擎客户。
+- 多引擎客户产生更高 ARPA，且 CAC 回收期没有变差。
+- 跨引擎推荐被采纳。
+- 系统能协调工作流，而不是靠服务团队手动拼接。
+
+## 停止或转向标准
+
+出现以下情况应暂停平台扩张：
+
+- 广度带来的实施成本增长快于收入。
+- 多引擎采用是销售驱动，但真实使用很浅。
+- 客户不信任跨引擎推荐。
+- 团队无法在多个引擎上维持准确性和证据质量。
+
+---
+
+# 阶段 4：首席增长官平台与可选第二曲线
+
+## 阶段命题
+
+先成为美妆增长决策的记录系统，再决定是否扩展边界。
+
+只有在公司已经赢得美妆防御性滩头阵地，并证明多引擎经济性后，本阶段才可信。
+
+## 时间盒
+
+36 个月及以后。
+
+## 平台能力
+
+- 在合规前提下统一天猫、京东、抖音、小红书、私域和线下来源。
+- 记录市场信号、推荐、草稿、执行和结果的完整决策历史。
+- 在批准护栏内运行更高自主度工作流。
+- 对受限任务采用人工监督的运行模式。
+- 覆盖产品创新、内容、广告学习、用户运营、客服智能和产品反馈闭环的增长决策基础设施。
+
+此时，信任阶梯第 5 级才变得合理：在受限工作流内运营。
+
+## 行业情报谨慎原则
+
+跨品牌匿名化或聚合情报可能成为第二收入曲线，客户包括：
+
+- 原料供应商。
+- OEM/ODM 厂商。
+- 投资机构。
+- 大型消费品集团。
+- 零售渠道。
+
+但它必须被视为高风险后期选项，而不是既定资产。它需要：
+
+- 客户明确 opt-in。
+- 监管审查和许可。
+- PIPL 安全的聚合与匿名化。
+- 合同授权。
+- 与客户机密战略清晰隔离。
+- 渠道冲突分析。
+- 对现有品牌客户进行信任测试。
+
+如果它削弱核心品牌产品的信任或私有数据锁定，就不要推进。
+
+## 可能的第二垂直
+
+只有在以下条件成立时，才考虑第二行业：
+
+- 美妆打法可重复，且不依赖创始人个人经验。
+- 知识图谱方法可迁移。
+- 数据获取在法律和经济上可行。
+- 扩张不会拖慢美妆楔子。
+
+---
+
+# 推荐路线图摘要
+
+## 阶段 0：数据门与礼宾式验证
+
+证明合法数据获取、PIPL 就绪、知识图谱来源、ICP 痛点、付费设计合作伙伴需求和战略意图过滤器。
+
+## 阶段 1：产品机会雷达与产品创新引擎
+
+发布聚焦楔子：机会卡、证据层、决策工作台、AI 分析师、每周简报，以及入门层轻量私有数据连接。定价基于 CAC 回收期和销售动作经济性。
+
+## 阶段 2：深化楔子，加内容激活
+
+先深化楔子，再将内容激活作为唯一相邻引擎。证明洞察 -> 推荐 -> 起草 -> 反馈，并要求可衡量的多引擎协同后才投入更多引擎。
+
+## 阶段 3：多引擎增长系统
+
+只有当客户数据证明更好留存、ARPA 或回收期时，才一次增加一个引擎。编排应来自真实重复工作流，而不是平台野心。
+
+## 阶段 4：完整 CGO 平台与可选行业情报
+
+成为美妆增长决策的记录系统。只有在明确授权、监管许可，以及不会伤害信任的证据存在时，才考虑行业数据或第二垂直。
+
+---
+
+# 关键战略选择
+
+公司不应先用这句话取胜：
+
+> 我们是一个什么都能做的 AI 首席增长官。
+
+而应先证明：
+
+> 我们能帮助美妆品牌从真实消费者、竞品和私有数据信号中发现更好的产品机会，并解释清楚为什么每条建议值得行动。
+
+当这个楔子变得可信、有留存且经济性成立后，更大的首席增长官愿景才可信。

roadmap/product-roadmap-unified.md

@@ -0,0 +1,849 @@
+# Chief Growth Officer - Unified Product Roadmap
+
+_A consolidated roadmap for the Chief Growth Officer product. The core
+principle is evidence before expansion: each phase advances only when the
+company has proven the risk that phase exists to retire._
+
+---
+
+## Reader and Post-Read Action
+
+This roadmap is for founders, product leaders, growth leaders, and investors
+deciding what the Chief Growth Officer product should prove and build next.
+After reading it, the team should be able to sequence work, define phase gates,
+price the early product, and reject premature platform expansion.
+
+---
+
+## Roadmap Philosophy
+
+Do not start by building a full "AI Chief Growth Officer." The long-term vision
+is credible only if the first product is narrow, trusted, legally durable, and
+commercially provable.
+
+The product should evolve through three layers:
+
+1. **Product Opportunity Radar** - find high-quality product and messaging
+   opportunities from market signals.
+2. **Decision Intelligence System** - help brands decide what to build, change,
+   test, or communicate.
+3. **Growth Orchestration Platform** - coordinate product, content,
+   advertising, user operation, and service workflows.
+
+The first 12 months should prove one question:
+
+> Can we reliably help beauty and personal-care brands discover actionable
+> product opportunities earlier and more accurately than they can on their own,
+> with a legally durable data pipeline and viable unit economics?
+
+If yes, the broader platform can grow naturally from the wedge. If no, adding
+more engines will compound risk instead of creating a moat.
+
+---
+
+## Guiding Principles
+
+1. **Sequence by risk, not by feature.** The risks that can kill the company are
+   data access, customer trust, willingness to pay at a workable price, and
+   whether private-data connection can be made easy enough for the entry
+   product.
+2. **Make Phase 0 a hard data gate.** Do not assume access to competitor
+   reviews, social comments, platform signals, knowledge-graph sources, or
+   private customer data. Prove PIPL readiness, data licensing, API feasibility,
+   and off-limits areas before productizing the wedge.
+3. **The wedge must stand alone.** The Product Innovation Engine has to work as
+   a gross-margin-positive business on its own. It cannot depend on a future
+   multi-engine upsell to justify its economics.
+4. **Trust is earned in steps.** Start as decision support. Move toward
+   orchestration and operation only after customers repeatedly act on the
+   system's output.
+5. **Depth before breadth.** Build depth in one beauty sub-category, then one
+   adjacent engine. Do not fund a five-engine platform until there is proof that
+   multi-engine usage improves retention, ARPA, or CAC payback.
+6. **Private-data lock-in starts at entry.** The hero question, "Where are we
+   weaker than competitor X?", requires customer data. The entry tier must
+   include a light private-data connection so the flagship experience is real
+   where most customers land.
+7. **Industry intelligence is optional and risky.** Anonymized or aggregated
+   cross-brand intelligence is a later-stage revenue option that requires
+   explicit opt-in, regulatory clearance, and careful management of channel
+   conflict. It is not an assumed asset.
+
+---
+
+## Trust Ladder
+
+The trust ladder runs across all phases. Pricing power and the "Chief Growth
+Officer" positioning are earned by climbing it, not asserted at launch.
+
+| Rung | Product role | What the AI does | Primary phases |
+| --- | --- | --- | --- |
+| 1 | Insight | Surfaces signals, ranks pain points, flags anomalies | Phase 0-1 |
+| 2 | Recommendation | Proposes specific actions with reasoning, evidence, and confidence | Phase 1 |
+| 3 | Draft | Produces ready-to-use artifacts for human approval | Phase 2 |
+| 4 | Orchestration | Coordinates multi-step workflows across functions with human-in-the-loop control | Phase 2-3 |
+| 5 | Operation | Executes bounded workflows inside approved guardrails with human-on-the-loop oversight | Phase 3+ |
+
+Every product decision should state which rung it serves. Features that imply a
+higher rung than the evidence supports should be deferred.
+
+---
+
+## Initial ICP
+
+Start with Chinese beauty, skincare, personal-care, or haircare brands with
+annual GMV between RMB 30 million and RMB 300 million, selling mainly through
+Tmall, Douyin, Xiaohongshu, JD, or private channels.
+
+Prioritize brands with at least three of these traits:
+
+- Frequent product launches or SKU iteration.
+- Heavy reliance on ingredients, efficacy, texture, or functional claims.
+- Founder or product lead directly involved in product decisions.
+- Strong competitor pressure.
+- Existing pain around identifying new product opportunities.
+- Customer feedback scattered across multiple platforms.
+- Willingness to connect lightweight private data for better diagnosis.
+
+---
+
+# Phase 0: Data Gate, ICP, and Problem Validation
+
+## Phase Thesis
+
+Earn the right to build the Product Opportunity Radar.
+
+This phase replaces any assumed data access with a hard feasibility gate. The
+team should spend as little engineering effort as possible while proving the
+business-critical assumptions.
+
+## Timebox
+
+0-8 weeks is a useful planning estimate, but transition is evidence-based, not
+timeline-based.
+
+## What to Prove
+
+- Competitor review, e-commerce Q&A, social content, and social comments can be
+  obtained legally and durably through official APIs, licensed providers, direct
+  partnerships, customer-authorized exports, or other compliant channels.
+- PIPL and data-handling requirements are understood for private-data ingestion,
+  including customer service conversations, refund reasons, SOV data, CRM tags,
+  and sales/loss data.
+- A narrow beauty knowledge graph can be seeded from reliable sources in one
+  sub-category, such as skincare serums, sunscreen, or haircare repair.
+- The target ICP has urgent product-opportunity pain and will pay for an
+  expert-reviewed, AI-assisted output before a full SaaS product exists.
+- The strategic-intent filter changes outputs in a way customers notice and
+  value.
+
+## Workstreams
+
+### 1. Data Access and Compliance Investigation
+
+Map each core data source:
+
+- Competitor product reviews.
+- E-commerce Q&A.
+- Xiaohongshu posts and comments where legally and technically available.
+- Douyin product and content signals where legally and technically available.
+- Public social mentions.
+- Customer-provided service conversations, refund reasons, surveys, CRM tags,
+  SKU sales, and regional return data.
+
+For each source, document:
+
+- Legal basis and consent model.
+- API or provider feasibility.
+- Rate limits, durability, and cost.
+- Whether scraping is prohibited or too fragile.
+- Data retention requirements.
+- PIPL obligations.
+- Whether the source is allowed in product, only allowed in concierge analysis,
+  or off-limits.
+
+### 2. ICP and Problem Validation
+
+Run 20-30 customer interviews with founders, product leads, and growth leads.
+Validate:
+
+- How they currently find product opportunities.
+- Which product decisions are expensive when wrong.
+- What data they already trust.
+- How often they review competitor reviews, social content, and customer service
+  conversations.
+- What would make them trust an AI-generated recommendation.
+- Whether they would pay for opportunity cards, weekly briefings, or a decision
+  workspace.
+
+### 3. Concierge Product Opportunity Radar
+
+Deliver a done-with-you version to 5-10 design partners. Charge real money.
+Use AI behind the scenes, but keep expert review in the loop.
+
+Deliverables:
+
+- Ranked pain points.
+- Opportunity cards.
+- Weekly founder briefing.
+- Evidence layer with representative feedback, source, time window, confidence,
+  and risk level.
+- Strategic-intent variants, such as premium ingredient-led vs value-led
+  positioning.
+
+### 4. Knowledge Graph Seed
+
+Build depth in one narrow category. Capture:
+
+- Ingredient taxonomy.
+- Efficacy claims.
+- Texture vocabulary.
+- Common complaints.
+- Regulatory and claim-risk notes.
+- Competitor SKU map.
+- Brand-positioning dimensions.
+
+## Continue Criteria
+
+Advance to Phase 1 only if all are true:
+
+- A repeatable, legally defensible data path exists for the core radar.
+- PIPL and private-data handling requirements are documented clearly enough to
+  shape product architecture and sales promises.
+- At least 5 paying or semi-paid design partners participate.
+- At least 70% of interviewed brands confirm product opportunity discovery as a
+  real pain.
+- At least 3 brands provide historical launch, feedback, or customer-service data
+  for testing under a compliant process.
+- Founders or product leads at 5+ brands change, prioritize, or seriously
+  discuss a real decision based on the output.
+- There is clear willingness to pay at a price that can support the expected
+  sales and onboarding motion.
+
+## Stop or Pivot Criteria
+
+Pause or pivot if:
+
+- Core external data cannot be acquired legally, durably, or affordably.
+- Customers refuse even lightweight private-data connection, making the hero
+  diagnosis too shallow.
+- The output is treated as interesting research but does not affect decisions.
+- Willingness to pay cannot support CAC payback and service cost.
+- The strategic-intent filter does not improve perceived quality.
+
+---
+
+# Phase 1: Product Opportunity Radar and Product Innovation Engine
+
+## Phase Thesis
+
+Build one product that pays for itself.
+
+Phase 1 productizes the validated concierge workflow into a focused SaaS-plus-
+service wedge. It should operate at trust ladder rungs 1-2: insight and
+recommendation.
+
+## Timebox
+
+Months 2-12 as a planning estimate. Transition depends on the evidence gates
+below.
+
+## Positioning
+
+Use "Product Opportunity Radar" for the first sellable package and "Product
+Innovation Engine" as the broader Phase 1 product. Keep "Chief Growth Officer"
+as the parent vision, not the day-one promise.
+
+Core promise:
+
+> We help beauty brands discover new product, texture, ingredient, efficacy, and
+> messaging opportunities from competitor feedback, consumer pain points, market
+> signals, and lightweight private data.
+
+## Product Form
+
+Start hybrid, then progressively productize:
+
+- AI-powered data analysis.
+- Expert-reviewed opportunity cards.
+- Weekly opportunity briefing.
+- Lightweight web workspace.
+- Push alerts for urgent competitor signals.
+- Light private-data connection in the entry tier.
+
+Do not start with a dashboard-heavy BI system. The core product object is the
+opportunity card.
+
+## Core Modules
+
+### 1. Strategic Configuration Center
+
+Customers configure:
+
+- Category.
+- Price band.
+- Brand positioning.
+- Core competitors.
+- Product lines.
+- Hero ingredients.
+- Key claims.
+- Target consumers.
+- Strategic priorities.
+
+### 2. Market Signal Radar
+
+Monitor:
+
+- Competitor review changes.
+- Complaint spikes.
+- Ingredient trends.
+- Texture and usage feedback.
+- Social content themes.
+- Emerging pain points.
+- E-commerce Q&A signals.
+
+### 3. Light Private-Data Connection
+
+Include this in the entry experience, not only the professional tier. Keep scope
+small enough for fast onboarding:
+
+- A CSV or platform export upload.
+- A limited customer-service conversation sample.
+- Refund/return reason sample.
+- Post-purchase review sample.
+- Basic own-brand SKU feedback.
+
+This enables the hero comparison:
+
+> Where are we weaker than competitor X, and what should we fix first?
+
+### 4. Competitor Review Intelligence
+
+Track 20-50 competitor SKUs per customer. Identify:
+
+- Rising negative feedback.
+- Repeated complaints.
+- Texture issues.
+- Efficacy doubts.
+- Packaging problems.
+- Ingredient concerns.
+- Price/value complaints.
+- Usage confusion.
+
+### 5. Pain Point Ranking
+
+Rank pain points by:
+
+- Frequency.
+- Growth rate.
+- Severity.
+- Relevance to brand positioning.
+- Competitive whitespace.
+- Evidence quality.
+
+### 6. Opportunity Card System
+
+Each card should include:
+
+- Opportunity name.
+- Consumer pain point.
+- Representative evidence.
+- Signal strength.
+- Competitor weakness.
+- Own-brand comparison where private data is connected.
+- Brand fit.
+- Suggested product direction.
+- Suggested claim or messaging angle.
+- Recommended next action.
+- Confidence level.
+- Risk level.
+- Decision status.
+
+### 7. Decision Workspace
+
+Teams can:
+
+- Save, compare, reject, prioritize, and archive cards.
+- Assign opportunities.
+- Add comments.
+- Vote or score.
+- Export internal briefs.
+- Track whether an opportunity was adopted.
+
+Decision statuses:
+
+- New.
+- Under review.
+- Testing.
+- Adopted.
+- Rejected.
+- Archived.
+
+### 8. AI Analyst and Exportable Briefs
+
+Support natural-language questions:
+
+- "Where is Competitor A weakest recently?"
+- "What complaints are rising in sunscreen?"
+- "Where are we weaker than competitor X?"
+- "What product opportunity fits our sensitive-skin positioning?"
+- "Which opportunity is most suitable for our next launch?"
+
+Generate:
+
+- Founder weekly summary.
+- Product concept brief.
+- Product improvement brief.
+- Content strategy bridge.
+- Competitor response brief.
+
+## What Not to Build Yet
+
+Do not build:
+
+- Full AI agent orchestration.
+- Five-engine platform.
+- Automated ad optimization.
+- Broad category coverage.
+- Fully automated decision-making.
+- Industry data products.
+
+## Packaging and Pricing
+
+Pricing must be co-designed with CAC payback and sales motion economics, not
+only willingness-to-pay interviews.
+
+### Entry
+
+For small and mid-market brands landing in the wedge.
+
+Includes:
+
+- Competitor monitoring.
+- Pain point ranking.
+- Weekly opportunity briefing.
+- Limited opportunity cards.
+- Light private-data connection.
+- Basic own-brand vs competitor diagnosis.
+
+The entry price should be high enough to support onboarding, data costs, and a
+target CAC payback under 12 months. RMB 19,800-29,800/year may be a starting
+hypothesis, but it should not be accepted unless the actual motion works at that
+price.
+
+### Professional
+
+For brands using the product in recurring product reviews.
+
+Includes:
+
+- More competitor SKUs.
+- Deeper private-data connection.
+- Product decision workspace.
+- Monthly strategy review.
+- More detailed evidence layer.
+- Feedback loop from decision to outcome.
+
+RMB 59,800-99,800/year is plausible only if gross margin and payback survive the
+required service layer.
+
+### Strategic Co-Creation
+
+For brands that want deeper analysis and expert involvement.
+
+Includes:
+
+- Custom taxonomy.
+- More private data.
+- Expert review.
+- Monthly strategic workshops.
+- Custom opportunity reports.
+
+RMB 150,000-300,000/year is a useful starting hypothesis, but the main gate is
+whether the package creates repeatable learning for the product instead of
+becoming bespoke consulting.
+
+## GTM
+
+- Founder-led sales.
+- Design-partner referrals.
+- Content marketing built from the founder's growth methodology.
+- Beauty-founder community education.
+- High-touch onboarding only where pricing supports it.
+
+Explicitly instrument:
+
+- CAC by channel.
+- Sales cycle length.
+- Onboarding cost.
+- Data integration cost.
+- Gross margin by tier.
+- CAC payback.
+- Expansion rate.
+- Recommendation action rate.
+
+## Continue Criteria
+
+Advance to Phase 2 only if all are true:
+
+- 50-80 paying customers, or a smaller number with clear retention and pricing
+  evidence sufficient to fund focused expansion.
+- Net revenue retention at or above the team's target.
+- CAC payback under 12 months on the entry product alone.
+- Positive gross margin for Phase 1 in isolation.
+- 60%+ monthly active account rate or another explicit activity benchmark tied
+  to product reviews.
+- 40%+ of customers export or share at least one brief per month.
+- Documented evidence that at least 20 real business decisions were influenced
+  by the product.
+- Customers with private-data connection retain or engage materially better than
+  public-data-only customers.
+
+## Stop or Pivot Criteria
+
+Pause expansion if:
+
+- The product is consumed as a report but not used in decisions.
+- The entry tier cannot include private data without breaking onboarding
+  economics.
+- Gross margin depends on unscalable expert work.
+- CAC payback requires an unrealistic sales motion.
+- Customers ask for generic AI content before trusting the opportunity engine.
+
+---
+
+# Phase 2: Depth Plus One Adjacent Engine
+
+## Phase Thesis
+
+Prove synergy with two engines before claiming a multi-engine moat.
+
+This phase climbs from recommendation to draft and early orchestration. It adds
+one adjacent engine only after the product innovation wedge has proven retention
+and decision impact.
+
+## Timebox
+
+Months 12-24 as a planning estimate. Advancement is gated by proof of
+multi-engine value.
+
+## Build Order
+
+### 1. Deepen the Product Innovation Engine
+
+Before adding breadth:
+
+- Expand the knowledge graph to more beauty sub-categories.
+- Improve confidence calibration.
+- Improve evidence traceability.
+- Strengthen claim-risk awareness.
+- Improve private-data feedback loops.
+- Track the relationship between market signal, brand decision, execution, and
+  business result.
+
+The strongest proprietary asset is not raw data. It is the historical
+relationship:
+
+> Market signal -> brand decision -> execution -> business result.
+
+### 2. Add Content Activation as the First Adjacent Engine
+
+Content is the natural second engine because product insights become content
+angles.
+
+For an opportunity such as "consumers complain sticky sunscreen feels heavy,"
+the system can draft:
+
+- Xiaohongshu post angles.
+- Douyin short video scripts.
+- Live-stream selling points.
+- Product page copy.
+- Influencer briefs.
+- Founder explanation scripts.
+- Comparison claims.
+- FAQ and objection-handling copy.
+
+### 3. Build the Product-to-Content Workflow
+
+Each opportunity card can become:
+
+- Product concept.
+- Selling point.
+- Content campaign.
+- Influencer brief.
+- Launch message.
+
+Track which drafts are used and how they perform. This is the first real
+orchestration proof: insight -> recommendation -> draft -> execution feedback.
+
+### 4. Add Claim Risk Check
+
+Help brands identify risky, exaggerated, unsupported, or non-compliant claims.
+This protects trust and reinforces the difference between strategic AI and
+generic content generation.
+
+## Pricing and Expansion
+
+- Treat conversion from Product Innovation to Content Activation as a measured
+  cohort variable, not an assumption.
+- Avoid a 25-40x price cliff between tiers.
+- Price bundles based on observed synergy value, implementation cost, and
+  payback impact.
+- Track whether multi-engine accounts have better retention, higher ARPA, or
+  shorter payback than single-engine accounts.
+
+## Multi-Engine Proof Gate
+
+Do not commit engineering resources to the next engine until at least one of
+these is proven with customer data:
+
+- Multi-engine customers retain materially better than single-engine customers.
+- Multi-engine customers produce meaningfully higher ARPA without worse payback.
+- Content Activation improves the frequency or quality of Product Innovation
+  usage.
+- Product-to-content workflows create adopted outputs that generic AI content
+  tools do not.
+- Expansion sales are repeatable without bespoke consulting.
+
+## What Still Not to Build
+
+Do not build full ad automation yet. Advertising optimization is a separate
+market with stronger incumbents, higher complexity, and clearer performance
+accountability.
+
+Do not build user operations, full-chain operations, or industry intelligence
+unless the proof gate says breadth is improving economics.
+
+## Continue Criteria
+
+Advance to Phase 3 only if all are true:
+
+- Content Activation attach rate reaches the team's target, with 30%+ as an
+  initial benchmark.
+- 50%+ of activated customers export or use content briefs monthly.
+- Multi-engine customers materially out-retain or out-expand comparable
+  single-engine customers.
+- Multi-engine ARPA or CAC payback is better, not merely larger revenue with
+  higher service burden.
+- Cross-engine recommendations are adopted by customers.
+- The company has a defensible reference base in beauty.
+
+## Stop or Pivot Criteria
+
+Pause further engine expansion if:
+
+- Content Activation behaves like a generic AI copy tool.
+- Content users do not retain better or expand more than wedge-only users.
+- The second engine distracts from Product Innovation retention.
+- Each expansion sale requires bespoke workflow design.
+- Claim-risk concerns create liability or trust issues.
+
+---
+
+# Phase 3: Multi-Engine Growth System
+
+## Phase Thesis
+
+Expand only where breadth improves retention, ARPA, or payback.
+
+Phase 3 moves toward orchestration across multiple growth functions, but each
+additional engine must pass its own value gate.
+
+## Timebox
+
+Months 24-36+ as a planning estimate. Do not enter on timeline alone.
+
+## Conditions Before Entering
+
+All must be true:
+
+- Product Innovation has strong retention.
+- Content Activation has meaningful attach rate and measurable synergy.
+- Customers use the system for real decisions, not just reports.
+- Private-data integration is working.
+- The company has enough implementation capacity.
+- There is clear customer pull for the next engine.
+- Multi-engine usage has improved retention, ARPA, or payback.
+
+## Candidate Engine Expansion Order
+
+Choose the next engine based on customer pull and economic evidence.
+
+### 1. User Feedback and Retention Engine
+
+Use post-purchase, community, private-domain, and CRM feedback to identify:
+
+- Repeat-purchase drivers.
+- Dissatisfaction points.
+- Churn signals.
+- Product improvement tasks.
+- Segment-specific objections.
+
+### 2. Advertising Learning Engine
+
+Start with learning and diagnosis, not autonomous ad optimization:
+
+- Winning message analysis.
+- Creative angle diagnosis.
+- Ad comment mining.
+- Landing-page objection analysis.
+- Message-to-product feedback.
+
+### 3. Full-Chain Operations Engine
+
+Only later, and only where customer data supports it:
+
+- Return reason analysis.
+- Customer service issue clustering.
+- Delivery or regional anomaly detection.
+- Product quality feedback loops.
+- Service-to-product improvement tasks.
+
+## Orchestration Layer
+
+The orchestrator should emerge after multiple engines are used by the same
+customers. Its job is to connect workflows:
+
+- Product opportunities become content briefs.
+- Content performance becomes product insight.
+- Customer complaints become product improvement tasks.
+- Ad objections become landing-page or product messaging improvements.
+- Return reasons become product or service fixes.
+
+This phase operates mainly at trust ladder rung 4: orchestration with
+human-in-the-loop control.
+
+## Continue Criteria
+
+Advance toward the full platform only if:
+
+- 25%+ of customers use at least two engines, or another explicit threshold is
+  met with stronger economics.
+- Multi-engine customers retain materially better than single-engine customers.
+- Multi-engine customers generate higher ARPA without worse CAC payback.
+- Cross-engine recommendations are adopted.
+- The system can coordinate workflows without a services team manually stitching
+  everything together.
+
+## Stop or Pivot Criteria
+
+Pause platform expansion if:
+
+- Breadth increases implementation cost faster than revenue.
+- Multi-engine adoption is sales-led but usage is shallow.
+- Customers do not trust cross-engine recommendations.
+- The team cannot maintain accuracy and evidence quality across engines.
+
+---
+
+# Phase 4: Chief Growth Officer Platform and Optional Second Curves
+
+## Phase Thesis
+
+Become the system of record for growth decisions in beauty, then decide whether
+to expand the surface.
+
+This phase is credible only after the company has earned a defensible beachhead
+in beauty and proven multi-engine economics.
+
+## Timebox
+
+36 months and beyond.
+
+## Platform Capabilities
+
+- Cross-platform data unification across Tmall, JD, Douyin, Xiaohongshu, private
+  channels, and offline sources where compliant.
+- Full decision history across market signal, recommendation, draft, execution,
+  and outcome.
+- Higher-autonomy workflows under approved guardrails.
+- Human-on-the-loop operation for bounded tasks.
+- Broader growth decision infrastructure across product innovation, content,
+  advertising learning, user operation, customer service intelligence, and
+  product feedback loops.
+
+This is where trust ladder rung 5 becomes plausible: operation inside carefully
+bounded workflows.
+
+## Industry Intelligence Caution
+
+Cross-brand anonymized or aggregated intelligence may become a second revenue
+curve for:
+
+- Ingredient suppliers.
+- OEM/ODM manufacturers.
+- Investment firms.
+- Large consumer groups.
+- Retail channels.
+
+Treat this as a risky later-stage option, not a planned asset. It requires:
+
+- Explicit customer opt-in.
+- Regulatory review and clearance.
+- PIPL-safe aggregation and anonymization.
+- Contractual permission.
+- Clear separation from customer-confidential strategy.
+- Channel-conflict analysis.
+- Trust testing with existing brand customers.
+
+Do not pursue it if it weakens the core brand product's trust or private-data
+lock-in.
+
+## Possible Second Vertical
+
+A second industry is an option only if:
+
+- The beauty playbook is repeatable without founder-dependent expertise.
+- The knowledge-graph approach transfers.
+- Data access is legally and economically feasible.
+- Expansion does not slow the beauty wedge.
+
+---
+
+# Recommended Roadmap Summary
+
+## Phase 0: Data Gate and Concierge Validation
+
+Prove legal data access, PIPL readiness, knowledge-graph sourcing, ICP pain,
+paid design-partner demand, and the strategic-intent filter.
+
+## Phase 1: Product Opportunity Radar and Product Innovation Engine
+
+Launch the focused wedge with opportunity cards, evidence, decision workspace,
+AI analyst, weekly briefings, and light private-data connection in the entry
+tier. Price from CAC payback and sales motion economics.
+
+## Phase 2: Depth Plus Content Activation
+
+Deepen the wedge, then add Content Activation as the only adjacent engine.
+Prove insight -> recommendation -> draft -> feedback, and require measured
+multi-engine synergy before funding more engines.
+
+## Phase 3: Multi-Engine Growth System
+
+Add engines one at a time only when customer data proves better retention, ARPA,
+or payback. Build orchestration from real repeated workflows, not platform
+ambition.
+
+## Phase 4: Full CGO Platform and Optional Industry Intelligence
+
+Become the growth decision system of record in beauty. Consider industry data or
+a second vertical only with explicit opt-in, regulatory clearance, and evidence
+that it will not damage trust.
+
+---
+
+# Key Strategic Choice
+
+The company should not try to win by saying:
+
+> We are an AI Chief Growth Officer that does everything.
+
+It should first win by proving:
+
+> We help beauty brands discover better product opportunities from real consumer,
+> competitor, and private-data signals, and we can show why each recommendation
+> is worth acting on.
+
+Once that wedge becomes trusted, retained, and economically sound, the broader
+Chief Growth Officer vision becomes credible.