2026.09.24 | 9K | META | Muse Impact on Meta Ads

音频信息

音频标题: 260924 - Merit Muse对于Meta广告收入的影响

相关标的：undefined

议题：

完整音频链接：

音频纪要总结

Meta广告业务、Muse个人智能体与商业化前景专家访谈纪要

1. Meta广告业务表现与增长驱动

a) Meta Q3广告收入表现如何，未来几个季度的增速预期是多少？

• Q3广告收入表现略高于此前预期：

1. 原预计同比增长28%，目前上调至29%。

1. Q4增速预测由29%上调至30%。

1. Q1最新预计同比增长27%。

• Q1增速回落至27%并不代表业务发生重大变化，主要是由于同比基数升高。

• 若观察两年复合增长趋势，Meta广告业务的基础增速仍在逐季度提升，整体经营表现良好。

b) Meta广告收入的主要增长驱动因素有哪些？

• Reels仍是重要驱动力，但边际贡献正在减弱：

1. 此前对广告收入增速的贡献约为5个百分点。

1. 目前贡献约为4个百分点或略高于4个百分点。

1. 预计明年贡献可能降至约2个百分点。

• 表单、点击和消息类广告继续增长：

1. 越来越多广告主付费引导用户点击广告并进入WhatsApp。

1. 该类广告目前约贡献1个百分点的增速。

• 即时购物广告约贡献1个百分点：

1. 该广告形式不再只是把用户导向外部电商网站。

1. 其增长主要依靠Meta与Shopify、Amazon等电商平台的集成。

• 覆盖式广告，即显示在内容上层的广告，约贡献3至4个百分点。

• WhatsApp Business等可直接衡量的项目也在贡献增长。

• 上述可识别产品和广告位的增长贡献合计约为13至14个百分点，其余增长主要来自：

1. 用户规模增长。

1. Meta在广告主预算中的市场份额提升。

1. 广告价格特别是CPM上涨。

c) Meta广告CPM为何显著加速？

• Meta全平台CPM增速持续加快：

1. Q1同比增长5.3%。

1. Q2同比增长11.6%。

1. Q3预计增长16%至17%。

• CPM上涨部分来自Click to Messenger、购物广告等新广告形式，因为这些广告的商业意图更强，CPM天然更高。

• 更深层的原因是Meta持续使用算法、数据和AI提升：

1. 广告个性化程度。

1. 用户点击率。

1. 购买转化率。

1. 广告主投资回报率。

• Meta的广告系统正在从“持续展示广告”转向“识别购买意图并集中促成交易”：

1. 当系统发现用户对某类商品表现出明显兴趣时，会实时调整推荐内容。

1. 系统可能增加相关广告的展示频次和广告负载。

1. 当用户点击广告或在广告上的停留时间超过普通内容时，算法会进一步转向帮助用户完成购买。

• AI提升了每次广告曝光的商业价值，是广告主愿意支付更高CPM的核心原因。

d) Meta Audience Network是否成为新的增长点？

• Meta Audience Network过去2至4年在样本广告支出中的占比一直稳定在约2.1%，季度波动范围约为2.0%至2.2%。

• 自4月以来，其占比快速升至2.6%。

• 虽然仅提升0.5个百分点，但基本全部属于增量收入。

• 这一变化可能意味着Meta正加强游戏广告及互联网外部流量领域的布局，并对AppLovin形成更直接的竞争。

• 目前该趋势能持续到什么程度仍有待观察。

e) Meta Business Agent的采用情况及广告价值如何？

• Meta Business Agent自6月正式推出，仍处于早期推广阶段。

• 在约300家广告主样本中：

1. 截至5月，约30家参与测试。

1. 到9月，采用数量已增至约60家。

1. 当前产品功能较5月更加丰富。

• 广告主采用Business Agent通常经历三个阶段：

1. 将自身数据接入Business Agent，首先获得运营价值，但不会立即增加广告支出。

1. Meta算法利用相关数据提升广告ROI，推动广告主逐步增加预算。

1. 广告主启用Meta Business Messaging或WhatsApp Business后，获得更多销售转化，并进一步提高预算。

• 该产品有望成为新的广告预算增长驱动因素，但短期贡献仍较有限。

2. Muse的产品定位与市场表现

a) Muse为何能在美国和加拿大取得较高下载量？

• Muse成功的主要原因包括：

1. 产品体验本身较好。

1. 定位与ChatGPT、Claude、Gemini存在明显差异。

1. 能够利用Meta庞大的分发渠道和品牌认知度。

1. 切中了普通消费者尚未充分使用AI管理生活的市场空白。

• 当前大众对AI的使用仍较浅：

1. 约5%至10%的人基本不使用AI。

1. 约60%的人只把通用AI当作增强版搜索引擎。

1. 约20%至25%的人开始构建智能体改善生活。

1. 能够构建智能体工作流并衡量ROI的人通常仅占0%至2%。

• 因此，70%以上消费者尚未使用AI管理电子邮件、日历、账单、税务和收据等日常事务，这为Muse提供了较大市场空间。

b) Muse与现有聊天机器人最核心的差异是什么？

• Muse没有将自己定位为更强的搜索引擎，而是定位为“个人生活管理助手”。

• 其核心场景包括：

1. 管理用户日常生活。

1. 监控电子邮件和日历。

1. 寻找折扣及优惠码。

1. 帮助用户比价购物。

1. 充当私人购物顾问。

• ChatGPT、Claude和Gemini过去更多强调问答、搜索和生产力工具，尚未以如此明确的方式面向普通消费者宣传“生活管理”能力。

• Muse向用户传递的是一次产品形态升级，而不是对现有聊天机器人的简单复制，因此参与的是一场不同的竞争。

c) Muse的核心目标用户是谁？

• Muse明显是一款消费级产品，而非以B2B为主的工具。

• 典型目标用户包括：

1. 主要通过手机处理事务的用户。

1. 需要同时兼顾孩子、工作和家庭的中产阶层母亲。

1. 没有大量时间使用电脑，但存在日程安排、购物、休闲活动等复杂需求的人群。

• Claude受到的直接威胁可能较小，但ChatGPT和Google应对Muse保持警惕。

d) Meta的分发和数据能力将如何帮助Muse？

• Meta可以通过Facebook和Instagram识别用户的实时购买意图。

• 当用户持续浏览或点击某类商品广告时，Meta未来可以自然地提示：

1. 用户是否需要Muse帮助筛选商品。

1. 是否需要Muse提供多个候选选项。

1. 是否需要Muse进行比价、寻找优惠或完成后续交易。

• 这种从“兴趣识别—广告互动—Muse介入—交易完成”的衔接较为自然。

• WhatsApp未来也可能成为Muse的重要入口，但由于涉及私人对话和隐私，相关功能需要更加谨慎。

3. Muse商业化路径与广告收入潜力

a) Meta可能如何推进Muse商业化？

• Muse短期内可能不会以变现为首要目标，Meta更可能优先提升产品实用性和用户规模。

• 参考WhatsApp、Stories和Reels的发展路径：

1. 前6至12个月可能基本不商业化。

1. 随后逐步引入风险较低的品牌广告。

1. 再增加效果广告、交易抽佣和其他商业模式。

1. 整个商业化成熟过程可能持续4至5年。

• Meta并不缺少短期收入，因此有条件先培养用户习惯，再逐步变现。

b) Muse有哪些潜在商业模式？

• 主动式广告

1. Muse可以根据用户日历、邮件和消费需求主动推荐相关商品或服务。

1. 例如，发现用户周六没有为孩子安排活动时，可推荐儿童俱乐部。

1. 发现用户计划聚餐但尚未预订餐厅时，可主动提供餐厅建议。

• 电商GMV抽佣

1. Muse可作为私人购物助手提供不同品牌的候选选项。

1. 当用户完成交易后，Meta可按GMV的一定比例收取佣金。

1. 该模式已被OpenAI和Google采用，市场接受度正在提高。

• 订阅模式

1. Meta已提供免费层级和付费层级。

1. 对于每天可节省0.5至1小时的高频用户，每月收费20至30美元具有可行性。

1. 但消费者长期习惯使用免费互联网产品，因此订阅未必成为最大收入来源。

• 支付服务

1. Meta可参与用户与企业之间的支付和交易。

1. 按每笔交易收取固定费用或一定比例佣金。

• 商家消息收费

1. 品牌已经愿意为WhatsApp上的每条商业消息支付几美分。

1. Muse与WhatsApp融合后，商家端消息收费可能成为重要变现方式。

c) Muse能否提升Facebook和Instagram广告转化率？

• Muse有望帮助Meta获得比过去更深入、更实时的商业意图数据。

• 传统电商网站平均转化率约为3%，即每100名访问者约有3人完成购买。

• 在AI充分了解用户购买记录、价格敏感度、品牌偏好和使用场景后：

1. 广告素材可以完全个性化。

1. 推荐商品和价格可以与用户偏好匹配。

1. 用户只会访问少数高度相关的候选商家。

• 如果AI将候选范围缩小至3个商家，理论转化率上限可接近33%；考虑用户延迟购买、分享链接等因素，实际转化率达到10%具有一定可能性。

• 即使不采用10%的激进假设，AI推动转化率提升至原来的2倍也较容易实现。

• 转化率提升2至3倍后，广告主愿意支付的CPC也可能提升至原来的2至3倍。

d) Muse能否支撑Meta广告收入翻倍甚至增长至3倍？

• 专家认为，从长期看完全存在可能。

• 增长来源包括：

1. 更高的点击率和转化率。

1. 更强的购买意图识别。

1. 更高的CPC和CPM。

1. 主动式广告。

1. 电商GMV抽佣。

1. 支付和商家消息收入。

• 但该增长不会短期兑现，预计需要经历用户培养、产品完善和逐步商业化的多年过程。

4. AI Agent电商模式与产品设计

a) OpenAI早期电商产品为何效果不佳？

• OpenAI第一代电商产品过度简化了购买流程：

1. 用户看到商品后直接进入结账页面。

1. 页面只有“立即购买”和信用卡选择。

1. 缺少配送时间、退货流程、条款、数量调整等必要信息。

• 这种设计让消费者感到过于突然，也无法确认交易责任主体，因此几乎无人点击。

• 其推荐质量也不够准确，例如用户寻找跑鞋时，系统可能推荐偏时尚属性的运动鞋。

• 核心问题包括：

1. 商品定向不准确。

1. 结账设计过于简单。

1. 缺少消费者信任机制。

b) OpenAI和Google目前的电商方案有何改进？

• OpenAI已转向更合理的方案：

1. 将用户引导至商家落地页。

1. 通过类似iframe的方式将商家结账页面嵌入ChatGPT。

1. 让用户确认实际商家、配送和退货责任。

• Google的方案从一开始就明确展示商品对应的商家，流程更顺畅，产品完成度更高。

• Google目前的问题并非产品质量，而是推广速度较慢，现阶段仍更专注于改善Gemini体验。

c) Meta进入Agent电商时应避免哪些问题？

• 不应直接设置缺乏解释的“立即购买”按钮。

• 应保留明确的确认页面，让用户核对：

1. 商品及推荐理由。

1. 真实商家身份。

1. 商品价格。

1. 配送时间。

1. 退货政策。

1. 条款和条件。

• 调查显示：

1. 约65%的美国消费者愿意让聊天机器人帮助研究商品。

1. 只有约14%的消费者愿意在不了解具体情况的情况下，让智能体直接代为购买。

• 因此，AI可以完成研究、筛选和推荐，但最终购买决定仍应由消费者确认。

• Meta与Shopify的整合有助于提高消费者信任，并可支持商品购买、B2B服务预约等场景。

5. AI Agent对电商聚合平台与广告竞价模式的影响

a) AI Agent是否会取代Amazon、Booking等聚合平台？

• 未来5至10年可能是聚合平台面临较大压力的时期。

• 个人Agent将能够访问用户的：

1. 日历和电子邮件。

1. 购买历史。

1. 银行账户。

1. WhatsApp对话。

1. 登录过的网站和浏览记录。

• Agent因此能够更加深入地理解用户偏好，并成为私人购物助手。

• 当Agent筛选出候选餐厅、酒店或商品后，可以：

1. 直接连接终端商家。

1. 通过商家官网完成交易。

1. 在商家网站体验较差时，再使用Booking等聚合平台。

• 聚合平台不会被完全淘汰，但部分流量会被Agent直接分流至终端商家。

• 网站和交易体验优秀的品牌更可能绕过聚合商；中小商家仍可能依赖聚合平台提供标准化交易能力。

b) AI Agent时代传统广告竞价是否仍然有效？

• 传统搜索和聚合平台通常允许大量品牌竞价进入搜索结果首页。

• 个人Agent会先筛除大量不适合用户的商品，因此进入竞价环节的品牌数量将显著减少。

• 未来可能只保留3个最佳候选项，由消费者最终选择。

• 竞价机会仍然存在，且单个候选项的商业价值可能更高：

1. 用户购买意愿可能已达到80%。

1. 进入3个候选项意味着理论成交概率约为27%。

1. 品牌会愿意为这种高意向流量支付更高费用。

• 因此，未来竞价广告可能从“大量低相关广告主竞争”转向“少数高相关品牌竞争”。

c) Agent平台还可以采用哪些收费模式？

• 可按GMV的一定比例收费，并根据行业利润率调整：

1. 毛利率50%的商品，广告主可能愿意支付GMV的20%。

1. 毛利率10%的商品，广告主可能只愿意支付GMV的3%。

• 酒店目前通常向Booking.com支付GMV的15%至20%，AI Agent也可能复制类似模式。

• 最终可能形成“候选项竞价+GMV抽佣”的混合商业模式。

• 即使用户选择价格最低、无法承担高竞价成本的商品，平台仍可通过GMV抽佣获得收入。

6. Muse对Google搜索的影响与竞争格局

a) Muse会对Google搜索造成多大威胁？

• Muse会对Google构成长期威胁，但影响将从小规模和特定品类开始。

• 最先可能被分流的场景包括：

1. 100美元以下的低价商品。

1. 周末休闲活动。

1. 餐厅和儿童俱乐部。

1. 时尚和美妆。

1. 口红、快时尚等轻决策商品。

1. 短途旅行和生活方式服务。

• 相对安全的领域包括：

1. B2B采购。

1. 政府事务。

1. 严肃医疗服务。

1. 教育项目比较，例如MBA课程。

1. 高专业度、长决策周期的需求。

• 如果Muse成功实现规模化，5年后Google收入可能因此下降3%至5%，但该估计仍存在较大不确定性。

b) Google在个人Agent竞争中有哪些优势？

• Google拥有庞大的分发渠道和成熟用户习惯。

• Google可以自然整合：

1. Gmail。

1. Google Calendar。

1. 浏览记录。

1. Google Maps历史数据。

1. 搜索和购物数据。

• Apple目前可能使用Gemini支持Siri，这进一步增强了Google的分发能力。

• Google理论上最具备构建全能型个人助手的数据基础，但其推出独立消费级产品的时间可能晚于Muse。

• Google仍可通过Gemini相关产品快速反击，竞争结果尚未确定。

c) Meta、Google、OpenAI、Claude未来可能如何分工？

• Meta和Muse更可能在以下领域占据优势：

1. 生活方式。

1. 娱乐与休闲。

1. 时尚和美妆。

1. 短途旅行。

1. 游戏。

1. 基础消费者健康。

• Google、ChatGPT和Claude可能继续在以下领域保持优势：

1. B2B。

1. 严肃医疗。

1. 教育。

1. 政府事务。

1. 专业采购和复杂研究。

• 整体来看，个人Agent趋势：

1. 对Google搜索构成阻力。

1. 对OpenAI也可能形成一定竞争压力。

1. 对Meta则是明确利好。

d) Muse对Meta而言是新增业务还是防御性产品？

• Muse不仅是新增收入来源，也是在防御未来社交媒体使用时长下降。

• 美国和欧洲用户的屏幕使用时间已接近上限，数字娱乐主要由社交媒体、视频和游戏占据。

• 与聊天机器人交流、利用生成式AI规划生活，正在成为新的数字娱乐形式。

• 美国用户目前平均每天使用OpenAI约16分钟，其中既包括工作，也包括心理健康、生活规划和休闲用途。

• 如果5年后普通人平均每天使用聊天机器人30分钟，社交媒体、视频和游戏的使用时间可能相应减少。

• Muse有望帮助Meta把一部分用户时间留在自身生态内，其增量价值可能超过聊天机器人对传统社交媒体时长造成的负面影响。

7. Meta支付体系与Apple平台关系

a) Meta支付业务目前进展如何？

• Meta已在全球约3至5个国家推出支付服务，可能包括墨西哥、巴西、印度尼西亚、印度或沙特阿拉伯等市场。

• 目前尚未覆盖美国、英国等主要市场。

• Meta已公开表示计划将支付服务推广至全球。

• 考虑监管、牌照和基础设施建设，全球覆盖可能需要2至3年。

• Meta也可能通过与Visa、Mastercard、Stripe等机构合作缩短推进时间。

• Stripe首席执行官加入Meta董事会，反映出：

1. 支付业务对Meta具有较高战略重要性。

1. Meta未来可能与Stripe开展更深入合作。

b) 如果Muse成为主要入口，Meta能否像Apple一样对交易抽成？

• 如果Muse掌握用户的流量入口和购买决策，Meta理论上可以通过以下方式变现：

1. 广告收费。

1. GMV抽佣。

1. 支付手续费。

1. 订阅收费。

1. 商家消息收费。

• 但Meta仍依赖Apple和Android等终端平台，因此无法完全复制Apple控制iOS生态的模式。

• Apple未来可能通过以下方式从Muse获利：

1. 抽取部分订阅收入。

1. 推广Apple Pay。

1. 对推荐位或预装入口收费。

1. 与Meta协商其他形式的收入分成。

• Meta短期内可能先免费推广Muse，尚不存在明确的商业分成基础。

• 用户习惯形成并开始商业化后，Apple可能要求参与分成。

• 专家判断Apple每年可能从Muse生态获得数十亿美元，但达到200亿美元的可能性较低，因为Muse只是热门应用，而不是类似默认搜索引擎的系统级入口。

• Meta预计仍将保留Muse相关收入的大部分收益。

8. Muse与WhatsApp的融合前景

a) Muse未来可能如何与WhatsApp整合？

• 目前Muse与WhatsApp尚未实现明显整合，但未来存在较大协同空间。

• 可能的整合方式包括：

1. Muse以虚拟朋友或私人助手的形式直接出现在WhatsApp中。

1. 用户无需切换应用，即可通过WhatsApp安排活动、购物或处理日程。

1. Muse既可以作为独立应用，也可以成为WhatsApp内的交互入口。

• 这种整合能够显著降低用户使用门槛，并提升Muse黏性。

b) Muse与WhatsApp Business如何形成商业闭环？

• Muse帮助用户完成筛选后，如果仍需向商家咨询，可以直接将用户转入WhatsApp Business。

• 商家可通过企业员工或AI智能体回答用户问题。

• Meta最终可能同时掌握：

1. 代表消费者的Muse智能体。

1. 代表商家的Business Agent。

1. 用户与商家之间的通信渠道。

• 这意味着Meta可能控制交易对话的双方，并从消息、广告、交易和支付等多个环节变现。

• Meta需要建立严格的防火墙，避免商家追求利润与消费者追求最佳交易之间产生利益冲突。

c) WhatsApp对话中的商业链接有什么价值？

• Meta未来可能把WhatsApp对话中提到的品牌、酒店、餐厅或商品转化为可点击的实时链接。

• 例如：

1. 朋友讨论某家餐厅时，餐厅名称可直接变为链接。

1. 用户讨论酒店时，可直接查看并预订。

1. 系统也可提示用户让Muse进一步调查。

• Meta可对这些点击、消息或后续交易收费。

• 该模式可能进一步分流Google搜索流量。

• Muse主动读取对话并提出建议虽然有价值，但必须谨慎处理隐私问题，避免让用户产生被监控的不适感。

9. WhatsApp与Muse的长期市场空间

a) WhatsApp本身可能形成多大的收入规模？

• 专家认为，即使不考虑Muse，WhatsApp长期也可能发展为每年2000亿美元规模的业务，并存在达到3000亿至5000亿美元的可能性。

• 核心基础是WhatsApp拥有约40亿日活跃用户。

• 商业化逻辑包括：

1. 品牌向用户发送商业消息并按条付费。

1. 每条消息收费数美分。

1. 一次商业对话即可产生多条收费消息。

1. 假设平均每位日活用户每天收到约5条品牌消息，收入规模即可非常可观。

• 上述判断属于长期潜在市场空间测算，实际结果取决于：

1. 商业消息渗透率。

1. 用户接受度。

1. 各地区监管要求。

1. Meta对用户体验和广告负载的控制。

1. Muse、Business Agent与支付体系的协同进展。

b) Muse能否证明Meta大规模AI资本支出的合理性？

• Muse可能通过以下方式为AI资本支出提供回报：

1. 提升现有广告系统的点击率和转化率。

1. 推动CPM和CPC上涨。

1. 打开主动式广告市场。

1. 建立Agent电商与GMV抽佣模式。

1. 扩大WhatsApp商业消息收入。

1. 切入支付和交易服务。

1. 防御社交媒体使用时长被其他AI产品分流。

• 从长期潜力看，Muse及相关AI产品存在推动Meta广告和交易收入实现倍增的可能。

• 但收入释放预计是渐进式过程，不能仅根据未来1至2年的直接收入判断AI资本开支回报。

结构化总结表格

新版prompt提炼之后的转译

提问者：Meta Q3的整体业绩表现如何？具体来看，广告收入增速大概是多少，高于还是低于此前预期？主要变动原因是什么？

回　答　者：略高于此前预期。Q3初期，我们预计广告收入增长28%，目前已将预测上调至29%。这一上调也延续到了Q4，我们目前预计Q4增长30%，此前的预测为29%。对于Q1，我们最新预计增长27%。

增速降至27%并不意味着出现了任何重大变化，主要是因为同比基数更高。从2年增长率来看，基础增速仍在逐季度提升，因此目前整体表现良好。

此前的利好因素依然存在，同时还出现了2项新的利好因素。就原有利好因素而言，Reels带来的推动作用依然较强，但强度有所减弱。此前约贡献5个百分点，目前约贡献4个百分点或略高于4个百分点。这一推动作用开始减弱并不意外，预计未来还会继续减弱。例如，明年Reels带来的推动作用可能只有约2个百分点，而今年为5个百分点。

其他现有利好因素还包括表单、点击和消息广告。越来越多的广告主付费吸引用户点击广告并进入WhatsApp，这一因素仍带来约1个百分点的推动作用。

即时购物广告也带来约1个百分点的推动作用。此类广告不同于用户点击后跳转至电商网站的广告，主要依托与Shopify、Amazon等平台的集成。

此外，覆盖式广告，即显示在内容上层的广告，带来约3至4个百分点的推动作用。

回　答　者：根据月份不同，该项为2.1，我认为这一趋势将延续到明年。WhatsApp Business约为2.1。这些都是我可以主动衡量的项目，因为它们与广告形式和广告位等因素相关。

这些项目合计可能只有13%、14%，而我们的增速接近30。其余增长来自哪里？其中一小部分可能来自Meta用户增长，或者我们的广告主样本市场份额提升，但现有增长驱动因素中还有价格。

CPM增速正在加快。以季度数据来看，我也掌握了更详细的月度数据。第1季度，所有平台的CPM增速为5.3%，主要包括Meta和Instagram的信息流、Stories和Reels。第2季度为11.6，第3季度可能达到17，也可能是16，但会显著提高。

价格上涨部分源于此前提到的因素。Click to Messenger或购物等许多新广告形式的CPM要高得多，但这并非全部原因。更深层的原因是，Meta持续改进算法、数据应用和AI应用，以提高广告个性化程度、点击率和转化率等，这确实会推高价格。

一个简单的例子是，其系统正变得更加灵活。当系统发现某位用户正在选购某种特定产品，例如狗床，算法就会实时调整，展示更多狗床广告，并提高广告负载，因为系统发现该用户对狗床广告的互动程度远高于其他广告。Meta曾向我表示，当用户点击广告或在广告上的停留时间超过内容时，整个算法就会发生变化，转而帮助该用户购买产品。过去，用户只会继续滚动浏览并持续看到广告。广告虽然会逐渐改善，但系统不会集中资源促成购买。

以上是许多现有的增长驱动因素，我认为这些因素都会持续。明年，Reels带来的增长动力可能会减弱，但其他因素仍将延续。

此外还有2个新增因素，其中1个目前可以衡量，另1个很快也可以衡量。目前可以衡量的是Facebook Audience Network，即Meta Audience Network，主要是在互联网其他位置的游戏内容中展示广告。

过去2年或3年，甚至可能是4年，Meta Audience Network在我们广告支出中的占比一直维持在约2.1%，部分季度为2.2，部分季度为2.0，但总体约为2.1。自4月以来，这一比例快速上升，目前已达到2.6。

虽然半个百分点听起来不多，但这部分全部属于增量，同时也表明Meta可能正在游戏广告领域加强对AppLovin的反击。这一点很值得关注。至于这一趋势能够发展到什么程度，我还没有进行充分思考。最后我还要补充1点，相信我们之后会进一步深入讨论。

回　答　者：我们开始看到小型企业快速采用 Meta Business Agent。广告主在熟悉并认可这一产品后，确实会发现投资回报率有所提升，其中许多广告主开始以快于正常水平的速度增加预算。目前仍处于早期阶段，因为该产品直到6月才正式推出，至今仅3个月。

在我们约300家广告主中，截至5月，约有30家参与了测试；从5月到9月，这一数量已增至60家。与此同时，目前的产品功能也比5月丰富得多。

广告主采用这一产品通常需要经历一定的过程。第1步，广告主会将自己的数据接入 Meta Business Agent，这能够为其带来显著价值，但不会立即增加广告支出。随后，Meta 的算法开始发挥作用，提升广告支出的投资回报率，进而可能推动广告主增加投入。之后，广告主可以选择启用 Meta Business Messaging，即 WhatsApp for Business。此时，广告主会看到更多销售转化，往往也会相应提高预算。

这方面还可以展开讨论更多细节，但其核心优势在于，Meta 拥有许多有利因素，而且这些因素大多正在持续改善，而非恶化。

提问者：我们今天将重点讨论 Muse。我们注意到，Muse 最近在美国和加拿大的下载量非常高。您认为 Muse 为什么会在这些地区如此成功和受欢迎？

回　答　者：Muse 的表现也令我印象深刻。我认为，一方面是因为产品本身较好，另一方面是因为它能够利用 Meta 的分发能力和品牌认知度。不过，在具体分析之前，有必要先从更广泛的背景来看待这一问题。

如果身处特定圈层，很容易认为所有人都在充分使用各种强大的 AI 功能。我自己显然也处于这样的圈层，认识的大多数人都非常熟悉科技，他们通常是科技从业者或投资者。我猜各位可能也身处类似圈层。但实际上，并非所有人都在使用 AI 管理日常生活、整理电子邮件、追踪账单、处理税务或管理收据。我每天有大约70个或80个智能体为我工作，但大多数人并没有这样使用 AI。

我经常在世界各地发表主题演讲，每次都会询问听众目前与 AI 的关系，并提供4个选项。第1个选项是“我不太使用 AI”，通常占现场人数的5%或10%。第2个选项是“我使用通用 AI，但不会对其进行训练”，也就是把 ChatGPT 当作功能更强的搜索引擎，这类人约占60%。第3个选项是“我正在构建智能体，并利用它们改善生活”，这类人约占20%到25%。最后1个选项是“我正在构建智能体工作流并衡量投资回报率”。在一个有100人的会场中，选择这一项的通常只有0到2人。

因此，实际上有70%以上的人不会使用 AI 管理生活、整理电子邮件或管理日历。他们基本上只是把 AI 当作功能更强的 Google，偶尔用来总结信息或起草电子邮件等。

回　答　者：Muse 面向市场时，并没有将自己定位为一个功能更强的搜索引擎，完全没有强调这一点。它传递的信息是：我们可以管理你的生活，监控你的电子邮件，为你寻找折扣，查看你的日历，替你比价购物，成为你的私人购物顾问。

我认为，这是一种全新的定位，此前在 ChatGPT、Claude，以及某种程度上的 Gemini 身上都未曾真正看到。Gemini 曾在 I/O 开发者大会上表达过类似理念，但并未直接面向消费者进行宣传。因此，这确实是一种全新的定位。

从平台迭代和平台变革的角度来看，为什么 ChatGPT 曾经一度优于 Google？因为它明确告诉用户，自己实现了一次升级，而市场也理解了这种升级。Muse 现在传递的信息同样是：这是一次升级，是一种不同的产品。因此，它所参与的是一场不同的竞争。我认为，它的定位和推出时机都很好。

表面上看，人们可能已经形成了使用 ChatGPT、Gemini 或 Claude 的习惯。但对于许多消费者，尤其是主要使用手机、每天没有太多时间使用笔记本电脑的中产阶级母亲而言，Muse 非常符合她们的需求。

因此，这显然是一款消费级产品，而不是 B2B 产品。Claude 可能不会担心，但 ChatGPT 和 Google 应该对此有所警惕，因为 Muse 的产品表现很好，契合当前的市场时机，整体定位也非常新颖。

此外，Muse 还拥有 Meta 的分发能力，我认为这种分发能力会非常强大。举例而言，Meta 目前还没有这样做，以下只是我基于现有信息作出的联想。正如我在电话会议开场部分简要提到的，当 Meta 发现用户正在考虑购买某种商品时，通常会提高广告负载，向用户展示更多相关广告。

一个很简单的例子是，夏天即将结束，伦敦的天气开始有些寒冷。我当时想买一双斯堪的纳维亚风格的拖鞋，后来看到了一则广告并点击了它。由于我对这类广告的互动程度高于对普通内容的互动程度，Meta 随后开始向我展示越来越多关于斯堪的纳维亚羊毛拖鞋的广告。最终，我确实通过一则广告购买了这类拖鞋。

我认为，一个非常自然的衔接方式是，在某个阶段向用户提示：“你似乎正在选购斯堪的纳维亚风格的拖鞋，是否希望 Muse 成为你的私人购物顾问，为你提供一组选项？”这样的衔接非常自然，完全有可能奏效。

如果 Meta 未来能够通过 Facebook 和 Instagram 实现类似的衔接，我并不会感到意外。也许有一天还可以通过 WhatsApp 实现，但这会更加敏感。当系统判断用户需要完成一项复杂任务时，可以将其自然地引导至 Muse。

总体而言，我非常看好 Muse。我认为这是一款出色的产品，并未错过入场时机，尤其是它的定位非常新颖。

提问者：您认为 Muse 这款产品可能会对 Meta 的整体广告收入带来什么影响？对于广告主而言，Muse 又可能带来什么样的价值？

回　答　者：这是一个很好的问题。首先，我认为其影响会比较缓慢。虽然整个行业都在快速发展，但就 Muse 而言，Meta 短期内并不缺少相关收入。我认为，Meta 短期内可能会沿用过去的策略，先把产品做好。

提问者：Meta未来可能如何实现商业化？

回　答　者：Meta可能会优先考虑尽可能提升产品的实用性，而暂时不考虑盈利。WhatsApp就是一个典型案例，其在约10年间几乎没有产生收入。Stories和Reels也经历了类似过程：前9至12个月仅投放极少量广告，随后逐步引入风险较低的品牌广告，再增加效果广告，最终在4年或5年后实现全面商业化。因此，Meta可能会先用6至12个月暂不进行商业化，之后逐步推进，并用4年或5年完成整个商业化过程。

最简单、最直接，但未必是最佳的方式，是逐步插入广告。用户使用Muse时，页面上会有空间展示相关广告。Meta掌握完整的购买意向图谱，了解用户可能有意购买的产品，因此完全可以据此投放广告。例如，Muse在帮助用户安排日程时，可以提示：“我注意到您周六没有为孩子安排活动，这里有一个值得关注的儿童俱乐部。”这类广告与用户需求相关，因此不能排除这种可能性。这类似于一位优秀的行政助理主动思考如何帮助自己的上司改善生活。我认为未来会出现这类广告。

另一个主要方向是，Muse可能会成为许多人的私人购物助手。其定位包括帮助用户进行比价、寻找优惠、持续监测特定商品的折扣码，并在发现后通知用户。在这种情况下，Muse需要始终提供最佳答案，因此可以提供不同品牌的选项。作为私人购物助手，直接通过品牌选项实现商业化可能较为困难，但Meta可以在最终完成销售后，按照GMV的一定比例抽成。这种模式正日益被市场接受，OpenAI和Google都已有采用该模式的产品，因此Meta也可能采取这种方式，但目前尚不确定。

付费模式也值得关注。目前Meta已经提供免费层级和付费层级。用户形成使用习惯后，收费将成为可能，但未必适用于所有人，因为消费者已经习惯免费产品。不过，对于真正的高频用户，例如认为该产品每天能为自己节省0.5小时或1小时的中产阶层母亲，每月收取20美元或30美元可能是可行的。

此外，还有一种与按GMV比例抽成相关但有所不同的模式，即支付业务。Meta目前已在全球约5个国家推出Meta Payments，Stripe的首席执行官也已加入Meta董事会。Meta团队表示，计划在全球范围内提供支付服务。由于存在各种监管障碍，这一计划可能需要数年时间。如果能够实现，Meta还可以促成会员之间、用户之间或企业与用户之间的交易，并从每笔交易中按一定比例或收取少量费用。这也可能成为值得关注的商业模式。

Meta可以采取多种方式实现商业化。对Meta而言，最容易的方式仍然是广告。我认为，其可以通过友好的方式推荐广告，主动帮助用户进行消费。

回　答　者：不再只是把内容发布出去并期待用户点击，而是告诉用户：“我为你找到了这个，你感兴趣吗？”例如，如果我的妻子，甚至是我，收到 Muse 这类执行助理的建议：“我注意到你计划在周五聚餐，但还没有预订餐厅，需要我提供一些建议吗？”她会点击“是”。尤其是在第1次体验良好的情况下，她之后可能会主动提出：“你能否查看我的日历，找出我安排外出晚餐或午餐的其他时间，并帮我提供一些选择？”

我认为这里存在很多可行方式，但较容易落地的可能是主动式广告和支付。电商模式也可以按 GMV 抽取一定比例，订阅模式同样可行。

提问者：您刚才提到了电商。我们注意到，去年 ChatGPT 已经开展了电商业务，谷歌也发布了自己的 UCP 协议，但目前这些业务的规模仍然比较有限。现有的电商实践与电商智能体之间，主要有哪些区别？

回　答　者：这两者略有不同，我分别说明。OpenAI 的产品刚刚达到 10 亿美元的年化收入，其中一部分来自电商业务，但我认为他们并未单独披露具体数据。不过，他们提供的产品体验并不好。实际上，第1款产品并未成功，现在的版本有所改善。

如果从应当避免的做法来看，OpenAI 的第1款产品实际上是：假设用户正在寻找跑鞋并看到一则广告，平台希望用户点击“立即购买”，随后直接进入结账页面，页面上只有“立即购买”和选择信用卡的按钮。对消费者来说，这种体验过于突然，也容易令人不安。

在常规电商结账流程中，用户通常既可以“加入购物车”，也可以“立即购买”。即使选择“立即购买”，落地页上也会提供条款和条件、配送时间等链接，用户还可以增加或减少购买数量，例如选择购买 2 件，或者将商品从购物车中移除。这些信息和功能能够向用户提供保障，降低用户对交易出错的担忧，而 OpenAI 当时完全没有提供这些内容，只有“立即购买”按钮，因此没有取得效果，几乎没有人点击。

此外，他们还面临广告推荐质量不佳的问题。从消费者角度看，用户搜索的是跑鞋，但系统可能推荐更偏时尚属性的运动鞋，而不是适合慢跑的鞋，因为 OpenAI 没有做好相关准备。因此，当时的问题既包括定向不准确，也包括产品设计过于简单。

OpenAI 目前的发展方向至少合理得多：不再试图推动用户立即购买，而是将用户带到电商网站的落地页，同时有效吸纳其中的内容。

回　答　者：这种方式类似于 iframe，将结账页面嵌入 ChatGPT。这样一来，用户能够确认商家是亚马逊、沃尔玛或其他平台，因此可以建立信任。目前，这种方式的效果明显更好，也已经可以正常运行。

Google 的产品更好。Google 从一开始就意识到，只有一个简单且没有任何解释的“立即购买”按钮是行不通的。因此，Google 在产品中明确展示用户所购买商品对应的商家。虽然本质上仍是链接跳转，但整个流程较为顺畅。我不太清楚 Google 为什么没有更快扩大这一功能的应用范围。目前该功能确实在扩展，但 Google 似乎仍然更专注于持续改善 Gemini 的用户体验，而不是过度关注广告。这可能会成为 Google 未来的一个机会。

从这些经验来看，如果我是 Meta，就不会设置一个没有任何解释的简单“立即购买”按钮，因为这种方式不会奏效。我认为仍然需要将用户带到商家的落地页，即使该页面是通过 iframe 嵌入的。整个过程应尽可能简便，让用户不会觉得自己正在不熟悉的环境中反复跳转，同时也能确认自己是在向真实商家购买，并且商家具备明确的责任主体、实际配送时间、退货流程等。

与 Shopify 的整合也会有所帮助，因为大多数人信任 Shopify 商店，认为这些商店运营规范。Meta 应考虑如何以不逊于、最好是优于 OpenAI 和 Google 的方式推出这一功能。首先，可以跳过对即时结账模式的试验。

有一项调查，我不记得具体由谁发布，可能是 Visa 或其他支付系统。调查结果显示，65% 的美国消费者愿意让聊天机器人代为研究，但只有 14% 的美国消费者愿意在自己不了解具体情况的前提下，直接让智能体代为购买。几乎所有消费者都希望看到一个确认页面，了解系统推荐的内容，并核对条款和条件、退货政策及价格，之后再亲自点击“立即购买”。消费者并不希望计算机代替自己完成这一决定。我认为这是 Meta 需要解决的关键问题。

相较于 OpenAI，Meta 的一大优势是拥有更多商家，但这一优势相较于 Google 并不一定存在。一方面，Meta 可以更准确地推荐用户应该购买的商品；另一方面，Meta 已经能够构建更加顺畅的整合体验，尤其是与 Shopify 的整合，可以帮助用户购买商品；如果是 B2B 场景，也可以帮助用户预约服务或完成类似操作。

总体而言，OpenAI 公开经历了很多试错，这反而为 Meta 提供了经验，使其可以避免犯同样的错误。Meta 可能有能力做得比 Google 更好，但我认为 Google 目前的产品已经非常出色，因此不确定 Google 为什么没有更快推广。我们合作的所有广告主都对这一产品感到满意。

提问者：您是否认为，整个智能体阶段的电商购物，以及其他各类服务的购买方式，与聊天机器人阶段存在较大差异？在聊天机器人阶段，用户会根据获得的答案自行搜索，随后跳转到网页。

提问者：传统电商聊天机器人通常根据用户的询问给出建议，用户仍需自行点击搜索结果并完成结账；AI Agent 电商则在了解用户意图后，直接完成搜索和购买的整个闭环。过去，Expedia、Booking、Ddash 等信息聚合商主要作为入口，并通过竞价排名等形式收取广告费。由于人存在惰性，通常更愿意购买排名靠前的产品，但 AI Agent 不一定会购买排名第1的商品，而可能进行各种比较。在 AI Agent 代替客户作出购买决策时，它是否会取代这些中介平台，成为第1入口？如果它成为主要入口，过去广告主采用的竞价排名系统是否仍然有效？广告本身的价值是否也会发生变化？

回　答　者：这是2个非常重要且相互关联的问题。我认为，未来5年或10年，对聚合商或平台型企业而言，可能会是一段非常可怕，甚至令人沮丧的时期。无论是Amazon、Booking，还是其他平台，尤其是那些经营状况不佳的公司，都可能面临这种情况。

从用户体验来看，过去2年，我一直相信，随着时间推移，每1位拥有智能手机的人都会至少拥有1个自己偏爱的聊天机器人。有些人可能会使用多个聊天机器人，例如分别用于健康和购物；也有些人只会使用1个。这个聊天机器人可能是ChatGPT、Gemini、Claude、Muse，也可能是Elon Musk在某个阶段开发的产品。

这个聊天机器人将能够访问我们的日历、电子邮件，甚至可能访问银行账户。以我为例，它还可能访问我的WhatsApp，以及几乎所有我登录的网站等。因此，它能够深入了解用户，并与用户之间建立真正的信任与理解。在这种情况下，聊天机器人多数时候将成为用户的私人购物助手。接下来需要考虑的是，这究竟意味着什么？

回　答　者：第1，聊天机器人已经非常了解用户。比如，我的聊天机器人知道我所有孩子的姓名和生日，也了解我和妻子对餐厅等方面的偏好。因此，随着对用户的了解不断加深，它能够提供越来越准确的推荐。

同时，由于这种关系以对话形式展开，聊天机器人可以通过非常便捷的方式提出问题，帮助用户进行分类和筛选，无论用户购买的是实体商品、酒店客房、餐厅服务，还是其他产品或服务。

如果聊天机器人帮助用户筛选出一份优质候选清单，用户可能会说：“感谢你帮我考虑周五晚上的餐厅选择，请预订选项B。”显而易见的下一步是，聊天机器人可以直接前往餐厅指南聚合平台完成预订，绕过竞价页面等环节；如果商家网站足够完善，也可以直接连接商家、餐厅、店主或商业服务提供商。

因此，对于拥有优质网站和直观用户体验的商家，大多数聊天机器人会倾向于绕过聚合平台，直接连接这些商家。以酒店为例，如果四季酒店的结账体验非常出色，聊天机器人就可能直接连接四季酒店。对于经营不佳、网站体验较差的家庭式餐厅，则可能通过Booking.com预订，因为这样更加便捷。

我认为，这不会淘汰大多数聚合平台，但当聊天机器人直接连接终端商家更合理时，它会逐渐分流原本属于聚合平台的流量。

关于广告竞价，亚马逊、Booking等聚合平台通过竞价获得了大量收入。例如，用户在亚马逊或其他网站上搜索半程马拉松跑鞋时，各品牌会相互竞争，以进入搜索结果第1页。

但如果聊天机器人代表用户行事，这种模式就会变得棘手，因为用户并不希望看到其中的大多数品牌。假设有100个跑鞋品牌竞争进入第1页，这种竞价在某种程度上是虚假的，因为聊天机器人实际上应该提前筛除所有不适合的产品和服务，而这些产品和服务根本不应进入竞价环节。风险在于，由于分类和筛选更加精准，竞价压力可能会减弱。

如果聊天机器人的目标是帮助其服务对象，即用户，找到最合适的产品，那么它可以通过几种方式实现商业化。第1，竞价仍然可以存在，但不会展示30个广告主，而可能只展示3个。消费者最终仍然希望拥有选择，只是不希望经历繁琐流程或浪费时间。因此，在许多竞价场景中，聊天机器人可以提供3个最佳选项，由消费者作出最终决定，竞价机会依然存在。

即使传统竞价压力可能不大，这3个品牌也会知道，该用户的转化率很高。此人已经有购买意愿，购买概率达到80%；如果品牌进入3个候选选项的短名单，就意味着获得购买的概率为27%，因此品牌愿意支付较高费用。

另一种方式是按照GMV抽取一定比例，具体比例可能根据利润率而有所不同。

回　答　者：这种模式值得尝试，我认为 Meta 可以从中借鉴。他们实际上会告诉广告主，竞价金额可以设定为 GMV 的一定比例。例如，如果利润率为 50%，广告主可能愿意支付 GMV 的 20%，这笔销售仍然有利可图。如果利润率为 10%，广告主可能愿意支付 GMV 的 3%。

以酒店为例，大多数酒店会向 Booking.com 支付 GMV 的 15%至20%。因此，无论是聊天机器人、Google 还是 Meta，都可以采用这种模式。有些用户会购买最便宜的产品，但不应因为该产品无法承担竞价成本，就将其排除在外，因为平台仍然可以获得一定比例的 GMV。即使用户购买的产品比其他产品便宜 10%，平台仍可获得 GMV 的 X%。

未来 5 年对聚合平台而言将是非常艰难的时期。不过，我认为，掌握用户体验的平台仍然可以通过针对 3 个入选项进行竞价、收取一定比例的 GMV，或将两种模式结合起来，获得可观的收入。

提问者：接下来想了解 Agent 对 Google 广告的影响。通常搜索可以分为两类：一类是被动搜索，例如观看短视频，或在没有明确购买目标的情况下浏览电商平台。我认为这类搜索不会受到 Agent 太大影响。另一类是主动搜索，用户带着非常明确的问题进行搜索，尤其是在购买汽车等大件商品，或确定旅游目的地时。过去数十年，这类搜索通常发生在 Google 上，搜索引擎和浏览器一直是互联网的首要入口。如果 Amuse 开始扩大 Agent 的应用场景，抢占大量具有明确商业意图的主动搜索，这会对 Google 搜索及其广告收入产生什么影响？

回　答　者：这无疑会对搜索业务构成威胁。不过，需要逐个品类进行分析，因为不同品类面临的风险程度不同。例如，在 B2B 领域，我认为 Google 可能不会受到 Mews 太大影响，因为通过 Mews 处理 B2B 类事务的用户可能并不多。但在休闲等领域，情况会有所不同。

回　答　者：我认为，这一趋势会从小规模起步，首先覆盖价格较低的再商业消费，例如购买价格低于$100的商品；也可能从休闲领域开始，例如周末活动、餐厅和儿童俱乐部等，之后再逐步扩大规模。

短期内，Google可能不会受到明显影响。但长期来看，Muse可能会逐个切入Google的垂直领域，并从中分流部分业务。可以设想这样一类用户：忙于兼顾孩子、工作、家庭和生活的中产阶级母亲。她们经常在车内或出行途中使用手机安排各种事务，这些相关品类最有可能被Muse从Google手中分流。

休闲领域可能首当其冲，其次可能是时尚和美妆。时尚和美妆属于需要更多考虑的消费领域，但Muse会先从低价商品切入，例如口红或快时尚产品。

对Google而言，相对安全的领域可能包括B2B、政府事务，以及医疗和教育中更严肃的细分领域，例如消费者比较医疗保险或MBA课程等。不过，Muse仍可能从低价、轻决策消费逐步向这些领域拓展。我不确定这种轻决策、相对低价的消费者支出占全球GDP的具体比例，但规模可能相当大，或许占全部支出的20%左右。我并不是说Meta会获得这20%，但它可能从中取得相当可观的份额。

我尚未对此进行量化，但如果Muse成功实现规模化，5年后Google的收入可能会因此下降3%、4%或5%。当然，目前仍无法确定Muse能否成功实现规模化。

Google也在推出类似产品Gemini Spark。它在许多方面与Muse相似，但目前还不是独立应用，因此Google仍有反击的机会。在情景分析中需要考虑这一点。不过，Google必须迅速行动，因为Muse很可能会快速发展。

提问者：如果未来个人Agent成为大量消费者的标准配置，面向全球10亿级别的消费者，Google可能也会很快推出自己的Agent产品。在Meta和Google的Agent展开竞争的情况下，您认为哪家厂商在消费者市场更具优势？此外，除应用厂商外，许多手机厂商也在开发相关产品，例如苹果正在推进Siri，并可能希望将其打造为个人AI助手。未来有意进入这一领域的厂商较多，您认为谁会更具优势？

提问者：您认为哪一方更具优势？从更广泛的角度来看，除了 Google 和 Meta，Apple 等手机厂商及其他手机品牌也在开发嵌入手机产品的个人助理。因此，个人智能体领域可能会出现许多潜在参与者。在这一背景下，您认为谁的优势更大？

回　答　者：这将是一场很有看点的竞争。值得注意的是，Apple 似乎正在与 Google 合作开发相关产品，至少目前 Siri 由 Gemini 提供支持，这为 Google 增添了明显优势。

Meta 最终需要改变用户的一些固有习惯。很多人经常使用 Siri，也有很多人经常使用 Google。当 Google 推出自己的同类产品时，凭借其庞大的分发渠道，也会获得很大吸引力。Google 还可以自然地整合用户的日历、电子邮件、浏览记录和 Google Maps 历史数据，因为这些服务实际上都由 Google 掌握。因此，从理论上讲，Google 的产品可能最有能力提供帮助，但其入场时间会比较晚，将晚于 Muse。

Muse 所面对的一个问题是，人们心中已经形成了根深蒂固的认知：Facebook 和 Instagram 本质上属于娱乐渠道。人们感到无聊时会使用这些平台。无论是通过视频带来刺激，还是通过购物带来兴奋，只要能够让用户的生活体验有所改善，这就是 Meta 擅长的领域，也是用户使用这些平台时的主要心态。

当人们进行严肃搜索时，往往会使用 Google、Claude 或 ChatGPT，Meta 在这类需求中获得的份额可能相对较少。因此，我认为 Meta 和 Muse 可能会在生活方式和娱乐相关领域占据最大的市场份额，例如休闲、时尚、短途旅行、游戏，以及较为基础的医疗健康和消费者健康服务。这些内容通常都与生活方式有关，也是人们在无聊时愿意投入时间和金钱的领域。

相比之下，Google、ChatGPT 和 Claude 更有可能继续在严肃采购相关领域占据优势，包括 B2B，以及医疗、教育和政府等领域中更为严肃、专业的需求。

总体而言，这种趋势对 Meta 可能比对 Google 更有利。目前，Google 确实承接了大量相对琐碎的搜索需求，但随着时间推移，这类搜索将会减少，因为 Muse，以及可能包括 OpenAI 在内的其他平台，很可能会获得更多此类需求。

如果关键问题是这种趋势对谁构成阻力、又对谁形成助力，那么我认为，它对 Google 无疑构成阻力，对 OpenAI 可能也构成阻力，但对 Meta 无疑是一项利好。

回　答　者：Muse可能取得的成功，将缓解Meta原本可能面临的另一个问题。这听起来或许有些为时过早，但我一直有一个判断：未来几年，我们花在社交媒体上的时间可能会减少。

简单来说，至少在美国和欧洲，人们的屏幕使用时间已大致达到上限。典型的1天有24小时，其中8小时用于睡觉，8小时用于工作，大约2小时用于开车、洗漱或其他不使用屏幕的活动，剩余约6小时用于通过屏幕进行娱乐。

目前，数字娱乐主要分为3类：社交媒体、视频和游戏。三者的规模大致相当，游戏可能略小一些。如果出现另一种数字娱乐形式，现有类别中的某一类必然会受到影响。目前最明显受到影响的是专业制作的内容，例如Netflix的用户参与度正在放缓，我认为甚至可能已经开始下降，因为用户显然更偏好UGC。

未来，与聊天机器人交流、体验生成式AI，实际上可能成为一种娱乐形式。几个月前有一项调查显示，美国人平均每天使用OpenAI的时间为16分钟。其中一部分用于工作，但另一部分用于休闲。人们正在使用聊天机器人讨论心理健康、规划生活、获得反馈、安排周末以及处理其他事务。Muse、Spark及其他产品将加速这一趋势。

假设5年后，普通人平均每天花30分钟使用聊天机器人，这意味着他们每天在社交媒体、视频和游戏上的时间将减少30分钟。因此，从某种意义上说，Meta原本面临着一个非常严峻的阻力，例如聊天机器人是否会导致其平台使用时长下降10%。我认为Muse带来的增量可能会超过这一影响，但需要将这两方面结合起来考虑。对Meta而言，这并非只有增量价值，在某种程度上也是在抵消未来可能出现的阻力。这就是我对消费者时间分配的判断。

提问者：如果Meta拥有Muse这样的个人Agent，是否意味着它能够获得比过去更多的用户线索，并更深入地了解用户的商业意图，从而显著提升Facebook和Instagram广告的转化率？您如何看待这一点？

回　答　者：是的，提升会非常显著。我认为可以对此进行一定程度的量化。回到大约2年前，也就是AI尚未广泛应用，搜索引擎、Facebook广告等渠道对消费者了解有限的阶段，典型电商网站的转化率约为3%。也就是说，每100名访问者中，有3人会购买商品。

如果AI能够充分发挥其应有的作用，并掌握相关信息，那么可以进一步设想转化率可能达到的水平。

回　答　者：AI能够基于我们过去的所有购买记录，了解我们的价格敏感度，并对广告和落地页进行个性化呈现。以跑鞋为例，2年前，如果我搜索跑鞋，只会看到大量跑鞋和相当随机的推荐。现在，如果我准备购买跑鞋，看到的每一则广告都应该是一位50多岁的男性，他比我稍微英俊一些，腹肌也更好，在伦敦的公园里跑步，看起来快乐而帅气。推荐的鞋应来自我喜欢的品牌，价格也应处于我能够接受的范围，广告内容则应涉及半程马拉松、Barry's Bootcamp等。每一则广告、产品推荐、价格和购买方式都会针对个人进行定制。

在这种未来场景下，点击率和转化率都会大幅提升。以转化率为例，我认为未来消费者只有在商家进入3个候选对象的短名单后，才会真正访问该商家的网站。如果消费者开始搜索后最终完成购买，并且因为短名单中有3个候选对象而访问3个网站，理论上，如果他们从不改变主意，未来的转化率将达到33%。

现实中，消费者在购买过程中可能会访问网站，但并不准备当天购买，也可能只是为了查找链接并分享给朋友。不过，我可以设想未来转化率达到10%的情形，这将是过去水平的3倍。即使认为这一判断略显激进，在由AI主导的环境中，转化率提升至2倍也很容易实现。这意味着，未来几年企业愿意支付的CPC至少会提升至2倍，甚至可能达到3倍。因为转化率提升至2倍或3倍后，广告主自然愿意为这些流量支付更高的价格。

提问者：我关注到Meta也在布局支付业务。在Meta的整个商业链路中，唯一尚未形成闭环的环节就是支付，这可能与中国的情况不太一样。在美国，开展支付业务的难度有多大？我的理解是，Meta希望通过AI实现这样的目标：未来，包括电商在内的许多网站，其流量可能由AI提供，许多中间商也会转变为通过AI提供服务。Meta希望从这些交易中获得一部分抽成，但目前相关业务由Stripe等公司提供。Meta现在也在自建支付能力，在美国自建支付的难度究竟有多大？您认为Meta何时能够推出自己的支付系统？我认为Meta现在的地位有些类似于苹果。苹果掌握了iOS的分发渠道，因此可以对许多流水收取“苹果税”。未来，如果Meta的个人AI确实成为最大的个人AI，它在整个应用生态中的地位是否会接近苹果应用商店？无论是平台上发生的流量分发还是交易，Meta是否都可以从中收取一部分费用，而这部分费用可能需要通过支付完成？

提问者：第1个问题是，在Meta当前的商业模式中，唯一缺失的环节是支付。过去，尤其是随着AI的发展，未来AI可能成为许多广告主触达用户的主要渠道，因为大量流量将通过AI智能体到达终端消费者。目前，Meta在支付方面一直依赖Stripe，但我们也了解到，Meta正在建设自己的支付系统。您认为该系统何时能够准备就绪？第2部分问题是，苹果控制着iOS系统的整个分发渠道，因此可以对其中发生的每笔交易收取“苹果税”。如果Meta的个人智能体确实能够取得市场第1的份额，Meta是否也会拥有类似的地位，并对每笔交易收取佣金？

回　答　者：关于支付，Meta已经在部分国家推出了支付服务。我不记得具体是哪些国家，但可能与Meta较早推出商业智能体的国家相同，例如墨西哥、巴西、印度尼西亚，也可能包括印度。我还听说过沙特阿拉伯，全球大约涉及3、4、5个国家。Meta已经在部分国家推出支付服务，但尚未覆盖美国、英国等主要市场。

Meta曾公开表示，将把支付服务推广至全球。我并非支付领域的专家，但以Meta的能力，我认为其可能在2年或3年内覆盖全球。由于监管等因素，这一过程可能仍需这么长时间。不过，Meta也可能通过与Visa、Mastercard等机构合作找到捷径。尽管这并非我的专业领域，但可以认为Meta最终会使支付服务在全球范围内运作，因为这显然是其战略的重要组成部分。Meta将Stripe的首席执行官引入董事会，也说明了2点：第1，支付业务对Meta非常重要；第2，Meta最终可能与Stripe开展大量合作。这是对您问题第1部分的回答。

关于Meta与苹果之间的关系，即谁能够从中获取收益，理论上，Meta确实需要担心苹果通过收取GMV的一定比例，或采取类似与谷歌合作的模式，向其收取高额费用。据传，谷歌每年向苹果支付约200亿，用于在苹果生态中的各类入口，包括成为默认搜索引擎等。因此，Meta未来确实可能需要向苹果支付巨额费用。

不过，苹果也无法过度索取，因为Meta与谷歌一样具有强大的影响力。如果苹果表示不允许使用Muse，那么在未来几年用户形成使用习惯后，许多消费者可能会选择Android或其他手机，Meta甚至可能在未来自行制造手机。因此，这种收费不太可能达到极端程度。

短期内，以Meta一贯的策略来看，它可能不会通过该产品盈利，因此暂时也不存在与苹果达成商业分成协议的基础。对消费者而言，这实际上只是一款免费产品，苹果也无法从中获得收入。Meta会先培养用户习惯，随后开始商业化。届时，苹果意识到用户已经形成使用习惯，可能会尝试获得相应的收入分成，但具体会以何种方式实现，目前尚不明确。

回　答　者：理论上，Meta可能会允许Apple Pay与其自有支付方式并存，因为这对消费者更便利，Apple也可以从中获得一定收益。如果用户每月支付$20订阅，Apple或许可以从订阅收入中抽取一定比例。但我认为，大多数用户不会在下载应用时立即订阅，而会在之后选择订阅。

我确信Apple每年至少能从中获得数十亿美元，但不确定能否达到数百亿美元。因为如果Apple要获得如此高的收入，可能需要以涉嫌反竞争的方式特别针对Meta。因此，我认为Meta将保留Muse相关收入中的大部分收益，而Apple会设法从中获得一定收入。

具体方式可能包括抽取部分订阅收入，但订阅不会是Meta最大的收入来源；也可能包括接入Apple Pay。我认为Apple不太可能抽取广告收入，因为这种方式过于复杂。Apple还可能收取大额费用，以确保Muse在第1天就进入推荐应用名单或成为预装应用。

因此，Apple有多种获利方式，但我认为相关收入不太可能达到200亿美元。因为Muse不会成为默认搜索引擎，而只是一个非常受欢迎的应用，两者性质不同。相较于默认搜索引擎，Apple更难从热门应用中收取类似的费用。

提问者：您如何看待Muse与Meta旗下WhatsApp的融合？这2个软件的融合可能会对商家和用户产生什么影响？

回　答　者：据我了解，目前双方尚未进行任何整合，但未来可能会推进。这是一个很有意思的方向。我曾认为，Meta可能会在某个阶段在WhatsApp中加入类似虚拟朋友的功能，用户可以与其交流并讨论自己的需求，这种可能性目前依然存在。它实际上可以成为WhatsApp中的私人购物助手或其他类型的个人助手。

如果用户订阅了Muse，Meta未来可以让Muse直接出现在WhatsApp中，使用户无须切换应用即可向Muse提问。例如，用户可以告诉Muse：“我需要为孩子安排周日的活动，最好是户外活动，除非下雨。”这样，WhatsApp就会成为与Muse互动的另一种方式，从而增强Muse的用户黏性。Muse既可以作为WhatsApp中的虚拟朋友，也可以作为独立应用，用户能够在2种使用方式之间切换。

反过来，在Muse内部也可以进行类似整合。假设Muse帮助用户完成个人购物，但用户最后仍需要向商家咨询一些问题，Muse可以将用户转入WhatsApp Business，让用户通过WhatsApp与企业或企业的AI智能体沟通。Meta也在开发这类AI智能体。

这样一来，Meta最终可能同时掌握对话双方：一方是代表商家的聊天机器人，另一方是代表用户的聊天机器人。Meta也可以支持用户与商家聊天机器人沟通，或让商家人员与用户聊天机器人沟通。因此，Meta很可能最终控制对话双方。

在这种情况下，Meta需要格外谨慎，避免产生利益冲突，因为商家希望尽可能多地销售商品。

回　答　者：商家希望获得最佳利润，消费者希望获得最优惠的交易，因此需要建立良好的防火墙或类似机制，确保双方各自履行职责。这将非常有意义。

从变现方式来看，变现主要发生在商家端或品牌端。事实证明，品牌非常愿意为 WhatsApp 上的每条消息支付几美分，因此这可能是一个值得关注的方向。

此外，还有一种可能性，既可能通过 Meta AI 实现，也可能通过 Muse 实现。马克·扎克伯格在6月的一次会议上表示，未来当人们在 WhatsApp 中谈论某个品牌时，相关内容将转化为可以点击的实时链接。这个品牌可能是一家酒吧、一双跑鞋或一家酒店。例如，一群朋友讨论在哪里见面，有人提议去 Antelope，这个名称就会转化为链接；或者朋友们讨论旅行地点，有人建议尝试巴厘岛四季酒店，相关链接也会直接出现，用户可以点击。

这对 Meta 来说是一个巨大的机会，因为未来 Meta 很可能对来自这些链接的点击收费，同时也会对 Google 构成不利影响。另一种方式是，系统可以提示用户“让 Muse 调查一下”。Muse 也可能主动询问：“我看到你在 WhatsApp 上寻找周四晚上与朋友聚会的地点，需要我帮你研究吗？”不过，这种方式较为敏感，需要确保不会让用户感到不安。

因此，Meta AI、WhatsApp 与 Muse 之间可能存在大量跨场景衔接机会。商家按消息付费似乎是一种接受度较高的变现方式。可能还存在其他方式，但整体而言，重点或许并不在广告，而更多在于通过对话中的商家端实现变现。

提问者：Muse 可能为 Meta 带来多大的潜在市场空间？考虑到 Meta 过去几年在 AI 方面进行了大规模资本支出，Muse 带来的广告收入增长能否证明这些资本支出的合理性？其广告收入是否可能翻1倍甚至2倍，即提升至2倍或3倍？

回　答　者：简要回答是，完全有可能。即使不考虑 Muse，我认为 WhatsApp 至少可以成为每年2000亿美元规模的业务，而且很容易达到3000亿美元、4000亿美元或5000亿美元。

核心原因是，WhatsApp 拥有40亿日活跃用户，因此实现可观收入并不困难。每当用户收到品牌发送的消息时，品牌都可以支付几美分，而这种情况每天可能发生数次。举一个简单的例子，最初假设平均每位日活跃用户每天收到10条品牌消息，随后更正为5条；这并不难实现，因为一次对话就可能产生这些消息。

原文转译

本服务于有谦咨询客户，与受邀客户，与多研究交流，不构成投资建议。 相关人员应自主做出投资决策，并自行承担投资风险。 有谦咨询不因使用本次内容所导致的任何损失，承担任何责任。 专家发言内容仅代表专家个人观点，不代表本公司观点。 本次会议内容不得涉及国家保密信息、内部信息、未公开重大信息、商业秘密、个人隐私。 不得涉及可能引发不当炒作或股价异常波动的敏感信息。 不得涉及影响社会或资本市场稳定的言论。 未经有谦咨询事前书面许可，任何机构或个人不得以任何形式复制、录音、刊载、转载、转发、引用本次会议内容。 否则由此造成的一切后果及法律责任由该机构或个人承担。 本公司保留追究其法律责任的权利。 感谢您的理解和支持。 OK，好的，在咱们开始今天的访谈前，请您确认一下声明：我同意或我不同意，就是明知道保密信息的定义，并不会在访谈中透露任何信息及其他的保密信息。 Okay, dear expert, before we formally start, please acknowledge that any of the information and also any discussions we have in this meeting, you promise you will not disclose in any forms, and you will not, you will conform the confidentiality of this meeting. I confirm. OK，好的，那现在我们想先问一下就是说，Meta的Q3的一个业绩，业绩的话在，就是在，Meta的Q3的话，我们的整个广告投放的增速的话大概是多少，比我们之前的预期的，高了还是低了，然后主要变动的一个原因是什么？ 对。 Okay, first question on the Meta's Q3 performance, more specifically on the ad revenue growth rate. And like is that higher than or lower than our expectations and what are the underlying drivers here? Yeah, it's slightly higher. So we started the third quarter forecasting 28% growth and we're now forecasting 29. Uh, and that revision up has flowed into the fourth quarter. We're now forecasting 30% where it was 29. Uh, and then for the first quarter, we have a new number which is plus 27. And that drop to plus 27 is not anything significant. It's just we have higher higher um comps, you know, for the year. So it's a the the the underlying two-year growth rate keeps going up each quarter. So, um so far so good. Do you want me to pause there and then I'll give the reasons? Yeah, yeah, yeah. Please go on. Okay. So, um the the the the reasons that the same tail winds as before still exist. There are two new ones. Um so if I with through the old tail winds, um the tail wind coming from reals is still um strong. Not quite as strong. It was about a five-point tail wind, it's now about four. Uh or just over four. So it is starting to ease off, which is no surprise. I think it will all ease off. So, you know, next year, for example, the real tail wind may only be about two points um as opposed to five this year. Um other existing tail winds, the forms and clicks and messaging, so more and more people paying to get someone to click on an ad and go into WhatsApp is still about a one-point tail wind. The shopping ads, which is the instant shopping ads as opposed to um ads where you click and go to an e-commerce website. So this is the integrations with Shopify, Amazon, etc. That's about a one-point tail wind. Um then overlay ads. So those are the ads that appear over the top of the content. They're about a three to four point. 2.1 depending on the month, but I think that will continue into next year. And then you have um, WhatsApp for business, uh, which is about a 2.1. So those are all things I can actively measure because it's linked to formats and placements and things. Now, that um those only add up to maybe, I don't know, 13, 14% and we're growing nearly 30. So, where is the other growth coming from? Now, there may be a little bit of meta growing users and or our advertiser panel gaining market share, but the other one I can point out to um on these existing tail wins is is uh price. Now, um and the CPM growth rate is accelerating. So, to give you some data points, I'll I'll go in quarters where I've got monthlys to um with a bit more detail, but in the first quarter, across all platforms, so like I said, basically meta and Instagram newsfeed stories and reels, the CPM growth rate uh was uh let me get the right number for you. Um, where is it here? It's I've got a very big spreadsheet. Um, it was 5.3% year-on-year. In the second quarter, it was 11.6, and in the third quarter, it'll be something like 17, uh, maybe 16, but it's it's going to be significantly higher. Now, some of that increase in price will be um driven by the stuff I've already mentioned. A lot of those new ad formats like click to messenger or shopping have much higher CPMs, but that's not the whole story. I think the underlying story is meta keep improving their algorithms, their use of data, their use of AI for improved personalization of ads, improved click through rates, etc. conversion rates, and that does drive up price. Um, and as a simple example, um, they are their system is becoming more nimble. So when they notice someone's in the market for a particular product, um, you know, let's say it's um a dog bed or whatever. They the algorithms will tweak in real time to show more dog beds and actually increase ad load, uh because they've said, oh, this person's engaging with ads on for dog beds way more than others. Um, and and they've kind of like they're almost like saying when when someone, actually what they've said to me is when someone clicks on or dwells on an ad more than the content, the whole algorithm changes to say let's help that person buy that product. Whereas in the past, you just kept scrolling and kept seeing ads and yes, the ads would gradually improve, but it wasn't like a double down and try to sell something. Um, now there's two new, so that's a lot of existing tail wins. I think those will all continue. as I think the real tail wind next year will probably ease off, the rest will continue. The two new ones, one I can measure, one is um I will be able to soon. So the one I can measure is um the Facebook audience network or the Meta audience network. So these are the ads that appear typically in gaming content elsewhere on the internet. Um, now for two or three years, maybe actually maybe four years, the meta audience network was flatlining for us at about 2.1% of our spend. Uh some quarters 2.2, some you know 2.0 but about 2.1. Since April, it's been picking up quite quickly. It's now 2.6. Now that may not sound like a lot, half a point, but A, it's all incremental, but B, it is an indicator that meta is maybe um increasing its fight back against um Apploving in that space of gaming ads. So that's kind of interesting. Don't know how far it will go. I haven't given it enough thought yet, but that's interesting. And the final one I'll add, um which I'm sure we will go deeper on. is that we're starting to see quite rapid adoption amongst small businesses for meta business agents. or meta business agent. And that, uh, once advertisers get used to it and like it, they do see an improvement in ROI. And many of them are starting to increase their budgets faster than normal. Uh, now that's early days because really it only rolled out properly in June. Um, so we're three months in. But we've gone from uh of our 300 or so advertisers, we had about 30 on the beta test up until May. Between May and September, we've gone up to 60. But also there's way more functionality now than there was back in May. Um, but there is a sequence to adopting it. So typically step one for an advertiser is they incorporate their data into meta business agent, which is a great benefit for them, but doesn't instantly increase their spend. Then what happens is Meta's algorithms kick in and improve your ROI on your ad spend. That then can lead to an increase in spend. But then you can opt in to turn on meta um business messaging, i.e. WhatsApp for business. And then you start seeing even more sales and that's often when the budget's get increased. So, I'm sure we can I can give a lot more detail on that, but that is the underlying strength. Like Meta just has a lot of tail winds, and more of them are getting better than getting worse. Mm-hm. Yes. 好嘞，然后，哦对，然后我们今天的话可能会把话题的重点的话，更多的放在Meta的整个的muse上面。 对，然后想问一下就是，我们看到就是整个muse的话，在美国的话其实，呃，最近的话其实就是整个下载量非常的高，然后专家认为muse的话为什么就是在美国，包括像加拿大这些地区的话，会那么受欢迎呢？ 对。 Okay, and for thank you for the answer. and today's focus may mainly on Muse. Um because we notice that the downloads uh of Muse in the US and also Canada is extremely high. So what do you think are the reasons for its uh success and the popularities. Yeah. So, um I I'm impressed too. Um, I think it's both a good product and obviously they have access to meta's distribution and awareness. But so here's my my take and I think it's slight step back. I think it's important. One thing, um, it's easy to assume if you live in a bubble, which I certainly do, most people I know are very tech savvy. They're either techies or investors frankly. Um, and I'm guessing you guys may live in a similar bubble. But it's very easy to assume that everyone out there is using AI in all its powerful forms, you know, to manage their day, organize their email, you know, chase bills, you know, manage the tax, um, you know, receipt management, whatever it might be. You know, I've got like 70 or 80 agents working for me every day. Um, but most people aren't doing that. So, what do I mean by that? So I I do a lot of keynote speaking around the world and one of the things I do with every audience is say, what's your current relationship with AI? And I give them four choices. Choice one is I'm not using AI much. It's normally five or 10% of the room. Choice two is I'm using generic AI but I'm not training it, i.e. they're using chat GPT as a glorified search engine. That's about 60% of the room. Then the third choice is I'm building agents and having and and they're making my improving my life. Uh, that's like 20 to 25%. And then the final choice is I'm building agentic workflows and measuring the ROI. In a room of 100 people, that will be between zero and two people. So effectively, 70 plus percent of people do not use AI to manage their lives, to organize their email, to manage their calendar or whatever. They're using it as a glorified Google basically, you know, with a little bit more, you know, summarize this message, drop me an email, etc. Muse is going out to the market, not saying we're a glorified search engine. They don't talk about that at all. They say we can manage your life. We can, you know, monitor your emails, we can find you discounts, we can check your calendar, we can shop around for you, be your personal shopper. And I think it's just a totally different positioning that we haven't really seen from chat GPT or Claude and to a certain extent Gemini. Gemini have said things like that to the developer conference at IO, but they haven't said it to consumers. Um, so I think it's just a whole new positioning. And if you think about sort of platform iterations and platform change, why did chat GPT do better than Google for a while? because they said, hey, you know, we we are a step up and the world understood that step up. I think Muse is saying, this is a step up, this is different. So they they're playing a different game. Uh so I think that's the the positioning and the timing, I think is good. Um, yeah, it would be easy on the surface to say, I know people have a habit with chat GPT or Gemini or Claude. I think for a lot of consumers, especially people on their phone, like middle class moms, um who maybe don't have much time on a laptop every day, Muse is awesome for what they need. So this is definitely a consumer product, not a B2B product, and you know, Claude probably isn't worried, but Chat GPT and Google should be worried. um because it's good. It just it works in meets the timing and the whole positioning is fresh. Um, so yeah, and obviously to add to that, you you can see that they have Meta's distribution and I think that Meta's distribution will be pretty powerful. So something, for example, and they haven't done this yet, but this is me hopefully joining the dots. What something Meta I mentioned briefly in the opening part of the call, when they see you're in the market for something, they they basically start showing you a higher ad load with more ads. A very simple example, you know, the summer's kind of ending, it's getting a little bit chilly in London. I thought I wanted some some Scandinavian slippers and at some stage I saw an ad and I clicked on it. Meta started showing me more and more ads for Scandinavian sort of wool slippers. Um, and because I was engaging with it more than normal content. Now, in the end I did actually buy them from an ad. But I think a really good handover would be at some stage to say, hey, it looks like you're in the market for Scandinavian slippers, would you like Muse to be your personal shopper and bring you a selection? It would be such a natural handover, it could work. And it wouldn't surprise if Meta could do things like that with Facebook and Instagram, maybe WhatsApp one day, that's slightly more sensitive, to actually hand over to Muse when it looks like someone's, you know, got a a complex task to conduct. So, overall, um, I'm very bullish. Um, I think it's, you know, a great product. Um, it's not late to the party and I think the positioning in particular is just really fresh. OK,好的,然后我想问一下那么专家的话,认为就是Muse这个产品的话,它对于Meta的整个广告收入可能会带来什么样的影响?包括对于广告主的话,可能会带来一个什么样的价值呢? Okay, next question. So, what do you think uh will be the major impact on the ad revenue brought by Muse? And also what kind of values uh Muse can bring to the advertisers? Yeah. Yeah. So that's a great question. The first thing says I think it will be slow. Um, Now I I know the whole world's accelerating really fast, but for this one, short-term meta doesn't need the money. I think they might play their old playbook in the short term of just make it. useful as possible. Don't even worry about money. If you look at, uh obviously WhatsApp's a great example where they made no money for for for a decade or whatever. Um, but even things like um uh stories or reals, they took nine to 12 months with very few ads and then they started weaving in just um low um risk sort of brand ads and then they started picking up performance ads until the end, you know, they they they get four or five years later they get to full monetization. So it wouldn't surprise me if meta say, let's give this 6 to 12 months with no monetization and then we'll start um building it up and we'll take four or five years. Um now what could they do? Um the the simple and easy one, but not necessarily the best one, would be to um just start squeezing in ads, you know, so you're you're using muse and um there you know, there's real estate on the page to slip in ads for things that you think are relevant. And clearly using that they've got access to meta's whole uh purchase intent diagram, you know, they they know a lot is in the market for what. They they absolutely could do that. So for example, you know, muse is helping organize your calendar and it could say, hey, I you know, I've noticed that you've got nothing planned on Saturday for the kids. Um here's an interesting kids club or whatever. Some things ads that are kind of relevant. So I wouldn't rule that out. It's like like a good EA saying, you know, how can I proactively help, you know, my boss, as it were, have a better life. So I think there will be ads like that. Um the another major one is I do think news may end up becoming a lot of people's personal shopper because a lot of their positioning is, let me help you shop around. Let me help you find a deal. Let me monitor for discount codes for that product and I'll tell you when I find them. And so in that scenario, um in many ways they always want to give the best answer. So, um they could try to have options between brands, but it's hard if you're acting as a personal shopper, but what they could do is um have a percent take a percentage of GMV when they finally sell something. So that would be it's now increasingly accepted. Open AI and Google both have a product that do that. So I think, you know, meta could do that. They don't know, but they could. Um, I do think the whole paying for it is an interesting one. I I know they have, you know, a free and pay tier now, etc. but I think once people get hooked, I think charging, you know, will um be possible. Maybe not for everyone. I consumers are addicted to free products, but for the real power users, the middle class moms who say, oh my this product saves me half an hour a day or an hour a day. you know, charging $20, $30 a month probably is possible. Um, so I think there's some the the the other thing linked to percentage of GMV but different is you may have seen that meta is getting into payments. So they now have meta payments in about five countries in the world and the CEO of Stripe has joined the board of Meta, and the meta team have told us that you know, they are planning to offer payments um for the whole world. Now, that probably takes a few years because there's all sorts of regulatory hurdles. But if they do that, you could also facilitate, you know, member member-to-member or like user-to-user or business-to-user um transactions and just take whatever a percent or a small part of every transaction. That could be interesting too. So there's a lot of things they could do. As we know with meta, the easy thing is ads, and I do think proactively helping people spend their money by recommending ads in a friendly way. So not even just like become putting it out there and hoping people click, but saying, I found this for you, is it interesting? Um, you know, for example, you know, my my wife, maybe even me, you know, if if if she got a proposal from her, you know, from an EA, you know, in the form of muse, hey, here's a, I've noticed you've you're planning a dinner on Friday, you haven't booked a restaurant, would you like some suggestions? She'd click yes. And especially if the first experience is good, she might go back and say, hey, can you look through my calendar, find any other times I've blocked, you know, dinner or lunch out, and help me find some choices. So I think there's a whole load of ways, but I think the easy ones will be proactive ads and probably payments. But I think taking a percentage of GMV in the e-commerce model could work, and I think uh subscriptions could work too. 明白。 那想问您一下,就是刚才,首先我可能有几个问题了,首先就是,刚才专家先讲到了那个电商,对,然后我们有注意到其实去年GPT的话也做过电商,包括像那个谷歌的话也在推自己的UCP的一个协议,那么像现在这种agent的一个电商和过去谷歌和和GPT正在做的电商的话,它会有什么,会有很多本质上的一个区别吗?因为这两个的话其实现在的整个的一个规模也没有做,也没有做的特别的大,对。 Okay, so next question, you mentioned about e-commerce, and we know that last year, chat GPT and also uh has been doing their own e-commerce and Google has uh released their UCP protocols. Um, but they have not grow much like the size or the scale is quite limited. Yeah. So, what is the major difference here between those e-commerce practices with the, like the e-commerce agents? Yeah. Yeah. So, um, I'll I'll I'll take those one by one. With, um, because they're slightly different. So Open AI's um product, they're not growing like they they have just hit a a $1 billion run rate, you know, some of which is with the e-commerce. They they don't break it out, I don't think. Um, but it's, it's not great what they offer. In fact, the first product broke. Uh the one now is better, but so if I talk about what they you don't want to do, um, Open AI's first product, effectively, if if you saw an ad, so let's say you were looking for I know running shoes, and you saw an ad, you they wanted you to click a button and say buy now, and literally the next stage was the checkout page with a buy now button, pick a credit card button. That's way too scary and shocking for consumers. Uh if you think about the normal e-commerce checkout process, you often have add to basket as well as buy now, but even on the buy now on the landing page, you are told things like there there are links to the terms and conditions, the delivery times, you can increase the volume or decrease the volume. I I want two of them or, you know, I I I want to take it out to the basket. Um, you get all these sort of assurances that if this, you know, this won't go wrong. And Open AI had none of that. It was literally buy now, and it didn't work. Like no one clicked on it. They also had the additional problem that a lot of ad recommendations were bad. So just on the consumer side, you know, they would search for a running shoe, but they'd be offered more of a fashion running shoe rather than a jogging shoe because OpenAI didn't do the homework. So there were a mixture of bad targeting and a really simplistic product. Now, where they're going now, which at the moment at least makes much more sense is, you know, don't try and drive an instant purchase, take people to the landing page of the e-commerce website, but effectively suck it in. So it's it's like an iframe where it brings the the the the checkout page into ChatGPT. Um, and then, um, that way they're getting all the assurances that this is, you know, Amazon or Walmart or whoever it might be, um, and therefore I can trust it. That seems to be working a lot better. Um, so it is working now, but something I would actually I'll I'll switch to Google then I'll come back a little bit. So Google's product is much better. They realized from the beginning that, you know, having a very simplistic buy now button with no explanations wouldn't work. So effectively, they've built it in so that if you do buy, you you know you're still buying from which merchant. You know it's the link, but it's kind of a seamless process. I don't really know why Google isn't scaling it more. It is scaling, but I think they're almost still thinking keep improving the, you know, the Gemini user experience rather than, um, obsessing over ads. So that's probably an opportunity for Google down the line. So if we take those lessons and say, what does that mean for Meta? If I was Meta, I wouldn't have a simplistic buy now button, um, with no explanations. I think it wouldn't work. I think you do still need to take people to the landing page of the merchant, even if it's sucked into an iframe. So make it as easy as possible. You don't feel like you're clicking around in an unfamiliar territory, but you are buying from a real merchant with real accountability and real delivery times and returns processes and all of that. Um, I think that would help a lot. Uh things like the integration with Shopify will help, you know, most people trust Shopify stores because they're well run and and what have you. Um, so I think where how will Meta maybe roll that out as well as or ideally better than OpenAI and Google. I think, A, skip the whole experimentation with instant checkout. Um, there's a survey by the way, I I don't know if you guys have read it. I forget who it's from. Um, that might have been Visa or someone, one of the payment systems, but it was something like 65% of American consumers will happily let a chatbot do research for them, but only 14% of American consumers would happily say to an agent, buy it without me, without me understanding. So almost all consumers want a check page to say, let me see what you're recommending, let me check the terms and conditions, the returns policy, the price, um, and then I want to click yes, buy it now. I don't want to computer doing that on my behalf. So I think that's the key thing they will need to do. The the one thing I'll add that Meta has a big advantage over um, OpenAI, not necessarily over Google, but Mata has so many more merchants. So it can A, make better recommendations for what people should buy, but B, um, it can um, it it it can already build a more seamless integration, um, as it especially with Shopify, I don't want to obsess over Shopify, but you know, to help people buy something or book an appointment if it's B2B, um, or or things like that. So, yeah, I think overall, OpenAI made loads of mistakes in public, which is great. I think Meta will not make those. They may be able to do better than Google, but I actually think Google's product is a really good one. So I I'm not sure why they're not rolling it out faster than they are. all of my advertisers that we work with it are happy. 明白。 啊对，然后想问一下就是专家是否认为就是整个agent的一个电商或者说其他的一种各种服务的一些购买的方式其实是和整个chatbot的阶段是其实不太一样的，因为在chatbot的阶段的话，其实是人的话，他其实根据他了解到的一个答案的话，最后去搜索，然后的话，跳到网 页上面的话去完成购买整个的闭环,但是在A政子这个阶段的话是A政子的话, 就是了解你的一个意图之后的话,他直接完成了整个搜索以及购买的一个闭环。 那其实的话就是说,以前的话很多的信息的聚聚合商啊,包括像很多的像那个Expedia, booking啊,包括像Ddash啊这些的话,就是 他们其实是作为一个入口,然后的话他们可以通过这种形式去收一个广告费的。 对,然后包括这个广告费很多时候它是以竞价排名那个形式。 对,那么首先就是在A政策的一个它的一个代替客户去做各种购买决策的时候,首先我想问一下就是, A政子是否会替代这些中介,它成为一个第一入口? 对,然后就是说,如果说它变成了一个主要的入口的话,那么过去很多的广告主,他们的这种竞价排名的系统, 在A政策这里是否还是有效的?因为人他其实是说 我会看谁哪个产品,哪个广告排在最前面,我更愿意去买,因为人是有有一个惰,有一个惰性的,但A政治它不一定,它不一定会买排在第一个的商品。 对吧,因为它可能会做各种各样的比较,那么这里的话广告的价值它是否会有一些变化呢?想问一下专家。 Okay, next question. So we noticed that there might be some differences. I'm not sure if you agree. Uh like the traditional way of the e-commerce practice on the chatbot where uh based on your on your inquiries and the answers, then the agent uh the chatbot give you some advices and you still click the search and the complete the checkout. So that's the traditional way. And for this agent e-commerce, it's it does the research for you. So traditionally, we have the information aggregators like Expedia booking. Yeah. And they they are mainly like a gateway where you are looking through all those options, and they also have this auction-based uh bidding or ranking systems. So, because human beings tend to give, you know, tend to buy those things that rank top uh more, right? But not necessarily for the agents. So, uh what do you think uh that could have like an impact on those advertisers, like, does the agent possibly replace those uh gateway websites and also the in in like the innate value of the advertising, like the auction based might also be changed as a result. Yeah, two very good and related questions. Um, so I think it's going to be a very scary and probably um um depressing time to be a let's call it an aggregator or a marketplace in the next five or 10 years. Yeah, and and you could assume that's Amazon or booking or whatever, and especially like the not so well run what companies. Because if you think about it and start with the user experience, I genuinely believe and I have done now for a couple of years that overtime every single person with a smartphone will have at least one favorite chatbot. And some people will have multiple ones, they have a different chatbot for health versus shopping. Some people will just have one. And so and that chatbot, whether it's chat GPT or Gemini or Claude, or you know, Muse or, you know, something that Elon Musk will build at some stage. Yeah, I think we will have one. And that, you know, particular um chatbot will have access to our calendar and our email, possibly our bank account, you know, in my case, my my WhatsApp, my, you know, almost every website I log into, etc. So you really can go deep and there's a real amount of trust and understanding. And in that world, your chatbot will become your personal shopper in most instances. And so if you then think, well, what does that mean? Number one is the chatbot already knows you pretty well. Um, like, like, my chatbot knows all my kids' names, birthdays, whatever, my wife and I's preferences for restaurants, all sorts. And so it just makes, it gets, it makes better and better recommendations just by knowing you well. But it's also because it's a conversational relationship, it can ask you questions in really easy ways to help you to sort and filter. You know, whether you're buying physical goods or hotel rooms or restaurants or whatever it might be. Um, and then if you say, well, if your chatbot gets you down to a great short list, and you say, thanks for doing all the thinking for my restaurant on a Friday night, please book, you know, option B. The obvious next step is to go straight from the for the chatbot to go either to the restaurant guide aggregator just to book, bypassing any, you know, auction page and and and what have you, or if the website's good enough, go straight to the merchant, straight to the restaurant, straight to the shop owner, straight to the, you know, um, the the the business service provider. So I think for those businesses that are merchants or whatever, who have great websites with really intuitive experiences, most chatbots will want to bypass any aggregator and go straight to those. Um, you know, let's say I'll I'll make it up, but let's say if you're looking at a hotel, you might say, you know, do you not the four seasons checkout experience is amazing. So I'll go straight from the chatbot to four seasons. Now there might be a badly run, you know, a family run, you know, restaurant with a bad website, you might book that through booking.com, uh, because it might just be easier. So I don't think this is going to kill most aggregators, but I think it's going to increasingly peel off any traffic where it makes more sense for the chatbot to go straight to the end merchant. Um, now, coming back to, I think you're also asking about the ad auction. If you think about those aggregators like Amazon and Booking and others, they do make a ton of money by having an auction. Um, you know, so someone searches for, I don't know, running shoe for a half marathon on Amazon or any other website, and brands, you know, compete with each other to get page one. That does get tricky because if the chatbot is working on behalf of the user, you don't want to see most of those. So it's kind of a false, um, um, auction to say, hey, I'm you 100 brands for running shoes, who wants to be on page one, because realistically, you should have sorted and filtered all the products and services that weren't suitable. So they shouldn't even be in the auction. Uh so the risk would be less auction pressure because you've done a better job of sorting and filtering. So for to to how so how does any chatbot make money if you're trying to get the the your your boss, the user, to have the perfect product. I think there's a couple of ways. Number one is, I think you can still have an auction, but you won't show 30 um, advertisers in an auction. You might show three. I think consumers ultimately still want choice, they just don't want hassle or time wasting. So you could say in in many auctions, I will bring you the three best options and you can make the final decision. So there's still an opportunity for an auction. And even though there may not be much traditional auction pressure, those three brands will know that this is going to be a high conversion rate. This person's in the market, there's a, you know, 80% chance they purchase, and you and and if you're on the short of three, that means you've got a 27% chance of a purchase. You're happy to pay pretty good money. So that's one way of doing it. The alternative is to take a percentage of GMV, um, and it might vary by margin. Now, what Google is is playing with and I think Meta could learn from this is they effectively say to advertisers, your auction bid can be a percentage of GMV. So let's say for example, you have 50% margins, you might say I'm happy to pay 20% of GMV. That's still a profitable sale. If you have 10% margins, you might say, I'll be I'll happily I'm happy to pay a 3% percentage of GMV. And so, effectively, if we take hotels as an example, most hotels, you know, with booking.com, pay between 15 and 20% of GMV to booking. Um, and so what, you know, um, the chatbot, Google whatever meta can do, is they can say, okay, some people will buy the cheapest product, um, but but that's okay. I don't want to block it out because they can't afford the auction, but I'll still get a percentage of GMV. So someone buys something that's say 10% cheaper than the alternatives, but you're still going to get X percent of GMV. So I've slightly waffled a bit there, but hopefully that's clear. I do think it's a terrible time to be an aggregator the next five years, um, but I do think there's ways for whoever owns the user experience to make good money through a combination of an auction for the shortest of three or capturing a percentage of GMV or a bit of both. 明白，明白，OK，然后，对，然后我，对，然后我还想问一下，就是，呃，这个事情的话，它对于谷歌的一个影响，对，因为过去的话，其实整个的谷歌的搜索和浏览器，其实是互联网的第一入口，对吧，然后如果说现在Agent它变成一个第一入口的话，这个对谷歌的影响是什么样的？因为我觉得，就是，人在互联网上搜索，其实一共有两种嘛，一种是无意识的搜索，比如说我刷短短短，我刷短视频也好，有的时候我去淘宝上逛一些购物也好，有的时候我不知道自己要要要买些什么，那这种搜索我觉得其实不会受到Agent很大的影响，但还有一种搜索是我明确知道自己要干嘛，我有一个很强的目的，我带着问题去搜索的，那这种搜索其实过去大部分是被谷歌，它它是作为第一个低入口的，包括很多谷歌上面比较之前的Query，包括旅游，包括像很多的购买汽车这种比较和决策，其实是通过谷歌搜索去去去去搞定的，那如果说Amuse的话，它现在开始把Agent这个场景做大，它开始抢占了很多有意识的带，带有这种商业意图的搜索，它会不会对于谷歌的广告，它会形成一些分流或者说是一些，或者说会拿走它一定的份额呢？对。 这样子怎么看？ Okay, okay, next question on the impact on Google ads because typically we have two types of search, like the passive ones when we are watching short videos or when we are going through the e-commerce platforms without a specific purpose or thing to buy. I I don't think that has much impact from the agent side. However, for active search with a very clear questions there, especially when we are trying to make a decision in buying big items like vehicles or determine your travel destinations and so on. Those kind of search usually over the past several decade always occur on Google, like the search engine and browser are like the top one entries. So what could be the possible impact from the agent towards the Google search and also the S revenues. Oh yeah, and this is this is definitely a threat to to to search. Now, you you almost need to go through category by category because it'll be a bigger risk to some. So for example on B2B, I think Google's probably safe from Mews. I doubt, you know, many people do B2B type stuff. Um, but on things like leisure, Um, yeah, I think it will start small. It will start with the cheaper end of re-commerce, like buying stuff for under $100. or leisure, things to do at the weekend, you know, restaurants, kids clubs, that kind of thing. Um, but then it will scale from there. Um, and so, um, yeah, in the short term, will Google see a dent? Probably not. But over time, I would say it'd almost like go through Google vertical by vertical and say which of these could muse peel off some, um, some stuff. And I think if if if you want if you needed a sort of a a clear sort of mental model, imagine middle class mums who are busy juggling kids and work and family and and and you know, uh, life, they spend a lot of their time in the car or traveling on their phone trying to organize stuff. And I think those categories are the most likely to be cannibalized from Google. Um, so it says here leisure is at the top one. Uh possibly some things around fashion and beauty. Again, fashion and beauty is a bit more of a thoughtful thing, but it will start at the cheap end, like a lipstick or a, you know, sort of fast fashion. Um, I so yeah, the ones that will be safe for Google will be B2B, probably government, um, the more serious ends of health and education, where people are shopping around for health insurance or for, you know, um, uh, MBA courses or whatever. Um, but it'll work up from there, but I don't know what percentage of, you know, global GDP is, you know, sort of light touch, relatively low price consumer spending, but it's probably quite a big amount. Um, you know, I don't know what it is, 20% of all spending or something like that. Now, I'm not saying Meta will get 20, but you know, it will get a decent share of that. Um, so that's how I would think about it. Um, I haven't quantified it, but, you know, will Google's revenue be, I don't know, three, four, 5% lower in five years time because of this, probably, assuming it's successful at scale. You know, we still, you know, we don't know for sure that will be the case. Now, Google is rolling out something similar. It's called, um, Gemini Spark. In many ways, it's similar, albeit it's not, um, yet a a stand-alone app. So Google could fight back. Um, so I think we need to have that in our, in our scenarios, but they need to move pretty fast because I think Muse is really, you know, going to take off pretty fast. 呀，呃，OK，然后想问一下就是，呃那么就是如果说，呃，未来agent，个人的agent的话，它成为一个消费者，很多消费者的一个配置的话，尤其是可能它会面向全球可能10亿级别的这种消费者的话，我相信谷歌也很快会推出自己的一个agent的一个产品，那么在，呃，整个的Meta和谷歌，他们自己的agent会有一些竞争的情况下，那专家认为的话，可能哪个，呃，A，可能哪个厂商的话，他们会更加占有优势呢，在消费者这一侧。 对，同时我想很多的，呃，除了应用侧以外，很多的一些手机厂，他们，比如说像苹果对吧，他们也会，他们也在推自己的Siri，啊，那我觉得他们可能也有，也会想做个人的一个AI的一个助手，那么其实想做的人是比较多，然后专家怎么看，未来的话谁会更加占据一个优势呢？ 对。 Okay, so if this Muse, personal agent Muse, uh, really get a large scale, like for example, a billion users using this product, Google is very likely to, uh, launch its own agent. And then it will be a competition between this these two agents from Meta and also Google. Yes. So, uh, in this perspective... which one do you think uh enjoys an upper hand here? And also if we uh put it into a larger picture, where uh on top of those uh companies like Google and Meta, we also have the phone OEMs like uh like Apple and many other phone brands who are also kind of developing their own personal assistant embedded in the phone products. So we can see there are potential a lot of potential players in this personal agent, yeah, arena. So and if we put it in that perspective, who do you think enjoys the larger advantage there? Yeah, Crikey, it's going to be a a fun battle to watch, isn't it? Um, you're right. And and by the way something you probably noticed as well is that you know, obviously Apple seems to be building their offering in partnership with Google, which is kind of interesting. So, you know, Siri at least for now is powered by Gemini. Um, so, um, so so that gives Google a nice nice bit of strength there. Crikey, who's going to win here? Let's think that one through. And ultimately Meta will have to break some old habits. Some people use Siri a lot. Um, you know, some people use Google a lot. And if if if Google um, or when Google let's say launches its own sort of um equivalent, it will get a lot of traction too because obviously Google's got massive distribution. Um, Google is also a natural place to integrate your data from your calendar, your email, your browsing history, your Google Maps history because effectively Google owns all of those. Um, so theoretically, Google, a Google product could be the most helpful here. But they're going to be slow to the party, as it were. they're going to be later than muse. What does muse know there there is something which is I think is deeply held in people's minds which is and you said the words to this effect, Facebook and Instagram are for effectively entertainment. People go there because they're bored. and as long as you can make their lives a bit better, whether it's with a thrill of a video or the excitement of some shopping, that's meta's territory. That's the mindset people are in. When people are doing a serious search, they tend to go to Google or Claude or or Chat GPT, but probably meta will get it less. So if I think that one through, I think meta and muse will be able to probably be the have the biggest market share of stuff related to lifestyle and entertainment. And so as I come back to this like, it's leisure, it's fashion, it's um, short trips to to places. It might be gaming, low-level healthcare, like consumer health stuff. The stuff that's kind of just related to lifestyle and, you know, the things you do and spend money on because you're bored. Whereas Google and probably ChatGPT and Claude on that matter, are more likely to continue to win in the areas of serious purchase. So B2B or I say the serious ends of health care, education, government, um, etc. Now, on balance, that's probably better for meta than it is for Google because right now Google probably hold, I not probably Google does get a lot of searches for relatively trivial stuff. and over time, it will get less of those, um, because effectively, OpenAI sorry, muse, but possibly OpenAI too, um, will likely get more. So if if if the if the key question is who's this a headwind or a tailwind for? I think this is definitely a headwind for Google. Uh probably a headwind for OpenAI um as well. I it's definitely a tailwind for meta. Although one thing I would say is I do think part of Muse's likely success will mitigate another problem that Meta was going to have. And this might sound premature, but certainly I've had a hunch for a while that over the next few years, the amount of time we spend on social media will probably decline. Um, and why do I say that? So, simplistically, our screen time is roughly maxed out, at least in in the US and and and in Europe. And so if you're in a typical 24-hour day, you have let's say 8 hours in bed, 8 hours working, about two hours driving or washing or whatever, but doing other stuff without a screen. And then we have about six hours a day looking at a screen for the sake of entertainment. And the big three buckets are social media, video, and gaming. And they're roughly equal size, I think maybe gaming's a bit smaller, but roughly equal. So the the key thing in my mind is if another form of digital entertainment comes along, someone's going to suffer. Now, the obvious one at the moment is studio created content, like Netflix user engagement is slowing. Um, I think may even be declining, but it's certainly um, people prefer UGC. But over time, what I believe is chatting to your chatbot, messing around with generative AI will effectively be some form of entertainment. There was a survey out a few months ago saying the average American spends 16 minutes a day using OpenAI. Now, some of that will be work, but some of it's leisure. Um, and, you know, people are using chatbots to talk about their mental health, to plan their life, to get feedback, to plan their weekend, to do this that and the other. And I think Muse and Spark and other things will accelerate that. So if we imagine in say, five years time, the average person is spending 30 minutes a day on their chatbot, that's 30 minutes less per day um between social media, video and gaming. So, in some ways, I think Meta had a headwind that was quite scary. You know, would they lose, I don't know, 10% of time on site due to chatbots. And so that was a headwind. I think Muse will probably more than offset that, but you do need to think through those two together. This isn't only upside for Meta, in some ways it's countering a future headwind. Does that make sense? And feel free to challenge me on that, but that will be my take on consumer time. Okay. Right. 好的好的，对，然后想问一下就是 呃，在A，就是Meta如果说有了Muse做一个个人Agent的情况下，我不知道它是否可以比过去的话拿到更多的用户的线索，包括了解更多用户的商业意图，从而的话帮助他自己的一个Facebook和包括Insta上面的一个广告转化率会有一个比较大的帮助呢？然后不知道专家怎么看，对。 Okay, so what uh since now Meta has the muse, the personal uh agent, it's uh fair to say that it now has more access to personal lease and also to have better understanding of the intentions of the customers. And do you think it will help to drive up the conversion rate on on on its ads on Facebook and Instagrams? Yes. Yeah, massively. Um, it's uh, I I I've I think I can kind of help quantify this. So if you think about the old world, by which I mean like two years ago, because things are moving so fast, but in the world where there wasn't much AI or knowledge about the consumer in a search engine or an ad on Facebook or whatever it might be. A typical e-commerce website would have a conversion rate of about 3%. So for every 100 visitors, three people buy something. Now, let's imagine what that could be with AI doing everything it's supposed to do, with it knowing. us really well. Um, you know, with it, um, having seen all our past purchases, with it and understanding our price sensitivity, but also with ads and landing pages being personalized. So, if we take running shoes as an example, two years ago, if I search for a running shoe, I'd just see a bunch of running shoes and a pretty random selection. Now, if I am in the market for a running shoe, every ad should be a picture of a 50-something guy, slightly better looking than me with a better six-pack, running through a London Park or whatever, um, looking happy and handsome and whatever. And the shoes should be from brands that I like, at price points I can buy. And the the talk would be about, yeah, half marathon, Barry's boot camp, um, etc. But you know, every single ad would be personalized to the individual, but so would the product recommendations, the price, the the means of purchase. In that future world, both click through rate and conversion rate will go through the roof. And if we take conversion rates as an example, I think in the future, people won't really visit a merchant site until that merchant is on the short list of three. And so if you then say, what will conversion rates be one day? In theory, if people never change their mind, if they started a search and ended up buying, and if they visit three websites because that's the short list, in theory one day conversion rates will be 33%. Now, in reality, there will be visits as part of the buying journey but you're not ready to buy the same day. There will be visits to uh you know, find a link and share it with a friend, etc. But I can envisage a world where conversion rates one day hit 10%. So that would be three times higher than the old world. Now, even if you think that's a bit aggressive, doubling sounds easy to me in a world dominated by AI. And so what that does mean is that the the the the CPCs that that um companies are happy going to pay will at least double over the next few years and maybe triple. because the simplicity of conversion rate doubles or triples, advertisers are happy to pay more for that traffic. 我我就是 我就是 关注到就是Meta的话，他其实也在做支付，因为其实Meta的所有的整个商业链路里面，它唯一没有闭环的那个环节其实就是支付， 对，这个可能和中国的情况是不太一样的。 对，然后我想问一下就是说，在美国的话，做支付的难度可能有多大， 因为我理解就是Meta通过AI这件事情想做的是说，未来的话很多的网站，包括像像很多的那种电商啊等等，他们的流量的话，可能是AI去提供的， 然后的话，其实未来的话很多的，很多的中间商的话，他们会变成像AI的话去服务， 对，然后的话那Meta的话想从这些交易里面的话获取一部分的那个抽成，对吧， 对，然后，呃，然后但是现在的话这件事情是是给Stripe他们去做的， 对，那我他其实现在也在自建支付，我不知道就是在美，我自建支付这件事情的难度到底有有有那个多大， 对，然后的话专家认为就是说什么时候的话Meta可以自己推出来一个支付系统， 因为我觉得就是现在Meta的地位其实有点像苹果，就是苹果掌握了IOS的一个分发，那其实它可以对很多的流水去收一个苹果税， 那未来的话Meta是不是在整个的一个应用生态里面，它有点，如果说个人AI它真的是最大的话，那是不是有点接近于苹果的这个苹果商店的一个地位， 它其实可以对上面发生的任何的流量的分发也好，对于它的交易也好，都会抽一笔税， 那这个税可能是需要通过支付去完成的，对。 Okay, next question. I think, uh, there are several folks here. Uh, the first one, so in in the current business patterns of Meta, the only thing that's been missing here is the payment. And so in the past uh and especially with the uh join of AI, AI in the in the future may be the target uh for a lot of the advertisers to serve because a lot of the traffic would go through AI agent to the uh end consumers. So, uh currently, Meta has been relying on Stripe for the payment part. But, yeah. And but we we also know that Meta has been building its own uh payment systems. When do you think that would be ready? And second part of the question is, uh just like Apple who uh controls the whole distribution on the iOS system, it can, you know, charge the Apple tax on every transaction happening there. Uh, if Meta really has that like top one market share for personal agent, then does Meta enjoys a similar status here, uh and also he's able it is able to uh charge a commission uh for every transaction here. Yeah, um good questions. So, on the payment side of things, um Meta has rolled out payments in a few countries. I forget which ones, but it's something like, I think it might be the same ones that where they rolled out meta business agents early, but like Mexico, Brazil, Indonesia, maybe India. I think I've heard Saudi Arabia, but there's like three, four, five countries around the world. So they've done it in some, but not the big ones, like US, UK, etc. But I think I've been told by Meta, you know, this is public, they're bringing payments everywhere. Now, I'm not an expert on what it takes to roll out payments, um but Meta being Meta, I think they could probably roll out to the whole world within two or three years. I think it still takes that long because of um just, you know, regulations and stuff. Um, but maybe there's a shortcut, you know, by partnering with Visa or Mastercard or something. I'm again, not my expertise, but I I you know, I think we can assume that Meta will make it work everywhere. It's clearly a big part of their strategy and the fact they've brought the CEO of Stripe onto their board, you know, suggests that uh and two things, I guess. One is that payments are a really serious thing for them and two is that they'll probably end up doing a lot of partnerships with Stripe. Um so that's the that's the first part of your uh question. In terms of um Meta versus Apple, as it were, so who will be able to extract rent from this? Um, I guess that's a very real worry, I guess, for meta, in theory, is that can Apple, you know, charge them a fortune somehow by either um taking a percentage of GMV or having some kind of deal like they have with Google. I think I think Google's rumored to pay Apple like 20 billion a year, um for all their sort of placements in in Apple, being the default search engine, that kind of thing. So, is there a chance that Meta ends up paying a fortune to Apple? Yes. Apple won't be able to be too greedy. The Meta like Google is is incredibly powerful. I think if Apple said we're not having muse, um a lot of consumers once they're hooked, this will be a few years out, would say, well, fine, I'll go and use an Android or some other phone or maybe Meta will even make their own phones one day. Um, so I think it it won't be, you know, um, sort of extreme, but, um, because I think I think Meta again, Meta being Meta, in the short term will make no money. So there's not even a commercial deal to do with Apple. Effectively, this is just a free product for consumers. There's no money for Apple to make. Meta will get people hooked and then start making money, and then Apple will realize that people are hooked and probably try and get their their their their sort of their share of the revenue. Um, exactly how. that would happen I don't know in theory, Meta might allow Apple Pay alongside their own pay because that is convenient for consumers. Apple could make a bit of money that way. Um, you know, could Apple capture a percentage of the subscription revenue if people pay, you know, $20 a month or whatever? Maybe, but my instinct says most people won't do that upon taking the app, they'll do it at a later date. Um, so, I don't I I I'm sure Apple will make at least a few billion a year from this. Um, but I don't know if it'll be like tens of billions because I think for Apple to do that would be singling out meta in a probably anti-competitive way. So I think meta will keep the majority of the upside from any money they make on muse. Uh, but Apple will find a way, um, as say it could be a percentage of subscription revenue which won't be that, it won't be the biggest revenue source for meta. It could be, um, you know, integrating with Apple Pay. Um, I doubt it would be a percentage of ads, that would be too complicated. it could be some just big fees for like making sure that muse is on the recommended app or in the pre-installed app on day one. So there's a few ways for for Apple to make money. I doubt it would hit that 20 billion number because Muse wouldn't be the default search engine, it would just be a very popular app. Um, and that's a sort of different, it's harder to tax, as it were. 哦，然后的话想问一下就是专家的话怎么看就是 Muse 和 Meta 的那个 WhatsApp 的一个融合， 对，然后的话，包括这个他们这个两个软件的一个融合的话，它可能对商家，包括对用户，可能会产生一个什么样的影响？ Okay, uh next question is on your uh perspective on the integration between Muse and WhatsApp and uh if they have successfully integrated, what's the potential uh impact would it have on both merchants and also consumers. Yeah, good question. So, to my knowledge, they haven't done any integration yet. So, um, yeah, but that could come. It's an interesting one because something I thought meta might do at some stage is have, uh, and and this is still possible, by the way, it is have, um, almost like a virtual friend within WhatsApp that you could chat to to discuss your needs. Um, so effectively, you know, could you have your personal shopper or whatever it might be within WhatsApp. Now, something that meta could do at some stage is if you're subscribed to Muse, have Muse appear in your WhatsApp so that you don't need to switch apps to ask Muse a question. You can say, hey Muse, you know, um, I need some activities for my kids on Sunday, ideally outside, um, unless the weather is raining or whatever. And you kind of you could do it and so be another way to interact with Muse that makes Muse more sticky because it can be your best friend in WhatsApp or it can be a standalone app and you can flip between the two. So I think there are ways to do that, likewise within muse. Let's say muse does some personal shopping for you, but at the end you need to ask some questions to the merchant. Muse could flip you into WhatsApp for business. Um, and, you know, you could then chat to a company over WhatsApp or to a company's AI agent over WhatsApp. And obviously meta's building those AI agents too. So meta could end up having both sides of the conversation. You know, there's a a a a chatbot representing the merchant, you know, and then there's a chatbot representing the the human, but Meta could also allow, you know, human to merchant chatbot or human merchant to, you know, human chatbot. So there is a good chance meta ends up controlling both sides of the conversation. They'd need to be very careful in that situation that they don't have a conflict of interest, because obviously merchants want to sell the most for the best margin. Consumers want to get the best deal. Um, so they'd need to have good sort of, um, firewalls or whatever so that both sides do their job properly. Um, but that would be really interesting. Now, if you then look at that and say, well, how does that monetize? It's certainly monetizes on the consumer, oh, sorry, on the on the merchant side, on the on the brand side. They they are proving very happy to pay a few cents per message on WhatsApp. Um, so that could be interesting. Uh, you uh there may be other stuff too. Um, uh, let me think what would be a good one. Something that might happen, but this could just be Meta AI, but it could be Muse. So something Mark Zuckerberg said at a conference back in June is he said at some stage when people are talking about a brand in WhatsApp, that will turn into a live link that you can click on. So what do I mean by that? as, you know, it could this could be a pub or a running shoe or, you know, um, a hotel, whatever. But, you know, a bunch of friends say, hey, where should we meet? Um, you know, does anyone want to go to, I don't know, the antelope, and that turns into a link. Or, you know, you're talking about places to go with some friends and someone says, you know, try the Four Seasons in Bali, or whatever. So you basically, you have the those links appear and then you can click on them. Now, first of all, that's a massive opportunity for Meta because, you know, they can probably one day charge for those any clicks coming from those links. Uh, also a headwind to Google by the way. Um, but alternatively, there could be something saying, get Muse to investigate. Um, or you could say to Muse, Muse could say, hey, I saw on WhatsApp, you know, you again, this is a sensitive one. You'd need to check people don't get spooked out. I saw on WhatsApp you're looking, you know, for somewhere to go with your friends on on Thursday night. Uh, do you want me to research it for you? So I think there's a lot of like cross-hardovers possible between them, and I think merchants paying per message seems to be a very acceptable way to monetize. There probably are others, but I think it's probably maybe a little bit less about ads and a bit more about monetizing the merchant side of the conversation. 我看你，那，好，专家的话认为就是Muse的话可能会给Meta带来有多大的一个潜在的市场空间呢？ 对，他专家认为包括像Meta这两年在AI上的投入的话，能够让它的收入有一个翻，可能翻一倍甚至两倍的一个空间吗？ 对。 收入，收入，你你问的是市场规模还是收入的？ 是是是它的一个收入的一个，它的广告收入的一个空间它有没有可能会有一个很大的一个增长，对。 Okay. So we know that the AI CAPEX of Meta has been quite massive over the several past several years. So do you think, um, the the the ad revenue increase brought by the Muse could potentially justify the CAPEX like can can the ad revenue increase by two or three times? Yeah. So the the quick answer is yes, it definitely can. Now, this is even before Muse. I believe WhatsApp, um, will be at least a 200 billion a year business and it could easily be 3, 4, 500 billion. Um, I'm happy to go into the detail, but but in essence just when you've got 4 billion daily users, it's not hard to make, um, you know, a bit of money. Like, you know, every time someone, um, you know, receives a message from a brand, they could, you know, the brand could pay a few cents and that could happen a few times a day. So, as a simple example, if your average daily user receives 10 messages, sorry, five messages of the day from brands, that's not hard, that could be one conversation.