广告
加载中

双11前品牌密集接入AI店小蜜:AI客服接更多 满意度为何反升

沈嵩男 2026-09-30 15:25
沈嵩男 2026/09/30 15:25

邦小白快读

EN
全文速览

AI客服今年双11前迎来明显升级,消费者接触到的客服不再是死板的问答机器,而是能看懂图片、识别情绪、主动推荐的数字员工。

1. 消费者咨询前可以尽量直接发商品截图或照片,AI能结合图片和文字理解问题,减少转人工等待;即使只说“这个怎么参加”这类不标准表达,AI也能接住。

2. 表达着急或焦虑时,AI会先安抚再回答,比如对新手妈妈说“别担心,刚开始都是这样的呢”,服务感受更像真人。

3. 部分品牌AI会主动询问宝宝月龄等关键信息,按需求推荐产品和优惠,提醒凑单和活动时间。

4. 企业数据显示,AI独立接待比例提升后,满意度不降反升,消费者对AI客服的满意度从89.5%提升到94.5%;涉及投诉升级、医学判断等复杂场景,仍会快速转人工,不用担心被机器敷衍。

品牌营销与用户信任:母婴品牌小皮把AI客服当作建立安全感、专业感的触点,在过敏原、体质适配等敏感场景共创应答边界,宁可转人工也不让AI越界,以此维护品牌信任。

1. AI主动询问宝宝月龄,根据月龄、体质推荐产品,并在优惠凑单时主动提醒,把售前咨询变成顾问式服务。

2. 消费者真正关心的不是便宜几块钱,而是安全感和专业感,品牌应把这一洞察落实到客服话术和策略中。

消费趋势与用户行为观察:消费者提问方式越来越不标准,发截图、图片和口语化问题增多,AI需要具备图文识别和情绪识别能力才能接住。

1. 新手妈妈用户常常表达焦虑,AI识别情绪后先安抚再回答,明显改善服务体验。

2. 售后拍图咨询多与真伪确认、包装变化、冲调结块有关,这类图文识别是品牌提升售后体验的机会点。

产品研发与数据反哺:客服会话中产生的对商品、活动、页面、物流的消费者反馈,应整理成分析并反馈给商品、运营和物流,让页面更清楚、活动机制更完整、流程更顺。

消费需求变化与机会:消费者不再按客服系统希望的格式提问,而是发截图、口语化表达,对情绪安抚和安全专业感要求更高。

1. 使用AI店小蜜高阶版后,小皮机器人询单转化率从25.9%升至29.6%,转人工率下降约13个百分点,AI满意度从89.5%提升到94.5%。

2. 卖家可以借鉴的做法是让AI主动询问关键信息,比如宝宝月龄,按需求推荐产品并提醒优惠,提升销售转化。

双11应对与准备:今年双11备战方式从单兵作战变成与AI产品团队共创策略,提前梳理历史大促问题,做场景模拟预演,把火源找在活动开始前。

1. 力争新品上新当天配置好客服答案,因为AI能从商品详情页自动抓取参数生成问答素材。

2. 售后从被动回答退换查,转为主动提醒和关怀,减少因信息不对称造成的退款、差评和纠纷。

风险提示与边界:AI能识别情绪不等于可以独自处理所有情绪,涉及过敏原、医学判断、敏感公共议题、投诉升级或高风险表达时,要设置红线马上转人工;不要为了省成本让AI往前走多一步。

产品设计需求:小家电消费者收到产品后不会安装,AI能识别用户着急的状态,先安抚再给安装步骤,并把其他消费者容易卡住的细节提前说出来。

1. 工厂可优化说明书、安装视频和商品页信息,把易卡点写清楚,减少售后咨询量。

商品信息与包装设计:母婴辅食消费者关注过敏原、体质适配,包装变化会引发真伪质询,冲调结块、果泥颜色差异会带来拍图提问。

1. 工厂在包装、标签、详情页上应做到信息清晰、升级有提示,降低消费者的不安全感。

数字化与电商启示:AI可从商品详情页自动抓取产品参数、卖点、配件信息和常见问题,批量生成问答素材,力争新品上新当天客服答案配置完成。

1. 工厂需要提供结构化、可被机器读取的商品数据,才能让AI客服快速响应新品。

2. 客服会话中产生的商品和物流反馈应整理成分析,反哺生产设计和流程优化,实现数据驱动改进。

行业趋势与客户痛点:AI客服正从FAQ工具升级为能看图、懂情绪、会主动推荐的数字员工,客户痛点集中在客服成本高、新品维护重、大促准备复杂。

1. 客服相关销售约占整体销售30%到35%,AI接走基础接待后,更关键的是人机一起把交易留在客服链路。

2. 小皮数据中,机器人询单转化率上升3.7个百分点,转人工率下降约13个百分点,说明AI可承担更深服务。

解决方案:AI店小蜜高阶版支持图文联合识别、情绪识别和主动服务,还能自动抓取商品详情页生成问答素材。

1. 把客服团队拆成AI数字员工、AI训练师、分析策略师和精英客服,让基础接待更便宜,高质量服务和策略设计更值钱。

2. 精英客服的高转化话术和复杂服务经验沉淀为机器可执行规则,人的价值从亲自回复变成生产好经验。

3. 敏感公共议题、情绪升级、投诉或高风险表达需设置边界并立即人工介入。

数据与双11:多数部署品牌的人机客服满意度达92%以上,部分高端品牌超过95%,12小时解决率达87%以上。

1. 双11前把商品、运营、客服、物流拉通,用历史问题做场景模拟预演,减少事后救火。

2. 将会话数据整理成VOC分析,反馈给商品、运营和物流,形成数据闭环。

平台最新做法:阿里AI店小蜜推出高阶版和Agent版,通过双11备战工作坊、蜂芒奖、服务商共创计划等方式支持商家。

1. 平台组织备战工作坊,让商家与AI产品团队一起搭策略,明确产品能力和模型边界,不再把AI当黑箱。

2. 百秋作为阿里AI店小蜜服务商首家共创伙伴,反映平台重视标杆共创和生态建设。

商业对平台的需求和问题:商家需要平台提供能接住“不标准问题”的能力,包括图文识别、情绪识别、主动服务,以及从商品详情页自动生成问答的工具。

1. 商家担心AI接得越多体验变差,平台需要用数据证明:AI自身满意度从89.5%提升到94.5%,人机客服满意度普遍在92%以上。

2. 商家要求AI在敏感场景有边界,能快速转人工;平台需要在产品机制上支持投诉升级、高风险表达识别和人工接管。

运营管理与风向规避:平台可引导商家把客服从成本中心变成销售中心,通过人机客服销售占比提升5个百分点以上来体现价值。

1. 双11前平台应提供场景模拟预演、售后主动关怀、订单工单联动等功能,帮助商家减少退款、差评和纠纷。

2. 平台应鼓励将聊天数据整理为分析反馈给商品、运营和物流,形成全链路优化。

产业新动向:电商客服产业正从单纯人工服务转向人机协同,AI独立承接比例提升,满意度反而上升;客服岗位重构为AI数字员工、AI训练师、分析策略师和精英客服等新角色。

1. 小皮和百秋的案例显示,AI客服可以提升询单转化率、降低转人工率,同时将客服部门从成本中心推向销售中心和数据中心。

新问题与治理:当AI开始接住更多顾客,人的价值需要被重新定义;企业发现相关客服人数阶段性减少,但总人力成本没有同步下降,薪酬结构和能力要求发生改变。

1. 优秀客服的隐性经验必须变成机器可理解和执行的规则,经验萃取成为新的组织能力。

2. AI能识别情绪不等于应该独自处理所有情绪,企业主动设置红线和人工接管机制,涉及医学判断、敏感公共议题、投诉升级时必须转人工,这为AI服务治理提供了参照。

商业模式与数据闭环:服务商可将客服定位为销售中心,人机客服销售占比比纯人工模式提升5个百分点以上,纯机器人销售可占全店5%到8%,证明AI投入有明确回报。

1. 平台与服务商共创(如百秋成为首家共创伙伴)形成新型合作模式。

2. 会话数据被整理成VOC分析反馈给商品、运营和物流,形成从客服到全链路优化的数据闭环。

3. 双11备战方式从单兵作战变为多方共创和前置预演,是运营模式创新的方向。

返回默认

声明:快读内容全程由AI生成,请注意甄别信息。如您发现问题,请发送邮件至 run@ebrun.com 。

我是 品牌商 卖家 工厂 服务商 平台商 研究者 帮我再读一遍。

Quick Summary

AI customer service has seen a notable upgrade ahead of this year's Double 11. Consumers now interact with digital agents that can understand images, recognize emotions, and make proactive recommendations, rather than rigid Q&A bots.

1. Before asking, consumers can send product screenshots or photos directly. The AI combines visual and textual input to understand the question, reducing wait time for human transfer. Even non-standard queries like "how do I join this?" are handled seamlessly.

2. When users express urgency or anxiety, the AI first offers reassurance before answering. For example, it may tell a new mother, "Don't worry, it's always like this in the beginning," making the experience feel more human.

3. Some brand AIs proactively ask for key details such as baby's age in months, then recommend products and offers based on needs, reminding users about threshold-breaking purchases and campaign timing.

4. Enterprise data shows that as the proportion of AI-handled interactions increases, satisfaction rises rather than falls. Consumer satisfaction with AI customer service improved from 89.5% to 94.5%. In complex scenarios such as complaint escalation or medical judgment, the system quickly transfers to a human agent, so there is no concern about being brushed off by a machine.

Brand Marketing and User Trust: Little Freddie (a maternal and infant brand) treats AI customer service as a touchpoint for building a sense of safety and professionalism. In sensitive scenarios such as allergens and physical suitability, they co-create response boundaries with AI, preferring to transfer to human agents rather than let the AI overstep, thereby protecting brand trust.

1. The AI proactively asks about the baby's age in months, recommends products based on age and physical condition, and reminds users about promotional threshold-breaking purchases, turning pre-sales consultation into advisory service.

2. What consumers really care about is not saving a few yuan but feeling safe and professionally served. Brands should translate this insight into customer service scripts and strategies.

Consumer Trends and Behavioral Observations: Consumer questions are increasingly non-standard, with more screenshots, images, and colloquial phrasing. AI needs multimodal (image and text) recognition and emotion recognition to handle these.

1. New mothers often express anxiety; AI first recognizes the emotion, offers reassurance, then answers, significantly improving the service experience.

2. After-sales image-based inquiries often involve authenticity confirmation, packaging changes, or clumping during formula preparation. Such image-text recognition is an opportunity for brands to enhance the after-sales experience.

Product Development and Data Feedback: Consumer feedback on products, promotions, pages, and logistics generated in customer service conversations should be compiled into analyses and fed back to product, operations, and logistics teams to make pages clearer, campaign mechanics more complete, and processes smoother.

Changes in Consumer Demand and Opportunities: Consumers no longer ask questions in the format preferred by customer service systems; they send screenshots, use colloquial expressions, and demand higher emotional reassurance and a sense of safety and professionalism.

1. After adopting the advanced version of AI Dianxiaomi (Alibaba's AI customer service agent), Little Freddie's bot inquiry-to-order conversion rate rose from 25.9% to 29.6%, human handoff rate dropped by about 13 percentage points, and AI satisfaction climbed from 89.5% to 94.5%.

2. A practice sellers can adopt is to have the AI proactively ask for key details, such as baby's age in months, recommend products based on needs, and remind users of offers, thereby improving sales conversion.

Double 11 Preparation and Response: This year's Double 11 preparation shifted from working alone to co-creating strategies with the AI product team. Historical issues from past promotions were reviewed in advance, and scenario simulations were run to identify potential pain points before the campaign started.

1. The goal is to configure customer service answers on the same day a new product launches, because the AI can automatically extract parameters from the product detail page to generate Q&A material.

2. After-sales service moves from passively responding to returns, exchanges, and order status to proactively reminding and caring for customers, reducing refunds, negative reviews, and disputes caused by information asymmetry.

Risk Warning and Boundaries: AI's ability to recognize emotions does not mean it should handle all emotions independently. In cases involving allergens, medical judgment, sensitive public topics, complaint escalation, or high-risk expressions, red lines must be set and the matter transferred to a human immediately. To save costs, do not let the AI take one step too far.

Product Design Needs: For small home appliances, consumers may not know how to install the product after receiving it. The AI can recognize the user's anxious state, offer reassurance first, then provide installation steps, and proactively mention details that other consumers often get stuck on.

1. Factories can optimize manuals, installation videos, and product page information to clearly explain potential sticking points, reducing after-sales inquiries.

Product Information and Packaging Design: Consumers of maternal and infant complementary food pay attention to allergens and physical suitability. Packaging changes can trigger authenticity questions, and clumping during formula preparation or color differences in puree may prompt photo-based inquiries.

1. Factories should ensure clarity in packaging, labels, and detail pages, and provide notices for upgrades, to reduce consumer insecurity.

Digitalization and E-commerce Insights: The AI can automatically extract product parameters, selling points, accessory information, and common questions from the product detail page, batch-generating Q&A material, aiming to complete customer service answer configuration on the day a new product launches.

1. Factories need to provide structured, machine-readable product data so that AI customer service can quickly respond to new products.

2. Product and logistics feedback from customer service conversations should be compiled into analyses and fed back to production design and process optimization, enabling data-driven improvement.

Industry Trends and Client Pain Points: AI customer service is upgrading from an FAQ tool to a digital employee that can read images, understand emotions, and proactively recommend. Client pain points mainly center on high customer service costs, heavy maintenance of new products, and complex promotion preparation.

1. Sales related to customer service account for about 30% to 35% of total sales. After AI takes over basic inquiries, the key is to keep transactions within the customer service chain through human-machine collaboration.

2. In Little Freddie's data, the bot inquiry-to-order conversion rate rose by 3.7 percentage points, and the human handoff rate dropped by about 13 percentage points, indicating that AI can handle deeper service.

Solution: The advanced version of AI Dianxiaomi supports integrated image-text recognition, emotion recognition, and proactive service, and can automatically extract Q&A material from product detail pages.

1. Split the customer service team into AI digital employees, AI trainers, analyst strategists, and elite customer service agents, making basic handling cheaper and high-quality service and strategy design more valuable.

2. The high-conversion scripts and complex service experience of elite agents are distilled into machine-executable rules; human value shifts from personally responding to producing good experience.

3. Sensitive public issues, emotional escalation, complaints, or high-risk expressions require setting boundaries and immediate human intervention.

Data and Double 11: Most brands that have deployed human-machine customer service achieve satisfaction above 92%, some high-end brands exceed 95%, and the 12-hour resolution rate is above 87%.

1. Before Double 11, align product, operations, customer service, and logistics teams, use historical issues to run scenario simulations, and reduce firefighting after the event.

2. Organize conversation data into VOC analysis and feed it back to product, operations, and logistics to form a closed data loop.

Latest Platform Moves: Alibaba's AI Dianxiaomi launched an advanced version and an Agent version, supporting merchants through Double 11 preparation workshops, the Fengmang Award, and service provider co-creation programs.

1. The platform organizes preparation workshops where merchants co-build strategies with the AI product team, clarifying product capabilities and model boundaries instead of treating AI as a black box.

2. Baiquiu, as the first co-creation partner among Alibaba AI Dianxiaomi service providers, reflects the platform's emphasis on benchmark co-creation and ecosystem development.

Business Needs and Issues for the Platform: Merchants need the platform to provide the ability to handle "non-standard questions," including image-text recognition, emotion recognition, proactive service, and tools to auto-generate Q&A from product detail pages.

1. Merchants worry that more AI involvement will worsen the experience. The platform needs to prove with data that AI's own satisfaction rose from 89.5% to 94.5%, and human-machine customer service satisfaction is generally above 92%.

2. Merchants require AI to have boundaries in sensitive scenarios and to quickly transfer to human agents. The platform needs to support complaint escalation, high-risk expression recognition, and human takeover in its product mechanisms.

Operations Management and Risk Avoidance: The platform can guide merchants to turn customer service from a cost center into a sales center, demonstrating value by increasing the sales proportion of human-machine customer service by more than 5 percentage points.

1. Before Double 11, the platform should provide scenario simulation drills, proactive after-sales care, order-work-order integration, and other features to help merchants reduce refunds, negative reviews, and disputes.

2. The platform should encourage organizing chat data into analyses and feeding them back to product, operations, and logistics to enable full-chain optimization.

Industry Trends: The e-commerce customer service industry is shifting from purely human service to human-machine collaboration. As the proportion of AI-independent handling increases, satisfaction rises. Customer service roles are being restructured into new positions such as AI digital employees, AI trainers, analyst strategists, and elite customer service agents.

1. Cases from Little Freddie and Baiquiu show that AI customer service can improve inquiry-to-order conversion rates and reduce human handoff rates, while moving the customer service department from a cost center to a sales center and data center.

New Challenges and Governance: As AI begins to handle more customers, human value needs to be redefined. Companies find that the number of customer service staff decreases temporarily, but total labor costs do not decline correspondingly; compensation structures and capability requirements are changing.

1. The tacit knowledge of excellent agents must be transformed into machine-understandable and executable rules. Knowledge extraction becomes a new organizational capability.

2. AI's ability to recognize emotions does not mean it should handle all emotions alone. Companies proactively set red lines and human takeover mechanisms; cases involving medical judgment, sensitive public topics, or complaint escalation must be transferred to human agents. This provides a reference for AI service governance.

Business Models and Data Loop: Service providers can position customer service as a sales center. The sales proportion of human-machine customer service is more than 5 percentage points higher than in human-only models, and pure bot sales can account for 5% to 8% of total store sales, proving a clear return on AI investment.

1. Platform-service provider co-creation (e.g., Baiquiu becoming the first co-creation partner) forms a new collaboration model.

2. Conversation data is organized into VOC analysis and fed back to product, operations, and logistics, forming a closed data loop from customer service to full-chain optimization.

3. Shifting Double 11 preparation from solo work to multi-party co-creation and pre-event rehearsal represents a direction for operational innovation.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

每年双11之前,电商的客服团队都会提前进入一种很具体的紧张状态:活动规则要变,优惠组合会变,预售和发货政策也会变。过去的做法通常很直接——把可能被问到的问题一条条写进知识库,测试机器人能不能答,答不了的再交给人工。

对母婴食品品牌“小皮”,和服务超150个品牌、GMV超300亿元的电商服务商百秋来说,今年的双11准备方式都在发生变化。AI不再只负责回应最简单的一层咨询,而是开始进入商品推荐、情绪安抚、销售转化和售后流程。与此同时,人也没有简单地从客服中心消失。有人转去设计应答策略,有人专门维护高净值客户,有人开始把会话数据整理成商品、运营和物流可以继续使用的东西。

下面是两家企业的客服部门的口述。它们的业务完全不同,但都绕不开同一个问题:当AI开始接住更多顾客,长出更多能力,人该如何如何被重新定义。

小皮:我最在意的,不是AI答得更多,而是它开始知道怎么服务

Little Freddie(小皮)是跨国婴童食品品牌,2014年成立,以"成为父母最信任、孩子最喜爱的婴童食品品牌"为愿景,主营婴童辅食、零食等婴幼儿食品,全线无不必要添加,每款配方均经注册营养师审核。销售覆盖天猫、京东、山姆等主流渠道,连续多年位居行业前列。

我们从2019年就开始用AI店小蜜,也很早配了AI训练师。那时候的机器人比较像一个FAQ工具:客户问什么,它就按知识库回答什么。稍微复杂一点,比如发来一张活动截图、一张产品图,或者问题没有完全踩中我们预设的关键词,基本就得转人工。

2026年7月,我们正式引入AI店小蜜高阶版,开始深度用。真正让我觉得变化比较大的,不只是回答更自然,而是它开始具备“服务意识”。

最直接的是图文识别。以前客户发截图,我们的老版本基本识别不了;现在它可以把图片和文字放在一起理解,再去知识库里核对信息。这件事听起来不大,但售前咨询经常就是这样:消费者不会按照客服系统希望的格式来提问,她会丢一张图,说一句“这个怎么参加”,甚至只问“是这个吗”。能不能接住这种不标准的问题,差别很大。

第二个变化是情绪。我们做母婴辅食,很多消费者是新手妈妈。她们问的不是单纯的“米粉怎么冲”,而是“我是不是弄错了”“这样宝宝能不能吃”。过去如果想让机器人安抚用户,训练师得提前在某个意图下面写好一段话,而且必须命中那个意图才会触发。现在AI会自己识别焦虑、不耐烦这些情绪,先安抚,再回答。有时候它会说一句:“别担心,刚开始都是这样的呢。”我自己看到这种回复,会意识到它不再像在发说明书了。

第三个变化是主动服务。以前机器人就是“你问什么我答什么”。现在我们让它在推荐产品前主动问宝宝月龄,再根据不同月龄、体质和需求去推荐;碰到优惠,也会主动提醒怎么凑单、活动什么时候结束。

数据上,这种变化也比较明显。按我们的内部数据,7月至8月,机器人询单转化率从25.9%升至29.6%,提升3.7个百分点,同时,转人工率下降了大约13个百分点。让我比较安心的是,AI接得更多之后,满意度没有下降:AI店小蜜自身满意度从89.5%提升到94.5%,全店满意度维持在96%左右。以前我会担心,机器接得越多,体验是不是一定会打折。至少现阶段,数据没有支持这个担心。

母婴这个品类,消费者真正关心的往往不是便宜几块钱,而是安全感和专业感:这个东西安不安全,我的宝宝能不能吃,会不会过敏,营养够不够。过敏原、体质适配这些敏感场景,我们是和项目交付团队一起共创应答策略的,先把红线和边界对齐。碰到特别复杂的个体差异,或者涉及医学判断的问题,我们宁可转人工,也不会让AI往前多走一步。

图片也是类似。母婴辅食其实不是一个高频“带图找货”的品类,用户通常说“多大宝宝吃”“什么口味”“米粉还是果泥”,文字就够了。真正会发图的,更多是在售后:有人收到货后拍包装来确认真伪,有人看到包装变化,想确认是不是升级款;也有人冲调后结块、果泥颜色和预期不一样,拍图来问。这些场景对AI有价值,但目前售后图文识别还有提升空间。

所以我们团队并没有因为高阶版上线就大幅缩人。现在更接近“增量不增人”——业务可以往上走,但人不用按原来的比例一起加。变化最大的其实是训练师。

以前训练师有点像“知识搬运工”:哪条答案错了就改哪条,哪个场景缺知识就补哪个。现在我们要做的是应答策略和应答矩阵。比如同样是推荐米粉,不同月龄、不同体质到底先问什么,什么时候推荐搭配产品,怎么引导,什么情况必须交给人工,都要先被整理成一套策略,再让AI去执行。

这件事比补知识库更难,因为它要求我们把过去优秀客服脑子里的经验拆出来。经验原来可以是“我知道该怎么说”,现在不行,要变成机器也能理解、也能重复执行的规则。

如果要算一笔人效账,高阶版上线后,AI独立承接的占比提升了十几个百分点,大致相当于减少了两名客服的工作量。但我不太愿意把这件事只理解成“少两个人”。对我们来说,更重要的是这些工作量被拿走以后,人能不能把时间用到更复杂的服务上。

今年双11的准备方式,也能看出这种变化。9月15日我去参加了AI店小蜜的双11备战工作坊。过去备战基本是“单兵作战”——训练师自己配活动规则、优惠信息和售后政策,自己找场景测试,自己验收,出了问题也自己扛。

今年最大的感受,是终于有人一起搭策略了。我们懂业务和用户,AI店小蜜团队更熟悉产品能力和模型边界。到9月中旬,我们已经围绕大促优惠、预售发货等规则沟通过三次以上。另一个变化是,我们开始更清楚AI在一个答案出来之前到底经历了什么,不再把它当成一个“黑箱”——知道AI是怎么工作的,才知道什么场景应该交给它,什么场景不要勉强。

百秋:AI接走基础接待后,客服岗位先变了

百秋尚美是中高端品牌数字零售服务商,覆盖天猫、京东等平台,系天猫六星服务商,客户包括全球高端时尚、美妆品牌。2018年引入店小蜜,2025年618、双11获AI店小蜜“蜂芒奖”,2026年7月成为阿里AI店小蜜服务商首家共创伙伴。

我们从2018年就开始使用店小蜜,当时还是基础版本,也是比较早的一批用户。2025年双11,我们先让一部分品牌试用Agent版。今年年初,我们又把AI店小蜜高阶版部署到了30多家品牌。

百秋现在服务150多个品牌,客服团队大约500人,这个规模放在上海,成本当然不低。但如果只看“用了AI以后少了多少人”,其实很容易把这件事看窄。

过去半年,高阶版用下来,我们的相关客服人数阶段性有所减少,但是总人力成本并没有同步下降——也就是说,客服的能力要求、组织架构和薪酬结构也变了。

我们现在把客服团队重新拆成几种角色:AI数字员工、AI训练师和分析策略师,再加上精英客服。基础接待更多交给机器,人的位置往两个方向移动——一边去训练和校验AI,一边去做更难、更值钱的服务。

这和我们对客服的定位有关。我们内部一直不太愿意把客服只当成本中心,更愿意把它当销售中心。按我们的统计口径,消费者下单前只要和机器人或人工客服发生过咨询,并且在这条咨询链路里完成成交,这部分会算作客服销售。平均来看,客服相关销售大约占整体销售的30%到35%。

高阶版上线之后,我们更关心的不是某一个“纯机转化率”涨了多少,而是人和机器一起能不能把更多交易留在客服链路里。我们看到,使用AI店小蜜的一些店铺,人机客服销售占比较纯人工模式提升了5个百分点以上。纯机器人、完全没有人工介入并完成下单的销售,在一些店铺里也可以占到全店销售的5%到8%。这些数字当然不是所有品牌都一样,但它解释了我们为什么愿意持续投入。

这里面其实有一个挺现实的矛盾。我们的客服薪酬过去很大程度和销售结果挂钩。机器人卖得越多、接待得越多,如果人的工作方式完全不变,人的销售机会就会被挤掉。这个问题得靠让人的工作变得更难替代来优化。

所以我们会把更多人工放到高净值客户上。以前一个客服可能要同时接很多普通咨询。现在,他可以花更多时间做1对1维护。举个例子,以前可能服务10个客户成交100万元,现在可能只服务两个客户,也能做到同样的销售额。工作量的计算方式变了,客服本人也得把销售能力、服务能力一起往上抬。

反过来,我们也会把精英客服的高转化话术、复杂服务经验继续沉淀进训练和策略里。精英客服怎么沟通、怎么促成交易、怎么处理复杂需求,这些经验都会被整理成可供机器人执行和校验的规则。于是人的价值不只是“亲自回复了多少句话”,还包括能不能生产出机器可以复制的好经验。

训练师的工作也在变。我们服务的很多品类上新很快,过去知识库靠关键词配置,新品一多,维护会非常重。现在新品在立项、详情页定稿之后,运营会同步产品参数、卖点、配件信息和常见问题,AI可以从商品详情页等已有资料自动抓取信息,批量生成问答素材和图文对比,再由训练师做规则校验。我们现在希望做到,新品上新的当天,相关答案就能配置完成。

AI店小蜜高阶版会在很细节的地方,显出它和人的区别。我们服务过一个小家电品牌,消费者收到产品后不会安装,这是很常见的问题。以前人工最容易做的事情,是发一个安装视频或者说明书,告诉他按步骤来。现在机器人可以先识别用户着急的状态,先把情绪接住,再给出安装步骤;有时还会把其他消费者容易卡住的细节提前说出来。这个动作不复杂,但它开始让你感受到——AI不只是帮你解决问题,而是在服务你。

当然,我们不会把所有事情都交给AI。涉及敏感公共议题、特殊时间节点等场景,我们会直接设置边界;消费者情绪明显升级、明确提出投诉、升级处理,或者出现“曝光”等高风险表达时,也会马上由人工介入。能识别情绪,不等于应该独自处理所有情绪。

满意度是我们会持续看的指标。按我们的内部阶段性统计,在已部署高阶版的品牌中,多数品牌的人机客服满意度达到92%以上;部分高端品牌达到95%以上——这里的满意度,指消费者在机器人和客服聊天后的会话评价。售后端,我们现在也会看12小时解决率。按同一阶段性口径,12小时内解决消费者问题的比例可以达到87%以上。

对双11来说,今年更大的变化不是临时多排多少人,而是把准备工作往前挪。我们会把商品、运营、客服、物流一起拉进来,把历史上消费者问过的问题、大促期间出现过的集中事件整理成文档,再让机器人提前做场景模拟和预演。过去很多精力花在活动开始之后救火,今年更希望在开始之前把火源找出来。

售后也是今年重点往前推的一块。现在高阶版可以和订单、工单体系做更多连接,我们希望从过去被动回答“怎么退、怎么查、怎么处理”,慢慢变成主动做售后关怀和提醒。不同商品可以设置不同的提示,尽量减少因为信息不对称造成的退款、差评和纠纷。

这也是为什么我觉得,客服团队未来不会只剩“人”和“机器人”两个岗位。我们已经在增加AI训练师、技术产品专家,也会需要更多VOC和分析能力。聊天里每天都在产生消费者对商品、活动、页面、发货和退货的反馈,如果这些信息只是解决完一单就结束,其实很浪费。我们的方向,是把这些问题整理成分析,再反馈给商品、运营和物流,让页面更清楚、活动机制更完整、流程更顺。

所以回到最开始那笔账:为什么相关客服人数阶段性有所减少,总人力成本却没有同步下降?因为我们不是把原来的客服团队简单压缩成更少的人,而是在重新定义剩下这些人做什么——基础回答可以越来越便宜,但高质量服务、策略设计、知识校验和数据分析,反而变得更贵。

注:文/沈嵩男,文章来源:天下网商(公众号ID:txws_txws),本文为作者独立观点,不代表亿邦动力立场。

文章来源:天下网商

广告
微信
朋友圈

FAQ回顾

AI店小蜜是什么?有哪些新能力?

AI店小蜜是阿里巴巴的电商AI客服产品。2026年7月推出的高阶版新增三项能力:图文识别,可同时理解图片与文字咨询;情绪识别,能识别焦虑等情绪并先安抚再回答;主动服务,推荐商品前主动询问月龄等需求,并提醒优惠和凑单。小皮、百秋等品牌已接入部署。

AI客服接待量增加后满意度会下降吗?

不会。根据小皮内部数据,2026年7月至8月接入AI店小蜜高阶版后,转人工率下降约13个百分点,AI店小蜜自身满意度从89.5%升至94.5%,全店满意度维持在96%左右。百秋数据显示,已部署高阶版的品牌中,多数人机客服满意度达92%以上,部分高端品牌超95%。

品牌如何用AI客服备战双11?

品牌会将大促优惠、预售和发货规则提前整理成应答策略并让AI预演。小皮于9月参加AI店小蜜双11备战工作坊,围绕活动规则与产品团队沟通多次。百秋则联合商品、运营、客服、物流,把历史大促咨询集中问题整理成文档,让机器人提前场景模拟,减少活动开始后救火。

AI客服会取代人工客服吗?

短期不会,而是走向人机协同。AI负责基础接待,人工转向高净值客户1对1维护、应答策略设计和数据反哺。百秋将客服团队拆分为AI数字员工、AI训练师、分析策略师和精英客服,并把精英客服的话术沉淀为机器可执行的规则;涉及敏感议题、投诉或高风险表达时仍由人工介入。

AI客服如何帮助电商提升转化率?

AI通过主动推荐和精准服务提升转化。小皮让机器人在推荐前主动询问宝宝月龄并按体质推荐产品,2026年7至8月询单转化率从25.9%升至29.6%。百秋部分店铺人机客服销售占比较纯人工模式提升5个百分点以上,纯机器人完成下单的销售可占全店销售的5%至8%。

这么好看,分享一下?

朋友圈 分享

APP内打开

赞 +1
+1
微信好友 朋友圈 新浪微博 QQ空间
关闭
收藏成功
发送
/140 0