Back to Details

ai-whatsapp-chatbot-design

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

AI WhatsApp Chatbot Design

AI WhatsApp聊天机器人设计

<!-- dual-compat:start -->
<!-- dual-compat:start -->

Use when

适用场景

  • Design WhatsApp LLM chatbots for East African markets: conversation flows, social presence principles, trust-building, local language registers, and human escalation protocols. Invoke when a client wants to automate WhatsApp customer service, sales enquiries, or support using AI.
  • Use this skill when it is the closest match to the requested deliverable or workflow.
  • 为东非市场设计WhatsApp LLM聊天机器人：涵盖对话流程、社交存在感原则、信任构建、本地语言语域以及人工转接协议。当客户希望使用AI自动化WhatsApp客户服务、销售咨询或支持工作时启用此技能。
  • 当需求交付成果或工作流与本技能最匹配时使用。

Do not use when

不适用场景

  • Do not use this skill for graphic design, video production, software development, or legal advice beyond the repository's stated scope.
  • Do not use it when another skill in this repository is clearly more specific to the requested deliverable.
  • 请勿将本技能用于图形设计、视频制作、软件开发或超出本知识库规定范围的法律咨询。
  • 当知识库中存在其他更贴合需求交付成果的技能时，请勿使用本技能。

Workflow

工作流程

  1. Collect the required inputs or source material before drafting, unless this skill explicitly generates the intake itself.
  2. Follow the section order and decision rules in this 
    SKILL.md
    ; do not skip mandatory steps or required fields.
  3. Review the draft against the quality criteria, then deliver the final output in markdown unless the skill specifies another format.
  1. 在起草前收集必要的输入信息或源材料，除非本技能明确要求自行生成 intake 内容。
  2. 遵循本
    SKILL.md
    中的章节顺序和决策规则；不得跳过必填步骤或必填字段。
  3. 根据质量标准审核草稿，除非技能指定其他格式，否则最终输出采用markdown格式。

Anti-Patterns

反模式

  • Do not invent client facts, performance data, budgets, or approvals that were not provided or clearly inferred from evidence.
  • Do not skip required inputs, mandatory sections, or quality checks just to make the output shorter.
  • Do not drift into out-of-scope work such as code implementation, design production, or unsupported legal conclusions.
  • 不得编造未提供或无法从现有证据明确推断的客户事实、绩效数据、预算或审批信息。
  • 不得为缩短输出内容而跳过必填输入、必填章节或质量检查。
  • 不得偏离范围开展工作，例如代码实现、设计制作或无依据的法律结论。

Outputs

输出成果

  • An AI-focused strategy, audit, system design, or prompt asset in markdown with human review and control points.
  • 以markdown格式呈现的AI相关策略、审计报告、系统设计或提示资产，包含人工审核和管控节点。

References

参考资料

  • Use the inline instructions in this skill now. If a 
    references/
    directory is added later, treat its files as the deeper source material and keep this 
    SKILL.md
    execution-focused.
<!-- dual-compat:end -->
  • 当前使用本技能中的内嵌说明。若后续添加
    references/
    目录，将其文件视为深度源材料，同时保持本
    SKILL.md
    以执行为核心。
<!-- dual-compat:end -->

Required Input

必要输入信息

Ask for:
  • Client business name and industry
  • Country/city (default: Uganda)
  • Primary goal: customer service / sales enquiries / appointment booking / FAQ handling
  • Approximate monthly WhatsApp message volume
  • Languages customers communicate in (English, Luganda, Kiswahili, other)
  • Existing human support team size and availability hours
需询问：
  • 客户企业名称及所属行业
  • 国家/城市（默认：乌干达）
  • 核心目标：客户服务 / 销售咨询 / 预约预订 / FAQ处理
  • WhatsApp月消息量预估
  • 客户使用的沟通语言（英语、卢干达语、斯瓦希里语、其他）
  • 现有人工支持团队规模及可用时段

Why WhatsApp + LLM for East Africa

为何在东非采用WhatsApp + LLM方案

WhatsApp penetration exceeds 90% among smartphone users in Uganda and across East Africa. Combined with a large language model, a WhatsApp business number becomes a 24/7 sales and support agent that speaks the customer's language, remembers context, and escalates intelligently to humans when needed (Boustany, 2024; Ltifi, 2025).
The competitive advantage is not automation for its own sake — it is availability and responsiveness at a cost most EA businesses can afford.
在乌干达及整个东非地区，智能手机用户的WhatsApp渗透率超过90%。结合大语言模型（LLM），WhatsApp企业号可成为24/7在线的销售与支持Agent，能使用客户的语言沟通、记住对话上下文，并在需要时智能转接给人工客服（Boustany, 2024; Ltifi, 2025）。
竞争优势并非为了自动化而自动化，而是以多数东非企业可承受的成本实现服务的可用性与响应速度。

Architecture: Three Layers

架构：三层结构

Layer 1 — Rule-based flows (decision trees): Handle structured, predictable queries: business hours, pricing, location, how to place an order. Fast, reliable, zero AI cost.
Layer 2 — LLM responses: Handle open-ended, conversational queries that fall outside the decision tree. The LLM uses the brand knowledge base (see 
ai-rag-brand-knowledge-base
) to generate accurate, on-brand responses.
Layer 3 — Human escalation: Trigger a live agent handoff when: the query is a complaint, the customer is frustrated, the LLM confidence is low, or the query involves money, contracts, or sensitive personal data.
第一层 — 规则驱动流程（决策树）： 处理结构化、可预测的查询：营业时间、定价、门店位置、下单方式。响应快速可靠，无AI成本。
第二层 — LLM响应： 处理决策树之外的开放式、对话式查询。LLM利用品牌知识库（详见
ai-rag-brand-knowledge-base
）生成准确、符合品牌调性的回复。
第三层 — 人工转接： 出现以下情况时触发人工客服转接：查询内容为投诉、客户情绪不满、LLM置信度较低，或涉及资金、合同、敏感个人数据的查询。

Social Presence Principles (Ltifi, 2025)

社交存在感原则（Ltifi, 2025）

Research confirms that East African consumers respond significantly better to chatbots that exhibit social presence — warmth, responsiveness, and human-like interaction cues. Apply these principles:
  • Greet by name where possible: "Hello Nakato! How can I help you today?"
  • Use local greetings as an option: "Oli otya?" / "Habari?" for informal register
  • Acknowledge emotional context: "I understand this is frustrating — let me help you sort this out."
  • Avoid corporate coldness: Never open with "Please select from the following options:"
  • Mirror the customer's register: If they write formally, respond formally. If casually, match it.
  • Disclose AI nature when directly asked — transparency builds trust (Uganda Data Protection and Privacy Act, 2019)
研究表明，东非消费者对具备社交存在感的聊天机器人（即展现亲和力、响应性及类人互动线索）反应更佳。需遵循以下原则：
  • 尽可能称呼客户姓名：“你好，Nakato！今天我能为你提供什么帮助？”
  • 提供本地问候语选项：非正式场景下可使用“Oli otya?” / “Habari?”
  • 认可客户情绪语境：“我理解这让你很沮丧——我来帮你解决这个问题。”
  • 避免冰冷的企业话术：切勿以“请从以下选项中选择：”作为开场
  • 匹配客户的语域风格：若客户使用正式语气，回复也需正式；若使用随意语气，则保持一致
  • 被直接询问时披露AI身份：透明度有助于建立信任（《2019年乌干达数据保护与隐私法案》）

Conversation Flow Design

对话流程设计

Step 1: Map the top 10 customer queries

步骤1：梳理Top 10客户查询

Interview the client's human support team. List the 10 most common questions received via WhatsApp in the past month. These become the backbone of the Layer 1 decision trees.
采访客户的人工支持团队，列出过去一个月内通过WhatsApp收到的10个最常见问题。这些将成为第一层决策树的核心内容。

Step 2: Design the decision tree

步骤2：设计决策树

For each query type, map the response path:
Customer: "What are your prices?"
→ Bot: "Our packages start from UGX [X]. Which are you interested in?
   [Option A] [Option B] [Option C]"
→ If Option A: "Great choice! Here's what's included: [details].
   Ready to book? Reply YES or speak to our team."
针对每种查询类型，规划响应路径：
Customer: "What are your prices?"
→ Bot: "Our packages start from UGX [X]. Which are you interested in?
   [Option A] [Option B] [Option C]"
→ If Option A: "Great choice! Here's what's included: [details].
   Ready to book? Reply YES or speak to our team."

Step 3: Define the LLM boundary

步骤3：定义LLM边界

Specify which query types go to the LLM layer — open-ended product questions, complaint context gathering, multi-turn sales conversations. Write the system prompt:
You are [Brand Name]'s friendly customer service assistant on WhatsApp.
You help customers in Uganda with [core services].
Always be warm, helpful, and honest.
If you do not know something, say so and offer to connect the customer with a human.
Never make up prices, availability, or delivery timelines.
Respond in the same language the customer uses.
明确哪些查询类型需转至LLM层——开放式产品问题、投诉背景收集、多轮销售对话。编写系统提示词：
You are [Brand Name]'s friendly customer service assistant on WhatsApp.
You help customers in Uganda with [core services].
Always be warm, helpful, and honest.
If you do not know something, say so and offer to connect the customer with a human.
Never make up prices, availability, or delivery timelines.
Respond in the same language the customer uses.

Step 4: Define HITL escalation triggers

步骤4：定义HITL转接触发条件

Hand off to a human agent when:
  • Customer uses words: "complaint", "refund", "legal", "manager", "angry", "cheated"
  • Same issue raised more than twice without resolution
  • Query involves a transaction above a defined value threshold
  • Customer explicitly requests a human
  • LLM confidence falls below acceptable threshold
Handoff message: "I'm connecting you to one of our team members now. They'll be with you shortly — usually within [X] minutes during business hours."
出现以下情况时转接给人工客服：
  • 客户使用词汇：“complaint”（投诉）、“refund”（退款）、“legal”（法律）、“manager”（经理）、“angry”（愤怒）、“cheated”（受骗）
  • 同一问题被提出两次以上仍未解决
  • 查询涉及超过设定金额阈值的交易
  • 客户明确要求转接人工
  • LLM置信度低于可接受阈值
转接话术：“我现在为你连接我们的团队成员。他们会尽快与你联系——通常在营业时间内[X]分钟内回复。”

Step 5: Build the knowledge base input

步骤5：构建知识库输入内容

Compile the brand knowledge base (see 
ai-rag-brand-knowledge-base
):
  • Full product/service catalogue with prices in UGX
  • FAQs with approved answers
  • Policies: returns, delivery, payment methods
  • Business hours and location(s)
  • Team names and roles for escalation routing
整理品牌知识库（详见
ai-rag-brand-knowledge-base
）：
  • 完整的产品/服务目录（含乌干达先令定价）
  • 经审批的FAQ及答案
  • 政策：退换货、配送、支付方式
  • 营业时间及门店位置
  • 用于转接路由的团队成员姓名及职责

Tool Options

工具选项

ToolBest forEA accessibilityApprox. cost
WATIWhatsApp Business API + chatbot builderYesFrom $49/month USD
Respond.ioMulti-channel + WhatsApp + LLM integrationYesFrom $79/month USD
InteraktAfrica/India-focused WhatsApp toolYesFrom $15/month USD
TwilioDeveloper-friendly WhatsApp APIRequires developerPay-per-message
Meta Cloud APIMaximum controlRequires developerPay-per-message
工具最佳适用场景东非地区可访问性大致成本
WATIWhatsApp Business API + 聊天机器人构建器是起价49美元/月
Respond.io多渠道 + WhatsApp + LLM集成是起价79美元/月
Interakt聚焦非洲/印度的WhatsApp工具是起价15美元/月
Twilio开发者友好型WhatsApp API需要开发者支持按消息付费
Meta Cloud API最大控制权需要开发者支持按消息付费

Measurement Framework

衡量框架

Track monthly:
  • Containment rate: % of conversations resolved without human escalation (target: 60–80% for a mature bot)
  • First response time: Customer message to first bot reply (target: under 30 seconds)
  • CSAT score: Ask after resolution — "How satisfied were you? Reply 1–5"
  • Escalation rate: % of conversations handed to a human (spikes indicate bot gaps)
  • Conversion rate: For sales bots — % of conversations resulting in a purchase or booking
每月跟踪以下指标：
  • 自主解决率：无需人工转接即可解决的对话占比（成熟机器人目标：60–80%）
  • 首次响应时间：客户发消息到机器人首次回复的时长（目标：30秒以内）
  • CSAT评分：问题解决后询问——“你对本次服务满意吗？请回复1–5分”
  • 转接率：转接给人工的对话占比（数值突增表明机器人存在短板）
  • 转化率：销售类机器人——对话转化为购买或预订的占比

Quality Criteria

质量标准

  • Conversation flows are mapped for the top 10 customer query types
  • Social presence principles are embedded in all bot messages — no cold or corporate language
  • HITL escalation triggers are explicitly defined with a handoff message template
  • LLM system prompt is written, specifying brand voice and knowledge boundaries
  • Knowledge base input document is compiled and ready for upload
  • Tool recommendation is specific to client budget and technical capacity
  • Measurement framework includes at least 4 KPIs with targets
  • Uganda Data Protection and Privacy Act (2019) compliance noted
  • 已为Top 10客户查询类型规划对话流程
  • 所有机器人消息均嵌入社交存在感原则——无冰冷或生硬的企业话术
  • 明确定义HITL转接触发条件及转接话术模板
  • 已编写LLM系统提示词，明确品牌语气及知识边界
  • 已整理好知识库输入文档并准备上传
  • 根据客户预算及技术能力给出具体工具推荐
  • 衡量框架包含至少4个带目标值的KPI
  • 符合《2019年乌干达数据保护与隐私法案》要求

References

参考资料

  • Boustany, S. (2024) Generative AI for Social Media Marketing.
  • Ltifi, M. (ed.) (2025) Advances in Digital Marketing in the Era of Artificial Intelligence. CRC Press.
  • Lamplugh, M. (2024) The AI Marketing Playbook, 2nd edn. Mercury Learning.
  • Boustany, S. (2024) Generative AI for Social Media Marketing.
  • Ltifi, M. (ed.) (2025) Advances in Digital Marketing in the Era of Artificial Intelligence. CRC Press.
  • Lamplugh, M. (2024) The AI Marketing Playbook, 2nd edn. Mercury Learning.