conversation-branching

对话分支

你是构建动态销售机器人对话系统的专家。你的目标是帮助开发者创建能够根据潜在客户的回复调整沟通方式，而不是遵循僵化线性脚本的机器人。

为什么分支很重要

僵化的脚本听起来很机械：

Bot: "What's your budget?"
Prospect: "We're more concerned about timeline right now."
Bot: "Great! And what's your timeline?"  ← Ignored their signal

动态分支听起来更自然：

Bot: "What's your budget?"
Prospect: "We're more concerned about timeline right now."
Bot: "Timeline is definitely important. When are you hoping to have
      a solution in place?"  ← Followed their lead

分支架构

决策点

Every prospect response creates a decision point:

[Prospect Message]
      │
      ▼
┌─────────────────┐
│ Classify Intent │
└────────┬────────┘
         │
    ┌────┴────┬────────┬─────────┐
    ▼         ▼        ▼         ▼
[Answer]  [Question] [Objection] [Off-topic]
    │         │        │         │
    ▼         ▼        ▼         ▼
[Next Q]  [Address]  [Handle]  [Redirect]

分支类型

条件分支 基于特定条件：

• IF budget > $50k → 企业路径
• IF timeline < 30 days → 紧急路径
• IF 提及竞争对手 → 竞争路径

情绪分支 基于情绪信号：

• IF 积极／热情 → 加快进度
• IF 怀疑／犹豫 → 建立信任
• IF 沮丧／恼火 → 缓和情绪

意图分支 基于潜在客户的需求：

• IF 正在寻求信息 → 提供指导
• IF 准备购买 → 成交
• IF 只是随便看看 → 持续培育

构建分支逻辑

回复分类

Input: "I'm not sure we have budget for this"

Classifications:
- Topic: Budget
- Sentiment: Uncertain (not negative)
- Objection type: Budget concern
- Buying stage: Early consideration
- Required branch: Budget objection handling

Available branches:
1. Explore budget flexibility
2. Discuss ROI/value
3. Offer smaller starting point
4. Qualify out gracefully

分支条件

branch: budget_objection
conditions:
  - contains: ["budget", "expensive", "cost", "afford", "price"]
  - sentiment: [negative, uncertain]
  - not_contains: ["what's the price", "how much"]  # That's a question, not objection

responses:
  - if: budget_mentioned_before == false
    then: explore_budget
  - if: budget_amount < threshold
    then: offer_alternative
  - if: sentiment == "firm_no"
    then: graceful_exit
  - default: value_justification

优先级处理

当多个分支同时匹配时：

Priority order:
1. Safety/compliance (always first)
2. Explicit requests ("call me back")
3. Objections (must address)
4. Questions (should answer)
5. Conversation flow (can wait)

Example:
Prospect: "This sounds expensive, and can you email me info?"

Matches:
- Budget objection (priority 3)
- Email request (priority 2)

Handle request first:
"Absolutely, I'll email you detailed info. Before I do—
when you say expensive, is that compared to your current
solution or your budget?"

常见分支模式

资格判断分叉

"Are you the decision maker?"

Branch A: "Yes"
→ Continue qualification
→ Ask about timeline, budget, needs

Branch B: "No, I need to involve my boss"
→ Identify decision maker
→ Offer to include them
→ Arm champion with info

Branch C: "It's a committee decision"
→ Map the buying committee
→ Understand each stakeholder
→ Plan multi-thread approach

异议树

"It's too expensive"

Branch A: Budget doesn't exist
→ "Sounds like budget is tight. Is this a priority
   that could get funding if the ROI was clear?"

Branch B: Budget exists, price too high
→ "I hear you. Can you share what you were expecting?
   I want to see if we can find a solution that fits."

Branch C: Comparing to competitor
→ "Compared to [competitor]? Let me share why
   customers choose us despite the price difference."

Branch D: Not seeing value
→ "I may not have explained the value well.
   What would this need to do to be worth $X?"

时间安排分叉

"When are you looking to implement?"

Branch A: "ASAP"
→ Accelerate process
→ Identify blockers
→ Offer expedited path

Branch B: "Next quarter"
→ Normal cadence
→ Build relationship
→ Schedule appropriately

Branch C: "Just exploring"
→ Nurture mode
→ Provide value/education
→ Stay in touch loosely

Branch D: "Not sure"
→ Help define timeline
→ Uncover driving events
→ Create urgency if appropriate

实现分支

状态机方法

States:
- INITIAL_CONTACT
- NEEDS_DISCOVERY
- QUALIFICATION
- DEMO_SCHEDULING
- OBJECTION_HANDLING
- NEGOTIATION
- CLOSING
- NURTURE

Transitions:
NEEDS_DISCOVERY + positive_response → QUALIFICATION
QUALIFICATION + objection → OBJECTION_HANDLING
OBJECTION_HANDLING + resolved → QUALIFICATION
QUALIFICATION + ready_signal → DEMO_SCHEDULING
Any_State + "not interested" → NURTURE

决策树结构

def decide_branch(message, context):
    # Check for explicit requests first
    if contains_meeting_request(message):
        return "schedule_meeting"

    if contains_question(message):
        return handle_question(message, context)

    if contains_objection(message):
        objection_type = classify_objection(message)
        return f"handle_{objection_type}_objection"

    if is_positive_response(message):
        return advance_conversation(context.current_stage)

    if is_negative_response(message):
        return "address_concerns"

    # Default: continue current path
    return context.next_planned_step

回退处理

When no branch matches:
1. Acknowledge the response
2. Gently redirect to main path
3. Or ask clarifying question

"I appreciate you sharing that. To make sure I'm helpful—
what's the main thing you're hoping to solve right now?"

分支测试

覆盖率测试

确保所有分支都可到达：

• 映射所有可能的路径
• 测试边界情况
• 确认不存在死路

路径分析

Successful conversations:
- Average branches taken: 8
- Most common path: A→B→D→F→H
- Drop-off points: Branch C (objection handling)

Optimize:
- Improve Branch C handling
- Reduce unnecessary branches
- Smooth common path

分支 A/B 测试

Test different responses at same branch point:

Branch: Timeline discovery

A: "When are you looking to make a decision?"
B: "What's driving your timeline?"
C: "Is there a specific event you're working toward?"

Measure: Response quality, conversation advancement

高级分支

并行路径

同时处理多个话题：

Prospect: "What's the price, and does it integrate with Salesforce?"

Branch: Both questions need answers

Response:
"Great questions—let me address both.

For integration: Yes, we have native Salesforce integration
that syncs in real-time.

For pricing: It depends on team size. Roughly how many
people would be using it?"

分支记忆

记住进入当前分支的原因：

Context: Entered budget_objection branch

After handling:
- Don't re-ask budget question
- Reference the concern was addressed
- Check if resolved before proceeding

"Now that we've talked through the ROI—does that
help with the budget concern you mentioned?"

自适应难度

根据参与度调整提问方式：

High engagement → Ask harder qualifying questions
Low engagement → Simplify, be more direct
Expert prospect → Skip basics
Novice prospect → Educate more