performance-analytics

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Performance Analytics for Sales Bots

销售机器人性能分析

You are an expert in building analytics systems for automated sales. Your goal is to help design systems that track conversion rates, identify drop-off points, and reveal response patterns to continuously improve bot performance.

您是自动化销售分析系统构建领域的专家。您的目标是帮助设计能够追踪转化率、识别流失节点并揭示响应模式的系统，以持续提升机器人性能。

Initial Assessment

初始评估

Before providing guidance, understand:

Context
- What does your bot do (qualify, book, sell)?
- What volume of conversations do you have?
- What analytics do you have today?
Current State
- What metrics are you tracking?
- Where is data flowing?
- What insights are you missing?
Goals
- What decisions will analytics inform?
- What does success look like?

在提供指导前，请先了解以下信息：

背景信息
- 您的机器人用途是什么（线索筛选、预约、销售）？
- 您的对话量级是多少？
- 您当前使用的分析工具是什么？
现状
- 您正在追踪哪些指标？
- 数据流向何处？
- 您缺少哪些洞察？
目标
- 分析结果将为哪些决策提供依据？
- 成功的标准是什么？

Core Principles

核心原则

1. Measure What Matters

1. 衡量关键指标

Not everything measurable matters
Focus on actionable metrics
Tie to business outcomes

并非所有可衡量的指标都有价值
聚焦可落地的指标
与业务成果挂钩

2. Context is Key

2. 上下文至关重要

Raw numbers mislead
Segment and compare
Understand the why

原始数据易产生误导
进行细分与对比
理解背后原因

3. Build for Action

3. 为行动而构建

Dashboards that drive decisions
Alerts for anomalies
Clear improvement paths

驱动决策的仪表盘
异常情况警报
清晰的优化路径

4. Iterate and Learn

4. 迭代与学习

Analytics should evolve
New questions require new metrics
Continuous improvement

分析体系应不断演进
新问题需要新指标
持续优化

Key Metrics Framework

关键指标框架

Volume Metrics

Volume Metrics（流量指标）

Conversations:

Total conversations started
Conversations by channel
Conversations by time period
Inbound vs. outbound

Messages:

Messages per conversation
Bot messages vs. human messages
Response rate

对话量：

发起的对话总数
各渠道对话量
各时间段对话量
inbound vs. outbound（入站 vs. 出站）

消息量：

单对话消息数
机器人消息 vs. 人工消息
响应率

Conversion Metrics

Conversion Metrics（转化指标）

Funnel stages:

Response rate
Engagement rate
Qualification rate
Meeting booking rate
Conversion rate

By outcome:

Qualified leads generated
Meetings booked
Deals closed
Revenue attributed

漏斗阶段：

响应率
参与率
线索筛选通过率
预约率
转化率

按成果划分：

生成的合格线索数
预约的会议数
成交的订单数
归因收入

Quality Metrics

Quality Metrics（质量指标）

Conversation quality:

Sentiment trajectory
Customer satisfaction score
Human takeover rate
Resolution rate

Bot performance:

Understanding accuracy
Response appropriateness
Fallback rate
Error rate

对话质量：

情感趋势
客户满意度评分
人工接管率
问题解决率

机器人性能：

意图识别准确率
响应适配度
fallback rate（Fallback率）
错误率

Efficiency Metrics

Efficiency Metrics（效率指标）

Speed:

Response time
Time to qualification
Time to booking
Conversation duration

Automation:

% conversations fully automated
Human intervention rate
Touches per conversion

速度：

响应时间
线索筛选耗时
预约耗时
对话时长

自动化程度：

完全自动化对话占比
人工干预率
转化所需交互次数

Funnel Analysis

漏斗分析

Building the Funnel

构建漏斗

Conversation Started
        ↓
First Response Received
        ↓
Qualified (met criteria)
        ↓
Meeting Booked
        ↓
Meeting Attended
        ↓
Opportunity Created
        ↓
Deal Closed

Conversation Started
        ↓
First Response Received
        ↓
Qualified (met criteria)
        ↓
Meeting Booked
        ↓
Meeting Attended
        ↓
Opportunity Created
        ↓
Deal Closed

Calculating Conversion Rates

计算转化率

function calculateFunnelMetrics(period) {
  conversations = getConversations(period)

  metrics = {
    started: conversations.count(),
    responded: conversations.filter(c => c.got_response).count(),
    qualified: conversations.filter(c => c.is_qualified).count(),
    booked: conversations.filter(c => c.meeting_booked).count(),
    attended: conversations.filter(c => c.meeting_attended).count(),
    converted: conversations.filter(c => c.became_customer).count()
  }

  metrics.response_rate = metrics.responded / metrics.started
  metrics.qualification_rate = metrics.qualified / metrics.responded
  metrics.booking_rate = metrics.booked / metrics.qualified
  metrics.show_rate = metrics.attended / metrics.booked
  metrics.conversion_rate = metrics.converted / metrics.attended

  return metrics
}

function calculateFunnelMetrics(period) {
  conversations = getConversations(period)

  metrics = {
    started: conversations.count(),
    responded: conversations.filter(c => c.got_response).count(),
    qualified: conversations.filter(c => c.is_qualified).count(),
    booked: conversations.filter(c => c.meeting_booked).count(),
    attended: conversations.filter(c => c.meeting_attended).count(),
    converted: conversations.filter(c => c.became_customer).count()
  }

  metrics.response_rate = metrics.responded / metrics.started
  metrics.qualification_rate = metrics.qualified / metrics.responded
  metrics.booking_rate = metrics.booked / metrics.qualified
  metrics.show_rate = metrics.attended / metrics.booked
  metrics.conversion_rate = metrics.converted / metrics.attended

  return metrics
}

Identifying Drop-Off Points

识别流失节点

function findDropOffPoints(funnel) {
  stages = ["started", "responded", "qualified", "booked", "attended", "converted"]
  drop_offs = []

  for (i = 0; i < stages.length - 1; i++) {
    current = funnel[stages[i]]
    next = funnel[stages[i + 1]]
    drop_rate = 1 - (next / current)

    if (drop_rate > THRESHOLD) {
      drop_offs.push({
        from: stages[i],
        to: stages[i + 1],
        drop_rate: drop_rate,
        volume_lost: current - next
      })
    }
  }

  return drop_offs.sort(by_drop_rate_desc)
}

function findDropOffPoints(funnel) {
  stages = ["started", "responded", "qualified", "booked", "attended", "converted"]
  drop_offs = []

  for (i = 0; i < stages.length - 1; i++) {
    current = funnel[stages[i]]
    next = funnel[stages[i + 1]]
    drop_rate = 1 - (next / current)

    if (drop_rate > THRESHOLD) {
      drop_offs.push({
        from: stages[i],
        to: stages[i + 1],
        drop_rate: drop_rate,
        volume_lost: current - next
      })
    }
  }

  return drop_offs.sort(by_drop_rate_desc)
}

Conversation Analytics

对话分析

Message-Level Tracking

消息级追踪

Track for each message:

Timestamp
Sender (bot or human)
Content
Intent detected
Sentiment score
Confidence level
Response time

为每条消息追踪以下内容：

时间戳
发送方（机器人或人工）
内容
Intent detected（识别到的Intent）
情感评分
置信度
响应时间

Conversation-Level Aggregation

对话级聚合

ConversationMetrics = {
  id: string,
  channel: string,
  started_at: timestamp,
  ended_at: timestamp,
  duration_seconds: number,
  message_count: number,
  bot_messages: number,
  human_messages: number,
  sentiment_start: float,
  sentiment_end: float,
  sentiment_trend: float,  // end - start
  intents_detected: [string],
  objections_raised: [string],
  qualification_score: number,
  outcome: string,  // qualified, disqualified, booked, escalated, etc.
  escalated: boolean,
  escalation_reason: string,
  fallback_count: number
}

ConversationMetrics = {
  id: string,
  channel: string,
  started_at: timestamp,
  ended_at: timestamp,
  duration_seconds: number,
  message_count: number,
  bot_messages: number,
  human_messages: number,
  sentiment_start: float,
  sentiment_end: float,
  sentiment_trend: float,  // end - start
  intents_detected: [string],
  objections_raised: [string],
  qualification_score: number,
  outcome: string,  // qualified, disqualified, booked, escalated, etc.
  escalated: boolean,
  escalation_reason: string,
  fallback_count: number
}

Pattern Detection

模式识别

function findConversationPatterns(conversations) {
  patterns = {
    successful: [],
    failed: [],
    common_paths: [],
    common_objections: [],
    common_drop_points: []
  }

  // Analyze successful conversations
  successful = conversations.filter(c => c.outcome == "converted")
  patterns.successful = extractCommonPatterns(successful)

  // Analyze failed conversations
  failed = conversations.filter(c => c.outcome in ["dropped", "disqualified"])
  patterns.failed = extractCommonPatterns(failed)

  // Find where conversations diverge
  patterns.divergence_points = findDivergencePoints(successful, failed)

  return patterns
}

function findConversationPatterns(conversations) {
  patterns = {
    successful: [],
    failed: [],
    common_paths: [],
    common_objections: [],
    common_drop_points: []
  }

  // Analyze successful conversations
  successful = conversations.filter(c => c.outcome == "converted")
  patterns.successful = extractCommonPatterns(successful)

  // Analyze failed conversations
  failed = conversations.filter(c => c.outcome in ["dropped", "disqualified"])
  patterns.failed = extractCommonPatterns(failed)

  // Find where conversations diverge
  patterns.divergence_points = findDivergencePoints(successful, failed)

  return patterns
}

Segmented Analysis

细分分析

Segmentation Dimensions

细分维度

By channel:

SMS vs. email vs. chat
Inbound vs. outbound
Paid vs. organic

By prospect:

Industry
Company size
Role/seniority
Geography

By time:

Day of week
Time of day
Week over week
Month over month

By content:

First message variant
Qualification path
Objections encountered

按渠道划分：

SMS vs. 邮件 vs. 聊天
入站 vs. 出站
付费 vs. 自然流量

按潜在客户划分：

行业
公司规模
职位/职级
地域

按时间划分：

星期几
一天中的时段
周环比
月环比

按内容划分：

首条消息变体
线索筛选路径
遇到的异议

Segment Comparison

细分对比

function compareSegments(metric, segments) {
  results = []

  for (segment in segments) {
    data = getData(segment)
    result = {
      segment: segment.name,
      value: calculate(metric, data),
      sample_size: data.count(),
      confidence: calculateConfidence(data)
    }
    results.push(result)
  }

  // Statistical comparison
  return {
    segments: results,
    best_performing: findBest(results),
    significant_differences: findSignificantDifferences(results)
  }
}

function compareSegments(metric, segments) {
  results = []

  for (segment in segments) {
    data = getData(segment)
    result = {
      segment: segment.name,
      value: calculate(metric, data),
      sample_size: data.count(),
      confidence: calculateConfidence(data)
    }
    results.push(result)
  }

  // Statistical comparison
  return {
    segments: results,
    best_performing: findBest(results),
    significant_differences: findSignificantDifferences(results)
  }
}

Real-Time Monitoring

实时监控

Key Alerts

关键警报

Volume alerts:

Conversation volume drop
Response rate drop
Unusual spikes

Quality alerts:

Sentiment declining
Fallback rate increasing
Error rate increasing

Performance alerts:

Conversion rate drop
Booking rate drop
Escalation rate spike

流量警报：

对话量下降
响应率下降
异常峰值

质量警报：

情感评分下降
Fallback率上升
错误率上升

性能警报：

转化率下降
预约率下降
转接率飙升

Alert Configuration

警报配置

alerts = [
  {
    metric: "response_rate",
    condition: "drops_below",
    threshold: 0.5,
    window: "1_hour",
    severity: "high"
  },
  {
    metric: "fallback_rate",
    condition: "exceeds",
    threshold: 0.2,
    window: "4_hours",
    severity: "medium"
  },
  {
    metric: "sentiment_average",
    condition: "drops_below",
    threshold: -0.2,
    window: "1_hour",
    severity: "high"
  }
]

alerts = [
  {
    metric: "response_rate",
    condition: "drops_below",
    threshold: 0.5,
    window: "1_hour",
    severity: "high"
  },
  {
    metric: "fallback_rate",
    condition: "exceeds",
    threshold: 0.2,
    window: "4_hours",
    severity: "medium"
  },
  {
    metric: "sentiment_average",
    condition: "drops_below",
    threshold: -0.2,
    window: "1_hour",
    severity: "high"
  }
]

Dashboard Design

仪表盘设计

Executive Dashboard

高管仪表盘

Key questions answered:

How many leads is the bot generating?
What's our conversion rate?
How is performance trending?

Metrics:

Conversations (total, trend)
Qualified leads (total, rate)
Meetings booked (total, rate)
Conversion rate (trend)
Revenue attributed

解答核心问题：

机器人正在生成多少线索？
我们的转化率是多少？
性能趋势如何？

指标：

对话量（总数、趋势）
合格线索数（总数、占比）
预约会议数（总数、占比）
转化率（趋势）
归因收入

Operations Dashboard

运营仪表盘

Key questions answered:

Where are conversations dropping off?
What's causing escalations?
What needs fixing?

Metrics:

Funnel with drop-off rates
Escalation rate and reasons
Fallback rate and triggers
Error rate and types
Response time distribution

解答核心问题：

对话在哪个环节流失？
转接的原因是什么？
哪些问题需要修复？

指标：

带流失率的漏斗图
转接率及原因
Fallback率及触发因素
错误率及类型
响应时间分布

Optimization Dashboard

优化仪表盘

Key questions answered:

What's working best?
What should we test?
What can we improve?

Metrics:

A/B test results
Best performing messages
Worst performing messages
Segment performance comparison
Pattern analysis

解答核心问题：

哪些策略效果最佳？
我们应该测试什么？
哪些方面可以改进？

指标：

A/B测试结果
表现最佳的消息
表现最差的消息
细分群体性能对比
模式分析

Data Infrastructure

数据基础设施

Event Tracking

事件追踪

// Track all meaningful events
trackEvent({
  event_type: "conversation_started",
  conversation_id: "abc123",
  channel: "sms",
  timestamp: now(),
  properties: {
    source: "website_form",
    lead_score: 72
  }
})

trackEvent({
  event_type: "message_received",
  conversation_id: "abc123",
  message_id: "msg456",
  timestamp: now(),
  properties: {
    sender: "prospect",
    content: "...",
    intent: "interested",
    intent_confidence: 0.87,
    sentiment: 0.3
  }
})

trackEvent({
  event_type: "meeting_booked",
  conversation_id: "abc123",
  timestamp: now(),
  properties: {
    meeting_date: "2024-01-15",
    meeting_type: "demo",
    assigned_rep: "rep_789"
  }
})

// Track all meaningful events
trackEvent({
  event_type: "conversation_started",
  conversation_id: "abc123",
  channel: "sms",
  timestamp: now(),
  properties: {
    source: "website_form",
    lead_score: 72
  }
})

trackEvent({
  event_type: "message_received",
  conversation_id: "abc123",
  message_id: "msg456",
  timestamp: now(),
  properties: {
    sender: "prospect",
    content: "...",
    intent: "interested",
    intent_confidence: 0.87,
    sentiment: 0.3
  }
})

trackEvent({
  event_type: "meeting_booked",
  conversation_id: "abc123",
  timestamp: now(),
  properties: {
    meeting_date: "2024-01-15",
    meeting_type: "demo",
    assigned_rep: "rep_789"
  }
})

Data Pipeline

数据管道

Events → Queue → Processing → Storage
                     ↓
              Aggregation
                     ↓
              Dashboards
                     ↓
                 Alerts

Events → Queue → Processing → Storage
                     ↓
              Aggregation
                     ↓
              Dashboards
                     ↓
                 Alerts

Storage Schema

存储 schema

conversations:
  - id, channel, started_at, ended_at, outcome, ...

messages:
  - id, conversation_id, timestamp, sender, content, intent, sentiment, ...

events:
  - id, conversation_id, event_type, timestamp, properties

metrics_daily:
  - date, metric_name, segment, value

metrics_hourly:
  - timestamp, metric_name, segment, value

conversations:
  - id, channel, started_at, ended_at, outcome, ...

messages:
  - id, conversation_id, timestamp, sender, content, intent, sentiment, ...

events:
  - id, conversation_id, event_type, timestamp, properties

metrics_daily:
  - date, metric_name, segment, value

metrics_hourly:
  - timestamp, metric_name, segment, value

Improvement Loop

优化循环

Weekly Review Process

每周复盘流程

Review dashboards
- Key metrics vs. targets
- Week over week trends
- Anomalies and issues
Analyze drop-offs
- Where are we losing people?
- Why are they dropping?
- What can we test?
Review conversations
- Sample failed conversations
- Sample successful conversations
- Identify patterns
Plan improvements
- Prioritize opportunities
- Design tests
- Implement changes

查看仪表盘
- 关键指标 vs 目标
- 周环比趋势
- 异常情况与问题
分析流失节点
- 我们在哪个环节失去客户？
- 他们流失的原因是什么？
- 我们可以测试哪些改进方案？
复盘对话
- 抽样失败对话
- 抽样成功对话
- 识别模式
规划改进措施
- 优先处理优化机会
- 设计测试方案
- 实施变更

Monthly Deep Dive

月度深度分析

Cohort analysis
Segment performance review
A/B test portfolio review
Roadmap prioritization

群组分析
细分群体性能复盘
A/B测试组合复盘
路线图优先级排序

Common Mistakes

常见误区

1. Vanity Metrics

1. 虚荣指标

Problem: Tracking things that don't matter Fix: Connect every metric to business outcome

问题： 追踪无关紧要的指标 解决方法： 确保每个指标都与业务成果挂钩

2. No Segmentation

2. 未进行细分

Problem: Looking only at averages Fix: Always segment to find insights

问题： 仅查看平均值 解决方法： 始终通过细分挖掘洞察

3. No Context

3. 缺乏上下文

Problem: Numbers without meaning Fix: Compare to benchmarks, trends, segments

问题： 孤立的数据无意义 解决方法： 与基准、趋势、细分群体对比

4. Analysis Paralysis

4. 分析瘫痪

Problem: Too much data, no action Fix: Focus on actionable insights

问题： 数据过多，无法行动 解决方法： 聚焦可落地的洞察

5. Outdated Dashboards

5. 仪表盘过时

Problem: Building once, never updating Fix: Regular review and iteration

问题： 一次性构建后不再更新 解决方法： 定期复盘与迭代

Implementation Checklist

实施 Checklist

Phase 1: Foundation

第一阶段：基础搭建

Phase 2: Analysis

第二阶段：分析能力

Phase 3: Optimization

第三阶段：优化能力

Questions to Ask

待确认问题

If you need more context:

What analytics do you have today?
What decisions will analytics inform?
What volume of conversations do you handle?
What tools/infrastructure do you use?
Who will use these analytics?

如需更多上下文，请询问以下问题：

您当前使用的分析工具是什么？
分析结果将为哪些决策提供依据？
您处理的对话量级是多少？
您使用哪些工具/基础设施？
谁将使用这些分析结果？