agents-webinar-demos

内容来源：README.md（说明文档） · 原始地址 · 查看安装指南

原始内容

<div align="center">
  <img src="assets/funnelenvy-logo.png" alt="FunnelEnvy" width="400">
</div>

# Agents for Marketers: Demo Repository

<p>
  <strong>📺 <a href="https://youtu.be/LWD4VKthf1A?si=DwX2QwhiTm0hOdl6" target="_blank">Watch the Webinar [YouTube]: Agents for Marketers</a></strong>
</p>

Practical demonstrations of how AI agents (Claude Code, Codex, Gemini CLI) can automate analytics, CRO, and campaign workflows for growth marketers.

> **From Insight to Action**: This repository shows how agents can query data, generate hypotheses, and produce actionable outputs—without requiring engineering support.


---

## 🎯 What You'll Find Here

Six production-ready demos that show agents:

1. **Analyzing funnel drop-offs** and generating A/B test hypotheses
2. **Creating paid media audiences** (enhanced conversions, retargeting)
3. **Detecting anomalies** and sending Slack alerts with N8N automation
4. **Identifying "aha moments"** in trial-to-paid conversion data
5. **Auditing tracking implementation** and UTM parameter consistency
6. **Analyzing landing pages** using vision capabilities for CRO recommendations

Each demo includes:
- ✅ Complete prompt sequences
- ✅ Synthetic datasets with engineered patterns
- ✅ Example outputs showing expected results
- ✅ SQL queries and Python scripts (auto-generated by Claude Code)
- ✅ Setup instructions for integrations (Slack, ClickUp, BigQuery, etc.)

---

## 🚀 Quick Start

### 1. Read the Playbook (Recommended)

Start with **[PLAYBOOK.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/PLAYBOOK.md)** to understand:
- **Why agents matter** for marketers (strategic context)
- **The Insight → Hypothesis → Action framework** (mental model)
- **Real use cases with strategic narratives** (what becomes possible)
- **Getting started guide** (practical steps)
- **The vision** (where this is going)

Then come back here for technical implementation.

### 2. Choose Your Demo

Browse the [`demos/`](https://github.com/funnelenvy/agents_webinar_demos/tree/HEAD/demos) folder or read [DEMOS.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/DEMOS.md) for detailed descriptions:

- **New to agents?** Start with Demo 2 (Paid Media Activation) - clear inputs, tangible outputs
- **Want to see vision?** Try Demo 6 (Live Website Audit)
- **Love automation?** Check out Demo 3 (Anomaly Detection + Slack + N8N)

### 3. Set Up Your Environment

Follow [SETUP.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/SETUP.md) to:
- Install dependencies
- Configure environment variables (Slack webhooks, API tokens, etc.)
- Load synthetic data into BigQuery or DuckDB

### 4. Run a Demo

Each demo folder contains:
```
demos/01-funnel-hypothesis-brief/
├── README.md              # Overview and instructions
├── PROMPT_SEQUENCE.md     # Exact prompts to copy/paste
├── queries/               # SQL queries
├── scripts/               # Python scripts (auto-generated)
└── example_outputs/       # What to expect
```

**To run:**
1. Open your AI agent tool (Claude Code, Codex CLI, etc.)
2. Follow the `PROMPT_SEQUENCE.md` step-by-step
3. Compare your results to `example_outputs/`

---

## 📊 Synthetic Datasets

All demos use realistic synthetic data with engineered patterns:

- **Funnel data** with 40% drop-off at pricing page
- **Daily metrics** with 3 anomalies (bug, improvement, activation issue)
- **Trial users** with clear "aha moment" (configure_integration in first 3 days = 3.7x conversion)
- **Tracking data** with intentional gaps (missing events)
- **UTM data** with consistency issues (case variations, typos)

See [DATA.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/DATA.md) for complete schemas and sample queries.

---

## 🛠 Key Integrations

| Platform | Required For | Setup Difficulty |
|----------|--------------|------------------|
| BigQuery or DuckDB | Demos 1-5 | ⚡ Easy |
| Slack Webhook | Demo 3 | ⚡ Easy |
| ClickUp API | Demo 1 | ⭐ Medium |
| N8N Instance | Demo 3 (optional) | ⭐⭐ Medium |
| Playwright | Demo 6 | ⭐ Medium |

No integrations are *required* to explore the prompts and outputs—the data is all local CSV/JSON files.

---

## 🎓 Core Concept: Insight → Hypothesis → Action

These demos follow a three-step framework:

1. **Insight**: Agent queries data and identifies patterns
   - *Example*: "40% of users drop off at the pricing page"

2. **Hypothesis**: Agent generates optimization ideas
   - *Example*: "If we add social proof to pricing, conversion will increase because trust signals reduce purchase anxiety"

3. **Action**: Agent produces usable outputs
   - *Example*: Experiment brief in ClickUp, audience CSV for Google Ads, Slack alert with context

This pattern works across analytics, CRO, paid media, PLG, and marketing ops.

---

## 📁 Repository Structure

```
agents_webinar/
├── README.md              # You are here
├── DEMOS.md               # Detailed demo descriptions
├── DATA.md                # Synthetic dataset documentation
├── SETUP.md               # Environment setup instructions
├── AGENTS.md              # Project overview and principles
├── CLAUDE.md              # Claude Code specific instructions
├── demos/                 # 6 demo workflows
│   ├── 01-funnel-hypothesis-brief/
│   ├── 02-paid-media-activation/
│   ├── 03-anomaly-detection/
│   ├── 04-trial-to-paid/
│   ├── 05-tracking-qa/
│   └── 06-vision-audit/
├── data/                  # Synthetic datasets
│   ├── users.csv
│   ├── events.csv
│   ├── daily_metrics.csv
│   ├── trial_users.csv
│   └── ...
├── scripts/               # Data generation scripts
└── .claude/               # Agent definitions and skills
    ├── agents/            # Specialized agent configs
    └── skills/            # Reusable skill modules
```

---

## 🎯 Who This Is For

**Growth marketers** who are:
- ✅ Comfortable with analytics, experimentation, and data tools
- ✅ Technically curious but not developers
- ✅ Looking to automate workflows without engineering support

**You don't need to:**
- ❌ Write code (agents do that)
- ❌ Know SQL deeply (agents write the queries)
- ❌ Understand APIs (agents handle the integration logic)

**You DO need to:**
- ✅ Know what questions to ask
- ✅ Validate that outputs make sense
- ✅ Adapt prompts to your specific data and use cases

---

## 💡 Demo Highlights

### Demo 1: Funnel → Hypothesis → Experiment Brief
**Impact**: Turns funnel analysis into actionable A/B tests

**You'll learn**: How to go from "where are we losing users?" to a complete experiment brief in ClickUp, including hypothesis, metrics, and success criteria.

**Key output**: Markdown brief auto-posted to ClickUp project management.

---

### Demo 2: Paid Media Activation Suite
**Impact**: Creates audiences ready for Google Ads upload

**You'll learn**: How to identify high-value converters (enhanced conversions) and engaged non-converters (retargeting) and export properly formatted CSVs.

**Key output**: `enhanced_conversions.csv` and `retargeting_audience.csv` files ready for Customer Match upload.

---

### Demo 3: Anomaly Detection + Slack Alert
**Impact**: Proactive monitoring without manual dashboards

**You'll learn**: How agents can detect unusual metric changes, explain what happened, and trigger Slack alerts. Bonus: Auto-generate N8N workflows for scheduled checks.

**Key output**: Slack message with anomaly details + N8N workflow JSON.

---

### Demo 4: Trial-to-Paid + Feature Adoption
**Impact**: Identifies the "aha moment" that predicts conversion

**You'll learn**: Which features or behaviors correlate with trial conversion and how to focus onboarding efforts.

**Key output**: "Users who configure_integration in first 3 days convert at 70% vs 19% baseline"

---

### Demo 5: Tracking & UTM Audit
**Impact**: Catches tracking gaps and data hygiene issues

**You'll learn**: How to validate event tracking and identify UTM inconsistencies (case variations, typos, missing parameters).

**Key output**: QA report with missing events and suggested fixes.

---

### Demo 6: Live Website Audit (Vision)
**Impact**: Get CRO recommendations from page screenshots

**You'll learn**: How vision-capable agents can analyze landing pages for value prop clarity, CTA visibility, trust signals, and form complexity.

**Key output**: Prioritized CRO recommendations with specific changes.

---

## 🧠 Behind the Scenes: Specialized Agents

This repository uses **specialized agents** for platform-specific tasks:

- **`@agent-bigquery-ops`**: Runs SQL queries, exports CSVs, loads data
- **`@agent-playwright-browser`**: Captures screenshots, scrapes content
- **`@agent-slack-alerts`**: Sends formatted Slack messages
- **`@agent-n8n-workflows`**: Generates and deploys automation workflows
- **`@agent-hubspot-sync`**: Syncs contacts and lists to HubSpot

See [CLAUDE.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/CLAUDE.md) for full agent delegation guide.

---

## 📝 Scripts Note

All Python scripts in this repository were **auto-generated by Claude Code** during demo development. They serve as examples of what agents can produce and can be adapted for your own use cases.

Scripts include:
- Data generation (`scripts/generate_*.py`)
- Demo execution helpers (`demos/*/scripts/*.py`)
- Integration connectors (ClickUp, Slack, HubSpot)

---

## 🎬 Getting Started Paths

### Path 1: Explorer (No Setup)
**Time**: 15 minutes

1. Browse `demos/` folders
2. Read `PROMPT_SEQUENCE.md` files to see exact prompts
3. Review `example_outputs/` to see what agents produce
4. Pick the most relevant demo for your workflow

**No setup required** - just exploring what's possible.

---

### Path 2: Runner (Light Setup)
**Time**: 1-2 hours

1. Choose 1-2 demos that interest you
2. Follow [SETUP.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/SETUP.md) to configure those specific integrations
3. Run the demos with your AI agent of choice
4. Compare results to `example_outputs/`

**Minimal setup** - only configure what you need.

---

### Path 3: Adapter (Full Setup)
**Time**: Half day

1. Set up all integrations (BigQuery, Slack, ClickUp, etc.)
2. Run all 6 demos
3. Replace synthetic data with your own analytics data
4. Adapt prompts to your specific business context
5. Build custom demos for your unique workflows

**Full immersion** - make these agents part of your stack.

---

## 🤝 Contributing

This repository is open source to help marketers learn how to work with AI agents.

**Ways to contribute:**
- 🐛 Report issues or bugs in demos
- 💡 Suggest new demo use cases
- 📝 Share adaptations for your specific platforms
- ⭐ Star the repo if you find it useful

---

## 📚 Additional Resources

- **[DEMOS.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/DEMOS.md)**: Detailed descriptions of all 6 demos
- **[DATA.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/DATA.md)**: Complete dataset schemas and sample queries
- **[SETUP.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/SETUP.md)**: Step-by-step environment setup
- **[AGENTS.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/AGENTS.md)**: Project overview and key principles

---

## 🙏 Acknowledgments

These demos were created for the "Agents for Marketers: From Insight to Action" session, demonstrating practical applications of Claude Code, Codex CLI, and Gemini CLI for growth marketing workflows.

Special thanks to the growth marketing community for inspiring these use cases.

---

## 📄 License

This project is licensed under the MIT License - see the [LICENSE](https://github.com/funnelenvy/agents_webinar_demos/tree/HEAD/LICENSE) file for details. Feel free to adapt these demos for your own use.

---

**Ready to see what agents can do?** → Start with [SETUP.md](https://github.com/funnelenvy/agents_webinar_demos/blob/HEAD/SETUP.md) or dive into [`demos/`](https://github.com/funnelenvy/agents_webinar_demos/tree/HEAD/demos)

Agents for Marketers: Demo Repository

📺 Watch the Webinar [YouTube]: Agents for Marketers

Practical demonstrations of how AI agents (Claude Code, Codex, Gemini CLI) can automate analytics, CRO, and campaign workflows for growth marketers.

From Insight to Action: This repository shows how agents can query data, generate hypotheses, and produce actionable outputs—without requiring engineering support.

🎯 What You'll Find Here

Six production-ready demos that show agents:

Analyzing funnel drop-offs and generating A/B test hypotheses
Creating paid media audiences (enhanced conversions, retargeting)
Detecting anomalies and sending Slack alerts with N8N automation
Identifying "aha moments" in trial-to-paid conversion data
Auditing tracking implementation and UTM parameter consistency
Analyzing landing pages using vision capabilities for CRO recommendations

Each demo includes:

✅ Complete prompt sequences
✅ Synthetic datasets with engineered patterns
✅ Example outputs showing expected results
✅ SQL queries and Python scripts (auto-generated by Claude Code)
✅ Setup instructions for integrations (Slack, ClickUp, BigQuery, etc.)

🚀 Quick Start

1. Read the Playbook (Recommended)

Start with PLAYBOOK.md to understand:

Why agents matter for marketers (strategic context)
The Insight → Hypothesis → Action framework (mental model)
Real use cases with strategic narratives (what becomes possible)
Getting started guide (practical steps)
The vision (where this is going)

Then come back here for technical implementation.

2. Choose Your Demo

Browse the demos/ folder or read DEMOS.md for detailed descriptions:

New to agents? Start with Demo 2 (Paid Media Activation) - clear inputs, tangible outputs
Want to see vision? Try Demo 6 (Live Website Audit)
Love automation? Check out Demo 3 (Anomaly Detection + Slack + N8N)

3. Set Up Your Environment

Follow SETUP.md to:

Install dependencies
Configure environment variables (Slack webhooks, API tokens, etc.)
Load synthetic data into BigQuery or DuckDB

4. Run a Demo

Each demo folder contains:

demos/01-funnel-hypothesis-brief/
├── README.md              # Overview and instructions
├── PROMPT_SEQUENCE.md     # Exact prompts to copy/paste
├── queries/               # SQL queries
├── scripts/               # Python scripts (auto-generated)
└── example_outputs/       # What to expect

To run:

Open your AI agent tool (Claude Code, Codex CLI, etc.)
Follow the PROMPT_SEQUENCE.md step-by-step
Compare your results to example_outputs/

📊 Synthetic Datasets

All demos use realistic synthetic data with engineered patterns:

Funnel data with 40% drop-off at pricing page
Daily metrics with 3 anomalies (bug, improvement, activation issue)
Trial users with clear "aha moment" (configure_integration in first 3 days = 3.7x conversion)
Tracking data with intentional gaps (missing events)
UTM data with consistency issues (case variations, typos)

See DATA.md for complete schemas and sample queries.

🛠 Key Integrations

Platform	Required For	Setup Difficulty
BigQuery or DuckDB	Demos 1-5	⚡ Easy
Slack Webhook	Demo 3	⚡ Easy
ClickUp API	Demo 1	⭐ Medium
N8N Instance	Demo 3 (optional)	⭐⭐ Medium
Playwright	Demo 6	⭐ Medium

No integrations are required to explore the prompts and outputs—the data is all local CSV/JSON files.

🎓 Core Concept: Insight → Hypothesis → Action

These demos follow a three-step framework:

Insight: Agent queries data and identifies patterns
- Example: "40% of users drop off at the pricing page"
Hypothesis: Agent generates optimization ideas
- Example: "If we add social proof to pricing, conversion will increase because trust signals reduce purchase anxiety"
Action: Agent produces usable outputs
- Example: Experiment brief in ClickUp, audience CSV for Google Ads, Slack alert with context

This pattern works across analytics, CRO, paid media, PLG, and marketing ops.

📁 Repository Structure

agents_webinar/
├── README.md              # You are here
├── DEMOS.md               # Detailed demo descriptions
├── DATA.md                # Synthetic dataset documentation
├── SETUP.md               # Environment setup instructions
├── AGENTS.md              # Project overview and principles
├── CLAUDE.md              # Claude Code specific instructions
├── demos/                 # 6 demo workflows
│   ├── 01-funnel-hypothesis-brief/
│   ├── 02-paid-media-activation/
│   ├── 03-anomaly-detection/
│   ├── 04-trial-to-paid/
│   ├── 05-tracking-qa/
│   └── 06-vision-audit/
├── data/                  # Synthetic datasets
│   ├── users.csv
│   ├── events.csv
│   ├── daily_metrics.csv
│   ├── trial_users.csv
│   └── ...
├── scripts/               # Data generation scripts
└── .claude/               # Agent definitions and skills
    ├── agents/            # Specialized agent configs
    └── skills/            # Reusable skill modules

🎯 Who This Is For

Growth marketers who are:

✅ Comfortable with analytics, experimentation, and data tools
✅ Technically curious but not developers
✅ Looking to automate workflows without engineering support

You don't need to:

❌ Write code (agents do that)
❌ Know SQL deeply (agents write the queries)
❌ Understand APIs (agents handle the integration logic)

You DO need to:

✅ Know what questions to ask
✅ Validate that outputs make sense
✅ Adapt prompts to your specific data and use cases

💡 Demo Highlights

Demo 1: Funnel → Hypothesis → Experiment Brief

Impact: Turns funnel analysis into actionable A/B tests

You'll learn: How to go from "where are we losing users?" to a complete experiment brief in ClickUp, including hypothesis, metrics, and success criteria.

Key output: Markdown brief auto-posted to ClickUp project management.

Demo 2: Paid Media Activation Suite

Impact: Creates audiences ready for Google Ads upload

You'll learn: How to identify high-value converters (enhanced conversions) and engaged non-converters (retargeting) and export properly formatted CSVs.

Key output: enhanced_conversions.csv and retargeting_audience.csv files ready for Customer Match upload.

Demo 3: Anomaly Detection + Slack Alert

Impact: Proactive monitoring without manual dashboards

You'll learn: How agents can detect unusual metric changes, explain what happened, and trigger Slack alerts. Bonus: Auto-generate N8N workflows for scheduled checks.

Key output: Slack message with anomaly details + N8N workflow JSON.

Demo 4: Trial-to-Paid + Feature Adoption

Impact: Identifies the "aha moment" that predicts conversion

You'll learn: Which features or behaviors correlate with trial conversion and how to focus onboarding efforts.

Key output: "Users who configure_integration in first 3 days convert at 70% vs 19% baseline"

Demo 5: Tracking & UTM Audit

Impact: Catches tracking gaps and data hygiene issues

You'll learn: How to validate event tracking and identify UTM inconsistencies (case variations, typos, missing parameters).

Key output: QA report with missing events and suggested fixes.

Demo 6: Live Website Audit (Vision)

Impact: Get CRO recommendations from page screenshots

You'll learn: How vision-capable agents can analyze landing pages for value prop clarity, CTA visibility, trust signals, and form complexity.

Key output: Prioritized CRO recommendations with specific changes.

🧠 Behind the Scenes: Specialized Agents

This repository uses specialized agents for platform-specific tasks:

@agent-bigquery-ops: Runs SQL queries, exports CSVs, loads data
@agent-playwright-browser: Captures screenshots, scrapes content
@agent-slack-alerts: Sends formatted Slack messages
@agent-n8n-workflows: Generates and deploys automation workflows
@agent-hubspot-sync: Syncs contacts and lists to HubSpot

See CLAUDE.md for full agent delegation guide.

📝 Scripts Note

All Python scripts in this repository were auto-generated by Claude Code during demo development. They serve as examples of what agents can produce and can be adapted for your own use cases.

Scripts include:

Data generation (scripts/generate_*.py)
Demo execution helpers (demos/*/scripts/*.py)
Integration connectors (ClickUp, Slack, HubSpot)

🎬 Getting Started Paths

Path 1: Explorer (No Setup)

Time: 15 minutes

Browse demos/ folders
Read PROMPT_SEQUENCE.md files to see exact prompts
Review example_outputs/ to see what agents produce
Pick the most relevant demo for your workflow

No setup required - just exploring what's possible.

Path 2: Runner (Light Setup)

Time: 1-2 hours

Choose 1-2 demos that interest you
Follow SETUP.md to configure those specific integrations
Run the demos with your AI agent of choice
Compare results to example_outputs/

Minimal setup - only configure what you need.

Path 3: Adapter (Full Setup)

Time: Half day

Set up all integrations (BigQuery, Slack, ClickUp, etc.)
Run all 6 demos
Replace synthetic data with your own analytics data
Adapt prompts to your specific business context
Build custom demos for your unique workflows

Full immersion - make these agents part of your stack.

🤝 Contributing

This repository is open source to help marketers learn how to work with AI agents.

Ways to contribute:

🐛 Report issues or bugs in demos
💡 Suggest new demo use cases
📝 Share adaptations for your specific platforms
⭐ Star the repo if you find it useful

📚 Additional Resources

DEMOS.md: Detailed descriptions of all 6 demos
DATA.md: Complete dataset schemas and sample queries
SETUP.md: Step-by-step environment setup
AGENTS.md: Project overview and key principles

🙏 Acknowledgments

These demos were created for the "Agents for Marketers: From Insight to Action" session, demonstrating practical applications of Claude Code, Codex CLI, and Gemini CLI for growth marketing workflows.

Special thanks to the growth marketing community for inspiring these use cases.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details. Feel free to adapt these demos for your own use.

Ready to see what agents can do? → Start with SETUP.md or dive into demos/