Skip to main content

Build a Lead Generation & Outreach Agent

Learning Path

Level 4, Step 4.5 — Learning Path. Prerequisites: Human in the Loop, Structured Output.

This is one of the highest-value projects you can build with CiniterFlow. It directly generates revenue by automating the most time-consuming part of sales — researching prospects and writing personalized outreach emails.

What You'll Build​

A lead outreach system that:

  • Takes a list of prospect companies or contacts
  • Researches each prospect using web search
  • Generates personalized outreach emails based on the research
  • Presents each email for human approval before sending
  • Processes prospects in batch using iteration

Why This Matters​

A sales rep spends 2-3 hours per day researching prospects and writing emails. This agent does the same work in minutes, and the emails are often better because they're informed by real-time web research.

The Architecture​

Start (Form: prospect list) → LLM (parse list into structured data)
    → Iteration (for each prospect)
        → Agent (research prospect with web search)
        → LLM (generate personalized email)
    → Human Input (review all emails)
    → Iteration (send approved emails)
        → Tool (Gmail/SendGrid)

Step 1: Create the Flow​

  1. Create a new Agentflow named "Lead Outreach Agent"
  2. Configure the Start node:
    • Input Type: Form Input
    • Form Title: "Prospect Outreach"
    • Form Description: "Enter prospect details for personalized outreach"
    • Form Inputs:
      • Label: "Prospects", Variable: prospects, Type: String (textarea)
      • Label: "Your Product/Service", Variable: product, Type: String
      • Label: "Outreach Goal", Variable: goal, Type: String
    • Flow State:
      • prospectList: ""
      • emails: ""

The user will enter prospects like:

1. John Smith, CTO at TechCorp (john@techcorp.com)
2. Sarah Lee, VP Marketing at GrowthCo (sarah@growthco.com)
3. Mike Chen, Head of Operations at ScaleUp (mike@scaleup.io)

Step 2: Parse Prospects into Structured Data (LLM Node)​

Model: GPT-4o-mini (cheap, fast — just parsing text)

System Message:

You are a data parser. Extract prospect information from the user's input into a structured format. Handle various input formats gracefully.

Input Message:

Parse the following prospect list into structured data:

{{ $form.prospects }}

JSON Structured Output:

  • Key: prospects, Type: JSON Array
  • Schema:
{
  "name": { "type": "string", "description": "Full name of the prospect" },
  "title": { "type": "string", "description": "Job title" },
  "company": { "type": "string", "description": "Company name" },
  "email": { "type": "string", "description": "Email address" }
}

Update Flow State: prospectList → {{ output.prospects }}

Step 3: Research Each Prospect (Iteration + Agent)​

Iteration Node​

  • Array Input: {{ $flow.state.prospectList }}

Agent Node (inside iteration)​

Model: GPT-4o-mini

System Message:

You are a sales research assistant. Your job is to research a prospect and their company to find information that will help personalize a sales outreach email.

Look for:
- What the company does and their recent news
- The prospect's role and likely responsibilities
- Pain points the company might have that our product could solve
- Any recent achievements, funding rounds, or announcements
- Shared connections or interests

Keep your research summary concise — 3-5 bullet points of the most useful findings.

Tools:

  • Google Custom Search (or Serper): For web research
  • Web Scraper: To read company websites and LinkedIn profiles

Input Message:

Research this prospect for a sales outreach:
- Name: {{ $iteration.name }}
- Title: {{ $iteration.title }}
- Company: {{ $iteration.company }}

Our product/service: {{ $form.product }}
Our outreach goal: {{ $form.goal }}

Find information that will help us write a personalized, relevant outreach email.

Step 4: Generate Personalized Email (LLM Node, inside iteration)​

Model: GPT-4o (use the better model for the actual email — this is the output that matters)

System Message:

You are an expert sales copywriter. Write personalized outreach emails that feel genuine, not templated.

Rules:
1. Open with something specific to the prospect (a recent achievement, news, or shared interest)
2. Connect their situation to how our product/service can help
3. Keep it under 150 words — busy executives don't read long emails
4. End with a clear, low-commitment call to action (e.g., "Would a 15-minute call next week make sense?")
5. Tone: Professional but warm. Never salesy or pushy.
6. Never use phrases like "I hope this email finds you well" or "I wanted to reach out"

Subject line: Make it specific and curiosity-inducing. Never generic.

Input Message:

Write a personalized outreach email for:

Prospect: {{ $iteration.name }}, {{ $iteration.title }} at {{ $iteration.company }}
Email: {{ $iteration.email }}

Research findings:
{{ agentAgentflow_0 }}

Our product/service: {{ $form.product }}
Our goal: {{ $form.goal }}

JSON Structured Output:

  • subject (string): "Email subject line"
  • body (string): "Email body"
  • recipientEmail (string): "Recipient email address"
  • recipientName (string): "Recipient name"

Step 5: Review All Emails (Human Input)​

After the iteration completes, add a Human Input node:

  • Description Type: Dynamic
  • Prompt: "Present all the generated outreach emails for review. Show each email with the recipient name, subject line, and body. Ask the reviewer to approve or request changes."
  • Enable Feedback: Yes

This pauses the flow and lets a human review every email before anything is sent.

If rejected → Loop back to the email generation step with feedback If approved → Continue to sending

Step 6: Send Approved Emails (Iteration + Tool)​

Second Iteration Node​

  • Array Input: The approved emails from the previous step

Tool Node (inside iteration)​

  • Tool: Gmail (or SendGrid for higher volume)
  • Action: Send Message
  • Input Arguments:
    • to: {{ $iteration.recipientEmail }}
    • subject: {{ $iteration.subject }}
    • body: {{ $iteration.body }}

Testing​

  1. Start the flow with 2-3 test prospects (use your own email addresses for testing!)
  2. Enter your product description and outreach goal
  3. Watch the agent research each prospect
  4. Review the generated emails — are they personalized? Relevant? Professional?
  5. Approve and verify emails are sent correctly

Example Input​

Prospects:

1. Lisa Park, VP of Engineering at DataFlow Inc (lisa@dataflow.io)
2. James Wilson, Head of AI at CloudScale (james@cloudscale.com)

Product/Service:

CiniterFlow — a visual platform for building AI agents without code. Helps companies automate customer support, data processing, and internal workflows.

Outreach Goal:

Book a 15-minute demo call to show how they can build AI agents for their team.

Tips for Production​

  1. Start small: Test with 5-10 prospects before scaling up
  2. Personalization is everything: The research step is what makes these emails effective. Give the research agent good tools and clear instructions
  3. A/B test subject lines: Try different styles and track open rates
  4. Respect sending limits: Gmail has daily sending limits. For high volume, use SendGrid or a dedicated email service
  5. Track responses: Create a follow-up flow that handles replies
  6. Compliance: Make sure your outreach complies with CAN-SPAM, GDPR, or other relevant regulations. Include an unsubscribe option

Taking It Further​

  • Add a CRM integration (HubSpot, Airtable) to log outreach and track responses
  • Build a follow-up agent that sends a second email if no response after 3 days
  • Add LinkedIn research via web scraping for deeper personalization
  • Create a scoring system that prioritizes high-value prospects
  • Build a response handler that classifies replies (interested, not interested, out of office) and routes accordingly