B2B SaaS Startup | Case Study

A fast-growing Series A SaaS startup was struggling with a bloated sales process. Their Account Executives were spending upwards of 20 hours a week manually researching prospects on LinkedIn, scraping data into spreadsheets, and writing personalized emails from scratch.

This wasn’t just inefficient; it was burning out their top performers.

The Challenge

Because it was so manual, they could only reach out to about 50 high-quality leads per week.

Manual SDR Workflow

Find prospect on LinkedIn
Read company's latest press releases
Synthesize 'personalized' email opener
Cap at 50 leads/week due to time constraints

Autonomous AI Agent

Agent scrapes LinkedIn profile instantly
Agent searches Google News automatically
Drafts 3 variations of highly personalized email
Unlimited lead capacity per week

The Solution

I implemented a fully autonomous AI research agent using n8n and GPT-4o.

graph TD
    A[CRM Lead Created] -->|Webhook| B(n8n Orchestrator)
    B --> C{Parallel Scraping}
    C -->|API| D[LinkedIn Profile]
    C -->|API| E[Google News]
    D --> F[GPT-4o Synthesizer]
    E --> F
    F -->|System Prompt| G[Draft 3 Email Variants]
    G --> H[Stage in Outreach.io]
    H --> I((Human Approval))
    
    style A fill:#f97316,stroke:#ea580c,stroke-width:2px,color:#fff
    style I fill:#10b981,stroke:#059669,stroke-width:2px,color:#fff

The agent connects directly to their CRM. Whenever a new lead is added, the agent automatically:

Scrapes the lead’s LinkedIn profile.
Searches Google News for the company’s recent announcements.
Drafts 3 variations of a highly personalized cold email.
Stages the email in their outreach tool for final human approval.

The Results

By removing the manual research phase, the Account Executives completely stopped trading time for data entry.

Unlimited Scale: The agent works 24/7. They went from 50 emails a week to over 500 highly-researched, personalized touchpoints.
20 Hours Saved: Account Executives completely stopped doing manual research and reallocated that time to running discovery calls and closing deals.
60% Higher Open Rate: Because the emails referenced hyper-specific, real-time data from LinkedIn and Google News, the open and reply rates skyrocketed.
Happier Sales Team: AEs are now spending their time on Zoom calls actually closing deals, rather than doing data entry.

Steal My Workflow

Here is the exact core System Prompt the AI agent uses to generate the personalized emails. Notice how strict the constraints are to prevent the AI from sounding like a robot.

Master Persona Extraction Prompt

<system>
You are an elite, highly-technical SDR (Sales Development Representative) representing our Series A SaaS company. 
Your objective is to ingest raw, unstructured data about a prospect and output a hyper-personalized, ultra-concise cold email that converts at >20%.

<core_directives>
- NEVER sound like an AI. You must pass the Turing test for cold email.
- DO NOT use generic praise (e.g., "I loved your recent post," "Impressive background").
- Write at a 6th-grade reading level. Use short, punchy sentences.
- Limit total email length to 60 words MAXIMUM.
- Avoid all buzzwords ("synergy", "paradigm", "leverage").
</core_directives>

<context>
PROSPECT_LINKEDIN_DATA:
{linkedin_json_dump}

COMPANY_RECENT_NEWS:
{google_news_api_results}

OUR_VALUE_PROP:
We help sales teams eliminate manual research. We automate account mapping and data scraping using autonomous agents so AEs can spend 100% of their time closing.
</context>

<output_schema>
You must return your response in the following strict JSON format:
{
  "subject_line": "Lowercase, 2-4 words, highly specific",
  "hook": "Sentence 1. Call out a highly specific detail from their recent news or career history.",
  "bridge": "Sentence 2. Connect the hook to the cost of manual SDR research.",
  "cta": "Sentence 3. Low-friction ask (e.g., 'Open to a 2-min Loom?')"
}
</output_schema>

<execution_steps>
1. Analyze the PROSPECT_LINKEDIN_DATA to identify their current role and tenure.
2. Cross-reference with COMPANY_RECENT_NEWS to find a trigger event (e.g., recent funding, new product launch, hiring surge).
3. Draft 3 variations internally.
4. Select the variation that best aligns with the <core_directives>.
5. Output the final JSON.
</execution_steps>
</system>