Sales success starts with great lead research—but let’s face it, the process can feel like an uphill battle. Sifting through endless tabs, chasing down decision-makers, and piecing together insights from company websites can quickly turn into a time-consuming grind. And after all that effort, many outreach attempts still miss the mark, lacking the personal touch needed to stand out and make a real connection.
In a sales world that moves at lightning speed, efficiency isn’t just a nice-to-have—it’s a must-have. By automating the most tedious tasks, sales teams can reclaim their time and focus on what truly matters: building genuine relationships and closing more deals.
In this technical guide, we’ll explore how I build an AI outreach automation for a marketing agency that:
Gathers insights from LinkedIn, websites, and social media,
Automatically evaluates and qualifies leads, and
Generates tailored outreach materials that help you stand out in crowded inboxes.
Let’s get started!
Why Sales Outreach Needs AI Automation?
Sales outreach is crucial for business success, yet many teams face challenges due to outdated methods. Here are key reasons why automation is essential:
Time-consuming research limits sales professionals to manual data gathering, reducing time for actual lead engagement.
Inconsistent qualification occurs when processes rely on incomplete data or subjective decisions.
Lack of personalization makes generic outreach efforts easy to ignore, diminishing client engagement.
Scalability issues arise as manual processes struggle to keep up with growing outreach demands.
Missed opportunities happen when leads slip through the cracks due to delayed follow-ups or incomplete insights.
Simplifying Lead Outreach with AI Automation
To tackle the complexities of lead research, qualification, and outreach for an AI marketing agency (ElevateAI), I created a fully automated system that use multiple AI agents to accelerate and streamline the entire process. Here’s a high-level breakdown:
High-Level Overview
CRM Integration: Connects with platforms like HubSpot, Airtable, and Google Sheets to streamline lead data access and synchronization.
-
Lead Research: Gathers insights on leads and their company, including:
- LinkedIn data (profile, roles, company size, industry).
- Company website, blogs, and social media activities.
- Recent company news for relevant outreach.
- Identification of challenges and opportunities.
Analysis Reports: Create detailed analysis reports using AI from all gathered information, helping sales teams better understand the lead and make informed decisions.
Lead Qualification: Scores leads based on factors like online presence, industry fit, and growth potential.
-
Personalized Outreach:
- Custom Outreach Report: Generate a company audit report that outlines the challenges identified, how your services can address them, and highlights relevant case studies.
- Craft a personalized email using all collected insights, including a link to the custom outreach report, to engage the lead effectively.
- Generate custom interview scripts to facilitate productive calls and meetings with the lead.
Data Management: All generated research reports are saved to Google Drive and CRM is updated with the collected data for easy access.
Key Benefits
Automated Lead Research: Gathers insights from LinkedIn, websites, social media, and more, ensuring thorough lead evaluation.
Higher Reply Rates & Conversions: Attach detailed audit reports to outreach emails, increasing the likelihood of positive responses.
Time-Saving & Improved Efficiency: Automates research and report generation, freeing up time and streamlining team efforts.
Technical Deep Dive
To build this automation, I used Langgraph and LangChain. Langgraph facilitates the development of complex workflows, while LangChain enables integration with different LLMs (Google Gemini, GPT-4o, and LLaMA 3). We use the power of LLMs to interpret insights, generate analytical reports, craft personalized outreach materials, and qualify leads.
Below is a glimpse of the automation graph structure developed using Langgraph:
# Entry point of the graph
graph.set_entry_point("get_new_leads")
graph.add_edge("fetch_linkedin_profile_data", "review_company_website")
graph.add_edge("review_company_website", "collect_company_information")
graph.add_edge("collect_company_information", "analyze_blog_content")
graph.add_edge("collect_company_information", "analyze_social_media_content")
graph.add_edge("collect_company_information", "analyze_recent_news")
graph.add_edge("analyze_recent_news", "generate_digital_presence_report")
graph.add_edge("generate_digital_presence_report", "generate_full_lead_research_report")
graph.add_edge("generate_full_lead_research_report", "score_lead")
graph.add_conditional_edges(
"score_lead",
nodes.check_if_qualified,
{
"qualified": "generate_custom_outreach_report", # Proceed if lead is qualified
"not qualified": "save_reports_to_google_docs" # Save reports and exit if lead is unqualified
}
)
# Outreach material creation
graph.add_edge("generate_custom_outreach_report", "create_outreach_materials")
graph.add_edge("create_outreach_materials", "generate_personalized_email")
graph.add_edge("create_outreach_materials", "generate_interview_script")
graph.add_edge("generate_personalized_email", "save_reports_to_google_docs")
# Save reports and update the CRM
graph.add_edge("save_reports_to_google_docs", "update_CRM")
Now, let’s break down each component of this automation in detail:
Fetching Leads from CRM
The process begins with extracting lead information (e.g., name, email, phone number) from CRM platforms such as Airtable, HubSpot, or Google Sheets. The system connects to these platforms using a standardized class structure, ensuring seamless integration and making it easy to extend support for new CRMs without disrupting the automation.
Here’s an example of how it connects to Airtable:
class AirtableLeadLoader(LeadLoaderBase):
def __init__(self, access_token, base_id, table_name):
# Use the access_token instead of api_key
self.table = Table(access_token, base_id, table_name)
def fetch_records(self, lead_ids=None, status_filter="NEW"):
"""
Fetches leads from Airtable. If lead IDs are provided, fetch those specific records.
Otherwise, fetch leads matching the given status.
"""
if lead_ids:
leads = []
for lead_id in lead_ids:
record = self.table.get(lead_id)
if record:
# Merge id and fields into a single dictionary
lead = {"id": record["id"], **record.get("fields", {})}
leads.append(lead)
return leads
else:
# Fetch leads by status filter (based on "Status" field)
# You can choose your own field for filter with different naming
records = self.table.all(formula=match({"Status": status_filter}))
return [
{"id": record["id"], **record.get("fields", {})}
for record in records
]
def update_record(self, lead_id, updates: dict):
"""Updates a record in Airtable"""
# Fetch the current record to ensure it exists and get its fields
record = self.table.get(lead_id)
if not record:
raise ValueError(f"Record with ID {lead_id} not found.")
# Merge current fields with updates, adding any new fields
current_fields = record.get("fields", {})
updated_fields = {**current_fields, **updates}
# Update the record in Airtable
return self.table.update(lead_id, updated_fields)
In this example, the fetch_records function retrieves new leads based on their status, while the update_record function saves any insights gathered during the outreach process directly into the CRM.
Enriching Lead Profiles via LinkedIn
Once the basic lead information is retrieved, the system enhances their profiles by gathering insights from LinkedIn. This involves finding and scraping LinkedIn profiles for both the lead and their company.
Here’s how the system retrieves company details:
def research_lead_company(linkedin_url):
# Scrape company LinkedIn profile
company_page_content = scrape_linkedin(linkedin_url, True)
if "data" not in company_page_content:
return "LinkedIn profile not found"
# Structure collected information about company
company_profile = company_page_content["data"]
return {
"company_name": company_profile.get('company_name', ''),
"description": company_profile.get('description', ''),
"year_founded": company_profile.get('year_founded', ''),
"industries": company_profile.get('industries', []),
"specialties": company_profile.get('specialties', ''),
"employee_count": company_profile.get('employee_count', ''),
"social_metrics": {
"follower_count": company_profile.get('follower_count', 0)
},
"locations": company_profile.get('locations', [])
}
What’s happening here? The function scrapes the company’s LinkedIn profile to extracts key details such as industry, company size, and description.
A similar approach is used to gather details about the lead’s professional background, including their roles, skills, and experience.
Scraping the Company Website
The automation goes a step further by scraping the company’s website to gather deeper insights. This is done using the following function, which scrapes a URL and converts its HTML content into markdown, making it more readable for LLMs:
def scrape_website_to_markdown(url: str) -> str:
headers = {
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.77 Safari/537.36",
"Accept-Language": "en-US,en;q=0.5",
"Accept-Encoding": "gzip, deflate"
}
# Make the HTTP request
response = requests.get(url, headers=headers)
if response.status_code != 200:
raise Exception(f"Failed to fetch the URL. Status code: {response.status_code}")
# Parse the HTML
soup = BeautifulSoup(response.text, "html.parser")
html_content = soup.prettify()
# Convert HTML to markdown
h = html2text.HTML2Text()
h.ignore_links = False
h.ignore_images = True
h.ignore_tables = True
markdown_content = h.handle(html_content)
# Clean up excess newlines
markdown_content = re.sub(r"\n{3,}", "\n\n", markdown_content)
markdown_content = markdown_content.strip()
return markdown_content
We then use AI to summarize the website content, focusing on key information such as the company's mission, products, and services. Additionally, it extracts relevant links, including those to social media accounts and blogs.
WEBSITE_ANALYSIS_PROMPT = """
The provided webpage content is scraped from: {main_url}.
# Tasks
## 1- Summarize webpage content:
Write a 1500-word comprehensive summary in markdown format about the content of the webpage,
focusing on relevant information related to the company mission, products, and services.
## 2- Extract and categorize the following links:
1. Blog URL: Extract the main blog URL of the company.
2. Social Media Links: Extract links to the company's YouTube, Twitter, and Facebook profiles.
# IMPORTANT:
* Ensure the summary is organized in markdown format.
* Ensure that only the specified categories of links are included.
* If a link is not found, its value is an empty string.
* If the link is relative (e.g., "/blog"), prepend it with {main_url} to form an absolute URL.
"""
If the company operates a blog, we also scrape it and use an AI agent to extract relevant insights, such as:
Total Number of Blog Posts: Measures the volume of published content.
Posting Frequency: Identifies patterns (e.g., consistent, irregular, or inactive).
Key Themes and Topics: Highlights recurring subjects and main themes.
Analyzing Social Media Metrics
To better understand the company’s social media presence, we examine its activity on popular platforms such as Facebook, Twitter, YouTube, and TikTok. This involves scraping key metrics—posts, videos, followers, and more—followed by AI analysis to generate detailed reports on content themes, engagement levels, and actionable opportunities for improvement.
Example: Analyzing YouTube Performance
For YouTube, the following function is used to collect key statistics from a company’s channel:
def get_youtube_stats(channel_url):
"""
Collects statistics and metadata from a company's YouTube channel.
"""
channel_name = extract_channel_name(channel_url)
channel_id = get_channel_id_by_name(channel_name)
result = get_channel_videos_stats(channel_id)
return {
"total_videos": result['total_videos'],
"subscriber_count": result['subscriber_count'],
"average_views": result['average_views'],
"recent_videos": [
{"title": video["title"], "published_at": video["published_at"]}
for video in result["last_15_videos"]
]
}
This function retrieves key details, including:
Total number of videos
Subscriber count
Average views per video
Details of recent videos
The gathered data is passed to an AI marketing agent instructed to generate a detailed analysis report:
YOUTUBE_ANALYSIS_PROMPT = """
# **Role:**
You are a Professional Marketing Analyst specializing in evaluating YouTube channel
performance and identifying actionable insights to improve content strategies.
---
# **Task:**
Analyze the provided YouTube channel's content and generate a detailed performance report.
This report will evaluate the channel's activity, relevance to the company’s services,
and opportunities for improvement.
# **Specifics:**
Your report will include the following sections:
## **Channel Summary:**
* **Number of Videos:** Count of videos provided for analysis.
* **Activity:** Describe the frequency of uploads (e.g., consistent, irregular, or inactive).
* **Engagement:** Summarize key engagement metrics (e.g., average views, likes, and comments
per video).
* **Summary of Topics:** Highlight the main themes and subjects covered.
* **Examples:** List five representative video titles and descriptions.
## **Scoring:**
Evaluate the channel in these categories: **Volume of Videos**, **Upload Activity**,
**Engagement Levels**, **Relevance to the Company’s Services**
Combine these to calculate a total performance score.
## **Opportunities for Improvement:**
* Identify content gaps or underexplored topics.
* Suggest new themes and innovative content formats (e.g., shorts, live streams, tutorials).
## **Action Plan:**
Provide actionable recommendations to boost engagement, relevance, and activity.
"""
This analysis evaluates the channel’s activity, audience engagement, and alignment with the company’s goals. It also provides targeted strategies to enhance visibility, improve content relevance, and boost engagement
Recent News Review
In the final step of our research, we use the Google News API to fetch the latest news and announcements related to the company. This step provides valuable insights into the company’s recent initiatives, achievements, and potential challenges:
def get_recent_news(company: str) -> str:
# Define the payload and headers for the request
url = "https://google.serper.dev/news"
payload = json.dumps({"q": company, "num": 20, "tbs": "qdr:y"})
headers = {
'X-API-KEY': os.getenv("SERPER_API_KEY"),
'Content-Type': 'application/json'
}
# Make the POST request to the API
response = requests.post(url, headers=headers, data=payload)
# Check if the response is successful
if response.status_code == 200:
news = response.json().get("news", [])
# Prepare the string to return
news_string = ""
news.reverse() # Reverse the list to get the most recent news first
for item in news:
title = item.get('title')
snippet = item.get('snippet')
date = item.get('date')
link = item.get('link')
news_string += f"Title: {title}\nSnippet: {snippet}\nDate: {date}\nURL: {link}\n\n"
return news_string
else:
return f"Error fetching news: {response.status_code}"
Reports Generation
Each step of the analysis culminates in a detailed, AI-generated report. These reports serve two primary purposes:
Empowering Sales Teams: By providing rich insights to facilitate more informed and personalized engagements.
Supporting Outreach Preparation: By offering a clear foundation for crafting tailored materials for outreach.
The reports include the following:
Lead Profile Report: Generated from LinkedIn and company website data, detailing the lead’s professional background and company information.
Blog Analysis Report: Insights extracted from blog content to highlight themes, posting frequency, and engagement.
Social Media Analysis Reports: Detailed evaluations for each social media platform, covering activity, engagement, and strategic opportunities.
Recent News Analysis: A summary of the latest news, emphasizing recent achievements and challenges.
Digital Presence Report: Combines insights from blogs, news articles, and social media to provide a holistic view of the company’s online presence.
Global Research Report: Consolidates findings from all reports into a comprehensive document for internal use.
All reports are saved locally and synchronized to Google Docs, ensuring easy access and collaboration among team members.
Lead Qualification
Once the research process is complete, the next step is to qualify the lead. This is also done using an AI agent, which analyzes the comprehensive reports generated previously and determines if the lead is qualified against predefined criteria.
Given that I built this system for an AI marketing agency, the lead qualification requirements are based on that. Here is a glimpse of the AI lead scoring prompt:
SCORE_LEAD_PROMPT = """
# **Role & Task**
You are an expert lead scorer for **ElevateAI Marketing Solutions**, tasked with evaluating and scoring leads based on their digital presence, social media activity, industry fit, company scale, and marketing strategy.
# **Scoring Criteria**
- **Digital Presence:** Blog activity and website quality.
- **Social Media Activity:** Platform presence, engagement rates, and posting frequency.
- **Industry Fit:** Relevance to target industries and use of AI/automation.
- **Company Scale:** Growth signals and alignment with ElevateAI's focus.
- **Marketing Strategy:** Use of tools and consistency in messaging.
- **Pain Points & ROI:** Identifiable challenges and potential ROI from ElevateAI solutions.
# **Output**
Provide a final score (1–10) based on these criteria, with no additional explanation.
"""
For this implementation I choose to consider that if a lead gets a score 7.0 or above is considered qualified, marking them as high-priority for our outreach efforts.
Preparing Outreach Materials
For qualified leads, the system crafts personalized outreach materials to maximize engagement:
Custom Outreach Report
For every qualified lead, we task an AI writer agent to generate a Custom Outreach Report—a dynamic, data-driven presentation showcasing how ElevateAI can address their unique challenges.
The outreach report includes:
Business Overview: A concise profile of the lead’s company, detailing their industry, core offerings, and competitive position.
Challenges Identified: A breakdown of specific pain points in their current marketing or business strategy, such as limited digital presence, low engagement rates, or operational inefficiencies.
AI-Powered Solutions: Customized recommendations that leverage ElevateAI’s capabilities, highlighting how our solutions align with their business goals.
Tangible Results: Real-world impact examples, drawn from our knowledge base using Retrieval-Augmented Generation (RAG). These examples feature case studies from similar businesses, showcasing measurable outcomes, such as improved ROI, increased customer retention, or enhanced operational efficiency.
Clear Call-to-Action (CTA): A compelling CTA, such as scheduling a free consultation or demo, encouraging leads to take the next step in the sales funnel.
Here’s a sample of a Custom Outreach Report created for Relevance AI for example:
You can also view the Full Report if you want!
Personalized Email & Interview Script
For each qualified lead, we use another AI email writer agent to craft a highly personalized email designed to address their specific needs and challenges. The email highlights the value ElevateAI can deliver, includes a link to the custom outreach report, and emphasizes measurable benefits such as increased ROI, enhanced customer engagement, or operational efficiencies.
To support sales teams, the system also generates an interview script rooted in SPIN selling principles. This script includes strategic, open-ended questions to guide discussions, uncover the lead’s pain points, and position ElevateAI’s solutions as the ideal fit for their business objectives.
Save to CRM
Once the system processes a lead, all generated data is saved back to the CRM, including report links, the lead's score, and their updated status.
Make It Your Own
While this AI outreach automation was specifically designed and configured for the needs of an AI marketing agency, it's fully customizable to fit any business. Here's the flexibility you get:
Your CRM: Integrates with whatever CRM you use.
Custom Reports: Choose which reports to generate (e.g., prioritize social media, skip blog analysis) and modify their structure to focus on the data you need.
Your Ideal Lead: Adjust lead scoring criteria to match your target audience.
Your Brand Voice: Tailor email templates and interview scripts to your own business.
Expand as Needed: Add data sources, AI models, or outreach channels.
It's a powerful, flexible automation ready to be customized for your unique outreach strategy.
Conclusion
Sales outreach doesn’t have to be a draining process. With AI automation and AI agents, you can transform how you connect with leads, shifting from generic emails and manual research to hyper-personalized, data-driven strategies.
By leveraging Langgraph and Langchain, I built an AI automation can level up your sales outreach by:
Saving time on manual research.
Boosting conversions with hyper-personalized strategies.
Streamlining processes for smarter, more efficient workflows.
🎯 Want to see the full details? Check out the complete project on my GitHub and take your sales game to the next level!
Top comments (0)