Embedded Files
Automated Lead Intelligence
Executive Summary
Executive Summary
Objective: To reduce "Speed-to-Lead" latency and eliminate manual pre-call research for high-value sales prospects.
Tech Stack: n8n (Orchestrator), Apify (LinkedIn Scraper), Tavily (AI Search), OpenAI GPT-5 Mini (Data Synthesis).
Situation: The "Context Gap" in B2B Sales
Situation: The "Context Gap" in B2B Sales
In high-velocity sales environments, response time and information quality are critical. When a high-value lead arrives, sales representatives typically face a bottleneck: they must not only act fast but also have complete context and background information on potential customer.
The Problem:
Manual Inefficiency: Gathering data from LinkedIn, news sites, and company blogs is repetitive, low-leverage work.
Data Fragmentation: Critical insights (e.g., "Company X just raised Series B") are often missed in a quick Google search.
Task: Automating the Analyst Role
Task: Automating the Analyst Role
My goal was to engineer a "Zero-Touch" intelligence pipeline. The system needed to:
Ingest raw lead data from a web form.
Autonomously perform deep-dive research on both the individual (via LinkedIn) and the company (via live web search).
Synthesize the unstructured data into a structured executive dossier.
Deliver the report via email before the sales rep could even open the notification.
Action: Solution Architecture
Action: Solution Architecture
I designed an event-driven architecture using n8n to orchestrate multiple AI agents in parallel.
A. The Trigger (Data Ingestion)
The workflow initiates via a Webhook connected to a "Contact Us" form simulation. It captures four key data points: Name, Email, LinkedIn URL, and Company Website.
Image 1: The ingestion interface capturing raw lead metadata for Aadit Palicha, Zepto (for illustration only)
B. The Workflow Logic (Parallel Processing)
To ensure low latency, I implemented a "Fork-Join" pattern in n8n, splitting the execution into two simultaneous threads:
Figure 2: The n8n orchestration layer. Note the parallel execution paths to minimize runtime.
Path 1: Social Intelligence (Apify): I triggered a specialized Apify Actor to scrape Aadit’s LinkedIn profile. This agent extracted structured biographical data, including his education (Stanford University dropout), previous founding roles, and volunteer history.
Path 2: Market Intelligence (Tavily): Simultaneously, I utilized Tavily API, a search engine optimized for LLMs. Unlike standard search APIs which return SEO clutter, Tavily was configured to retrieve high-authority financial news, specifically looking for "Zepto funding rounds," "IPO plans," and "Quick Commerce regulatory challenges."
Image 2: The n8n orchestration layer. Note the parallel execution paths to minimize runtime.
C. Data Synthesis (LLM Transformation)
Both data streams (JSON payloads) were merged and fed into OpenAI’s GPT-5 Mini. I engineered a strict system prompt to force the model to act as a Market Analyst. It was tasked with discarding irrelevant noise and structuring the insights into three specific categories: Strengths, Industry Challenges, and Recent News.
Result: The Outcome
Result: The Outcome
The system successfully processed the lead in under 45 seconds, delivering a fully enriched dossier directly to the inbox.
Key Performance Indicators (KPIs):
Accuracy: The system correctly identified Zepto’s "10-minute delivery" model and its current valuation status (~$5B).
Formatting: The output was delivered as a clean HTML table, ready for immediate consumption on mobile or desktop.
Image 3: The final automated output. A structured briefing delivered instantly to the decision-maker.
Future Scalability & Roadmap
Future Scalability & Roadmap
This architecture serves as the backend for a scalable "AI SDR" (Sales Development Representative). The next phase of development involves:
CRM Integration (HubSpot/Salesforce): Replacing the email notification with a direct API POST request to the CRM. This would auto-enrich the Contact record with fields like Annual_Revenue, Last_Funding_Date, and Competitors, ensuring database hygiene.
Voice AI Dispatch: Connecting the workflow to a telephony agent (e.g., Vapi or Bland AI). Upon generating the report, the system could trigger an outbound call to the lead within 2 minutes, using the "Profile Summary" to personalize the script dynamically.
Hyper-Personalized Outreach: Adding a secondary LLM node to draft a cold email that specifically references the "Industry Challenges" identified (e.g., regulatory pressure on gig workers), saving the draft for human review.
Conclusion
Conclusion
This project demonstrates that automation is not just about speed—it is about augmentation. By offloading the cognitive load of research to AI agents, we empower sales teams to focus entirely on high-value relationship building.
Contact
Contact
Google Sites
Report abuse