THE PROBLEM
Two People Just to Understand the Leads
Vastav Intellect operates in a specialised space where understanding a prospect matters as much as finding one. Not every lead is qualified, and finding the signals that matter takes research.
Before working with Zobique Labs, two employees were dedicated to reviewing incoming leads, researching who the prospects were, and identifying what the company could offer them. At roughly ₹20,000 per employee per month, that was about ₹40,000 a month in staffing cost, and it grew with lead volume. The workflow was straightforward but labour-intensive:
- Review the lead list.
- Search the internet for information about each prospect.
- Understand the company's business, products, and activities.
- Identify potential needs and relevant service opportunities.
- Prepare a brief that helps the sales team decide what to do next.
THE REQUEST
Can an Agent Do This for Us?
The founder's requirement was clear: provide an Excel sheet of leads, and have an AI agent research each prospect across the internet, gather relevant information, identify opportunities, and return a concise brief on what Vastav Intellect could sell to them.
A general-purpose AI tool was the obvious starting point, but a tool that works well for a handful of prospects can become expensive across hundreds or thousands of leads. The question wasn't whether AI could research a lead. It was whether we could build a research process that made economic sense at scale.
THE INVESTIGATION
Calculate Before You Automate
Before committing to an architecture, we estimated what researching a lead would cost with different approaches. General-purpose AI research came to roughly ₹5-₹8 per lead for the depth required. A web-search-agent approach came down to roughly ₹2-₹3.
We believed there was still unnecessary cost in how the workflow was executed, so we broke lead research into its components, each with different performance and cost needs:
- Discover: search the web for relevant information about a prospect.
- Extract: retrieve useful content from the pages discovered.
- Interpret: classify the findings and identify potential business opportunities with an LLM.
THE SOLUTION
A Purpose-Built Lead Research Engine
Rather than relying on one expensive model for the whole process, we built a pipeline that assigns each task to a tool suited to the job.
- Web search with Serper: discovery handled as its own stage instead of being bundled into an all-in-one AI research call.
- Content extraction with Jina AI: turns discovered pages into material the analysis stage can use.
- Lead classification with Gemini: a lighter model identifies what matters about the prospect and its potential needs, without paying for premium reasoning on every step.
- Two-worker execution: web research and other processing progress without blocking each other.
- Django and MongoDB: the application backend and storage for lead data and generated research.
THE ECONOMICS
From ₹5-₹8 to ₹0.80 Per Lead
Breaking the workflow into specialised components cut our estimated per-lead research cost substantially. These are our own estimates for this workflow, not independently audited benchmarks.
Against the web-search-agent estimate, that is roughly a 60-73% reduction. Against general-purpose AI research, it is up to about 90%.
- General-purpose AI research: about ₹5-₹8 per researched lead.
- Web-search-agent approach: about ₹2-₹3 per researched lead.
- Zobique Labs' specialised pipeline: about ₹0.80 per researched lead.
- At 1,000 leads, that is roughly ₹800 against ₹2,000-₹3,000 for a web-search-agent approach (illustrative; actual cost depends on research depth, retries, usage, and infrastructure).
THE IMPACT
Research Before Outreach
The system turns raw lead lists into more actionable sales intelligence: it processes lead information, researches prospects online, classifies findings, and prepares concise briefs that highlight potential opportunities.
Instead of investigating every prospect from scratch, the sales team gets a repeatable, technology-driven research step, so attention can go to the prospects with the most promising opportunities.
Patent design and drafting automation is a niche, technical space, and getting it right requires real precision. This team took the time to understand the nuances of our process before writing a single line of code. The result has saved us hours of manual work and reduced errors significantly. Impressive execution.
