Most sales teams are drowning in leads they'll never close.
Not because there aren't enough good ones - because no one has the bandwidth to figure out which ones are good. The average SDR spends 40% of their week on leads that go nowhere. That's not a pipeline problem. That's a qualification problem.
So I built a bot. Here's exactly how it works.
The Problem It Was Solving
A client of mine - a B2B SaaS company - was generating ~300 inbound leads a month via their website form. Their sales team of 3 could meaningfully work maybe 80.
They were choosing which 80 manually. Based on gut feel. At 9am on a Monday. You can imagine the quality.
The ask was simple: build something that scores and routes leads automatically, without needing a 6-month CRM implementation.
The Architecture (Simpler Than You Think)
Here's the stack we used:
- Claude API (claude-sonnet) - the reasoning layer
- n8n - workflow automation and trigger handling
- Airtable - lead storage and status tracking
- Typeform - intake form (already existed)
When a new lead submits the form, n8n catches the webhook, pulls the lead data, and sends it to Claude via the API with a structured prompt.
The prompt does three things:
- Scores the lead (0-100) based on company size, role, use case fit, and urgency signals in their free-text answers
- Generates a 2-sentence qualification summary for the SDR
- Assigns a routing tag:
hot/nurture/disqualify
The whole thing runs in under 4 seconds. Total cost per lead processed: less than ₹1.
What the Prompt Actually Looks Like
You are a B2B lead qualification expert. Given the following inbound lead data,
score this lead from 0-100 based on ICP fit, urgency, and decision-making authority.
Return a JSON object with:
- score (integer)
- summary (2 sentences max, written for an SDR)
- tag ("hot" | "nurture" | "disqualify")
- reasoning (1 sentence)
Lead data: {lead_json}Claude returns structured JSON every time - no parsing nightmares, no hallucinated formats. We added a fallback validation layer using Pydantic just in case, but it's triggered less than 2% of the time.
Results After 60 Days
- SDR time on unqualified leads dropped by 61%
- Hot lead response time went from 6 hours average to under 40 minutes (routing made it instant)
- Pipeline quality score (as rated by the sales team) went up - more subjective, but they felt it
The system cost about 3 days to build end-to-end. The ROI conversation was over in 5 minutes.
The Takeaway
Lead qualification is one of the best first AI automation projects for a sales-driven business. The inputs are structured, the output criteria are definable, and the impact is immediately measurable.
You don't need a data science team. You don't need a 6-month roadmap. You need a Claude API key, a workflow tool, and someone who knows how to write a prompt that actually performs under production load.
That last part is where most teams get stuck.