An AI Agent That Qualifies Leads While You Sleep
We built an autonomous sales pipeline agent that scores inbound leads, enriches CRM data, and drafts personalized outreach — reducing time-to-first-contact from 6 hours to 11 minutes.
The Bottleneck
A B2B SaaS company with 3 sales reps was losing deals before they even started. Inbound leads from the website sat in HubSpot for 6+ hours before anyone touched them. By then, the prospect had already booked a demo with a competitor.
The math was brutal:
| Metric | Before | Industry Best |
|---|---|---|
| Time to first contact | 6.2 hours | < 5 minutes |
| Lead response rate | 34% | 78% |
| Qualified leads per week | 12 | — |
| Rep time on qualification | 18 hrs/week | — |
"We knew we were leaving money on the table. We just didn't have enough hands." — VP of Sales
What We Built
An autonomous agent pipeline with three stages, orchestrated through n8n and powered by Claude:
Stage 1: Enrichment
When a lead submits a form, the agent immediately:
- Pulls company data from Clearbit
- Scrapes their LinkedIn company page
- Checks for existing CRM history
- Estimates company size and revenue tier
{
"company": "Acme Corp",
"employees": 142,
"revenue_tier": "$5M-$10M",
"industry": "eCommerce",
"icp_score": 87,
"signals": ["recently_hired_vp_ops", "series_b_funded"]
}
Stage 2: Scoring & Routing
The agent runs a scoring model:
- 90-100: Hot lead → immediate Slack alert to senior rep
- 70-89: Warm lead → auto-enroll in personalized email sequence
- Below 70: Nurture track → weekly educational content
The scoring isn't a black box. Every score comes with a reasoning summary so reps can override it:
Score: 87/100
Reasoning: Mid-market eCommerce (ICP match), recently hired VP Ops
(buying signal), visited pricing page 3x in 2 days (intent signal).
Recommended action: Priority outreach within 1 hour.
Stage 3: Personalized Outreach
For warm and hot leads, the agent drafts a first-touch email that references:
- The specific page they visited
- Their company's industry challenges
- A relevant case study from our library
Each draft is reviewed by the sales rep before sending. The agent learns from edits — if a rep consistently changes the tone, it adapts.
The Architecture
Website Form → Webhook → n8n Workflow
├── Clearbit API (enrich)
├── Claude (score + reason)
├── HubSpot API (update CRM)
├── Claude (draft email)
└── Slack (notify rep)
Every step logs to a shared Google Sheet so the team can audit decisions and spot patterns.
Results After 90 Days
| Metric | Before | After | Change |
|---|---|---|---|
| Time to first contact | 6.2 hrs | 11 min | 97% faster |
| Lead response rate | 34% | 71% | +109% |
| Qualified leads per week | 12 | 23 | +92% |
| Rep time on qualification | 18 hrs/week | 4 hrs/week | -78% |
| Pipeline value (monthly) | $180K | $340K | +89% |
Lessons Learned
- Don't automate the close — the agent qualifies and drafts, humans close. Trust is built person-to-person.
- Transparency matters — showing the scoring reasoning made reps trust the system within 2 weeks instead of fighting it.
- Start with the bottleneck — we could have built a full sales AI. Instead, we solved the one thing killing deals: slow response time.
The Takeaway
This wasn't a $500K AI transformation project. It was a focused, 3-week build targeting one metric: time-to-first-contact. The ROI was visible in the first month.
The best AI agents don't replace your team. They remove the bottleneck that's holding your team back.