How I Built a Voice AI System That Qualifies Leads While You Sleep

The Problem: Manual Lead Qualification Doesn't Scale

Jay runs an MCA (Merchant Cash Advance) lending operation. His team was spending 6+ hours daily calling leads, asking the same qualification questions:

• How long have you been in business?
• What's your average monthly revenue?
• How much funding are you looking for?
• Do you have any existing merchant cash advances?

At $25/hour for a caller doing 15-20 calls/hour, each qualified lead was costing $8-12 in labor alone. And the bigger problem was inconsistency — different reps captured data differently, creating chaos in the CRM.

The Solution: Voice AI + Automation Stack

Component 1: Retell AI for Voice

Retell handles phone conversations. The key is writing prompts that sound natural while extracting exact data points:

# MCA Lead Qualification Agent

ROLE: You are Sarah, a funding specialist.
OBJECTIVE: Qualify business owners for MCAs.

QUALIFICATION CRITERIA:
1. Business age: Minimum 6 months
2. Monthly revenue: Minimum $10,000
3. Funding need: $5,000 - $500,000

CONVERSATION FLOW:
→ Warm greeting (use their name)
→ Confirm they're the business owner
→ Ask about business age naturally
→ Transition to revenue question
→ Determine funding needs
→ If qualified: schedule callback

Component 2: Make.com for Data Orchestration

Make.com routes data between systems. When Retell completes a call, it triggers a webhook that:

• Parses the call transcript for qualification data
• Updates HubSpot with structured fields
• Calculates lead score based on criteria
• Triggers appropriate follow-up sequence
• Sends Slack notification for hot leads

Component 3: HubSpot for CRM & Lead Scoring

HubSpot serves as the single source of truth. The lead scoring model:

• +30 points: Business age 2+ years
• +25 points: Monthly revenue $25K+
• +20 points: No existing MCAs
• +15 points: Funding need $25K-$150K
• +10 points: Expressed urgency

The Results: Before vs. After

• Call volume: 15-20/day → 100-120/hour
• Cost per call: $1.25-1.67 → $0.12-0.15
• Document collection: 5 days → 1.5 days
• Qualified leads: 3x increase, same ad spend

Key Lessons Learned

1. Prompt Engineering is Everything

I spent 40% of project time on prompts. The difference between good and great is 20% vs 45% qualification rate.

2. Build for Failure

The AI will make mistakes. If it can't extract data, flag for human review rather than guessing.

3. Start with High-Volume, Low-Stakes

Lead qualification is perfect — calls are short (2-4 min), questions predictable, and mistakes don't cost deals.