AI Lead Scoring
How Despora's Gemini-powered AI analyzes every phone call transcript to score intent, estimate revenue, rate agent performance, and filter spam.
TL;DR: Every phone call transcript is analyzed by Google Gemini AI. Each call receives a lead score (0.0 – 1.0), qualification tier, service identification, revenue estimate, call summary, and agent performance ratings — all automatically, in under 10 seconds.
How Lead Scoring Works
When a call transcript is ready (from Deepgram or CallRail), Despora sends it to Google Gemini along with your project's service catalogue. The AI returns a structured analysis:
| Field | Type | Description |
|---|---|---|
| Lead Score | 0.0 – 1.0 | Probability the caller becomes a paying customer |
| Qualification | Tier | High Intent, Medium Intent, Low Intent, or Not Qualified |
| Service Identified | Text | Matched to your service catalogue (e.g., “Water Heater Installation”) |
| Sales Value | Dollar amount | Estimated revenue based on your service pricing |
| Summary | Text | 1-2 sentence overview of the conversation |
| Agent Name | Text | Automatically extracted from the conversation |
| Professionalism | 0 – 10 | Agent professionalism score |
| Listening Skills | 0 – 10 | Agent listening score |
| Closing Ability | 0 – 10 | Agent closing score |
Service Catalogue Matching
You configure a service catalogue for each project with service names, keywords, and pricing ranges. When Gemini analyzes a call, it matches the caller's inquiry to the closest service and uses your configured pricing for revenue estimation.
Example service catalogue:
| Service | Keywords | Avg Value | Min | Max |
|---|---|---|---|---|
| Water Heater Installation | water heater, hot water, tank | $2,800 | $1,500 | $5,000 |
| Drain Cleaning | drain, clog, backed up | $350 | $150 | $800 |
| Emergency Repair | emergency, leak, burst, flood | $1,200 | $500 | $3,000 |
Pre-Analysis Filtering
Before consuming an AI credit, Despora automatically detects and rejects non-leads:
- Spam / Solicitors — Regex pattern matching on common spam phrases
- Job Seekers — “I'm calling about the job posting” and similar patterns
- Empty Voicemails — Machine greeting with no caller message (Deepgram speaker analysis)
- Ultra-Short Calls — Answered and dropped in under 5 seconds
- Fallback Transcripts — Synthetic placeholders when no real transcript is available
None of these consume a credit. You only pay when the AI actually analyzes a real conversation.
Agent Performance Monitoring
Every analyzed call generates agent scores. Over time, Despora builds a performance profile for each agent, showing trends in professionalism, listening, and closing ability. This helps you:
- Identify which agents need training
- See if marketing-generated leads are being wasted by poor phone handling
- Track improvement over time after coaching sessions
Human Oversight & Trainable AI
Despora's AI isn't a black box. The scoring prompt includes your project name and service catalogue, so it understands your business context. A “qualified lead” for a plumber is different from a “qualified lead” for a law firm — and the AI adapts accordingly.
The Human-in-the-Loop Advantage
Every AI-scored lead is presented to you for review. When you approve, reject, or adjust a lead's qualification or value, that feedback sharpens the system:
- Approve a lead — confirms the AI's scoring criteria and revenue estimate for similar future calls
- Reject a lead — teaches the AI that this call pattern (topic, duration, caller type) shouldn't be counted
- Adjust the value — calibrates revenue estimation to match your real-world pricing and close rates
Over time, this creates a continuously improving feedback loop. The AI gets smarter with every review cycle — better at qualifying leads, more accurate at estimating revenue, and sharper at filtering noise. Human judgment and AI speed work together, not against each other.