The Problem
A B2B SaaS company with a 5-person sales team receives 50 incoming inquiries a day across various channels (contact form, email, LinkedIn). Every inquiry is read, assessed and categorized by hand — one sales rep spends 2–3 hours a day just on that.
The core problem: 30% of the inquiries are spam or clearly non-commercial. Another 40% are low-quality leads (wrong company size, wrong budget, no fitting use case). Only 30% of the inquiries are real A-leads — but the sales team only sees them after working through all the others.
- 50 inquiries/day — only 15 of them (30%) are real A-leads
- 3h a day per sales rep for manual qualification
- 30% spam/non-sales — pure waste of time
- Response time for A-leads: 4–8 hours on average (because they sink into the queue)
The Solution
An AI-driven lead-scoring system that rates incoming inquiries against the BANT framework (Budget, Authority, Need, Timing):
- Intent detection agent: Analyzes the inquiry email for buying signals, company size, industry and a concrete problem statement. Generates an intent score (0–100).
- BANT scoring agent: Rates the available information by budget signals, decision-maker status, concrete need and timing indicators. Builds a scoring matrix with reasoning.
- Routing agent: A-leads (score >70) are escalated to sales immediately with a full context briefing. B-leads go into the standard queue. C-leads and spam are filtered out.
Stack: Claude API for intent analysis and scoring, a Python webhook for email integration, CRM connection via REST API for automatic lead creation.
Results
| Metric | Before | After | Improvement |
|---|---|---|---|
| Manual qualification effort | 3h/day | 0.5h/day | −83% |
| Response time A-leads | 4–8 hours | 20 minutes | 95% faster |
| Spam filter rate | manual | 95% automatic | complete |
| Close rate | 22% | 30% | +35% (relative) |
| Leads auto-qualified | 0% | 60% | +60pp |
Implementation Timeline
- Week 1Analyze the last 3 months of incoming inquiries. Define patterns for A/B/C leads. Work out BANT criteria specific to the company.
- Week 2Build the intent detection agent and test it on historical data. Optimize precision/recall — goal: A-leads are never classified as C.
- Week 3Set up CRM integration, routing logic and the email webhook. Soft launch: the system runs in parallel with manual qualification.
- Week 4Compare AI scoring vs. human judgment. Calibrate the score thresholds. Full rollout.
Learnings & Best Practices
- Avoid false negatives: Better to rate a B-lead as A than the other way around. When in doubt: score higher, sales sees more — but never loses a real lead.
- The context briefing is decisive: The routing agent gives sales not just a priority but a 3-sentence briefing: Who? What? Why now? That saves another 10 minutes per A-lead.
- CRM hygiene: Automatic lead creation with a score field in the CRM enables later analysis: does the AI score correlate with close probability? After 8 weeks: yes, with a 0.73 correlation.
The sales team was skeptical at first. After two weeks of parallel operation, the numbers won them over: the AI score matched human judgment 85% of the time — at zero effort.
This case study is illustrative. The metrics shown are based on typical implementation experience and industry averages. It is not a real customer project and uses no real customer data.