Context
B2B fintech with R$ 500k/mo MRR, 200+ conversations daily incoming via WhatsApp. A 4-person SDR team spent 80% of their time qualifying repetitive leads: "what's the rate", "is there a fee", "how do you calculate", "do you serve company X". Average response time during business hours was 8 minutes. Outside business hours, leads went cold.
The problem
Leads coming in overnight simply vanished. SDRs burned out on repeated questions. By the time a real analysis came, the lead was tired. No off-the-shelf chatbot platform handled product-specific rules without going generic.
The call
Build a custom agent trained on real product rules + rate tables + approval criteria. Claude API with tools. Human handoff when judgment is required.
The solution
- Custom agent with a product knowledge base (rates, fees, approval flow, rules by company size).
- Z-API integration for WhatsApp + automatic handoff to humans via tagging.
- Qualification metrics baked into the prompt: every conversation ends with a lead score + category.
- Feedback loop: humans flag conversations the agent got wrong; the prompt is tuned weekly.
The outcome
- Average response time dropped from 8min to 2min.
- 70% reduction in human SDR time spent qualifying.
- Overnight leads stopped dying — agent runs 24/7 and does first-pass qualification.
- Lead → customer conversion up 23% per the sales team (better-qualified leads reaching the call).
Why it worked
Not a "any chatbot." The agent knew the product better than a junior SDR after 2 weeks. And the human handoff was the rule: the agent escalated whenever a question stepped outside the base. That protects the brand and the client, while scaling support.
"In 4 weeks Kode delivered what we'd been trying to build with 3 different platforms over a year."
— Head of Operations