Internal project / 2026
Prospecting pipeline: our method, applied in-house
An internal B2B prospecting system with AI classification and a weekly learning cycle. Built to prove the method before selling it.
- Client
- Heuristik
- Services
- Strategy, System Architecture, Development, Automation
- Timeline
- Ongoing project
- Industry
- Internal project
The challenge
The problem, stated precisely
Manual prospecting doesn't scale, and automated prospecting without criteria burns through the list. The challenge was to build a system that personalized outreach, classified replies consistently and improved on its own, without turning contact into spam.
Approach
The decisions that shaped the project
Every decision below had an alternative that was discarded. Recording the why is what makes the reasoning reusable in another context.
- 01
Separate memory, engine and intelligence
The CRM stores state and history, the orchestrator executes and the model decides. One responsibility per layer makes the system debuggable, which is the difference between a pipeline and a black box.
- 02
Record every interaction as historical truth
If it wasn't recorded, it didn't happen. Every message sent and received becomes an event tagged with the prompt version used, which makes it possible to attribute results to the decision that produced them.
- 03
Close the learning loop
A weekly routine reads the history, identifies reply patterns by stage and proposes a new version of the outreach. The system improves through accumulated evidence, not the intuition of whoever maintains it.
Solution
What was built
A pipeline with four flows in production: personalized prospecting, classification and reply, weekly analysis and rollout of improvements, running on our own infrastructure with auditable state end to end.
Stack
- n8n
- PostgreSQL
- Docker
- Caddy
- AI API
- CRM
- Flows in production
- 4Flows in production
- Pipeline stages tracked
- 15Pipeline stages tracked
- Automated learning cycle
- WeeklyAutomated learning cycle
Result
The system demonstrates in practice what the consultancy advocates: explicit process, traceable decisions and improvement through evidence. It serves as an architecture reference for client automation projects.
Services applied
What was hired in this project
Other case studies
View all- Small business management
Giro
Heuristik's own SaaS that tells a small business whether the month turned a profit. Sales are logged in the dashboard or in a WhatsApp message, and Giro does the math.
- Sports events
NovoTicket
A full review of the platform that sells race registrations across Brazil, from the runner's cart to the race organizer's dashboard.
- Beauty subscription
HairClub Brasil
A beauty subscription platform designed for two audiences at once: the people who subscribe to a plan and the partner salons that serve them.
Next step
Your product is losing users somewhere. The question is where.
Describe the context and we get back to you within 2 business days with an initial read of the problem. If Heuristik is not the right fit for your case, we tell you on the first call.