The problem
Finding SMBs that actually run SAP Business One — and reaching the right person with a reason to talk — is slow, manual, and easy to get wrong. The goal was a sales pipeline that could find, qualify, verify, enrich, and prioritise leads on its own, surfacing only the ones worth a human's time, with evidence for every score.
The approach
A fully autonomous multi-agent pipeline built on OpenClaw on GCP. Six specialised agents run on cron schedules and hand off through a shared pipeline database — no human in the loop after setup:
- Scout — scrapes leads from partner case studies and the open web.
- Qualifier — composite scoring across identity, event, and contact dimensions.
- Verifier — confirms the prospect actually runs SAP B1.
- Enricher — discovers contacts via an Apollo / Hunter / theHarvester / Perplexica cascade.
- Signal Hunter ("Pulse") — intent scoring from live web signals (hiring, funding, complaints, migration) using Gemini / Groq.
- Closer — drafts the outreach email.
Architecture
Each agent is independently schedulable and writes to a SQLite pipeline database that tracks a lead's state through the funnel. A real-time dashboard renders the pipeline; a WhatsApp gateway pushes notifications; signal sweeps run on cron. Built from scratch in roughly three weeks and deployed to a VPS.
What it produced
The run scored and classified 1,084 leads end to end, of which 644 landed on the client track (avg score 7.8/10), 569 were fully researched by the agents, and 75 were verified as confirmed SAP B1 users — all with zero manual intervention after the initial setup. (This pipeline later became one of the automations that informed the broader AgentOne platform.)
