Lead Gen Agent Improvements · Searcher School

Lead Gen Agent Improvements

Manus is good but their token costs calculations are obscure, and there isn't much room available for optimizations besides prompt length and other basic items. I ended up re-creating the initial sourcing pipeline in OpenClaw. New system runs entirely in OpenClaw using Sonnet 4-6 for Enrichment and Analysis, but using lower tier models for basic and routine tasks and has dramatically reduced costs

New features:

Sourcing (moved from Manus to OpenClaw. initial pass that gathers the first set of leads using ICP criteria)
Ingest (discards non qualified leads)
Enrich (deep pass that enriches qualified leads and verifies emails. Scores quality for instantly enrollment)
Campaign enrollment (qualified leads get assigned to an Instantly campaign)
Reply (alerts me via WhatsApp whenever there is a reply)
Metrics (pulled from Instantly via API)
Insights (new tab that collects performance such as click rates and things that could be useful for improvements)
Analyst agent (reviews metrics and insights. Proposes updated to copy, campaigns, delivery, etc to increase conversion)
Guardian agent (maintains health of the system. Flags issues and warnings)
Self- Healing agent (reviews flags from guardian, metrics and runs Sonnet for improvements. Self corrects if it finds an issue during any of the pipeline runs)

Next steps: