I built an AI agent that books B2B SaaS sales calls while I sleep. And I'm going to share it with y'all. Here's the exact 5-stage workflow.
After 6 months of cold-email burnout, I built an AI agent to do my prospecting for me. It now books between 8–12 sales calls per month with zero manual outreach.
I'm not going to gatekeep the build. Here's the architecture:
TRIGGER — Every Monday at 9am, a scheduled Zap kicks off the workflow.
FETCH — Pulls 50 new companies from LinkedIn Sales Navigator + Crunchbase matching my ICP (SaaS, 50–500 employees, North America, uses customer support software).
ENRICH — Each company gets routed to Apollo for emails + BuiltWith for tech stack. Anything that comes back "Unknown" gets routed to a human-review tab.
SCORE — A ChatGPT prompt applies my scoring model:
- +5 for each must-have (industry, size, geography, role)
- +3 for each nice-to-have (tech stack, funding stage)
- +2 for warm signals (visited pricing page, downloaded a case study)
- Leads scoring 15+ are Qualified. Others archived.
OUTREACH — Qualified leads enter a 3-email sequence in Mailshake. Each email is personalized by GPT using the lead's company name, recent activity, and tech stack. Email 3 includes a Calendly link.
Three things that actually mattered:
— Don't let the AI invent emails. Every prompt I use ends with "If unsure, write Unknown." Cut my bounce rate from 11% to under 2%.
— Score breakdown column. Force the AI to show its math. I caught 3 misclassifications in week one by reading the breakdowns.
— Domain warm-up before you launch. I ramped from 20 → 50 → 100 emails/day over 3 weeks. The friends I know who skipped this all got their domain flagged.
The whole thing took me about 6 hours to set up the first time. Now it runs itself. Happy to answer any questions about the build.
(If anyone wants the full step-by-step including the prompts I use — I packaged it as a $7 PDF on Gumroad. Link in profile. Not necessary though — the architecture above is the whole system.)