How I built an AI cold caller that books meetings while I sleep
When I first started running Facebook ads and cold emails, I was generating leads, but I wasn’t following up fast enough.
Leads would come in, I’d tag them in a sheet, then eventually someone (usually me) would start calling. By then, half of them were cold.
So I built a system to call every lead automatically.
It runs on n8n, Vapi, and Google Sheets.
Upload a list of leads, and the AI caller starts dialing.
It sounds completely human, books meetings, takes notes, and updates the sheet in real time.
It can make around 600 calls an hour.
At first, I only used it inside my agency to follow up on ad leads. Then clients started asking if they could use the same system, so I packaged it into a template anyone can import and run.
What surprised me most was how many people don’t want the service, they just want the system. Something they can run without waiting on anyone else.
Lessons so far
Speed matters more than personality. Calling a lead within minutes beats any perfectly crafted script.
Productizing your systems saves time. I don’t rebuild this setup for every new client anymore; they just import it.
AI doesn’t replace the rep. It replaces the first step: qualifying and booking. The rep still closes.
I’m planning to post a technical breakdown next: how Vapi handles the calls, how n8n tracks the results, and how booked leads get passed to the team.
Would that be helpful to anyone here?
This is brilliant, Mike 👏
The “AI-first step, human closer” mindset is exactly how automation should be used — augmenting, not replacing, the sales process.
The n8n + Vapi + Sheets combo is such a smart lightweight stack too.
I’d definitely be interested in the technical breakdown — especially how you’re managing voice quality and context handoff between calls. 🚀
Appreciate it. I see it the same way. The AI handles the first two minutes and the human closes the deal. This works because everything stays clean and simple. For voice quality and the handoff, most of it comes down to prompting and how the call flow is built inside n8n. I will post a full breakdown next so you can see exactly how it runs.
This is impressive! I'm curious how you handle caller ID and compliance with local laws when making automated calls. Also, how do you ensure the AI conversation feels personalized enough to keep prospects engaged? I'm building a UX audit tool and exploring how to incorporate AI into outreach, so your experience is invaluable.
Great questions. For caller ID, it is just registering the number inside Twilio and verifying the business so the number gets a green check mark and does not show up as spam.
For personalization, each lead gets called by their first name. We also have info on the leads before calling them like location, job title, and company name so the AI can stay specific.
The system also keeps the call short and mirrors the way the prospect talks, which is what makes the conversation natural.
Your AI caller is the perfect second step. The first is getting the leads. We use targeted cold emails to book meetings with decision makers. Your system calling a lead our system just found is a killer combo.
Happy to go for a partnership here :) prospectai.dev
Exactly. Email is the first touch. The AI caller is the fast follow up. If your system finds the decision makers, this is a strong second step. I will take a look at your link and see if it is something we can test.