How to Automate Lead Generation with AI Agents (Step by Step)

A sales development rep spending three hours a day on prospecting, research, and writing is not unusual. That is roughly 60 hours a month on the same loop: find a business, find the right person, visit the website, take notes, write a personalized email. Repeat.

Tools like Apollo or ZoomInfo help with enrichment, but they still leave you doing research and writing by hand. Most manual prospecting either stops at info@ or burns time on LinkedIn guessing email formats. Neither approach holds up at 100 leads with real personalization.

An AI pipeline solves both problems at once, and applies the same depth to lead 100 as it does to lead 1. For a wider map of agent use cases beyond lead gen, see What can AI agents do for my business?.

Your plain-English brief goes to a Supervisor (interprets and monitors), then a Prospector (finds leads and contacts), an Analyst (researches pain points), and a Copywriter (writes personalized outreach). The finished output lands in a Google Sheet.

lead-gen-pipelineSTATIC

No agent tries to do everything. That is where single-model approaches fail. Each specialist does one job well and hands off cleanly. The orchestration layer (often n8n) connects them and passes structured data between stages. For how agent teams work in general, see What is an AI agent team?. You do not see the wiring. You see a finished output.

You do not configure a form or set parameters in a dashboard. You describe what you want the same way you would tell a colleague: "Find me 100 digital marketing agencies in Singapore. We offer AI automation systems. Goal is to book a discovery call." The Supervisor parses that, extracts the brief, and dispatches the crew.

Throughout the run, the Supervisor watches for issues (a search returning fewer results than expected, a website that fails to load, an agent returning incomplete output) and handles them without stopping the pipeline. If something needs a decision, it surfaces it. Otherwise, it resolves it and continues.

Stage 2a: Business discovery

The Prospector searches for companies matching your target niche and location, handles pagination across multiple results pages, and deduplicates so the same business never appears twice. For each company it returns: business name, website URL, phone number, and physical address.

Stage 2b: Decision-maker contact finding

Rather than stopping at info@ or hello@, the Prospector runs targeted web searches combining the company name with role keywords: CEO, founder, owner, director. It applies a name-and-domain matching protocol to confirm the person belongs to that company and that the email format matches the domain.

Generic contact emails are still captured as a fallback on the main lead row. Verified decision-maker contacts take priority and feed the Outreach tab first.

• Business name, website, phone, address
• Contact email and source (scrape or POI search)
• Decision-maker name and role on Outreach rows (when found)
• Unique place identifier (prevents duplicates across runs)

This step is where most manual prospecting falls short. A researcher under time pressure skims a site in 90 seconds and moves on. The Analyst reads properly, cross-references what it finds against the offer in the brief, and surfaces only what is actually relevant.

• No contact email on the site, only a form
• Blog last updated more than 12 months ago
• No case studies, testimonials, or client results visible
• Services page with vague copy and no specifics
• Missing or broken social links

The hallucination guardrail is non-negotiable. The Analyst cites only what it can read from the page. If the site is clean and nothing maps to the offer, the output is "no pain point found", and the Copywriter handles that case differently.

1. Personalized opening: name when available, first line references something specific from their site.
2. Cost connection: one sentence on what that gap likely costs them.
3. Offer: one sentence connecting your service to a fast resolution.
4. CTA: a single, low-friction ask tied to the goal in the brief.

When no decision maker was found, the email opens with a professional generic salutation and the pain-point hook does the heavy lifting. The Copywriter runs once per contact in the batch. One hundred leads can mean dozens of unique outreach rows, each written around that lead's context, in the same run.

Leads tab: one row per company

• businessName, website, phone, address (Prospector)
• contactEmail, emailSource (scrape or enrichment)
• Social profiles: x, instagram, facebook (when found on site)
• placeId: upsert key so reruns update instead of duplicate

Outreach tab: one row per contact

• addressee, role, targetEmail (Prospector: POI path when found)
• painPoints (Analyst: bullet points per lead)
• emailSubject, emailBody (Copywriter: personalized per contact)

Total run time for 100 leads: approximately 10 minutes end to end. You review the sheet. You decide what gets sent and to whom. The system does not send automatically. That is by design.

• Finding the right contact per lead: 5-15 min manual vs. seconds in parallel
• Website research per lead: 3-5 min manual vs. seconds in parallel
• Email writing per lead: 5-10 min manual vs. seconds in parallel
• 100 leads total: 8-15 hours manual vs. ~10 minutes automated
• Monthly cost at scale: $2,000-5,000+ salary or VA vs. build fee + retainer

The build cost for a Lead Gen Engine is a one-time fee, plus optional dev hours at $50/hr for any custom integrations. See Scope & Cost for the full breakdown, or our Solutions page for the product catalog.

The system is not a replacement for a salesperson. It replaces the research and first-draft work that a salesperson should not be spending time on. The conversations, the relationships, and the close stay human.