Cold email still works — but only when it feels personal. AI lets small business operators send messages that read like they were written one at a time, even when you’re reaching hundreds of prospects. Here’s how to build a cold email workflow that combines AI speed with human judgment.

What to Automate

Focus AI on the research and drafting work that eats your time: pulling prospect details from LinkedIn or a website, writing a first-line personalisation hook, generating subject line variations, and formatting the email body. Do not automate the final send decision or the reply handling — those need you.

Which Tools to Use

ChatGPT or Claude for drafting and personalisation. Apollo.io or Hunter.io for verified prospect data. Instantly or Smartlead for sequencing and deliverability. Clay if you want to pull data from multiple sources and pipe it straight into your copy. Most of these connect via CSV export if you prefer to keep things simple.

Step-by-Step Workflow

1. Export a prospect list (name, company, URL, role) from Apollo or your CRM into a spreadsheet.
2. Open ChatGPT or Claude and paste a template prompt: “You are a copywriter. Write a 3-sentence cold email opening for [name] at [company] who works as [role]. Reference something specific from their website: [URL]. Keep it conversational, no buzzwords.”
3. Run the prompt for 5–10 prospects manually first to calibrate tone. Adjust the prompt until the output sounds like you.
4. Once the prompt is dialled in, use Clay or a simple Python script to batch-generate personalised openers for your full list.
5. Paste the generated openers into your sequencing tool (Instantly, Smartlead) and attach your pre-written follow-up steps.
6. Set daily send limits (50–100/day per inbox) and monitor reply rates in the tool dashboard.

Where to Keep a Human in the Loop

Review every batch of AI-generated emails before they go into your sequence — at least a random 10% sample. AI occasionally hallucinates details about a company or produces a tone that’s off-brand. You should also personally handle every reply that comes back: a warm response to a cold email is a sales conversation, and no AI should be running that for you yet. Check deliverability metrics weekly and pause any sending if open rates drop below 30% — that’s a signal something needs a human fix.

Quick Wins to Expect

Most operators see research time drop by 70–80% once this workflow is running. Subject line A/B testing becomes easy because AI can generate 10 variants in seconds. The real gain is consistency: your outreach keeps moving even during busy weeks when you’d normally let it slide.