Generic outreach doesn't work anymore. Buyers spot copy-paste in seconds. Automate sales outreach with AI the right way: use AI to research and write personalized lines at scale, then send through a human-reviewed layer. This playbook walks you through a five-step workflow—scrape, enrich, research, write, send—without code.
The Problem: Generic Outreach Doesn't Work
Spray-and-pray cold email gets ignored or flagged. Personalization is the only lever that still moves reply rates. Doing it manually doesn't scale. Automating sales outreach with AI means: keep the human in the loop for sending and tone, but let AI handle research and first-draft personalization so you can run 10x the volume without 10x the hours.
What AI-Personalized Outreach Looks Like
Each prospect gets: a relevant hook (recent news, role, or pain), a clear reason you're reaching out, and a low-friction ask. The body reads like you wrote it for them—because the AI drafted it from real signals. You review, tweak, and send. The system handles lead sourcing, enrichment, and draft generation; you handle the final touch and relationship.
The 5-Step Workflow
- Source leads — Apollo, LinkedIn, Clay, or your CRM.
- Enrich with AI — Company and contact data plus research.
- Write personalized lines — GPT-4o with a strict prompt and structured input.
- Review layer — Optional human checkpoint before send.
- Send — Instantly, Smartlead, or your email tool.
Step 1: Source Leads (Apollo, LinkedIn, Clay)
Start with a list. Options:
- Apollo: Filter by title, industry, company size; export or connect via API.
- LinkedIn Sales Nav: Build lists and use a connector or enrichment tool to get emails.
- Clay: Build lead lists with enrichment built in; export or trigger workflows.
Output: a table or CSV with at least company name/domain, contact name, title, and email. This feeds the next step.
Step 2: Enrich With AI (Apify + OpenAI)
Raw leads need context. Run each through:
- Apify (or similar): Scrape company page, LinkedIn profile snippet, recent news if needed.
- OpenAI: One call per lead with a prompt like: "Given this company and role, list: 1) one recent company signal (funding, product, news), 2) one likely pain point, 3) one sentence hook for cold email."
Store the output (e.g. in Airtable or your AI lead research system) so the writer step has structured input.
Step 3: Write Personalized Lines With GPT-4o (Exact Prompt)
Use a single GPT-4o call per lead. System prompt:
You are an expert B2B cold email writer. Write one short email (under 100 words) that: 1) opens with a specific hook from the prospect's company or role, 2) states one clear reason we're reaching out, 3) ends with one soft ask (e.g. 15-min call). Tone: direct, professional, no hype. No greetings like "I hope this finds you well." Do not use bullet points. Output only the email body.
User message: Pass the lead's name, title, company, and the enrichment output (hook, pain point, signal). Example:
Prospect: Jane Doe, VP Ops at Acme Corp. Hook: Recently raised Series B. Pain point: Scaling support. Signal: Hiring for support lead.
The AI returns a unique body. Map it to your sending tool's subject and body fields.
Step 4: Review Layer (Optional Human Checkpoint)
Before sending, either:
- Option A: Send drafts to a shared inbox or Notion/Airtable view; a human approves and clicks "Send."
- Option B: Send to a "Drafts" folder in Gmail/Outlook; rep reviews and sends in batches.
I recommend at least a light review until you're confident in the prompt and data quality. After that, you can automate send for segments that match strict criteria.
Step 5: Send via Instantly or Smartlead
Connect your workflow (Make or n8n) to Instantly or Smartlead (or similar). Map: To = email, Subject = from a template or AI-generated subject line, Body = AI output from Step 3. Respect daily limits and warm up domains; use the review layer to avoid sending bad drafts.
Results You Can Expect
Teams that automate sales outreach with AI with this structure typically see: higher reply rates than generic templates, 5–10x more personalized emails per week per rep, and consistent messaging quality. Scale by adding more leads and tightening the prompt; keep the review layer until you're comfortable.
Want the exact workflows and system prompts? Work With Me →
FAQ
Should I fully automate sending or always review?
Start with review. Once prompt quality and data are stable, you can auto-send for segments (e.g. SMB with high-confidence enrichment). Keep review for enterprise and high-value accounts.
What if I don't use Apollo or Clay?
Any source of leads with company + contact + email works. Spreadsheet, CRM export, or API—the workflow is the same: enrich, then write, then send.
How do I avoid AI slop or obvious templates?
Use a strict system prompt, pass real enrichment data into every request, and cap length. Review a sample weekly and refine the prompt. Avoid generic phrases; the hook must be specific to the prospect.
Can I use this with cold LinkedIn messages?
Yes. Same flow: source → enrich → write with GPT-4o → review → send via LinkedIn automation tool (respect platform rules and limits).