How Should Cold Outreach Emails Actually Be Written? The Logic Behind AI-Generated Drafts

Writing cold outreach emails is a near-daily task for export sales reps, and getting it right is harder than it looks. Too templated, and prospects spot the mass-send at a glance — straight to spam or ignored. Too personalized, and a single rep can only hand-write a few dozen a day, nowhere near enough to keep pace with incoming leads. What AI-generated outreach is really solving is the tension between personalization and scale. The first step is signal extraction. Before drafting anything, the system reads public information about the target company — its website, product pages, LinkedIn profile — and pulls out a handful of key signals: what the company mainly sells, which markets it targets, and any recent signs of expansion or new product launches. These signals are the raw material for an email that doesn't read like a template. The second step is pain-point matching. For Trade LeadGen, we built a pain-point library organized by industry and product category. Selling packaging equipment to a food company and selling the same equipment to a pharmaceutical company involves completely different compliance requirements and use cases. Based on the prospect's identified industry, the system picks the matching angle rather than applying one script to everyone. The third step is structured generation with variable filling. Outreach emails follow a fairly fixed structure — opening line, value proposition, social proof or case reference, a clear call to action. Within that structure, the extracted signals and pain points get filled into the right slots to produce a draft. The system also generates several subject-line variants and avoids wording and formatting patterns known to trigger spam filters. The fourth step is cadence sequencing. A single email rarely gets a reply, so the system plans a multi-touch follow-up sequence at the same time — a second and third email spaced a few days apart, each with a slightly different angle, so the sequence doesn't feel mechanically repetitive. One thing worth emphasizing: what AI produces is a draft, not a finished piece ready for blind sending. We kept a human review and editing step in the product on purpose, because a misspelled company name or a wrong industry term is a far more damaging mistake than an email that just reads as merely average — and that is a level of detail-checking AI cannot fully take over yet.