AI Agent 生成超个性化冷邮件，回复率提升至 31%
A 15-person outbound sales team was sending 500 cold emails per day with a generic template. Reply rate sat at 9% — industry average, but not good enough. Hiring more SDRs to personalize at scale wasn't feasible.
They built a two-stage agent pipeline. Stage 1 is a "research agent" that takes a prospect's name and company, searches LinkedIn for their recent posts, checks the company's news feed for announcements, and scans their job postings for signals (hiring 5 engineers = growing, needs tools). Stage 2 is a "writing agent" that takes the research JSON and generates a personalized opening line and value prop tied to a specific signal.
They kept humans in the loop for sending. Every email gets a 10-second human review before it goes out. This caught the 3% of cases where the agent misread a signal (e.g., treating a company's layoff announcement as a growth signal). The cost of that review is low; the cost of sending a tone-deaf cold email is high.
Partially. Clay.com + ChatGPT can handle the research and writing pipeline with no code. The limitation is research depth — Clay's enrichment is broader but shallower than a custom agent that actually reads full LinkedIn posts and news articles.