About the project
The AI layer of a LinkedIn prospecting tool
From September 2023 to January 2025, I worked with a French SaaS that helps sales teams prospect on LinkedIn. My part was the AI layer: the prompts that turn prospect information into usable outreach, and the analytics that show whether that output is actually used.
The Challenge
A prompt that works in a demo is not a prompt that works at scale
A prompt that works once in a demo is easy. A prompt that works for 200 prospects a day, across different profiles and contexts, is a different animal. Every gap in the instructions shows up at scale, and users notice it before you do.
HOW WE FIX ISSUES
Prompt architecture and usage analytics
I designed the prompt architecture: structured instructions, reusable templates and rules that keep the output on topic. Then I followed usage analytics to see which outputs people kept and which they rewrote, and adjusted the prompts from there. The throughput reached 200 prospects a day.
The lesson I now apply to product catalogs: AI amplifies the structure you give it, good or bad. More on that in AI without data governance is noise.
The Toolbox
Technology Stack
OpenAI
Claude
Bruno
