A client emails you a spreadsheet of contacts. Three tabs, no consistent column order, some rows missing a title, some missing a company. LinkedIn wants it in a specific template before it will match anyone, so somebody opens two windows side by side and starts copying columns across. Do that for fifty clients once a month, and you have lost a week.
The obvious fix is to hand the file to Claude or ChatGPT. The problem is what is in the file. Names, work emails, job titles, employers, and sometimes worse. Once that leaves your machine, you are trusting a checkbox in somebody’s settings page, and if the company holding it gets breached, the client whose data you uploaded has a claim against you.
AJ Wilcox runs B2Linked and has managed over $200 million in LinkedIn Ads spend. In this session, he builds the workaround live: an agent that routes client data to a model running on hardware in his basement, never to a cloud provider, plus the one step most people skip that stops the agent cheating. Then he takes the cleaned file into Campaign Manager and shows what LinkedIn does with it.
07:17 Telling the agent to hand off the file and not read it
13:40 What goes into a LinkedIn brand kit, and what stays out
The interesting part of AJ’s setup is not the local model. It is the handshake he forces before any file moves. He writes the instruction, sends it on its own, and waits for the agent to repeat the terms back before he attaches the spreadsheet.
Source link
