one in fifty... why ai should draft your sales emails, not send them

monday we walked the team through the crm and outreach platform we're moving onto.

it does a lot. one inbox across the outreach tool, gmail and whatsapp. prospect scoring. a show pipeline that cuts ai shorts. booking. a revenue dashboard that pulls stripe, bitcoin, wise and stablecoins into one view. a team leaderboard, where charles sits at $140k closed and i'm right behind him at $130k.

but the design choice that matters most is the boring one.

the ai drafts. it doesn't send.

everyone asks me why. the models are good now. they write better first drafts than most sdrs. why put a human in the way?

because of one number. one in fifty.

say your agent gets it wrong one time in fifty. wrong name, wrong company, a reply to the wrong thread, a line that reads fine to a machine and sounds insane to a person. that's a two percent error rate. on a slide, that's a great model.

now do the multiplication.

a warmed up outbound setup gets to around a thousand emails a day within a week or two. one in fifty of a thousand is twenty.

twenty wrong messages. every day. to real people, with your name on them.

that's a hundred in a working week. call it four hundred a month. some of those people are investors you'll want to talk to next year. some are customers who were about to say yes. some will screenshot it.

you would never hire a salesperson who sent twenty bad emails a day and tell yourself, well, the other nine hundred and eighty were fine.

so the rule we built into the system is simple. ai writes, a human approves. every outbound message sits in draft until someone looks at it.

and the funny part is the draft mode doesn't slow you down. it speeds you up, because the ai stops being a sender and starts being a coach.

in the middle of the demo it caught me. i'd let a reply slide. the system flagged it, more or less... you forgot to reply, you should draft an email. so i drafted it, approved it, done. thirty seconds.

and my ai's like, yo... you forgot to reply, so you should draft an email. so i go draft it, done.

that's what you want from an agent right now. it notices what you missed. it does the first ninety percent. you do the last ten, which is the part where your reputation lives.

clients never see the raw sending tool, either. they see the conversations that came out the other side. the machinery stays in the kitchen.

i've written before about the email i never approved, and about reading every email the robot writes. this is the arithmetic underneath both of those posts. once you see the multiplication, you can't unsee it.

will the error rate drop? sure. one in a hundred, one in a thousand. but volume grows too. the agents that send without review will send more, not less, and the wrong ones will scale right along with the right ones.

long term, i think the founders who win with ai outbound are the ones who treat approval as a feature, not a bottleneck. the speed comes from the draft. the trust comes from the human.

and if you're the one deciding, start with a simple question. what does one bad email cost you? not the average email. the worst one. for most founders it's a relationship that took a year to build. price the mistake, not the message.

if you want the longer version of how i think about putting agents to work without handing them your name, it's in the ai book.

so before you flip the switch on auto send... how many wrong emails a day are you willing to put your name on?