Don't automate the mess. Rebuild one job beside it.
A home-services shop books jobs like this. A form lands in an inbox. Someone copies it into the calendar, texts to confirm, texts again when two jobs collide. Put AI on that and you get the same mess, faster. MIT's 2025 study found 95 of 100 company AI pilots showed no measurable return (chart 1).
Salim Ismail's diagnosis: when TV arrived, the studios put radio announcers in front of the camera. Nobody used the medium. Same now. AI gets injected into legacy bottlenecks and fails on schedule. His fix is not to move the process. Copy it. Rebuild the copy AI-first, on the side, where it cannot sink the revenue engine. Switch when it wins.
“You're moving AI into legacy organizations and automating the legacy human bottlenecks. Of course they're going to fail.”
Salim Ismail, Moonshots EP #258, 9:31
| Do this | Not this | |
|---|---|---|
| 1 | Pick one flow. Bookings, say. | Pick a department. |
| 2 | Copy it to the side. | Edit the live one. |
| 3 | Rebuild it AI-first: the page answers, checks the slot, confirms. | Add a bot to the inbox. |
| 4 | Run both. Count real bookings. | Trust the demo. |
| 5 | Retire the old one when the copy wins. | Switch on day one. |
A big company has decades of wiring to fight. A three-person shop has one inbox. Pick the booking flow. Let the page answer, check the slot, and confirm. Keep the old way running until the new one has booked real jobs. Then retire it. One real win this month beats a hundred slides.