RH Dépannage is a vehicle recovery business in Lyon: two trucks, a phone that rings at terrible hours, and — like every trade that’s been going a few years — a customer list nobody had touched since the day it was exported from the old system.
A list everyone had given up on
1,842 past customers and old enquiries, sitting in a spreadsheet. Every recovery firm has this file. Nobody works it, because working it is a fortnight of admin: dedupe it, figure out who’s still relevant, write something that doesn’t sound like spam, send it in batches, log who replied, book the jobs. The maths never works when the person doing it is also the person out on a recovery.
The instruction, verbatim
“Go through the old list and get us some pre-winter vehicle checks booked. Don’t annoy anyone we’ve worked for recently.”
That’s it. One sentence, typed by the owner, with two rules already live from week one:
The plan it came back with
Datify read the list against its own records first. Of 1,842 rows, 214 were duplicates and 96 had dead email addresses. 118 had recent work and were removed by the 90-day rule before any message existed. What was left got split — past breakdown customers first, everything else behind them — with a priced plan and an expected return, held for approval:
The owner read it, changed nothing, and pressed approve. Total time invested so far: the length of one coffee.
What actually happened
The sends went out paced over eight days, each one a short, plain message about a pre-winter vehicle check — no discount, because nobody asked for one. Replies landed against the right customer record, and anyone who answered dropped out of the follow-up sequence automatically. Two people asked to be left alone; they were marked so it stays true forever.
Twenty-three vehicle checks, booked straight into the diary, from a file that had been dead weight for two years. At RH Dépannage’s call-out rates that’s a five-figure return on a list they already owned — and on roughly an hour of the owner’s attention across the whole fortnight.
What changed afterwards
The interesting part isn’t the 23. It’s that the list is no longer dead — it’s a maintained asset now. The 104 conversations are logged, the bad addresses are gone, and next September the same instruction takes thirty seconds, because the system remembers what worked this time. The reactivation rate is already part of its arithmetic for the next plan.
The work was never hard. It was just fourteen hours of admin nobody had. One instruction, two rules, a plan read before it ran — and the dead list paid for the software for a couple of years.
Writes about what actually happens when you put an AI in charge of real business operations — including the parts that go wrong.