Cadillac's Formula 1 team was built from a blank page, with its records set up properly from the first day, and it still ended up with wrong ones. In the rush to get a car on the grid, some parts that should have carried serial numbers did not, and others turned out later to need them. The team is now using an AI tool to find and correct those records.
Why an old mistake costs so much to fix
A serial number sounds like a small thing. At Cadillac it feeds the financial reports, the engineering work and the checks that keep the car safe to run. Once a part has been fitted into an assembly, and that assembly into a larger one, every later transaction depends on the first record being right.
So a fix cannot be made in one place. Someone has to trace transactions that sit 20 to 50 layers deep, reverse them all, correct the part and replay everything in order. That took the team about two weeks for a single part.
This is why most businesses carry their errors. A mistake costs little on the day it is made, and the cost of fixing it grows with every entry that comes after. The cost of living with it feels small and spread out, so the repair never wins a place on anyone's list.
What the AI changes
The tool works because the team's system keeps the full history of every part. That lets the AI do the reversing, correcting and replaying that a person did by hand. The team is still testing it, and it expects each part to take two to five minutes instead of two weeks.
It also does something no one had time for before. It applies the rules for classifying parts across the whole system and flags items that look misfiled, which turns up problems the team did not know it had. For a business, that second use matters more than the first, because most bad records are ones nobody has reported.
There is a lesson in why this is possible. A system that records every step can be unwound, and a spreadsheet that gets overwritten in place cannot. So an AI plan is partly a plan for how you keep your records, and a history of changes is what makes repair possible later. The same point holds at Cadillac: its team treats its AI plan as a data plan.
What it looks like in an ordinary business
Take a distributor with 4,000 products. For two years, some supplier items were entered by the case but sold by the piece, so the stock counts and margins on about 200 products are wrong. Checking the history of each by hand, at an hour apiece, is five working weeks. Nobody does it, and the buyer simply learns not to trust the count on those items.
A tool that can read the purchase and sales history could propose a correction for each one, and the buyer could approve them. That is a day or two of work. The decision about what the right unit is stays with a person; the weeks of tracing do not.
Stopping the next mistake
The Cadillac team has noticed that errors come mostly from people who use the system only now and then. Its plan is a chat-style way to order parts, so that someone who rarely logs in does not have to find their way through the system's screens. It is a plan, not something running yet.
The same fix suits any shop. The person who places an order once a month is the one who picks the wrong unit, and a tool that asks "cases or pieces?" catches it before it becomes an old mistake.
A useful first step is to write down the three wrong things in your records that everyone around you already knows to work around. Those are the first jobs to hand to an AI tool.