AI Won't Tell You Where the Car Is
Every vendor at every conference this year is selling AI to dealerships. Most of it is aimed at a problem you don't have, and none of it solves the one you do.
A model can draft your follow-up emails, summarise a service history, or guess which lead is worth calling first. Those are real uses. But when a customer is standing in your showroom asking whether their car is ready, no amount of intelligence helps if nobody wrote down that it is.
AI needs a record to reason over. In most stores, that record is a whiteboard, a group chat, and whatever the detailer remembers.
01 What the data actually looks like
Out The Lot runs a franchise dealership in Ontario. I can see the status history of every vehicle that passes through it. Here is every retail delivery whose date has already passed — 1,258 of them, across four and a half months:
Those 1,050 cars were delivered. The customers drove away weeks ago. Nobody went back into the system to say so, because once the keys are handed over there is no reason to. Twenty-two people did go back and close the record. That is 1.7%.
The last useful status is the one before the last one.
Ask a model how many deliveries are still open and it will answer 1,236. The real number is close to zero. The model isn't wrong — the data is.
02 Fix the input, not the interpreter
We rebuilt our own reporting around this. "Active" is now defined by whether the delivery date has passed, not by whether someone marked the car delivered. That one change made every dashboard honest overnight, and it required no machine learning at all.
Predictive models tell you which deliveries will run late, based on your historical status data.
Your status history says a thousand cars are still sitting on the lot waiting to be handed over. There is nothing to predict from.
The order matters. Get a system where updating status is faster than not updating it — a phone in the service drive, not a desktop in an office. Once a year of clean history exists, prediction becomes worth buying. Before that, it's an expensive way to be confidently wrong.
03 Where AI genuinely earns its place
Not nowhere. Drafting the customer's ready-for-pickup message in their own language. Summarising six weeks of notes on a problem vehicle. Answering "which detailer is free this afternoon" in plain English instead of three filters. All of those sit on top of a record that already exists.
That's the test worth applying to any AI pitch you hear this year: does it need data your store doesn't currently produce? If the answer is yes, the tool isn't the next thing you need to buy.
Start With the Record
Seven stages, one screen, updated from a phone on the lot floor. See it on the kind of deliveries your store actually runs.
Request a demo