$1,700 in missed gross doesn’t usually look like an AI problem. It looks like a service customer with 91,000 miles, $4,800 in equity, two declined maintenance lines, and a trade your used car manager would have paid up for — except nobody saw it until the customer sold it to somebody else on Saturday.
That’s where most dealership AI conversations get off track. Operators hear “AI” and picture some black-box closer that fixes response times, mines equity, writes emails, books appointments, predicts defections, and maybe makes the coffee. Then the demo ends, the contract gets signed, and six weeks later the GM is asking why the tool is sending lease-end messages to cash buyers and service offers to customers who traded out two years ago.
Todd Smith, founder and CEO of QuoreAI, put it plainly in an Auto Remarketing interview around the J.D. Power Auto Summit: dealers need a crawl, walk, run evolution with data as the foundation. He’s right. And frankly, I’d argue a lot of stores are trying to sprint on a sprained ankle.
There’s no plug in AI and it’s magic.
— Todd Smith, QuoreAI, via Auto Remarketing
The AI Problem Is Usually Not the AI
I’ve watched stores blame vendors for things that were really process rot. Duplicate customers in the CRM. Dead email addresses marked as active. Service op codes used five different ways across advisors. Equity tools nobody trusts because payoff data is stale. BDC notes that say “left message” 11 times and nothing else.
AI does not make that cleaner. It makes it faster.
That is the part dealers need to respect. A mediocre employee with messy data can only create so much damage in a day. An automated system with messy data can touch 3,000 customers before lunch. If the targeting is wrong, the offer is wrong, or the handoff is unclear, you didn’t modernize the store. You industrialized the same old leakage.
Crawl, Walk, Run Means Different Things in a Store
The phrase “crawl, walk, run” gets thrown around in software meetings until it loses meaning. On the retail floor, it should mean something very specific: don’t automate judgment until you’ve cleaned up observation. Don’t personalize until you can segment. Don’t optimize until the basic follow-up actually happens.
| Stage | What it looks like in a dealership | What has to be true first | Where stores usually trip |
|---|---|---|---|
| Crawl | AI flags customers worth attention: high-mileage RO, possible equity, lease maturity, declined work | Customer identity, vehicle history, mileage, and contact permission are reliable | Too many duplicate records and bad phone numbers |
| Walk | AI drafts or triggers outreach, routes opportunities, summarizes conversations, prompts advisors or BDC | Managers agree on who owns the next step and how fast it must happen | No one knows whether sales, service, or BDC is accountable |
| Run | AI continuously tests timing, message type, offer logic, and acquisition probability | Closed-loop reporting connects outreach to appointments, appraisals, purchases, and gross | Store measures activity instead of outcomes |
Most stores want to buy “run.” Most are operationally ready for “crawl” or maybe early “walk.” That is not an insult. Crawl done well can put real money on the board, especially in used-car acquisition and service retention.
Your Data Foundation Is Not an IT Project
Clean data sounds like something the controller, CRM admin, or vendor should handle. That thinking is why it stays broken.
The data that matters for dealership AI is operational data. It is created by salespeople, advisors, BDC reps, porters, F&I managers, and service cashiers. If the process rewards speed over accuracy, the database will show it. If advisors write vague notes to keep the lane moving, AI will inherit vague signals. If sales managers desk deals outside the system and update later, the machine is learning from leftovers.
Look at a simple service-lane acquisition case. The AI needs to know whether the customer still owns the car, whether the mileage is current, whether there is equity or at least a buyable payoff position, whether the vehicle fits your stocking plan, whether the customer has opted into SMS, and whether someone at the store will handle the reply within minutes instead of “when things slow down.”
Miss any two of those and the campaign gets noisy. Miss four and you teach the staff to ignore it.
The Back-of-Napkin Test: AI Readiness Per Dollar Touched
Here’s a simple way I’d score AI projects before signing anything: rank them by dollars touched and data confidence. Not excitement. Not demo quality. Not whether the OEM field rep liked it.
- High dollars touched, high data confidence: start here. Service-lane trade mining can fit if mileage, ownership, SMS consent, and appraisal workflow are solid.
- High dollars touched, low data confidence: fix the inputs first. Equity mining with bad payoff or ownership data will burn customers and staff.
- Low dollars touched, high data confidence: useful training ground. Appointment reminders, RO status updates, and basic follow-up can build comfort.
- Low dollars touched, low data confidence: leave it alone. That is where pilot projects go to die quietly.
This is also why customer communication is one of the more practical early AI lanes. The store already has the customer, the vehicle, the visit history, and a reason to reach out. Dealers using tools like AutoRelay in the service drive are not trying to replace the used car manager’s eye. They are trying to surface the right customer at the right moment, get a clean SMS conversation started, and make sure the opportunity does not disappear between the advisor’s screen and the sales tower.
That is a sane use case. It is narrow enough to manage, close enough to revenue to matter, and measurable without a 40-slide dashboard.
What I’d Audit Before Any AI Rollout
Pull 50 closed ROs from the last 30 days on vehicles you would have liked to own. Not all ROs. The ones that fit your used-car stocking profile.
- How many had a current mobile number?
- How many had SMS permission clearly documented?
- How many had mileage that looked accurate?
- How many customers were still tied to the correct VIN?
- How many had declined work or a service event that could create a buying conversation?
- How many received any trade/acquisition outreach within 72 hours?
- How many replies would have been handled by a named person, not a generic inbox?
If you cannot answer those questions from the DMS, CRM, and service records without a scavenger hunt, you are not ready to “run.” That’s fine. Crawl with discipline. Clean the fields. Assign ownership. Measure appointments, appraisals, purchases, and gross — not just messages sent.
AI is not a strategy. It is an amplifier. Make sure you like what it is about to amplify.
See how AutoRelay helps dealers acquire inventory from their own service drive → getautorelay.com