$412. That was the number a used car director showed me last month on a late-model SUV that sat three extra days waiting on recon photos, an internal RO update, and a pricing decision nobody wanted to own. Not $412 in theoretical “efficiency.” Real holding cost, floorplan drag, price movement, and a missed weekend shopper. Multiply that by 40 or 60 units a month and suddenly the AI demos we all rolled our eyes at don’t feel quite as cute.
At NADA Show earlier this year, AI was everywhere. No surprise there. What did surprise me was how quickly the conversation moved past chat widgets and BDC scripts. The better operators weren’t asking, “Can it answer a customer?” They were asking, “Can it shorten recon by 18 hours?” “Can it tell me which service customers are in equity before the auction rep gets my money?” “Can it keep my pricing manager from chasing yesterday’s market?”
AI Is Moving From the Front Door to the Back of the Store
For the last few years, most dealership AI lived in communication: chat, lead response, appointment reminders, review requests, and a lot of awkward “Hi, I’m your virtual assistant” messages that customers saw through in about six seconds. Some of it worked. Some of it created more noise for the BDC. A lot of it was just automation wearing a nicer name tag.
The new wave is different because it is showing up inside operational chokepoints. Recon status. Parts delays. Pricing moves. Appraisal mining. Service lane equity. Trade follow-up. Inventory merchandising. Those are not vanity projects. Those are places where a day lost has a dollar amount attached to it.
Look, I’ve seen stores blow money on technology because a dealer principal came back from a convention fired up. The vendor had a slick booth, the demo unit was clean, and nobody asked how dirty the store’s data was. Six months later, the tool was still “in pilot,” which is dealership code for dead but not cancelled. AI will create plenty of that.
But I’d argue that dismissing all of it as vendor theater is just as lazy. The margin pressure in used cars is too real, and the old playbook of buying deeper at auction and hoping the market bails you out is not as forgiving as it was when everything with wheels brought money.
The Used Car Department Has Too Many Clocks Running
Here’s the framework I’d use before signing anything with “AI” in the sales deck: identify which clock the tool is supposed to compress. If it does not shorten a clock, improve a decision, or reduce manual follow-up, it’s probably just another dashboard.
- The response clock: how long it takes to engage a customer after a lead, text, service visit, missed call, or equity trigger.
- The recon clock: how long a vehicle sits between acquisition, inspection, parts approval, detail, photos, and frontline-ready status.
- The pricing clock: how quickly the store reacts when market supply, condition, mileage bands, or competitive listings change.
- The acquisition clock: how fast the store identifies and contacts owners of vehicles it actually wants before they hit wholesale channels.
- The accountability clock: how long managers wait before they know a process is stuck.
That last one gets ignored. In a decent-sized store, a unit can be “almost ready” for four days. Everybody thinks somebody else has it handled. The used car manager is looking at the board. Service is waiting on approval. Detail is waiting on service. Photos are waiting on detail. By the time the car goes live, the market has moved and the first price cut is already baked in.
Recon AI Sounds Boring. That’s Why It Matters.
The flashiest AI demos still tend to be customer-facing. The money, in my opinion, may be in the dull stuff: reading repair orders, flagging stalled units, predicting parts delays, routing approvals, and telling a manager which five cars need attention before lunch.
A store doing 120 used units a month does not need a robot to write poetry. It needs fewer vehicles sitting in “service pending” because nobody noticed the estimate crossed the approval threshold. It needs the used car manager to know that a $1,900 recon decision on a 74-day-old sedan is different from the same decision on a fresh truck with three appointments on it.
The data doesn’t fully support every claim vendors are making yet, but the direction makes sense. If AI can read messy internal notes, summarize status, and surface exceptions faster than a manager walking the back lot with a clipboard, that is useful. Not glamorous. Useful.
Pricing Is the Dangerous One
Dynamic pricing gets dealers’ attention because it feels close to the gross. It also makes me nervous when stores treat it like autopilot. Manheim data has continued to show wholesale values moving unevenly, with segments behaving differently and retail demand not always matching what buyers are paying upstream. That is exactly the kind of market where bad automation can make a manager faster at being wrong.
A pricing tool can process more listings than a human. It can catch stale units. It can spot that your 2023 compact SUV is now surrounded by lower-mileage certified units within 50 miles. Good. But it cannot know, unless you feed it properly, that your unit has new tires, a minor Carfax ding, no second key, and a service customer coming in Saturday who already drove it once.
That is where operators need discipline. AI should recommend. Managers should decide. If every recommendation gets accepted because the desk is busy, you haven’t improved pricing. You’ve outsourced judgment.
The Acquisition Angle Is Where Stores Can Actually Separate
Auction buying is not going away. Digital wholesale platforms are not going away. But if your first shot at a vehicle is the same feed every other dealer is watching, you are already paying retail for wholesale access. The better opportunity is still the customer you already know: service history, mileage pattern, equity position, loyalty, declined work, and ownership cycle.
AI matters here because the service lane throws off more signals than most stores can manually chase. A 62,000-mile SUV in for brakes. A lease customer servicing outside the original selling store. A declined $2,400 repair on a vehicle your used car department would buy today. A customer with positive equity who has not submitted a lead anywhere. Those are acquisition opportunities hiding inside fixed ops traffic.
This is also where communication quality matters. If the first text reads like a blast from a shared lead provider, you’ll get ignored. Dealers using platforms like AutoRelay are trying to make that outreach more timely and specific: the right service customer, the right vehicle need, the right SMS follow-up, without asking an advisor to remember 19 side tasks during morning write-up.
A Simple Math Test for AI Spend
Before buying any AI product, I’d put it through a gross-preservation test. Forget the demo metrics for a minute. Use your own store.
- Pull 90 days of used retail sales.
- Find average days from acquisition to frontline-ready.
- Find average days from frontline-ready to sale.
- Estimate daily holding cost per unit, including floorplan, depreciation risk, insurance, lot expense, and opportunity cost.
- Calculate what one saved day is worth across your monthly volume.
If you retail 140 used units a month and one saved operational day is worth even $45 per unit, that is $6,300 a month before you touch improved close rate, fewer markdowns, or better acquisition cost. If the tool cannot credibly point to where that day comes from, press harder.
Same logic applies to service lane sourcing. Compare cost-per-acquired-unit from your own drive against auction fees, transport, arbitration exposure, and recon surprises. Most stores do not have that number cleanly. They should. You cannot manage the spread if you only track what you paid at the block.
What I’d Audit First
Do not start with the AI budget. Start with the leak. Pull 30 vehicles currently in stock and trace the timeline: acquisition date, inspection date, estimate approval, RO close, detail, photos, online listing, first price change, first lead, sale date. You will find at least one ugly handoff. Probably four.
Then pull 50 closed service ROs from the last two weeks on 5- to 9-year-old vehicles with decent mileage and customer-pay work. Count how many would have been worth an acquisition conversation. Count how many got one. That gap is not a technology problem yet. It is an operator’s map.
AI is not going to fix a store that refuses to define ownership. It can, however, make the misses harder to hide. For a used car manager or service director, that may be the most valuable part.
See how AutoRelay helps dealers acquire inventory from their own service drive → getautorelay.com