Someone at your dealership has probably already connected an AI tool to your customer data. Does your Qualified Individual know? Under the FTC Safeguards Rule, once customer data flows from your systems into an AI tool, that tool is part of your information security program. So is every vendor in between. Five things to settle before anyone turns it on: Which account? A personal ChatGPT login is not the same as a business plan with training off and retention set. Who can see what? Can a salesperson pull credit applications? A coworker's pay plan? Is there an audit trail? Can you see who asked what, and what came back? Is every provider under contract? The warehouse, the connector, the AI. All of them. What's the plan if it leaks? Who calls whom, and how fast? This is not a reason to slow down on AI. I want dealers moving faster. But the dealers who win with AI will be the ones who set it up right the first time, not the ones explaining it to the FTC later. I put a one-page checklist in the comments. Forward it to your controller.

AI will create more value in auto retail than anything in 100 years. Most dealers won’t keep it. The The Presidio Group made a bold call: within five years, AI becomes the most productive operating technology this business has seen in a century. I think they’re right. I also think it’s the second most important question. The first one: who gets paid? Electric motors reached factories in the 1890s. Productivity didn’t jump until the 1920s. Owners bolted the motor where the steam engine had been and kept the same floor plan. The gains came only when someone redesigned the factory around the motor. Dealers are in the motor swap phase. 82% use AI. 9% report more profit. That isn’t failure. It’s the dynamo gap. Bolt AI onto a 1995 process and you get a 1995 P&L. Now the history nobody likes to talk about. The internet made car buying far more efficient. Who captured it? Not dealers. Third party sites built multibillion dollar companies selling dealers access to shoppers who used to walk in the door. The efficiency was real. The margin moved. AI is the same movie with a bigger budget. Personnel runs about 45% of dealership gross. Every point AI takes out goes one of two places: the dealer’s bottom line or a vendor’s subscription line. Cox just bought Fullpath. Every platform is racing to embed AI. Embedded means the intelligence lives in their house. When you switch, sell, or they get acquired, it stays there. This is where it hits valuations. Blue sky gets paid on earnings and assets that transfer. Two groups, same AI spend. One owns its data layer. One rents it. Only one built enterprise value. Call it the Intelligence Dividend. AI will pay it. The only question is whose name is on the check. The tell is in Presidio’s own piece. David Hudson pointed AI at a 3,500 line DMS invoice and found charges that didn’t match his contract. The first thing AI did for a dealer who owned the question: catch a vendor taking margin. Three moves: Own the layer underneath. Your data, in a store you control, readable across every department. Redesign the work. Don’t decorate it. Time saved and never redeployed is just slack. Measure in gross, not demos. A third of dealers aren’t measuring at all. The arms race is real. Arms races make arms dealers rich. Full Presidio piece in the comments. Brodie Cobb, are you seeing the dynamo gap show up in valuations yet?

Ask your AI how many cars you sold last month. Then pull your financial statement. I'll bet the numbers don't match. It's not because the AI is bad. It's because nobody told it: Whether a deal counts when it's booked, delivered, or funded. What to do with the unwind that hit the following month. Whether dealer trades and fleet count. That John Smith in the DMS, J. Smith in the CRM, and the guy on your website are the same person. Plug an AI straight into raw tables and it writes its own queries. It makes every one of those calls for you. Silently. Two managers ask the same question. They get two different numbers. The dangerous AI answer isn't the obviously wrong one. It's the confident wrong one nobody can check. So ask every vendor one question: "When your AI gives me a number, how do I know it ties to my financial statement?" If they don't have a clear answer, you have your answer. Run the test this week. Tell me in the comments if your numbers matched.

My Saturday thoughts with Green Tea 🍵 Every dealer I know can tell you their payroll to the dollar. Not one has told me their token budget. That's about to change. And I believe it will be the largest shift in budget allocation in the history of business. For a hundred years, a dealership had one way to buy capacity. You hired a person. Need more calls answered? Hire a BDC rep. More ROs written? Hire an advisor. Every growth plan started with headcount, and personnel became the biggest controllable expense on the statement. Now there's a second way to buy capacity. Tokens. A token budget is not a software budget. Software is a tool your people use. Tokens do the work. So this money doesn't belong next to your DMS in data processing expense. It belongs next to payroll. The two budgets behave nothing alike. Human budget: Fixed. Hired one at a time. Months to ramp. Can walk out the door. Token budget: Variable. Scales up at 9pm on a Saturday and back to zero by Monday. Never quits. Gets cheaper every quarter. So the planning question changes. It stops being "how many people do we need?" It becomes "what's the right mix?" Three things I'm telling dealers right now: Every token needs a manager. A token workforce with no one accountable for the output is just an expensive way to make mistakes faster. Tokens burned on bad data are wasted wages. If your DMS and CRM disagree about who your customer is, you're paying AI to argue with itself. The cheapest token is the one running on clean data. The human budget doesn't shrink. It moves. From doing to judging. You pay people for the relationship, the judgment call, and the accountability. The tasks go to tokens. My prediction: within five years, the 20 Group composite has a token line. CFOs will benchmark token cost per RO and per unit retailed the same way they benchmark personnel as a percent of gross today. The dealers who build that muscle now will be running the comparison while everyone else is still asking what it means. So, a question for the dealer principals and CFOs here: Do you know your token budget? Who owns it? Tell me in the comments.

If your AI has to figure out what your data means, it will figure it out differently tomorrow. A franchised dealership runs on 10 to 15 systems. Each one defines the same things differently. Is a deal sold when it's contracted, delivered, or funded? Does front gross include pack? Is that RO closed or posted? Is Jon Smith in the DMS the same person as John Smith in the CRM and J. Smith on the service drive? Those answers aren't in the data. They live in how each DMS, each CRM, and each department actually uses it. Some new data platforms promise to go "schemaless," which skips that step. Nothing tells the AI which field holds front gross, or whether pack has already come out of it. So it works that out from column names every time someone asks a question. Monday it picks one field. Friday it picks another. Your GM gets two different gross numbers for the same month and stops trusting both. In a dealer group it gets worse. Store 1 runs one DMS and store 12 runs another, so the AI guesses differently at each store. The group report ends up comparing numbers that don't mean the same thing. That isn't intelligence. It's a guess with good grammar. Here's what we do instead: → Map every field from every source system to one canonical dealership model → Resolve customers and vehicles across systems, so one person is one record → Clean every record at onboarding (about 17,000 for the average rooftop), then process only the changes → Define each metric once, so front gross means the same thing at store 1 and store 12, on Monday and on Friday The AI is only as good as what it reasons over. A schema isn't overhead. It's where the meaning of your business lives. Dealers: how many different answers does your store have for "units sold last month"?

Saturday thoughts with Green Tea 🍵 . Here's the question I keep sitting with this morning. If you opened a franchised dealership tomorrow with a blank sheet of paper, how much of what you do today would you actually keep? Be honest. For 25+ years we have built our processes around our technology. More accurately, around its limits. The CRM doesn't talk to the DMS, so someone re-keys the deal. The DMS can't tell you who is in an equity position, so someone builds a spreadsheet that turned into a separate software tool. The service schedule doesn't know which tech can do which job, so the shop foreman carries it in his head. The F&I menu doesn't know what the customer looked at online, so the box starts from zero. None of these are processes. They are workarounds. We have hired, trained, and paid good people to be the glue between systems that were never built to work together. We got so good at the workaround that we forgot it was one. Now we're entering the agentic age. Software can read, decide, and act across systems. The biggest risk I see isn't dealers ignoring it. It's dealers bolting agents onto the hacks and automating the re-keying instead of asking why anyone is re-keying at all. A faster broken process is still a broken process. The dealers who win the next decade will pull out a blank sheet and ask one question about every step: does this exist because the customer needs it, or because our technology needed it? Keep the first. Challenge everything in the second. So I'll ask you. What's one process in your dealership that only exists because your systems can't talk to each other? Tell me in the comments.

I've spoken with more than 200 dealer groups this year. The most common thing I hear isn't excitement about AI. It's hesitation. "We're not ready." "We tried a tool and it didn't deliver." "We don't know who to trust." All fair. Here's what I tell every one of them. You don't have to decide on AI today. You do have to get your data ready. When AI disappoints a dealer, it's rarely the model. It's the data underneath it. Duplicate customers. Missing service history. A DMS and a CRM that disagree on who bought what. Put AI on top of messy data and you get confident wrong answers. I've been saying this for over a year. So we made it its own product. QoreVault does one job. It pulls your data out of your DMS, CRM and inventory systems, cleans every record and keeps it current. No AI layer. Nothing new for your people to learn. Think of it as getting the books in order before you hire the CFO. When you're ready for AI, the hardest part is already done. If you're on the fence about AI, start here. What's holding your dealership back on AI right now: the data, trust, or time?

Your AI phone agent thinks that call went great. Every dealership AI phone agent demos the same way. Natural voice. Books the appointment. Works at 9pm on a Sunday. Then it goes live, and 90 days later the store quietly turns it off. Two reasons. Neither one is the voice. 1. The agent does not know your store. It is answering from a script and a website scrape. It does not know the RO history, the open recall, the part that has been on backorder for eleven days, or that this caller has bought four vehicles from you. A phone agent without the store's data is a receptionist with amnesia. The customer figures that out in about fifteen seconds. 2. The agent does not hear your customer. Almost nobody in our industry is talking about this one. The standard stack turns audio into text, then hands the text to a model. Tone, pace, emphasis, hesitation, sarcasm, anger. All of it thrown away before the model ever sees the call. So the customer says "Oh, perfect." Furious. Dripping with sarcasm. The transcript reads as satisfaction. The call gets logged as resolved. Your dashboard says the agent is performing. That is why the consumer numbers keep moving the wrong direction while the models keep getting better. SurveyMonkey's 2026 data has the share of people who would rather deal with a human than a chatbot rising from 83% to 85%, while preference for AI slipped from 7% to 5%. Zendesk found 74% get frustrated having to repeat information they already gave. A Parloa poll found 43.9% of Americans resort to yelling "human" or "person" into the phone to get away from an AI agent. A better model does not fix either problem. Owning your data fixes the first. Listening to the audio, not the transcript, fixes the second. Right now most dealers are measuring calls the AI handled. The caller who gave up and hung up counts as one of those.

Everybody tells you AI changes the dealership. Nobody will name a job. Two years of conferences, vendor decks and LinkedIn posts. Not one has said which seat in your building goes away, or when, or what you are supposed to do with the person sitting in it. So I named all twenty. I built a model on NADA Data 2025. 66 people, five departments, every position in a franchised rooftop. Then I ran what actually happens to each one. Some of it I expected. A lot of it I did not. Your payroll goes up, not down. Your techs turn more hours and flat rate pays them for it. The most exposed seat in the building is your warranty administrator. Nobody has ever walked into a 20 group worried about that seat. There is no BDC line on a factory statement. No F&I department either. Two of the three most exposed groups in your store are the two nobody has ever managed as a department. And the one that should bother you most. Leave your pay plans alone and you cut your own return in half. Same software. Same spend. One answer. It is interactive. Put your own headcount in and the whole thing rebuilds around your store, seat by seat. Go find yours. Link in the comments.

That showroom is from 1977. Everything in the photo has been replaced. The carpet, the wood paneling, the Granada, the LTD, the guy who sold them. Everything except the operating model. New, used, F&I, service, parts. Gross attributed by department. Pay plans written against department gross. The DMS took the org chart that already existed, encoded it into software, and the factory standardized the financial statement on top of it. That structure has barely moved in fifty years while everything else about the business has. Then we spent three years bolting AI onto the outside of it. Here is a test worth running with your managers in the room. Cancel every AI product in your store tomorrow. What does Monday look like? It looks like 2019, running a little slower. Not one process missing. Not one decision orphaned. Nothing structural to rebuild, because nothing structural was ever built. That is not a failure of the tools. Bolt-on was the only thing available. You cannot build something native on data you do not hold, using operations you cannot call. What has changed is that the constraint is now addressable. New whitepaper: The Dealership Nucleus. 24 pages on what an AI-native store actually looks like, the three constraints that belong to our business alone, and the build order that gets you there. Including the one nobody says out loud: your org chart is not just routing information, it is clipping gross. Link in the comments.

Happy 4th of July!

Really proud of our team at QoreAI and what we continue to build. prweb.com/releases/qoreai-la…

Read the truth about Dealership AI toddsmith.ai/whitepapers/ado…

How AI is reshaping dealership operations, data strategy, the intelligen... youtu.be/YtnyAJZUSPk?si=tlnL… via @YouTube

Every dealership meeting starts with the same ritual: Someone opens a dashboard. Then the room debates whether the numbers are correct. That’s not a reporting problem. That’s a data architecture problem.

Dealerships don’t have a data problem. They have a decision problem. The average store has 10+ systems recording information. Almost none have a system turning that information into clear decisions. That’s the real AI opportunity.

If your vendor resists API or SFTP… Ask yourself why. Data companies shouldn’t fear data movement.

AI trained on bad data doesn’t get smarter. It gets more confident. That’s worse.

Every vendor says: “You own your data.” Try moving it freely. Watch what happens.

Dealers don’t have a data problem. They have a data mobility problem. Big difference.