Why I left selling cars

The Saturday I knew I had to leave, we lost a deal at a Honda store in Sarasota over a fourteen-dollar payment difference. Not fourteen hundred. Fourteen. The customer had been in three times across two weeks. She had two kids in the third row of the test-drive Pilot. She had her trade keys on the desk and a finance pre-approval letter folded into her purse. We had her.

What we didn’t have was a CRM that knew it. The be-back hadn’t been logged. The desk had no record of the prior pencil. F&I didn’t see the trade payoff we’d already validated. So when the salesperson re-quoted, he re-quoted blind, and the number was fourteen dollars off the one her husband had written down two Tuesdays ago. She stood up, looked at me, and said one of the truest things anyone has ever said to me in a dealership: “You don’t remember me.”

She walked. We never saw her again. The salesperson, a twenty-three-year-old who had been with us four months and was carrying the desk that day, went into the break room and cried — quietly, the way men cry when they’ve stopped being surprised by how the day ends. I followed him in. I told him it wasn’t his fault. He said it was the software, and that he was going to find a different store with better software, and I told him every store had the same software, because that was the truth.

I drove home that night and sat in my driveway for forty minutes without going inside. I had been a GSM for six years at that point and a salesperson and finance manager for fourteen years before that. Twenty years of watching the same scene play out in slightly different rooms. The names change. The CRM never does.

What I tried to fix from inside

I am not the kind of operator who blames the tools without first trying to make the tools work. For four years before I left, I tried to fix it from inside. I rewrote the lead assignment rules three times. I wrote a twelve-page T.O. playbook and trained every closer on it. I changed the follow-up cadence twice and the SMS templates four times. I paid for the “manager intelligence” bolt-on. I paid for the call-recording bolt-on. I paid for the data-quality bolt-on. I paid for the equity-mining bolt-on. I sat through every webinar.

None of it touched the actual problem. The actual problem was never the cadence. The actual problem was that the underlying system did not understand what a customer was. It understood what a lead was — a row in a table with a source and a status and a stale assignment. A lead is the artifact a CRM creates to justify its existence. A customer is the human who walked in three times across two weeks with her kids in the third row.

I tried bolt-ons. Every bolt-on assumed the database underneath it was telling the truth. The database was not telling the truth. The bolt-on, in good faith, just amplified the lie. The reports got prettier. The forecasts got confident-sounding. The deal still walked over fourteen dollars.

I tried training. I have done more rep training than I can count. Reps absorbed it on Monday and lost it by Thursday because the system they spent eight hours a day inside of actively punished the behavior I had asked for. If the CRM requires nine clicks to log a be-back, no amount of “always log the be-back” is going to win. The interface is the policy. Whatever the interface makes hard, does not happen. Whatever the interface makes easy, becomes the culture.

I tried switching. Twice. From one of the big four to another one of the big four. Same database underneath, same assumptions, same fourteen-dollar miss waiting in a different zip code. I do not think any of those companies are evil. I think they are old, and I think old software protects itself, and I think the people running them have not sat in a deal in a decade. That is not a moral failing. It is a structural one.

Why the incumbents can’t be fixed from outside

I’m going to name them, because pretending you don’t know who I mean is its own kind of dishonesty. CDK. Reynolds. VinSolutions. Elead. These companies have collectively done more for the dealership industry than any one rooftop ever will. They built the rails the modern franchise system runs on. The June 2024 CDK outage was a reminder of how much sits on those rails. None of this is said in spite.

But here is the structural reality: those products were architected in an era when a dealership’s information system was a digitized filing cabinet. The job was to replace paper. They did that job. They did it well. The job changed twenty years ago and the architecture did not. The schema still treats a lead as a row, the customer as a second-class join, and the conversation as an unstructured log line. You cannot retrofit a real-time AI-native speed-to-lead engine on top of a schema designed for batch paperwork. The bones don’t bend that way.

The other thing nobody on the inside will tell you: when your customer base is the entire installed franchise system, you ship for the slowest GM in the room. The slowest GM is the one who would rather not change. So your roadmap becomes the lowest common denominator of risk tolerance across twenty-thousand rooftops, and the operators who are ready for something genuinely better get told to wait three years for a beta. They wait. Three years pass. The beta is a re-skin.

That is not a fixable problem from inside the org chart. It is a structural problem about who the customer is, what the schema was, and how a public-company quarterly cycle treats risk. The only way out is a new schema, a new company, and an explicit commitment to ship for the fastest operator in the room, not the slowest.

The decision

On May 2nd I stepped off the floor. It was not a dramatic day. I told the dealer principal in the morning. I worked my last up in the afternoon. I helped one of the F&I guys close a back-end deal that needed a co-signer, watched the attach come in at one point eight, shook everyone’s hand, and drove home. No speech. No cake. I have always thought speeches were for people who were not coming back.

I am coming back, in a way. The whole point of building Gladius is to come back to every floor in the country — as the software that finally remembers the customer. As the CRM that finally treats the rep like a professional. As the system that, when the woman walks back in with the two kids and the trade keys, says: “She was here on the seventeenth and the twenty-fourth. Her husband’s number was four-oh-three. She told us the trade payoff was eighteen-two-fifty and we verified it. Don’t re-quote blind. Here’s the deal she already loved.”

That is the entire product brief. Everything else in Gladius — the speed-to-lead engine, the live pencil, the service-to-sales bridge, the rep passport, the AI brain, the AWAIS defense layer — is a corollary. It all comes from the same root sentence: the system has to remember the customer the way a great salesperson would remember her, and then it has to make it easy for any rep on the floor to stand on that memory.

What’s different now

Three things. None of them are the kind of thing the incumbents can copy in a quarter, which is the only test that matters.

One: it is multi-vertical from day one. Gladius runs auto, detailing, landscaping, stone fabrication, and BDC on the same engine. One schema, one rep passport, one AI brain. That sounds like a scope-creep mistake; it is the opposite. It is the only way to build a CRM that genuinely understands what a customer is, because the moment you have to make the same data model work for a stone slab order and an F&I attach, the lazy CRM-isms fall away. House leads is a concept that translates. So is be-back. So is follow-up cadence. The verticals share more than the incumbents will admit, and we use that to ship faster than any single-vertical shop ever can.

Two: AWAIS. The Autonomous Web Application Intelligence System is the category we built because nobody else was going to. It is a defense layer that sits between the customer-facing surface and the database, and it does in real time what your IT team wishes it could do at three a.m. Sixty-six rules, fifty-three signatures, two hundred sixty-one root-cause patterns. After June 2024 nobody in this industry gets to pretend the runtime layer doesn’t matter. AWAIS is the answer I would have wanted as a GSM whose store got knocked offline for ten days.

Three: the roadmap is public. Every feature we are building, every milestone we are committing to, every product we have not shipped yet, is on a page anyone can read. No NDAs. No beta-program theater. When I tell you we will have thirty signed accounts by Q4 2026, the number is on the website and you can hold me to it. When the number is wrong, you will see the number be wrong before I send you a press release saying it was right.

The contrarian bet

The contrarian bet is that the dealership doesn’t need ten more bolt-ons. It needs one operating system that knows the customer, the rep, the deal, the trade, the service history, and the conversation — all in one place, all in real time, all wired to an AI brain that has been trained by twenty years of watching the floor lose deals over fourteen dollars.

If I am wrong, the incumbents win the next decade and I go back to wearing a tie and walking a desk. That is a fine outcome. I love selling cars. The whole reason I am building this is that I love selling cars. I want the next generation of salespeople to have what we never had: a system that stops costing them deals.

If I am right — if a real operating system, built by an operator, for operators, ships fast enough to outrun the quarterly cycles of the incumbents — then the next time a customer walks in with two kids in the third row, the software will say her name before the rep does. The rep will stand on it. The deal will close. Nobody cries in the break room. The salesperson stays in the business another ten years.

That is what I left for.