Your Dealership Was Already Disrupted. The Invoice Arrives in 2028.

This is not a prediction. It is an argument built on thirty-five years of firsthand experience inside this industry: the dealership is not dying, but it is bifurcating. One version survives. One version does not. The difference between them will be decided in the next eighteen months by a small number of specific decisions that most dealer principals are not yet treating as urgent.

I am writing this because I have been on both sides of every disruption that got us here. I sold cars before there was a computer on my desk. I consulted on the first tools that started giving away the information I had been paid to protect. I have walked hundreds of stores in the past three years watching the next disruption arrive. This time the stakes are higher. The window is shorter. And the industry is running the same playbook it ran in 1999, when dealers fought internet pricing in meetings I sat through, against a direction that did not care how the meetings went.

If you are still running the 2023 dealership model, this paper is about you.

I sold my first vehicle in 1990. A truck to a man named Bill Wahl. My commission was $837.62.

There was no computer on my desk. There was no internet. There were no comparison tools, no third-party pricing sites, no way for a customer to know what I knew before they walked through the door. Bill Wahl came in, looked at the brochures we gave him, sat down across from me, and negotiated a deal I understood far better than he did. I knew the invoice price. I knew the holdback. I knew the floor. He knew what was on the sticker and what he had in his wallet.

That gap was not incidental to the transaction. It was the transaction.

In 1999 I started consulting on a project called AutoNation Direct. The concept was straightforward: put vehicle pricing online so consumers could see it before coming in.

Dealers hated it.

The meetings were contentious. I sat across from dealer principals who had spent twenty and thirty years building businesses on the information gap I had benefited from when I sold Bill Wahl his truck. They understood exactly what we were doing to them. Not every customer was online yet. The portion who came in with internet-researched pricing was small. But it grew fast. And once you saw it growing, you understood that the direction was not ambiguous.

The 2001 Toyota Buy A Toyota initiative followed the same template. The network resistance was the same. The direction was the same.

What I did not fully understand at the time was that I was consulting on the first step of a graduation that would not complete for another twenty-five years. The internet began the erosion of the information asymmetry I had been paid to maintain. Comparison tools accelerated it. AI is finishing it.

The dealers who fought internet pricing in 1999 and won the argument lost the decade. The dealers who understood that the direction was set regardless of the argument built better businesses and are still open today.

The argument playing out now will end the same way. The difference is the window to act is eighteen months, not ten years.

Ask most people what a car dealership sells and they will say cars. That is correct the way it is correct to say a hospital sells surgeries. Technically accurate. Not where the value actually lived.

The American franchised dealership was not primarily in the business of moving metal. It was in the business of solving three problems that customers genuinely could not solve for themselves.

The first problem was information.

When I sold that truck to Bill Wahl in 1990, there was no source of truth available to him outside of what I chose to share. The invoice price was not public. The manufacturer-to-dealer incentives were not public. The holdback, the regional adjustment, the true floor below which the negotiation would not go. None of it was accessible. He came in with a brochure I gave him and a number in his head. I came in with every number that mattered. That asymmetry was not incidental to the business model. It was the business model.

The second problem was geography.

Franchise law in all fifty states created legally protected territories. You could not comparison-shop the same brand across town because the competing dealer did not exist in your territory. Buying a car was already exhausting. Most people were not going to drive four hours to save three hundred dollars. The territory was the moat.

The third problem was complexity.

A vehicle transaction is not one transaction. It is five or six layered on top of each other. Together, in a room where you have already been sitting for two hours, they become overwhelming. Dealers monetize overwhelm. The F&I office exists to convert complexity into margin. F&I gross profit per vehicle at publicly traded dealer groups topped $2,501 per vehicle in 2025, the highest in years, even as front-end margins compressed.

The financial architecture of the business reflects where the real value lives. Fixed operations generates just 13% of the average dealer's revenue yet produces more than half of total profitability. The front-end drives traffic. Fixed ops is where the margin actually lives.

Was the cost of that task your moat?

For automotive retail, the answer is yes, three times over. The dealership's moat was built on the cost of tasks the customer did not want to perform. AI automates those costs toward zero.

Information: the asymmetry is nearly gone.

In early 2026, researchers deployed an AI shopping agent to contact 100 dealerships about the same vehicle. Of the 92 that responded within 24 hours, only 33 provided a full out-the-door price breakdown. What the agent did with those 33 was something no human consumer had ever been able to do. It retained, organized, and compared every line item across every dealership simultaneously. It produced a ranked shortlist in minutes, without fatigue, without the social pressure of the showroom, and without any of the psychological mechanisms the traditional sales process had been engineered around for a century. That is not a faster consumer. That is a different kind of counterparty.

Geography: the moat is leaking.

Carvana removed the friction of distance for used vehicles. In early 2026, Carvana acquired a network of Stellantis franchises and applied the same model to new vehicles. The franchise laws require manufacturers to sell through franchisees. They do not require franchisees to behave like traditional dealerships.

A note worth making here: some OEMs maintain primary marketing area requirements that mandate a percentage of new vehicle sales originate within the dealer's designated territory. That partially constrains the pure cross-market model for certain brands. But it does not address the more fundamental erosion. The geographic moat is leaking from the inside, not just from Phoenix. Price transparency does not care about territory lines.

Complexity: the F&I office meets something that finds nothing tedious.

Electric vehicles are accelerating this graduation from the other direction, though at a pace the market has revised downward. New EV sales in the US fell 28% in Q1 2026 compared to the same period in 2025, driven primarily by the expiration of the $7,500 federal tax credit. EV market share retreated from a peak of 7.5% in Q3 2025 to around 5.8% by early 2026. The immediate pressure on the service lane is therefore more gradual than the most aggressive projections of eighteen months ago.

But do not read the slowdown as a reversal. Hybrid sales surged 27.6% in 2025 to over 2 million units. New EV inventory stands at 130 days' supply versus 89 days for ICE vehicles. The structural thesis still holds. The F&I product shelf is shrinking. The direction of travel is set.

The fourth force: the OEM has been watching, and the data is worth $750 billion.

For a hundred years, the manufacturer made the vehicle and the dealer owned the customer. The connected vehicle changed that math. Today, an OEM with a modern platform knows where the car is, how it drives, when it needs service, and what features the driver uses.

McKinsey estimates the connected car data value pool will reach $450B to $750B worldwide by 2030, representing up to $310 in annual revenue per connected vehicle. General Motors has publicly stated a target of $20 to $25 billion per year in software-enabled sales and services by 2030. The OEM controls every one of those revenue streams. The dealer's share of the connected vehicle economy is the repair order when something requires a physical fix.

That is not nothing. It is also not $750 billion.

The dealer who does not understand this distinction is planning for a business that captures the labor and parts margin on the vehicle while the manufacturer captures the data economy built on top of it. For a hundred years that arrangement was invisible because there was no data economy to speak of. There is now. And none of it is flowing through the franchise.

In April 2026, Cox Automotive acquired Fullpath, an AI-native customer data platform. The largest technology ecosystem in automotive retail had to acquire a startup founded in 2018 to get the AI-native data layer it needed. That is a description of an industry that has been accumulating data for twenty years without building the infrastructure to use it.

The CDK cyberattack in June 2024 confirmed the substrate problem. A ransomware attack took half of automotive retail offline for two weeks. Dealerships were doing deal jackets by hand, the way it was done when I sold Bill Wahl his truck. That is not a substrate ready for the intelligence layer being sold on top of it.

The average salesperson in America sells 10 cars a month. NADA data shows that number has not moved in 75 years, not through computers, CRMs, or digital retailing tools. Technology layered on top of an unchanged model does not change outcomes. The AI nucleus is not another technology layer. It is the model changed.

There will be fewer dealerships in America. NADA counted 17,446 franchise dealerships in 2025. Not all of them will make it through the next decade.

Glenn Mercer has tracked average dealer net profit before tax for fifty years. In that span, the metric has held in a remarkably tight band of 1.5% to 2.5% of revenue. Not once in fifty years has the average dealer in America lost money.

“Not once in fifty years has the average dealer in America lost money. That band is about to break.”

That band is about to break. The question is which direction it breaks for your stores.

Haig Partners data puts the average blue sky multiple at five times net profit before tax. At $2.5M net profit, the average rooftop is worth $12.5M. A dealer who doubles net profit to $5M has a rooftop worth $25M. For a five-store group the difference is $62.5M against $125M in blue sky value.

Graduating the Deal · Todd Smith

The Financial Stake of Getting This Wrong

What the bifurcation means per rooftop, and per dealer group

Average Dealer Today

Running the 2023 model

Net Profit Before Tax$2.5M

Blue Sky Multiple avg 5x$12.5M

Per Rooftop Value$12.5M

5-Store Group Blue Sky$62.5M

AI Nucleus Dealership

Redesigned for 2027

Net Profit Before Tax$5.0M

Blue Sky Multiple avg 5x$25.0M

Per Rooftop Value$25.0M

5-Store Group Blue Sky$125M

+

$62.5M

additional blue sky value
per 5-store group

2x

per rooftop

Based on doubling net profit to $5M per store. NADA data. Haig Partners 5x multiple.

Sources: NADA 2025 Data Book · Haig Partners Q3 2025 Haig Report · Steve Greenfield investor presentation 2026 · QoreAI analysis

Version one: the dealership redesigned around an AI nucleus.

The AI nucleus dealership deploys agents where the traditional dealership deployed people. For a hundred years, the answer to every workflow deficiency was the same: hire someone. BDC not responding fast enough? Hire more reps. Service lane backing up? Add an advisor.

That model is over. The AI nucleus dealership deploys agents where the traditional dealership deployed people.

One person managing a well-configured fleet of agents replaces the function of eight to ten human positions in repetitive, process-driven roles. The tech-first employee is the human who manages this fleet. They are the most important hire you will make in the next eighteen months. Not because of what they do themselves, but because of what they deploy and manage. They are the agent operator. Find them. They are already in your building, under thirty, and have already automated something small without being asked. Promote them. Give them a title and four hours a week protected from everything else.

The AI nucleus dealership also goes after the opportunity the version-two dealership perpetually defers. The average RO written up is approximately $900. The average collected is $494. That gap, multiplied across 284 million ROs at franchise dealers last year, represents more than $115 billion in gross profit sitting in declined service work. Stone Eagle data shows roughly $80 billion in declined F&I products that were never revisited.

Version two: still fighting the wrong war.

Version two is a viable business today. It will be a smaller business in 2027. It will be an endangered business in 2028. Not because anyone defeats it. Because the problems it was organized to solve will have graduated further than it moved.

Graduating the Deal

The Dealership Redesigned Around an AI Nucleus

The AI nucleus is not a tech stack. It is a fundamental redesign: what does AI handle, and what do humans do because humans are uniquely better at it?

Version One: AI Nucleus

Data infrastructure at the center. Digital employees building the intelligence layer. Humans amplified, not replaced.

Version Two: Bolt-On Model

AI tools added on top of unchanged workflows. Faster version of a graduating model. Viable today. Endangered by 2028.

The competence is not the moat. The moat is what you build with it next. , Todd Smith

The most common reason dealers defer this decision is a number in their head that is too large.

What it actually costs.

Total first-year cost for a three-to-five store group: $80,000 to $150,000. Compare that to the $150,000 to $400,000 most dealers at that scale already spend on AI and technology vendor fees. The AI nucleus does not cost more than the current model. It costs differently. The money moves from vendor fees to infrastructure and talent. At the end of the year, the dealer owns the intelligence instead of renting it.

“The AI nucleus does not cost more than the current model. It costs differently. At the end of the year, the dealer owns the intelligence instead of renting it.”

The 18-month build sequence.

What Monday morning looks like.

I sold my first car in 1990 with no computer on my desk and complete control over every number in the room. By 1999 I was consulting on the tools that started giving those numbers away, watching dealers fight a direction they could not reverse. I am writing this in 2026 watching AI complete the job the internet started, at a speed the internet never managed.

The frozen human competence in your organization, everything your best people know about deals, customers, and service, is not what is being replaced. What is being replaced is the scarcity of that competence. AI makes it available everywhere. What AI cannot do is apply that competence with the judgment, the relationship, and the physical presence that your best people bring.

That is what you build around. That is the AI nucleus. That is what survives.

There will be fewer dealerships. The ones that remain will be the ones that stopped defending the 2023 model and started building the 2027 one before it was obvious they had to.

You have about eighteen months.

The clock is running. The canary is still alive.

Do the honest work first.

• 01Benedict Evans, AI Eats the World, May 2026. ben-evans.com.
• 02Dan Shipper, interview with Lenny Rachitsky, Lenny's Podcast, 2026. lennyspodcast.com.
• 03Steve Greenfield, investor conference presentation, 2026. Source for Glenn Mercer 50-year data, NADA 2025 statistics, labor cost breakdown, fixed ops profitability, declined service opportunity, and blue sky valuation math.
• 04McKinsey & Company, Monetizing Car Data and Unlocking the Full Life-Cycle Value from Connected-Car Data. mckinsey.com.
• 05General Motors investor communications, 2024 to 2025. $20B to $25B annual connected services target.
• 06Cox Automotive, Q1 2026 EV Sales Report.
• 07U.S. Energy Information Administration, Electric vehicle sales fell as hybrid vehicle sales continued to rise in 2025, 2026.
• 08Haig Partners, Q3 2025 Haig Report. F&I gross profit per vehicle and blue sky valuation multiples.
• 09Stone Eagle, F&I attach rate and gross profit per product data, 2025.
• 10Stella Automotive AI, dealer survey on missed inbound calls, 2025.
• 11Digital Dealer, AI shopping agent study, Q1 2026.
• 12Cox Automotive and Fullpath acquisition, April and June 2026.
• 13The Intelligent Dealership, Todd Smith, January 2026.