Everybody tells you AI changes the dealership. Nobody will name a job.
So I built the model. Every position in an average franchised rooftop, every department, on NADA Data 2025 ratios. Then I ran it three ways. The full interactive version is on toddsmith.ai. This is what it says.
The average rooftop
NADA counts 1,123,100 employees across 16,990 franchised dealerships. That is 66 people per rooftop, producing about $76.5 million in sales and $10.3 million in gross. Personnel expense takes 62% of that gross.
| Department | People | Share of the building |
|---|---|---|
| Service | 29 | 44% |
| New vehicle (including BDC and F&I) | 14 | 21% |
| Back office | 10 | 15% |
| Used vehicle | 7 | 11% |
| Parts | 6 | 9% |
Technicians are the biggest single group: 17 of the 66. One technician generates about $212,000 in gross against about $108,000 in cost. Remember that number. It comes back.
AI does not take a job. It changes the tasks inside it.
Take the service advisor. Between seven in the morning and close, your advisor does 14 things: greets on the drive, walks the vehicle, calms an upset customer, explains the work at delivery, recommends from the multi-point, quotes the repair, writes the warranty narrative, plans shop capacity, sends status updates, chases declined service, books the next appointment, orders parts, submits the warranty claim, builds the estimate.
AI shows up at three of those this year and two more next year. Same person. Same seat. Different day.
That is true of every position in the building. Nobody in your store has a job AI takes or does not take.
Three ways to run it
| Scenario | What the store does | People | Net impact per year |
|---|---|---|---|
| Bolt on (today) | Buys point tools and layers them on the current process | 66 to 66 | +$8,549 |
| Rewired (2028) | Owns its data record and rewrites pay plans | 66 to 66, 4 seats change | +$535,095 |
| AI native (2030) | Builds the store around its own model; org chart moves last | 66 to 62, 10 seats change | +$1,469,033 |
The bolt-on store spends $63,000 on technology and gains $119,334 in gross. After payroll moves and every other cost, it nets $8,549. You spent $63,000 and bought yourself almost nothing.
The rewired store runs the same headcount and makes 62 times more. Nothing about the AI is different. The data is owned and the pay plans are fixed.
Where the money shows up
In the bolt-on case, service drives most of the gross gain (+0.7 points), then new vehicle (+0.3), parts and used (+0.1 each). The back office adds nothing to gross. Taking cost out of the office shows up on the expense line, not the gross line. Saving half a person saves you nothing.
Pay pressure follows the work. Technicians see about +3%. Service advisors, the used car manager and the service manager see about +1%. BDC, reception, accounting clerks, warranty administration and marketing see about -2%. Sales consultants, F&I and the GM hold flat.
Why most stores will end up in scenario one
Here is the default path. A BDC tool goes in and one rep quits, and nobody backfills. Phone routing gets automated. The office gets quieter and deals post faster. The service manager asks for another tech and gets one, so AI just added a seat. Two years later you have fewer people, no documented decisions, and your vendors chose what got automated first.
Plenty of rooftops will come out of that fine. What they gave up was the say.
The build order
Own the record. Build the model. Fix the plans. Then move the chart.
- Own the record. Done when you can pull sales, service and inventory into one view without exporting.
- Build the store model. Done when the system knows capacity and inventory without anyone typing it in.
- Build the customer model. Done when one customer is one record across every system.
- Open it to people. Done when managers get answers instead of reports.
- Run one workflow end to end. One, not eleven, with a human approving.
- Rewrite pay plans. Nothing above this line sticks until this is done. It is the step everybody skips and the only one that costs nothing. Nobody in your building uses a tool that lowers their check.
- Move the org chart. By the time you get here, the chart is writing down what already happened.
Five questions for your store
Answer yes or no. Under three yeses and this model is still theory for your rooftop.
- Can you pull sales, service and inventory into one view without exporting?
- If you cancelled your biggest vendor, what would you still have?
- Is one customer one record, or four?
- Does your best tool take action, or hand somebody a list?
- Does any pay plan still pay for work a system already does?
If question two made you uncomfortable, start with the Data Ownership Scorecard. If question five did, that is what Blueprint is for.
Method: NADA Data 2025 ratios; average rooftop rounded to 66 people; each position modeled as one average person; no robotics, body shop, floorplan or competitive response modeled. Model version 0.4, September 2026. See the full interactive model.