Franchised dealers spent $9.96 billion on advertising in 2025, per NADA’s annual report. That works out to $586,246 per dealership, or $739 per new unit sold, up from $705 in 2024. Quote that $739 in a 20 Group meeting and heads nod because it sounds like a cost per sale. It isn’t one, and treating it like one is how dealer groups lose money on their last advertising dollars while the average tells them everything is fine.
The national number is a ratio, and any average, national or your own, hides the only question that matters: what did the last dollars buy, and did those cars earn more than the dollars cost? A good cost per vehicle sold is less than the contribution gross on the units those dollars added. Everything else is bookkeeping.
What does the $739 measure?
NADA’s chart is titled “Average Dealership Advertising per New Unit Sold.” It divides dealership advertising by new units only. NADA doesn’t break that advertising out by department, so you can’t tell how much of it chased used buyers or service customers, and none of those sales sit in the denominator. The result is useful for tracking direction over time. It isn’t a cost to sell a car.
NADA’s 2026 midyear report shows $791 per new vehicle sold June year-to-date, against $722 a year earlier, counting advertising and sales promotion net of rebates. Those figures cover January through June, so don’t compare the $791 to the full-year $739. A ratio is only as honest as its construction, and this one was never built to answer the question a general manager (GM) is asking.
So what should you compare your spend against? The temptation is gross profit per unit. Presidio-NCM’s Q1 2026 benchmark of more than 4,000 franchised dealerships shows gross profit per vehicle falling to an average of $1,781 new and $1,253 used. The coverage doesn’t say which gross that is. Resist the urge to divide one source by the other. They use different samples and different periods, so the result looks like insight and carries none.
I could find no independent, method-disclosed cost-per-sale benchmark for this industry. Most of what I found repackages NADA’s ratio.
Why does an average mislead?
An average blends two things that behave nothing alike: the base spend you would run no matter what, and the dollars you added last quarter. The base includes co-op commitments and the website and customer relationship management (CRM) fees you pay every month. The added dollars are the new campaign, the bigger budget on a third-party listing site, the connected TV test one store talked you into.
The average treats all of it as one pool. The added dollars are the only ones you control month to month, and the average can fall while those dollars lose money.
Run the arithmetic. If your base spend sells cars efficiently and you pile weak dollars on top, the blended number drifts up slowly enough that nobody flags it. Worse, if units rise for reasons that have nothing to do with advertising, such as a hot incentive from the original equipment manufacturer (OEM) or a competitor’s inventory drought, your cost per vehicle sold drops and the whole budget looks vindicated. The ratio moved. Nobody asked what caused what.
That question has a name. Google defines Conversion Lift as the increase in conversions caused by the presence of the ad. Caused, not correlated with. Your dealer management system (DMS) tells you a unit sold. It doesn’t tell you whether the ad made the sale happen or showed up at a sale that was happening anyway. That gap is why credit and cause need separate answers.
What goes in your number?
Before you can judge the number, build it honestly. Incomplete spend and generous attribution flatter every vendor on the account.
Include all of it: media, agency fees, and the vendor retainers that function as advertising whether or not accounting codes them that way. Net out co-op reimbursements, because the question is what the store spent, not what the invoice said. If a tool exists only to generate leads, its fee belongs in the number.
For the denominator, pull sold units from the DMS, not from vendor dashboards. Vendor-reported sales counts come with vendor-favorable matching logic. The same DMS records can teach Google and Meta which leads bought.
Then split three ways. New, used, and service each get their own number, because the economics differ. The Presidio-NCM averages show why: $1,781 on a new unit against $1,253 on a used one means the same advertising dollar clears a different bar depending on which lot it serves. A blended number can hide a used operation that’s carrying the whole budget.
And compute it by rooftop. A 12-store group running one blended cost per vehicle sold has decided not to know which four stores are burning money. The group average is a courtesy number for the bank. The store number is the one you manage.
What did the next car cost?
Take a hypothetical store with round numbers. It spends $60,000 a month on advertising and sells 120 units, so the average is $500 per unit. The GM adds $10,000 to the budget, and the store sells 126.
The new average is $556 per unit, which still looks healthy. Now walk the margin. The added $10,000 bought six units, so each added car cost $1,667. Hold that against the Presidio-NCM averages of $1,781 new and $1,253 used. If those six were new units, the added spend ate most of the gross. If they were used, it exceeded the gross before a salesperson got paid.
The average said $556. The margin said the last $10,000 barely paid for itself, and the margin was right.
One caution before you act on a month like that. A single-month swing is a clue, not proof. Weather, inventory, an incentive change, or plain noise can move six units. The way to know is a holdout test: pull the added spend from two comparable stores, keep it running at two others, and compare sold units over eight to 12 weeks. It’s painful to run because somebody’s pet campaign goes dark for a quarter. It’s still less expensive than funding a losing campaign for three years because the average never complained.
What’s a good number?
Less than the contribution gross on the units the spend added. That’s the whole answer, and it’s why no national figure can hand it to you.
Contribution gross means what an added unit earns the store: the gross on the deal, finance and insurance (F&I) income if your gross figure leaves it out, and the fixed ops relationship a sold unit seeds. You set the bar from your own financial statement, store by store, because your grosses aren’t the benchmark average. The same logic drives a budget built from gross instead of a copied percentage.
Notice what this definition rules out. It rules out “our cost per vehicle sold is below the NADA number, so we’re fine,” because the NADA figure is a ratio over new units only, built for watching trends. It rules out “the blended average fell this quarter, so the new vendor is working,” because blends hide margins. The standard is one sentence long, and it forces the right arithmetic: what did the last dollars cost per added unit, and what did those units earn?
A budget passes the test when every layer of spend clears that bar. It fails when the first $40,000 earns its keep and the last $20,000 rides along because nobody separated them.
Who owns this number?
Not the vendor who gets graded by it. An agency reporting its own attribution is writing its own report card. If three vendors touched one lead, each can claim the sale, so the dashboards can add up to more cars than the store sold.
Not the OEM either. Co-op programs shape where money can go, and compliance questions belong with your counsel and your factory rep, not with a vendor’s assurance that a campaign qualifies.
The number belongs inside the group, with one person who has budget authority across rooftops, no commission on media, and the standing to pull spend that fails the test. In a group of three to 15 rooftops, that role rarely exists. The controller owns the statement, each GM owns a store, and the vendors own the reporting. Nobody owns the arithmetic between spend and gross.
I’m a fractional chief marketing officer (CMO). I spent three and a half years running marketing for La Mesa RV, an 11-store RV dealer group, where I moved $500K a month out of broadcast that couldn’t be measured and into digital that could. I’ve never worked at an auto dealer, and the arithmetic doesn’t care: Spend is spend, units are units, and asking what the last dollars bought works anywhere big-ticket inventory sells off a lot. The job is building the number honestly, splitting it by store and department, running the holdouts, and holding every vendor to a bar none of them will volunteer. A dealer group that puts one accountable person on that arithmetic stops paying for the average and starts paying for the margin.