The Math Doesn’t Work: How the Cost to Sell the Same Car Tripled

For the first time in the history of the internet, humans are no longer the majority. According to Cloudflare Radar, automated traffic now accounts for 57.3% of worldwide HTTP requests to HTML content, compared with 42.7% for humans.

That single data point would be alarming in any industry that relies on digital measurement. In retail automotive, it helps explain a quieter, more expensive problem dealers have been living with for twenty years: we are paying nearly three times as much to sell the same number of cars.

The unit count that never moved

Start with the outcome nobody can spin how many vehicles actually sell.

Federal Reserve data and industry reporting show that U.S. new vehicle sales have lived in roughly the same band for a quarter century. Around 17 million units were sold in 2000. In 2025, total new sales were about 16.2 million.

There are year-to-year swing recessions, pandemics, supply shocks—but over the long arc, the line is flat. A full generation of “precision” marketing, real-time dashboards, and digital optimization has not produced more metal moved off the lot. If anything, we are selling slightly fewer new vehicles than we did at the dawn of the century.

The demand curve helps explain why. At any given moment, only a small slice of drivers are in-market for a vehicle—typically a low single-digit percentage, often around two to three percent. The remaining ninety-plus percent are not undecided; they are simply not buying this month. The pool of true buyers is finite, and it has not meaningfully expanded.

If the number of buyers is flat and the units sold are flat, any story about improved “efficiency” needs to show up somewhere else: cost.

The cost line that did move

Here the numbers tell a very different story.

Around the year 2000, a typical dealer spent $250 on advertising to sell a vehicle. By 2025, industry benchmarks and trade analysis put average advertising cost per sale around $700–$750 per unit, with many dealers reporting figures in that range or higher.

On paper, that’s an increase of nearly 3x in cost per vehicle sold. Even after accounting for inflation, most of the growth is real, not nominal. And the majority of that increase has gone to digital channels: paid search, paid social, programmatic, and an expanding stack of marketing technology. Today, roughly three out of every four dealer ad dollars end up on a screen.

Set the facts side by side:

• Similar annual new-unit volume: about 17 million in 2000 vs. about 16.2 million in 2025.bts+2

• Nearly three times the advertising cost to sell each unit: roughly $250 vs. roughly $739 per vehicle.

• A dramatic shift of budget into digital platforms that promised “optimization,” “efficiency,” and “more bang for the buck.”

If digital advertising delivered what it promised, one of two things should have happened: either the cost per sale should have fallen, or the number of units sold should have risen. The industry data shows neither.

The promise: optimization, efficiency, and Wanamaker’s ghost

For a century, retail advertisers lived with John Wanamaker’s famous lament: “Half the money I spend on advertising is wasted; the trouble is, I don’t know which half.” Digital platforms arrived as the answer.

Dashboards lit up with impressions, clicks, sessions, conversions, and cost-per-lead curves. For the first time, a dealer could watch “performance” move in real time. For general managers and marketing directors, especially those who did not own the store, this felt like salvation—a way to prove value in a business that had suddenly gone “digital” overnight.

The logic was straightforward: if everyone is using the same tools, we will simply become better at working those tools. Tune the campaigns. Improve the targeting. Drive down the cost per lead. Walk into the owner’s office with a dashboard full of rising numbers.

There was only one problem: what those numbers actually measured.

The quiet change in the meaning of “traffic” and “lead”

For most of the history of this business, the vocabulary of demand was human. Traffic was a person on the lot. An “up” was a live human being standing in front of a salesperson. A lead was a name, a phone number, an address—someone who had chosen to let the store find them again.

As digital channels expanded, the words stayed but the meaning drifted. Traffic became website visits. Ups became “unique visitors.” Leads expanded to include form fills, chat transcripts, and inferred intent signals.

None of this was announced in a meeting. There was no memo. The language simply slid, one definition at a time, until dealers were using the same words to describe a very different thing: activity, not identity.

Then came the statistic that clarified what many operators had felt for years: a growing share of what they were counting was never a person at all.

Cloudflare’s Radar data shows that bots now account for roughly 57.3% of worldwide HTTP requests to HTML content, with humans responsible for only 42.7%. In other words, more than half of the “traffic” hitting real web pages globally is non-human—automated agents, crawlers, and scripts.

Bot traffic does not map one-to-one to dealer ad spend, and major platforms attempt to filter non-human activity. But the headline is hard to ignore in a world where most web requests are generated by machines, any metric that equates “traffic” with “potential customer” is increasingly suspect.

The mirror that looked like a window

Inside the dealership, the effect is subtle but powerful.

On one side of the desk, a manager focuses on the dashboard: sessions climbing, click-through rates improving, cost-per-lead moving in the right direction. The screen looks alive. The work feels like progress. The metrics, all of them, are up and to the right.

On the other side, the dealer principal looks at the financial statement: advertising expense rising, gross under pressure, unit volume flat. The gut feeling is simple: something doesn’t add up.

Both perspectives are grounded in real data. They are simply looking at two different things.

Digital platforms report their own performance. They are designed to maximize and display the activity they generate impressions, clicks, and conversions as defined inside their systems. They are not designed to independently verify whether that activity represents a real, reachable person with the intent and capacity to buy a vehicle.\

In investigative terms, the dashboard is a mirror, not a window.

A mirror reflects motion—your campaigns are running; your metrics are moving. A window shows you the customer—who bought, whether you can reach them again, and how they came to your store.

Retail automotive has spent twenty years optimizing the mirror.

The finite 2% and the rising price of chasing it

Overlay the measurement problem on the underlying economics, and the cost spiral starts to look less mysterious.

• Only a small fraction of drivers are actively shopping at any given time—the “in-market” pool is finite and relatively stable.

• More dealers, OEMs, and vendors have piled into the same digital auctions, chasing the same small pool of intent signals: search queries, VDP views, retargeting lists.

• Platforms optimize for activity (clicks, impressions, conversions as they define them), not for verified, identity-based outcomes tied to actual sales.

When everyone is bidding to reach the same 2–3% through proxies, the price of those proxies goes up. Cost-per-click rises. Cost-per-lead rises. Cost-per-conversion rises.

Recent analysis in dealer-focused publications note exactly this: in 2025, paid search costs increased while click-through and conversion rates fell, leaving many dealers with higher bills and no corresponding lift in units sold.

The tools did optimize. They optimized interaction with the platforms’ inventory, not necessarily the dealer’s inventory.

Where the optimization promises break

Digital advertising was sold on a clear proposition:

• Better targeting

• More measurable results

• Less waste

• More sales for less money

If that proposition held at scale, the industry would see one of two unmistakable outcomes:

• Lower advertising cost per vehicle, or

• Higher units sold with similar spend

Instead, we see flat units and sharply higher cost per unit.

That doesn’t mean digital channels “don’t work.” It means the system is optimizing a layer that does not directly correspond to the outcome that matters in retail: selling a car to a person.

Clicks do not finance a vehicle. Sessions do not take test drives. Impressions do not sign paperwork.

You sell vehicles to human beings—with names, phone numbers, home addresses, credit histories, down payments, and reasons to want what’s on your lot. Those are the inputs of retail. Metrics that cannot be traced back to real, identifiable people are, at best, indirect indicators.

From an investigative standpoint, this is the crux: the industry quietly redefined success by numbers that no longer guaranteed a human being behind them.

The question dealers deserve to ask

For dealers, the core questions are not philosophical. They are practical and measurable:

• Of the “traffic” and “leads” I’m being shown, how many represent real, verified people I can reach?

• How many of my actual sales can I match back to those people and the channels that touched them?

• If I strip away non-human and non-identifiable activity, what does my true cost per sale look like?

Until those questions are answered with people instead of pixels, retail automotive is playing someone else’s game—spending more each year to move numbers on a screen that was not built to measure the outcome that matters most.

The way out: identity, households, and the 98%

The fix is not to abandon digital. The fix is to remember what every platform was supposed to be for: reaching people.

That means:

• Putting identity ahead of activity: prioritizing verified households and known individuals over anonymous clicks.

• Measuring sales outcomes first: matching back the vehicle sold to the person who bought it and the channel that reached them.

• Treating platforms as tools, not scoreboards: letting the customer data, not the platform’s dashboard, define success.

It also means looking beyond the 2–3% who are buying today.

If only a small fraction of the market is in the lane this month, then long-term growth depends on how consistently and credibly a dealer shows up for the other ninety-plus percent who are not yet shopping. That is brand work—patient, cumulative, and difficult to measure in real time. It does not fit neatly into a screen built to reward short-term activity.

The investigative conclusion is straightforward, if uncomfortable:

• We sell about the same number of cars today as we did in 2000.bts+2

• We pay roughly three times as much in advertising to sell each one.

• More than half of what the internet now calls “traffic” is non-human.

• The tools delivered optimization—but at the level of proxies, not people.

The question is not whether the platforms worked. It’s whether the industry defined “working” in a way that never guaranteed a real buyer at the end.

For dealers reading this, the challenge is clear: demand proof in people, not in pixels.

Ask where the names are. Ask where the addresses are. Ask how many of the “visitors” on the screen could ever walk through your door, pay for a car, and come back again.

If the answer is thin, the problem is not your store. It’s the game you were asked to play.

MuddVision helps dealers improve efficiency in advertising, lowering cost per sale and identifying REAL households who are ready to buy, then matching sales back to those people and the channels that touched them. Learn more.