The average price for a car model is close to the least useful number on a shopping page. Treat it as your benchmark and you'll end up doing one of two things: overpaying for the stripped-down version because "that's below average," or walking away from a loaded one that was never overpriced in the first place. Either way, you're negotiating against a number that was never describing your car to begin with.
I say this as someone who runs a pricing model for a living, on top of 1.13 million live listings across 73 regions. We flag roughly 12,800 of those as underpriced on any given day — a little over 1% of everything we track. If "below the model average" were our bar for a deal, that number would be ten times bigger and mostly wrong.
What the average is actually averaging
Here's the problem nobody puts on the sticker. A model name is not one car. It's a container for a dozen different cars wearing the same badge, and when you average their prices you're not describing the typical one — you're describing none of them.
Take a brand at the scale we actually see it:
| Make | Listings in our data |
|---|---|
| Toyota | 94,455 |
| Volkswagen | 84,970 |
| Mercedes-Benz | 63,516 |
| Hyundai | 61,724 |
| BMW | 57,457 |
Toyota alone covers close to 95,000 listings. Inside that one name you've got base Corollas with cloth seats and a four-cylinder, and Land Cruiser GR Sports with air suspension and a price tag five or six times higher — same brand, wildly different cars, averaged into a single figure that insults both of them. Average BMW's listings and you're blending a used 3 Series daily driver with an X7 that costs as much as a house. The number that comes out the other end is mathematically real and practically meaningless.
Age does the same thing, and so does mileage, and so does region — a GCC-spec version of a model and a cold-climate European version of the "same" car can carry genuinely different equipment and condition profiles before you even get to price. Average across all of that and the result isn't a midpoint. It's noise wearing the costume of a statistic.
What we compare a listing against instead
So we don't. When we decide whether a listing is underpriced, we compare it against its own cohort — similar age, similar mileage, similar trim and region, not the model's entire population smeared into one line. A base-trim five-year-old car gets judged against other base-trim five-year-old cars. A top-trim one-year-old car gets judged against its own kind. That's the whole trick, and it's not a clever one — it's just more work than printing "average price: $18,400" at the top of a search results page.
It's also why our deal rate holds around 1% instead of spiking every time a cheap base model shows up next to a stack of loaded ones. A stripped manual-transmission commuter car sitting below the nameplate's average isn't a deal. It's just a stripped manual-transmission commuter car, priced about right for what it is.
I learned this the annoying way years ago, chasing a Land Cruiser listed well under what every site was calling the "average price" for the model, convinced I'd found something the market missed. Drove out, looked at it, and it was a rental return with a salvage-adjacent service history and two missing option packages the average had no way of knowing about. The average wasn't lying, exactly. It just wasn't describing that car, or really any car.
Where the average still earns its keep
None of this means averages are worthless, and the people who lean on them aren't being lazy — they're usually being realistic about what they have time for. A model-wide average is a fine gut-check in the first five minutes of shopping, a way to tell "this market is roughly $15,000 cars" from "this market is roughly $60,000 cars" before you've narrowed anything down. For someone who just wants a sanity floor and ceiling before test-driving three cars on a Saturday, it does its job.
And cohort pricing has its own limit: it needs enough listings in a cohort to mean anything. In a region or trim combination that's thin — a handful of listings instead of hundreds — a cohort average is really just a small-sample average wearing a more precise name, and the discipline that makes it useful at scale stops paying off once the scale disappears. The fix for that isn't a better formula. It's more data, which is the one thing you can't shortcut your way into.
