
A regional team we were scoping last month had just finished their annual dealer survey. Weeks of calls, a spreadsheet passed around for sign-off, a number everyone felt reasonably good about. Then we ran the site-level count for the same geography off recent imagery and the two numbers didn't just differ, they told different stories about where the fleet actually sat.
It's close to the default outcome, and it's worth sitting with why.
Dealer surveys count intent and memory. A dealer reports what moved through the lot, what's on a service contract, what a sales rep remembers delivering to a job three counties over. That's useful information, but it drifts in predictable directions. Machines get counted at the dealer that sold them even after they've been resold, relocated, or scrapped. A unit working a job across the territory line gets claimed by two dealers in two different surveys, or by none, because nobody's sure whose column it belongs in. Add in the gap between "reported" and "verified," and a dealer number is really a number about dealer recordkeeping as much as it's a number about machines.
A pixel count has the opposite bias. It sees what's parked in the yard or staged on a site on the day the image was taken, nothing more and nothing less. A unit in a building, under a tarp, or tucked under tree cover on pass day doesn't get counted unless the correction work accounts for it. A unit that's been sold but hasn't moved yet, or one sitting idle at a rental yard between jobs, gets counted whether or not anyone would call it part of the "active" fleet. The image doesn't know what a machine is doing, only that it's there.
So when the two numbers land worlds apart, the gap isn't a sign that one method is broken and the other is right. It's the two methods measuring slightly different things and each carrying its own kind of error in opposite directions. The dealer number tends to run soft on verification and double-counts at the margins. The imagery-based number tends to run soft on occlusion and has no way of knowing a parked machine is dead inventory versus three weeks from a sale.
What's useful about seeing the gap is that it forces the real question into the open: what are you trying to measure a denominator for? If you're sizing a service and parts opportunity, you want physical presence, whether a machine is forgotten in a yard or working a job, because it can still break down and still needs parts. If you're sizing active market share you want something closer to units in use, and neither a dealer tally nor a raw count gets you there cleanly.
We've started treating the comparison itself as a diagnostic. When the two numbers are close, it tells you dealer records in that region are probably in decent shape and coverage holes in the imagery are minor. When they're far apart, it's worth asking which direction the gap runs and why, because that direction usually points at something concrete: a dealer territory overlap nobody's resolved, a stretch of canopy cover hiding a chunk of fleet, a region where resale channels move machines faster than anyone's updating a CRM.
None of this makes either number wrong on its own. It just means neither one should be the only input into a denominator you're going to defend in front of a strategy committee. If you're building that kind of regional figure and want to see how a site-level count holds up against what your dealer network is reporting, our installed fleet estimation approach is built for exactly that comparison.
Worth running the two side by side before you commit to either one as gospel.