Field notes: the quarter our regional total didn't match the board

Tractors and farm machinery parked together inside a large indoor storage building.

We'd run the same dealer-survey rollup for six quarters straight. Same regional splits, same attrition assumption carried forward from the fleet-age curve, same footnote about telematics penetration being too thin in two of the five territories to trust the raw counts. Nobody in the room had ever pushed back on the number before. This quarter they did.

The board deck had a slide, built by someone in corporate development, estimating total addressable units in the Southeast from shipment data and a generic replacement-cycle assumption. Their number was 22 percent higher than ours. That gap was big enough that our market share figure moved from a story about steady gains to a story about losing ground, depending on which denominator you believed.

I spent most of that week doing what every analyst in this seat eventually does: defending a number instead of building the next one. The uncomfortable part was that I couldn't prove our estimate was right. The dealer survey gives you a count of units a dealer network is aware of, which quietly excludes machines bought through a competing channel, machines moved in from another region, and anything still running past the age where a service department stops caring about it. All three of those failure modes push the true number up, which is exactly the direction the board's slide happened to lean.

So the real fight in that meeting came down to which blind spot was smaller. Shipment-based estimates assume a replacement cycle that holds steady across a region, which it rarely does once you've got pockets of extended-life equipment in low-utilization counties sitting next to high-turnover fleets near the ports. Dealer-survey estimates undercount anything that moves outside the branded service network. Neither method sees the actual metal sitting in a yard.

What finally settled the room was pulling up a handful of counties and counting machines visible in current imagery against what each model predicted for that same footprint. The shipment-based number ran high in counties with almost no active fleet. The survey number ran low in a cluster of counties where a regional rental outfit had clearly been buying secondhand units outside any dealer's radar. Neither model had it right, but you could finally see why each one was wrong, which is a very different conversation than arguing about whose assumption was more defensible on paper.

That's the part worth writing down for next quarter, and the quarter after. A denominator built entirely on channel data will always inherit the channel's blind spots, no matter how many years of history back it up. The fix is a count that starts from what's actually sitting on the ground and then corrects for what a single satellite pass can't see, like a canopy-covered yard or a machine parked behind a building, rather than another pass at survey cadence or replacement-cycle assumptions. That's the approach behind a site-level fleet count scaled up to a regional total, and it's the only version of this argument I've found that doesn't depend on trusting someone else's assumptions.

I don't expect the board to stop asking questions about the number. They shouldn't. Next quarter I'd rather walk in with a count I can defend down to the county. If your denominator gets challenged the way ours did, it might be worth asking what a ground-level count would show for your region.

Request early access

← Back to the blog