If you've built a regional installed-base number before, you've probably done it with a survival curve. Take ten or fifteen years of shipment data, apply a retirement assumption, and the curve spits out how many units from each shipment year are still presumably in the field. It's the standard approach because it only needs data most OEMs already have: unit sales by year and model.
The newer alternative doesn't touch shipment history at all. It counts machines that show up in high-resolution imagery at sites across a region, then scales that count up to account for what the imagery can't see, equipment under roof, in motion, obscured by other units parked close together. Two very different ways of arriving at the same denominator, and they can land in noticeably different places.
What the survival curve is actually betting on
A shipment-based attrition model is a bet on a retirement curve holding steady. You're assuming the average service life of a given machine class hasn't shifted much, that regional attrition tracks the national average, and that nothing in the last few years (a parts shortage, a rough commodity cycle, a rental-fleet buying spree) has pulled the retirement rate off its historical path.
That assumption usually holds well enough at a national level over a long horizon, which is exactly why survival curves are the default for ten-year fleet-age distributions. The trouble shows up at the regional level and on anything shorter than a full cycle. A basin that saw a surge of used-equipment imports two years ago, or a market where a competitor pushed an aggressive trade-in program, will have an attrition rate that looks nothing like the curve predicts. The model has no way to see that. It's extrapolating from sales, not observing what's parked on site today.
What a site count actually measures
A site-level count sidesteps the retirement-curve assumption entirely, because it isn't modeling how many units should still be alive. It's counting what's there right now, at quarries, yards, job sites, whatever the equipment class calls home, and then correcting for coverage gaps and occlusion to turn a sample of sites into a regional figure.
That gets you a number anchored to the present instead of a projection rolled forward from a sales ledger. The tradeoff is that this is a newer approach to the use case: the scaling method that turns site samples into a region-wide figure is still being proven out across different equipment classes and terrains. We're opening it up on a request-access basis so it gets tested against real analyst workflows, rather than publishing it as an index with quarters you could already pull.
Which one actually answers your question
They answer slightly different questions, and that's worth settling before you pick one. A survival curve estimates how many units from past shipments probably haven't retired yet. A site count estimates how many units are physically present in the region today. For a long-run, national-level installed base figure where you're comfortable with historical retirement assumptions, the survival curve is a reasonable, well-understood tool. For a quarterly, regional market-share denominator, especially in a market with unusual churn, a used-equipment inflow, or a competitor's fleet moving in and out, a site count gives you something the curve structurally can't: a read on what's actually sitting there this quarter, not what the sales ledger implies should be.
If your current denominator comes from a dealer survey or a shipment-based model and you want a second figure to check it against, our approach starts from imagery of the sites themselves rather than from historical sales assumptions. Request access if you want to see how that comparison holds up for your region.