| Customer | St | Models | Sq ft | Ht | Htr | Duct | PMR | Cost | Source |
|---|
| Used now | Source | Year | Customer | St | Loc | Models | Type | Sq ft | Ht | Htr | Duct | PMR | Install cost | Model says | Time frame quoted |
|---|
Installer PO rows are what the crew was actually paid. Quote rows are the quoted install line divided by 1.35, which the Job Cost/Profit sheets confirm was the markup on nearly every job. Model says = what this estimator predicts for that job, so you can see where the formula runs hot or cold.
Three sources, all parsed from the files on the G drive. First, every QuickBooks estimate PDF in Bidding Projects and Customer Projects since Jan 2025 that carries a Mechanical and Electrical Assembly line. Second, the 2023 and 2024 closed jobs, which use the older QuickBooks Proposal and Sales Order template and needed a separate parser. Third, and most important, the purchase orders Marathon issued to its installers, which are what the crew was actually paid rather than a quote divided by 1.35.
That comes to 238 jobs spanning 2023 to 2026. For quote rows, cost is the install line divided by 1.35, which the Job Cost/Profit sheets confirm was the markup on nearly every job. For PO rows, cost is the purchase order.
cost = (mobilisation + rate × sqft^pow + heater + mixing rooms + duct runs) × site factors
The add-ons are flat dollars, not percentages, and that change came from your own budgeting rules. A heater install costs about the same on a 364 sq ft booth as on an 864 sq ft one, and the paired quotes prove it: Jack's Auto, Lio's and Toy's 4 Trucks all came in at exactly +$7,500 for the heater regardless of booth size. The old multiplicative version scaled those with booth area, which was wrong.
The fit lands within 30% on about 91% of jobs. The range next to the estimate is one standard deviation around the midpoint.
Size is close to linear: mobilisation plus roughly $17 per square foot of booth floor.
Heater is a flat add. Paint mixing rooms and duct runs the same, each a flat figure per unit.
Height is a step, not a slope. Booths 14 ft and taller cost about 32% more to set, for lifts and scaffolding. Below 14 ft, height made no difference.
Outdoor rated (ORF/OSD) is the biggest single factor at about +71%.
Location: SoCal jobs run about 28% cheaper than travel jobs. Beyond that, distance did not matter. A Michigan job and an Arizona job for the same booth were quoted the same.
Booth type: once size, heat, ducts and outdoor rating are accounted for, side downdraft versus reverse flow versus modified downdraft made no measurable difference. If SD really takes the crew longer, it is not showing up in the numbers.
The formula alone was reading about $25,300 on a 14 by 27 heated side downdraft with two duct runs, when eighteen near-identical jobs in the book landed at $23,000 almost every time. A curve fit across 238 jobs of every shape will do that: it is balancing a blast booth in Alaska against a small open face in Temecula, and the most common job in the book gets averaged away.
So the estimator now looks for near-identical past jobs first. Same size within a few percent, same heat, same outdoor rating, same height class, same mixing room, same region, same duct count. When at least three turn up, their median is blended with the formula, weighted by how many there are and how tightly they agree. Eighteen jobs agreeing within 12 percent carries 58 percent of the answer; three scattered ones barely move it. Where nothing close exists, the formula carries it alone.
Tested leave-one-out across all 238 jobs, this cut the median error from 10.6 percent to 8.3 percent. On the standard side downdraft it now reads $23,057 at the midpoint, which is the number you have been quoting.
Past jobs are first put on a common footing before they are compared: escalated to the job year and adjusted for whether the figure came from a quote or an installer purchase order, so a 2023 PO and a 2026 quote can sit in the same median.
The model is unbiased, so at the midpoint about half of jobs land above it and half below. A margin on top is not about accuracy, it is about which direction you would rather be wrong in, and how big it needs to be depends on how much data stands behind a job of this shape.
The first thing I tested was the obvious one: does having lots of similar past jobs make the estimate better? On its own, barely. The correlation between comp count and how far off the model was came out near zero.
What does predict error is whether the combination of options has been done before. Booths whose mix of heat, outdoor rating, height, mixing room, multi-booth and dual skin appears five times or fewer miss by about 12 percent, against 6 percent for common combinations. Two other things widen it: the 400 to 700 sq ft band, which is the least predictable, and anything off the mainland.
The badge reports the typical miss for a job of this shape, and the margin is calibrated from it. Covering three jobs in four takes about +7 percent on a high-confidence job and +13 percent on a low-confidence one. Covering nine in ten takes +16 against +31. Those coverage numbers were solved against the actual residuals rather than assumed from a bell curve, so a setting that says three in four really did cover three in four.
One thing worth knowing: the margin does not grow with booth size. Percentage error actually shrinks as booths get bigger, from about ±23 percent under 200 sq ft to ±12 percent over 2,000. Big jobs have more moving parts but they get quoted more carefully.
The quotes print an Assembly Time Frame, and 179 of them parsed. Days track cost closely, which makes sense since both are driven by crew-days. The range shown is that fit plus or minus about a quarter, and it is calendar days with the crew working seven days a week, ten hours a day, the way the quotes state it.
Only 28 installer POs parsed cleanly, and a few jobs (Moz Design especially) came in far above model for reasons the booth spec does not explain. Quotes are quotes, not closeout costs; where the Internal Costing sheets had a post-install number it was usually within 10% of presale, but a few ran higher. Prevailing wage rests on two quotes and very large multi-booth systems have too few points to model, so use a crew-day buildup there. The single best thing that would improve this further is logging the installer PO on every job going forward, since those are the only rows here that are real cost.