The number everyone quotes for this spans 14x, because almost nobody has measured it
Ask about revenue per commercial building for the contractor who services it and you get a dollars-per-square-foot figure somewhere between $0.15 and $2.15. We hold real per-building revenue, so we measured it directly. An office building is worth $21,088 a year to one mechanical contractor, across 65 contractors and 2,167 buildings.
Annual revenue per building, by building type
The unweighted median of the 12 published segment medians is $24,966 per year for one contractor. It includes residential types, so it is not a typical commercial-building figure. Building type is a useful first grouping: the industrial-to-restaurant commercial comparison spans 3.4x, while floor area explains about 9% of the variance.
Presented in tiers rather than as a ranking of one through 12. Most adjacent steps do not separate at 95% confidence, so an ordered league table would claim precision the data does not support. The retained industrial-to-restaurant commercial comparison is about 3.4x.
| Building type | Median contractor / yr | Average contractor / yr | Contractors | Buildings | Tier |
|---|---|---|---|---|---|
| Industrial / manufacturing | $43,325 | $56,817 | 46 | 1,129 | Highest value per building |
| Hospitality | $42,333 | $56,312 | 16 | 211 | Highest value per building |
| Education | $29,120 | $35,037 | 52 | 1,542 | High value per building |
| Multifamily | $26,356 | $33,840 | 40 | 1,945 | High value per building |
| Government | $25,653 | $32,665 | 20 | 410 | High value per building |
| Worship | $25,370 | $33,730 | 12 | 217 | High value per building |
| Mixed use | $24,562 | $25,814 | 22 | 562 | High value per building |
| Office | $21,088 | $31,271 | 65 | 2,167 | High value per building |
| Healthcare | $20,914 | $36,782 | 57 | 1,339 | High value per building |
| Warehouse / distribution | $13,012 | $15,585 | 10 | 166 | Mid value per building |
| Restaurant | $12,645 | $16,487 | 21 | 505 | Lower value per building |
| Single-family residential | $6,573 | $8,708 | 34 | 2,013 | Lower value per building |
The headline is the MEDIAN of the contributing contractors, which is the right figure for judging what one relationship is likely to pay. The average across contractors is shown beside it and is higher in every single segment, by up to 1.76x, because a handful of firms hold genuinely enormous accounts. That is not an error and we are not hiding it: the equal-contractor-weighted mean shows how large accounts pull the average above the median. A territory value needs actual building counts and mix, not this unweighted mean. We lead with the median for one relationship.
We deliberately do not print a plus-or-minus percentage next to these. The standard error we can compute belongs to the average, not the median, and it understates true uncertainty in any case: splitting the contributing contractors into two independent halves moves individual cell values considerably more than the error bar predicts, because one contractor's whole book of a segment shares his pricing and market rather than being an independent draw. What DOES replicate strongly across those halves is the rank ordering. So read the tier and the ordering as the finding, and any single figure as approximate.
Three findings that are more useful than the table
These findings show where added detail helps and where it does not. We publish them because they are the parts nobody else has measured, and because each one is an enrichment expense a contractor or a software vendor does not have to pay for.
1. Square footage explains about 9% of it, and $/sqft makes the estimate worse
Floor area against annual revenue gives an r-squared of 0.093. That is not zero, and size does order buildings within a type: the largest third of offices bill 2.64x the smallest third, and multifamily runs 4.56x in the retained size comparison (the small cohort has five contractors, so that adjustment is withheld). But converting revenue to a per-square-foot rate ADDS roughly 19% more variance than it removes, because the denominator carries its own error and the relationship is not proportional. A size band works. A size rate does not. We publish measured gradients for 2 of 12 segments and withhold the rest, including restaurants, where the gradient is not even monotonic: the kitchen sets the mechanical load, not the dining room.
2. Office replacement work varies by age; other segments need their own evidence
This is the finding a total-revenue analysis cannot see, and it is why the conventional wisdom that age does not matter survives. Split the revenue and the two halves behave differently. Routine service in offices runs $9,690 in newer buildings and $8,921 in buildings 30 or more years old, a ratio of 0.92. Replacement work over the same split runs $8,970 to $15,646. Healthcare moves $11,020 to $11,882 on the same line. The office routine-service ratio is 0.92, while its replacement ratio is 1.74.
A separate public measurement provides context for age and maintenance. The Public Buildings Reform Board's March 2026 report on 1,958 federal buildings totalling 280 million gross square feet found deferred-maintenance liability stepping from $55 per gross square foot in buildings 1 to 20 years old to $130 in buildings 31 to 75 years old, and named building age the greatest single correlation factor with condition. That is roughly 2.4x deferred-maintenance liability, a different metric and population from Level's office replacement-revenue ratio. Level's published age bands do not establish a plateau after 76 years. The public comparison provides context; it does not validate Level's dollar coefficient.
The mechanism is staggered multi-system renewal, not a single replacement cycle. A facilities manager knows his 1968 building has new rooftop units, so "equipment fails at N years" is wrong on its face. ASHRAE's own service-life medians span 11 to 37 years across eight system families, so past 30 years the annual odds that something is due may rise. This is a mechanism to test, not an observed plateau; our two published age bands cannot establish where the curve flattens. Hospitality runs the other way in our data (hospitality at a ratio of 0.37). We ship hospitality with no age adjustment rather than assume a sign we cannot explain.
3. Multifamily and single-family are different businesses, and blending them is a several-fold error
Most property taxonomies fold apartments into "residential". Measured separately, multifamily bills $26,356 per building per year against single-family at roughly a third of that. A contractor quoting a 200-unit apartment complex off a blended residential figure would be several times low, and that is exactly the customer who notices. Our own client books, a different population on a different accounting system, independently flagged the same error: their "residential" segment read like commercial because it was actually multifamily.
Single-family is published through the standard validation lane at $6,573. Its published influence is below the standard ceiling; the cell covers 34 contractors and 2,013 properties, and its held-out result remained inside the fit band. The calculator therefore uses the median and shows the wide contractor range, with no age or size adjustment.
What the public housing data supports
The Census Bureau counted 85.261 million occupied single-family detached homes in the 2023 American Housing Survey, and the Energy Information Administration found that 90% of detached homes used air-conditioning equipment in its 2020 Residential Energy Consumption Survey. Those sources establish the scale and equipment exposure of the housing stock. They do not produce Level's dollar coefficient, which comes only from the anonymized active-account subset above.
Recurring versus replacement, per building per year
These are separate slices of revenue, not two halves of a whole. Jobs of $15,000 or more are counted as replacement, jobs under $5,000 as routine service, and work between the two sits in neither, so the columns do not add to the total. That gap is deliberate: a low-end 5-ton changeout and a compressor swap land in it, and forcing them into one bucket or the other would make the split look exhaustive when it is not.
| Building type | Routine service, newer buildings | Replacement, newer | Replacement, 30+ yrs | Age adjustment |
|---|---|---|---|---|
| Industrial / manufacturing | $12,757 | $26,339 | $29,252 | Withheld, not distinguishable from flat |
| Hospitality | $18,162 | $8,752 | $3,248 | Withheld, runs backwards (0.37x) |
| Education | $11,575 | $0 | $4,201 | Withheld, not distinguishable from flat |
| Multifamily | $15,302 | $10,643 | $26,537 | Withheld, not distinguishable from flat |
| Government | $9,315 | $15,588 | $29,083 | Withheld, not distinguishable from flat |
| Worship | $11,253 | $6,344 | $5,930 | Withheld, runs backwards (0.93x) |
| Mixed use | $14,968 | $11,631 | $8,600 | Withheld, runs backwards (0.74x) |
| Office | $9,690 | $8,970 | $15,646 | Ships, 1.74x |
| Healthcare | $12,156 | $11,020 | $11,882 | Withheld, not distinguishable from flat |
| Warehouse / distribution | $5,318 | $1,371 | $1,025 | Withheld, runs backwards (0.75x) |
| Restaurant | $11,350 | $6,141 | $3,390 | Withheld, runs backwards (0.55x) |
| Single-family residential | $5,176 | $3,256 | $2,727 | Withheld, runs backwards (0.84x) |
An age adjustment ships only where the ratio exceeds 1.2, both cohorts carry ten or more contributing contractors, and the standard error is under half the mean. That last test is what withholds government, which reads a 1.87x ratio on a standard error that is more than half its own value. Government's observed comparison is shown for context, but it is too imprecise to support an age adjustment, so the calculator does not apply one.
Tested against buildings the numbers were not built onHeld-out validation: 11 of the 12 published segments have a shown held-out result; worship has no published held-out result.
Eleven of the 12 published segments have shown held-out results from buildings the fit did not see, agreeing within their bands. Worship has no published held-out result. The relative errors are the honest headline, since wide bands make "agrees" an easy test to pass.
| Building type | Fit on training set | Held-out result | Held-out buildings | Difference |
|---|---|---|---|---|
| Industrial / manufacturing | $36,292 | $29,818 | 205 | -17.8% |
| Hospitality | $27,772 | $49,264 | 32 | +77.4% |
| Education | $29,321 | $25,661 | 300 | -12.5% |
| Multifamily | $22,744 | $22,926 | 388 | +0.8% |
| Government | $23,018 | $19,597 | 82 | -14.9% |
| Mixed use | $21,419 | $18,276 | 93 | -14.7% |
| Office | $20,188 | $17,719 | 400 | -12.2% |
| Healthcare | $22,158 | $24,155 | 259 | +9% |
| Warehouse / distribution | $13,090 | $4,662 | 26 | -64.4% |
| Restaurant | $12,604 | $14,463 | 82 | +14.7% |
| Single-family residential | $6,489 | $5,551 | 373 | -14.5% |
What we measured and did not publishThe cells that failed our own floors, and why. 4 segments were withheld rather than published thin.
We have numbers for these. They failed our own concentration test, which asks how far the published average moves if the single most influential contractor is removed. Above 20% the figure is one firm's book wearing an error bar, so it is named here rather than left silently out of the table.
Grocery / convenience store
Only 4 contractors (floor 10).
Data center
Only 6 contractors (floor 10). Removing one contractor moves the average across contractors by 41.5%, over our 20% single-firm influence ceiling.
Recreation / entertainment
Only 9 contractors (floor 10).
Retail
Removing one contractor moves the average across contractors by 31.7%, over our 20% single-firm influence ceiling.
A note on why this test replaced the more obvious one. A cell's top contractor can hold a large share of its dollars without moving the published figure at all, because every contractor contributes exactly one average regardless of size. Mixed-use is the clean example: its largest contractor holds 31.6% of that cell's dollars but 4.5% of the weight on the mean, and dropping it moves the published figure by 10.4%. Gating on dollar share would have rejected a healthy cell and told us nothing about the ones that are actually fragile.
How to reproduce thisThe unit of measurement, the window, the weighting and the exclusions, so readers can check published aggregates and calculations; private distributions cannot be recomputed from the download.
The unit
One HVAC or mechanical contractor's billed revenue from one building over a year. Not the building's total spend, and not a company average.
The window
Building must show at least 183 days between its first and last completed job; revenue is annualized on the actual window, not a calendar year.
The weighting
Contractor-weighted. Each contractor holding at least 5 buildings of a segment contributes one average, so no single firm can carry a cell. The published headline is the MEDIAN of those contractor averages, not their mean: the mean is pulled up by a right tail of outlier firms in every segment, and a contractor checking the figure against his own book would recognise the median and reject the mean. The mean is published alongside, labelled, because it is the right figure for valuing a whole territory of buildings.
Exclusions
- Non-operating (sandbox/training) tenants excluded by name.
- Contractors not conservatively classified as HVAC/mechanical are excluded. Conflicting high-confidence name and serviced-asset labels are withheld.
- Records averaging under $300 per job excluded as parts-counter sales.
- Invoices collapsed to one row per job before joining, because one job can carry hundreds of invoices.
Segments
Contractor-entered building type, normalized with an ordered regex taxonomy. Multifamily is tested before single-family residential so apartment complexes are not folded into houses.
Publication floors
Standard publication requires at least 10 contributing contractors, at least 25 buildings, and less than 20% leave-one-contractor-out movement in the mean across contractors. A large-sample lane also publishes a median when there are at least 50 contractors and 1000 buildings, the held-out result remains inside the fit band, and mean influence is below 25%.
What the figure includes
The figure pools relationships that persisted and relationships that ended. Buildings whose relationship ended before the extract bill 36% less on the mean, so this is the expectation for an average new account including churn, not what a good account is worth. Filtering to currently-active buildings inflates every segment by roughly a third.
Machine-readable version, with per-cell standard error, contractor and building counts, held-out results, the service and replacement split, and the size and age gradients: building-revenue-by-segment.json, version 1.0.5, CC BY 4.0.
Frequently asked questions
How much revenue does one commercial building generate for a contractor?
For HVAC and mechanical work, the MEDIAN contractor bills about $21,088 a year from an office building, measured across 65 contractors and 2,167 buildings. Within the commercial comparison used here, industrial buildings are the highest type at about $43,325, restaurants the lowest at about $12,645, a spread of roughly 3.4x. We report the median rather than the average on purpose: the average across contractors is 1.48x higher for offices, because a handful of firms in every segment hold genuinely enormous accounts and drag it above anything a normal contractor would recognise. Every figure is what ONE contractor bills, not the building's total mechanical spend across all vendors.
How much is an active single-family HVAC account worth per year?
The median contractor in Level's data bills $6,573 per single-family property per year, with the middle half of contractor averages running from $4,386 to $9,863. This is measured across 34 contractors and 2,013 properties. It is an active-account benchmark: every included property shows at least 183 days between its first and last completed job. It is not average HVAC spending across every U.S. house and must not be multiplied by the full housing stock.
Why does the industry quote dollars per square foot instead?
Because almost nobody holds real per-building revenue to calibrate against, and $/sqft is computable from a floor plan. The consequence is visible in the published figures: they span 14x, roughly $0.15 to $2.15 per square foot per year. We measured why the convention fails. Floor area explains only about 9% of the variance in one contractor's annual billed revenue from a building (r-squared 0.093), and dividing revenue by area ADDS about 19% more variance than it removes, so the per-square-foot form is actively worse than the raw number. Size does matter within a building type, and we publish those gradients, but as a band rather than as a rate.
Does building age change what a building is worth to service?
Yes, on the replacement half of the revenue only. Replacement work (jobs of $15,000 or more) rises materially with age in the segments that clear our publication gates, while routine service (jobs under $5,000) does not show a reliable age relationship. In offices the routine-service line moves from $9,690 to $8,921, a ratio of 0.92, while the replacement line moves from $8,970 to $15,646. The effect ships in 1 of 12 segments. It runs BACKWARDS in 6, including hospitality, so we publish those with no age adjustment rather than assume the sign.
Is this the building's total spend or one contractor's revenue?
One contractor's revenue, and the distinction is not a technicality. We tested whether total cross-vendor spend was recoverable and it is not: no two contractors in the data serve the same building, so there is no overlap to measure a building's full mechanical budget from. Read these as the revenue available to you from that building. The building's total is larger by an amount we cannot quantify, and anyone claiming to know it from data like this is guessing.
Does this apply to plumbing, electrical, or roofing?
No. The ordering of building types by value is measurably trade-specific, so carrying these numbers into another trade would be wrong in a way a practitioner would spot immediately. Against this HVAC ordering, electrical rank-correlates at rho 0.58 and refrigeration at only 0.25. Roofing follows roof area and replacement timing rather than mechanical load. The office HVAC replacement comparison is about 1.74x for buildings aged 30 years or more in the retained model. This is not a universal multiplier for every building segment or another trade. These coefficients are HVAC and mechanical only.
How was this measured?
Building must show at least 183 days between its first and last completed job; revenue is annualized on the actual window, not a calendar year. Contractor-weighted. Each contractor holding at least 5 buildings of a segment contributes one average, so no single firm can carry a cell. The published headline is the MEDIAN of those contractor averages, not their mean: the mean is pulled up by a right tail of outlier firms in every segment, and a contractor checking the figure against his own book would recognise the median and reject the mean. The mean is published alongside, labelled, because it is the right figure for valuing a whole territory of buildings. Non-operating (sandbox/training) tenants excluded by name. Contractors not conservatively classified as HVAC/mechanical are excluded. Conflicting high-confidence name and serviced-asset labels are withheld. Records averaging under $300 per job excluded as parts-counter sales. Invoices collapsed to one row per job before joining, because one job can carry hundreds of invoices. Standard publication requires at least 10 contributing contractors, at least 25 buildings, and less than 20% leave-one-contractor-out movement in the mean across contractors. A large-sample lane also publishes a median when there are at least 50 contractors and 1000 buildings, the held-out result remains inside the fit band, and mean influence is below 25%.
Does this include buildings the contractor eventually lost?
Yes, deliberately, and it matters more than it sounds. Buildings whose contractor relationship ended before our extract bill about 36% less on the mean and 44% less on the median than ones still running. Both groups are in these figures, so this is what an average new account is worth including the ones that will churn. Filtering to currently-active buildings is a tempting, clean-looking filter that would inflate every number on this page by roughly a third, and it would describe an expectation no contractor actually experiences.
Run this against your own building types
Pick a building type, size and age and get the band, split into recurring and replacement, with the arithmetic behind it.
Open the calculator →Analysis by Sam Yang, Founder, Level. Compiled from Level's proprietary contractor benchmark research.
Source: The Level Index, Annual Contractor Billed Revenue Per Building by Segment (Residential and Commercial), 12,206 buildings, 12 published segments. US contractors. Data vintage 2022 to 2026, last updated Q3 2026. Aggregated and anonymized.
Separate context on building age (different metric and population): Public Buildings Reform Board, The Cost of Inaction, Deferred Maintenance in GSA's Portfolio, March 2026, 1,958 buildings, 280 million gross square feet. Separate unpublished internal comparison of office and retail levels from contractor books on a different accounting system; it is not independent public validation.
Suggested citation: "The Level Index, Annual Contractor Billed Revenue Per Building by Segment (Residential and Commercial), Level (levelcfo.com), 2026." Method: see the benchmark methodology.
Operating review
Questions before adding a role
The margin, cash, and operating questions we check first when reviewing your numbers.
Does backlog require another person or a better handoff?
Booked work, available crew hours and invoice-ready work are different capacity measures.
What work would a supervisor release from the owner or crew?
A title and salary do not establish what capacity a new role creates.
Is the back office missing a bookkeeper, controller or decision owner?
Transaction processing, close review and operating decisions are separate responsibilities.
Simple pricing
Three tiers, one ladder.
$500+/mo
Bookkeeping
The clean data layer: monthly books, reconciliations, and organized financials AI can work with.
$1,500-$5,000/mo
Scale
The full AI operating layer: custom agents, weekly actions, and benchmarks to grow margin per hour.
Custom
Platform / Multi-Office
Multi-branch benchmarking and scorecards for PE-backed and multi-location groups.
Know which buildings are worth winning
Level rebuilds profit by job and by customer from your own books, so the work you chase is the work that pays.
No commitment. Real numbers, not generic advice.