The only test the building ever gets

Nearly every assurance instrument a capital project carries works by reading documents. An audit reads records. A gate review reads deliverables. A completion certificate confirms that other certificates exist. One instrument in the whole lifecycle switches the systems on, loads them, cuts the power to see what happens, and writes down what was measured rather than what was submitted.

That instrument is commissioning, and it is expensive. It puts specialist engineers on site for months, adds a party to the project whose job is to slow the handover down, and shows up as a visible line in a budget that is already under pressure at exactly the moment it is incurred. So it faces, and should face, the question this publication asks of every requirement: somebody is paying for this, what does it buy, and how would anybody know.

Commissioning is unusual among the requirements we test here, because the answer exists, in public, with a dataset attached.

What the word means when it is used properly

The word commissioning gets used for two different activities, and the price difference between them is the subject of this article.

The narrow activity is testing at the end: functional tests, balancing and certificates in the final weeks of construction, run by the contractor's own team, under the contractor's programme, while everyone involved is being measured on the handover date. The commissioning bodies call this traditional testing and commissioning, and it was standard practice for decades.

The wider activity is a process. ASHRAE Guideline 0 and Standard 202, NFPA 3 for fire and life safety systems, NFPA 4 for integrated systems testing, and the LEED rating system all define commissioning as a quality process that starts when the project starts. The owner's requirements are written down early, in testable values rather than adjectives. The design is reviewed against them. The commissioning requirements go into the construction documents, so bidders price them instead of discovering them. Then come the tests, and after occupancy the process continues: deferred seasonal testing, and a review of building operation months into use.

The difference between the two is not thoroughness at the end. The final-weeks tests exist in both. The difference is everything before and after: whether anyone wrote down what the building was supposed to do while it could still be changed, and whether anyone comes back to check once it is full of people.

Nine activities of the commissioning process as ASHRAE, NFPA and LEED define it, against what traditional testing and commissioning covers. The three shared rows are the tests at the end of construction. The process model adds four rows before them, while the design can still change, and two after occupancy, when the building can finally be measured in use.

Activity	Process commissioning	Traditional T&C
Owner’s requirements written as testable values, at planning
Design reviewed against those requirements
Commissioning scope priced into the construction contract
Commissioning plan and test procedures written before testing
Installation checks, factory and site acceptance tests
Functional and integrated systems tests
Full power-failure test with the building under load
Deferred seasonal testing after occupancy
Return review of operation, roughly 10 months in

Both columns run the final tests, so both produce certificates. What the dearer column buys is the six rows the cheaper one lacks, and the LBNL payback data in the article prices what those rows return.

The requirement with a price tag: independence

The costly clause in the process version is not the extra testing. It is the extra party. The standards require a commissioning authority who is independent of the delivery team and reports to the owner, and an independent specialist engaged from design development to a year after occupancy does not come cheap.

The reasoning behind the clause is published, and it is an incentive argument rather than a technical one. A commissioning agent inside the contractor's organisation reports to the party being measured on the handover date. When testing late in construction reveals that the wrong equipment was selected months earlier, the choice is between reopening a closed decision, which costs the deadline, and a workaround, which costs the operator for the life of the asset. A tester who answers to the deadline picks the workaround. The whole point of moving the reporting line to the owner is to make the other choice available.

One national manual built on these standards, the Saudi programme authority's testing and commissioning guideline, states it plainly: the traditional approach began in the last part of construction under the control and influence of the main contractor, owners' requirements documents rarely existed, and late-discovered selection errors were resolved through compromises because the project team's usual goal is the delivery deadline.

What is easy to miss is that the standards also scale this requirement, in writing. LEED v4 priced independence in tiers. For the baseline prerequisite, the commissioning authority may be an employee of the design or construction firm, provided they are not part of the project's own team and report directly to the owner. Only the enhanced commissioning credit, which is optional and scores points, requires full independence from both firms. And for projects under 20,000 square feet, a qualified member of the project team itself may do the job. A small building does not carry a specialist's fee, and the standard said so itself rather than leaving proportionality to the reader's discretion.

LEED v5, released in 2025, has moved. Its prerequisite no longer names who may employ the provider and points instead to ASHRAE 90.1's commissioning requirements without their small-project exceptions, so the allowance for small buildings is gone and the independence question for a firm employee now rests on a separate standard. Its enhanced credit requires an independent provider designated in predesign. The arrangement check computes both editions and says where v5's text stops deciding.

What the money buys, measured

For most assurance requirements, the honest answer to "what does it buy" is that nobody has measured. Commissioning is the exception, because commissioning produces measurements by nature, and because Lawrence Berkeley National Laboratory has spent two decades collecting them.

The 2009 meta-analysis covered 643 buildings across 99m square feet. Median commissioning cost was $1.16 per square foot for new construction, which was 0.4% of total construction cost, and $0.30 per square foot for existing buildings. Median whole-building energy savings were 13% for new construction and 16% for existing buildings, with median simple paybacks of 4.2 years and 1.1 years. The commissioning work across the sample surfaced over 10,000 energy-related deficiencies, in buildings that had either been accepted as complete or were being operated as if nothing was wrong.

Then the sample grew, and the numbers shrank. The 2020 update, drawing on roughly 1,500 buildings across three decades, found median energy savings for existing-building commissioning of 6.4% across all data and 6.0% in the newer 2018 cohort, where the median cost was $0.25 per square foot in 2017 dollars and the median payback 2.2 years. About two fifths of the earlier headline.

The caveats belong next to the numbers, and the study states them itself. The newer cohort is dominated by utility-sponsored programmes, which show systematically lower savings than owner-initiated projects. Projects that report data at all tend to be the better-run ones, so the sample is not representative of the building stock, and the bias runs in both directions at once: good projects are more likely to report, and the buildings that get commissioned at all may be the ones in most need of it.

Note what this dataset does and does not cover. It measures energy, because energy is what a meter can see. The deficiencies commissioning finds in fire systems, controls and access do not show up in a kilowatt figure, and no equivalent dataset prices them. So the published payback is a floor, computed from the one benefit that happens to be measurable, which is an unusual position for an assurance cost to be in. Most assurance requirements cannot show you any floor at all.

Where the instrument stops

Everything above concerns the asset, and the discipline is real: the power gets cut with the building under load, the fire systems get tested as an integrated whole, and the authority having jurisdiction repeats the critical tests before an occupancy permit is issued. By the day of taking over, the machine has been proven in the strongest sense the industry knows how to apply.

The organisation receiving it has not. Commissioning tests whether the building works, not whether the operator can work it, and no equivalent instrument, with an equivalent budget and an equivalent signature, exists on the other side of the boundary. We have written about that asymmetry separately, and about what an organisation-side readiness test would contain, so here we only mark the edge of the instrument's reach.

Why nothing tests the receiving organisation

, read: Completion describes the asset. Readiness describes the owner.Completion describes the asset. Readiness describes the owner.

What an organisation-side test would have to cover

, read: Operational readiness has seven conditions. Handover delivers two.Operational readiness has seven conditions. Handover delivers two.

What to ask

• Ask which commissioning model the project has actually bought. If commissioning first appears in the programme during construction, it is the traditional model regardless of what the specification calls it, and the design was never reviewed against testable requirements.
• Ask who the commissioning authority reports to. A reporting line into the contractor is not a variant of the process model. It is the thing the process model was created to remove.
• Ask for the owner's project requirements in the form the tests will reference. If they do not exist, the tests will verify the design against itself.
• Ask what happens after occupancy: whether seasonal deferred testing is in the contract and whether a return review is scheduled roughly 10 months in. These are the cheapest rows in the whole process, and their absence is the commonest saving.
• If the fee is challenged, put the LBNL numbers on the table with their caveats attached, and ask the challenger which other assurance line item on the project can produce any measured payback at all.

---

Sources. Lawrence Berkeley National Laboratory, Building Commissioning: A Golden Opportunity for Reducing Energy Costs and Greenhouse-gas Emissions, 2009, for the 643-building sample, the $1.16 and $0.30 per square foot medians, the 0.4% of construction cost, the 13% and 16% savings medians and the 4.2 and 1.1 year paybacks. Crowe, Mills, Poeling, Curtin, Bjørnskov, Fischer and Granderson, Building Commissioning Costs and Savings Across Three Decades and 1,500 North American Buildings, LBNL, 2020, published in Energy and Buildings 227, for the 6.4% and 6% savings medians, the $0.25 per square foot 2018-cohort median cost in 2017 dollars, the 2.2 year payback and the stated limitations of the sample. USGBC, LEED v4 BD+C, EA Prerequisite Fundamental Commissioning and Verification, for the commissioning authority qualifications, the 20,000 square foot threshold and the reporting requirement, and the enhanced commissioning credit for the stricter independence tier. USGBC, LEED v5 Building Design and Construction rating system, November 2025 edition, EAp3 and EAc5, read in full on 5 September 2026, for what v5 keeps, drops and defers. ASHRAE Guideline 0 and Standard 202, and NFPA 3 and NFPA 4, cited as the defining process standards. Two of those four are free to read, which this article got wrong until 4 September 2026: NFPA gives free online access to NFPA 3 and NFPA 4 in full, subject to a free account, and read-only, with no download, printing or copying. The ASHRAE pair are not. Guideline 0 is sold outright and absent from ASHRAE's preview list. Standard 202-2024 is listed for preview, but the preview link is gated to referrals from ashrae.org and leads to a purchase page rather than the text. Free to read is not the same as free to use: NFPA's free licence permits viewing and expressly excludes creating derivative works. A stronger claim about ASHRAE and AI stood here between 4 and 5 September 2026 and was withdrawn. It was taken from a banner on ashrae.org rather than from the policy that banner refers to. ASHRAE's Policy for Use of Artificial Intelligence, approved by its board on 9 February 2025, applies by its own scope clause to ASHRAE's employees, contractors, volunteers and third-party partners, and its prohibited-uses section covers documents not intended for public dissemination. It is internal governance, and it does not bind a reader. EXPRO, National Manual for Projects Management, Volume 10 Chapter 2, testing and commissioning guideline, EPM-KT0-GL-000003, for the stated rationale for the independent commissioning provider and the handover test sequence.