The rule
HVAC marketing runs on recycled statistics. A number gets quoted, the citation falls off, and a decade later it appears on a thousand contractor websites attached to claims its authors never made. We sell monitoring, which means our whole product is trust in numbers. So we hold every published figure to a sourcing test: name the agency, name the study, or show the math. Every statistic on this site carries its source inline or in a Sources section at the bottom of the page.
The savings range, built from cited parts
When we say monitoring can recover 16 to 30% of what a fault-laden unit spends on energy, both ends come from published, unit-level figures, each about the HVAC equipment itself, not the whole building:
| Fault or fix | Energy impact | Source |
|---|---|---|
| Charge and airflow, corrected together (the floor) | ~16% | Downey & Proctor 2002, ACEEE (13,000-unit field study) |
| Any single fault (undercharge is the largest) | under 15% | NIST, Effect of Common Faults on Vapor Compression Systems |
| Three to four simultaneous faults | up to ~34% | Controlled multi-fault laboratory study (heat pump) |
The floor is ~16%, the field-measured average from correcting charge and airflow together (Downey & Proctor). NIST's controlled study finds any single fault stays under 15%, with undercharge among the largest, and that faults do not simply add. But real units rarely carry just one: controlled laboratory studies of three to four simultaneous faults have measured COP reductions up to about 34%. So the worked examples feature a 25% multi-fault case, above the single-correction floor and below that measured maximum, and treat 30% as the ceiling. Every figure is the equipment's own energy line, applied only to the electricity the unit itself burns, never to the building's total bill.
On a commercial demand-billed meter we also count demand-charge relief, and here is why that is fair. A fault that makes a unit draw more power raises its peak draw, not just its energy. New York commercial demand is billed on the highest peak kW in the month, and that peak falls during business hours on hot afternoons, exactly when a rooftop unit runs hardest, so the fault-driven excess draw lands inside the metered window. We size it conservatively, a few kW of avoided peak on the example unit at the verified rate, and scope it to a unit that materially sets the building's summer peak.
The famous number we refuse to quote
The most-cited statistic in this industry says poorly maintained and improperly controlled equipment wastes 15 to 30% of energy. The primary source is Katipamula & Brambley (2005), a peer-reviewed review written at Pacific Northwest National Laboratory. It is a real, well-founded figure. It is also about energy used in commercial buildings, meaning whole-building energy, not HVAC energy. Restating it as “15 to 30% of your HVAC energy” changes what the authors measured, and most of the industry does exactly that. We don't. When we cite Katipamula & Brambley, we cite it as a whole-building commercial figure. Our own 16 to 30% HVAC-line range is a different construction: the unit-level fault figures in the table above, each part separately cited.
Shown math counts as a source
Some numbers have no agency to cite because they are our own arithmetic, so we show it. Our anchor operating figure, about $286 per ton per year of cooling OpEx in NYC, is tons times roughly 1.0 kW per ton, times about 1,600 equivalent full-load hours, times a deliberately conservative $0.18 per kWh marginal rate. Every input is on the page, every assumption is flagged, and the all-in commercial rate ($0.24 to $0.30 per kWh) that would raise the figure is stated next to it. The full derivation is in What HVAC really costs a NYC building per year. When we quote a floor, we say it is a floor.
What we claim about the technology
The detection engine is statistical baselining: per-unit, per outdoor-temperature-band mean and variance, with deviation scoring against the matching band. It is deliberately explainable, and you or your engineer can audit the math behind any alert. We do not claim deep learning, and we do not quote failure-prediction accuracy we haven't earned on a large deployed fleet. The mechanics are on How it works.
How pricing follows from the sourcing
Because the savings floor is built from cited figures, we can price against it honestly: each subscription is priced below the energy it can recover on a unit that carries these faults, so in the worked scenarios the customer keeps well more than half of what AirPulse finds. On a unit with no recoverable faults, the value is the monitoring and the early warning, not an energy refund, and we say so. On a demand-billed commercial meter we also show demand-charge relief, with the rate and the peak-kW assumption on the card, but the subscription still prices against the energy line alone, so it pays for itself before any demand relief. Longer equipment life, avoided emergencies, and LL97 headroom stay upside on top, never counted in the pricing test. The caveat ships with every claim: actual savings depend on your equipment, runtime, and utility rate.
Sources
- U.S. Department of Energy, Air Conditioner Maintenance (clogged filter: 5 to 15% of an AC's energy use): energy.gov/energysaver/air-conditioner-maintenance
- Downey, T. & Proctor, J. (2002), “What Can 13,000 Air Conditioners Tell Us?”, ACEEE Summer Study (about 16% average gain from correcting refrigerant charge and airflow together, the floor of our range): aceee.org/files/proceedings/2002/data/papers/SS02_Panel1_Paper05.pdf
- U.S. National Institute of Standards and Technology (NIST), “Effect of Common Faults on the Performance of Different Types of Vapor Compression Systems” (single faults stay under 15%, undercharge among the largest; multi-fault laboratory studies measured COP reductions up to about 34%, and faults do not simply add): tsapps.nist.gov/publication/get_pdf.cfm?pub_id=918162
- U.S. Department of Energy, Building America Program, “Measure Guideline: Air Conditioner Diagnostics, Maintenance, and Replacement” (field fault prevalence; 38% average delivered-capacity gain from correcting charge and airflow, Kuonen 2011): www1.eere.energy.gov/buildings/publications/pdfs/building_america/measure_guide_air_cond_diagnostics.pdf
- Katipamula, S. & Brambley, M.R. (2005), “Methods for Fault Detection, Diagnostics, and Prognostics for Building Systems, Part I,” HVAC&R Research 11(1), whole-building commercial figure: osti.gov/biblio/15011268
- U.S. Energy Information Administration, CBECS 2018 (commercial electricity end uses): eia.gov/tools/faqs/faq.php?id=1174