Untitled

Every dispatcher knows which tech gets the cream work. Senior guy with the clean van. The one who never complains. Whoever's closest to the owner's office. Over time, it shows up in paychecks.

The quiet assumption has always been that fair dispatch costs money. New research in Computers & Operations Research (2024) says otherwise.

The model balanced three objectives at once:

→ Firm profit → Customer satisfaction → Equity in technician earnings

Finding: fairness often improves alongside profit, not against it. The assumed tradeoff is smaller and more situational than industry intuition.

Why this matters for retention: the tech most likely to leave your shop in 2026 isn't your bottom performer. It's your solid mid-level tech who shows up on time, does clean work, and watches higher-revenue jobs route to two preferred colleagues. They see it. They have options now.

Simple audit you can run today:

Pull 90 days of jobs. Sum per-tech revenue, grouped by experience level. If the spread between your highest and lowest earner (comparable tier) exceeds 40%, you have a fairness problem and a retention problem building under it.

Three practical moves:

1. Measure. Most owners have never pulled this data. 2. Rotate the cream. High-revenue diagnostics and installs shouldn't default to the same two techs every week. 3. Explain the model. Techs don't expect perfect equality. They expect transparency. Resentment grows in silence when nobody knows why work gets assigned the way it does.

Fair dispatch isn't charity. It's retention math.

Ask any HVAC owner about their dispatch board at 7 a.m. and they'll describe the same chaos: 6 techs, 22 calls, three urgent, one needs a certification the Tuesday guy doesn't have, and the contract-labor partner is texting about the north end.

Research in the European Journal of Operational Research (2024) finally modeled that mess properly. Authors Nowak and Szufel built an optimization approach that accounts for technician skills, customer time windows, AND contract-labor availability together — not one at a time.

Result: meaningful drive-time reduction and higher jobs-per-day. Gains scale with fleet size. 4-6 techs: modest. 15+ techs: significant, because combinatorial complexity grows faster than human intuition.

Most shops plan routes by geography and appointment time. Skill gets handled via senior/junior labels. Contract labor is a last-minute text chain. Research says that leaves real efficiency on the floor.

At California tech labor of $50+/hr loaded, a 10% drive-time cut plus half a job per tech per day is material money.

The practical play doesn't require custom software. ServiceTitan, Housecall Pro, FieldEdge all have constraint-aware routing features. The question is whether you've configured them or are using the platform as a glorified calendar.

Three things to fix this week:

→ Skill matrix per tech (certs, brand training, specialties) → Honest time windows ("anytime" ≠ "before 2 p.m.") → Contract-labor availability entered the night before, not morning-of

Most AI fault-detection research for HVAC is bench data. This one isn't.

Case Studies in Thermal Engineering (2025) just published a three-year deployment study: 42 operating chillers in Taiwan, live production loads, real maintenance. A 1D Convolutional Neural Network with transfer learning hit 93% accuracy on three major fault types AND documented kW/RT energy savings on the real equipment.

That's the piece that's been missing from the predictive-maintenance pitch. Not accuracy scores. Proof of energy savings on machines that were actually running.

The faults it catches are the ones that quietly cost money:

→ Refrigerant undercharge → Fouled condenser tubes → Inefficient part-load operation

None of them trigger a conventional trouble alarm until they're bad. All of them show up in BMS data if you know how to read it. The model reads.

A chiller running 0.65 kW/RT instead of 0.58 is six figures a year on one machine in a data center or hospital. Flagging that early is direct money. Writing it into a service agreement with peer-reviewed backing is what separates you from the vendor who still pitches quarterly inspections.

Caveat: Taiwanese chillers, specific manufacturer and climate profile. The mechanism ports to the US. The exact 93% number has to be re-earned on local equipment.

Commercial HVAC contractors have been pitching "predictive maintenance" for years. New research in Buildings (2025) finally gives the pitch empirical teeth.

Researchers trained Long Short-Term Memory neural networks on multi-year Building Management System data from commercial buildings. The model predicts remaining useful life of HVAC components — compressors, fan motors, the stuff that fails — and outputs schedule-aware service actions, not just dashboard metrics.

That last part matters. Most academic ML research for HVAC stops at accuracy scores. This one made the outputs dispatchable.

Why it helps you sell commercial accounts:

Building owners have been burned by "predictive" pitches that turned out to be quarterly inspections with new branding and a higher price. Procurement managers want to know what's actually different. An LSTM model trained on years of on-site BMS data, producing per-component RUL estimates, is a real answer.

Don't have a data-science team? Good news. Johnson Controls, Siemens, Honeywell are all embedding ML-RUL models into commercial BMS platforms right now. Independent shops can configure those features and translate the outputs into service agreement language.

Realistic caveat: a model trained on Class A office towers won't run identically on a 1970s school boiler plant. Match model to building stock or you'll sell accuracy you can't hit.

First concrete step: ask your commercial accounts whether their BMS exports telemetry history. If yes, you have the raw material. If no, your first agreement upgrade is enabling that export.

Ever wonder why your expensive new dispatch software hasn't moved the needle on revenue? Research in Technovation (2024) studied 434 SMEs and landed on the reason: digital tools only lift firm performance when you also change how you package, price, and deliver service. Software alone is just a faster version of the same business.

Most HVAC shops still run a 1995 pricing model. T&M for repairs. Flat estimates for installs. An optional maintenance plan nobody renews. New dispatch tools on top of that speed up the same transactions without creating new revenue categories.

The path the research points to: use the tools to unlock products you couldn't sell before.

→ Subscription seasonal tune-ups → Tiered service agreements with predictive-maintenance callouts → Flat-rate repair bundles with parts guarantees → Customer portals showing equipment health, not just invoices

The software doesn't create the products. It makes them cheap enough to offer.

Simple discipline before your next software contract: write down the new service product you'll launch within 90 days of go-live. If you can't name it, you're not buying growth. You're buying faster status quo.

Good news for anyone training HVAC apprentices in California: the model your state is doubling down on just got a heavyweight empirical endorsement.

A randomized controlled trial in Review of Economic Studies (2025) found that pairing classroom vocational instruction with paid on-the-job training AND a wage subsidy:

→ sharply lifted apprenticeship take-up → substantially raised post-program earnings → produced near-zero displacement of existing informal apprentices

That last one matters. Critics have argued formal apprenticeships just shuffle workers between tracks. Data says no.

Caveat: study ran in Côte d'Ivoire. Institutional context is different. But the design is randomized, peer-reviewed, top-five economics journal. The mechanism is what California is already betting on through DAS-registered programs and Golden State Pathways.

The wage-subsidy detail is what American HVAC programs still miss. First-year apprentice pay is low enough that candidates often take warehouse work instead. The study found topping up apprentice wages during training dramatically lifts enrollment. WIOA and CTE grants can fund this — they're just underused in our trade.

The funding stack is the actual argument to make at your next industry meeting: DAS registration + WIOA vouchers + CTE grants = first-year apprentices at near-zero net cost, backed by peer-reviewed earnings evidence.

Helper pay is the early warning. New research out in the Journal of Management in Engineering (2025) analyzed 24 years of construction wage data and found unskilled and helper wages climbed 2.75-3.5% per year — and those moves statistically predicted the skilled-trade raises that followed.

Translation for HVAC shop owners: you've been watching the wrong number.

Most owners track competitor journeyman pay because that's who walks out the door for a $2 raise. Helpers feel captive. They're not. When unskilled construction wages climb, your green techs have options — solar crews, new-construction GCs, warehouse jobs. Backfill costs rise, compresses the gap to mid-level techs, and the whole ladder has to reprice.

The practical move:

Pull BLS Occupational Employment Statistics helper and laborer wages for your metro. Drop them in a spreadsheet next to your own helper pay. When the market number pulls ahead by more than 5%, you're about 12 months from losing a journeyman.

California already ran this movie. Minimum wage at $16.50, helpers at $22-26/hr. The next journeyman repricing cycle is already queuing up.

Caveat: construction-engineering journal, not pure labor economics. Signal is strongest in active commercial-construction metros. But if you're in Austin, Phoenix, Nashville, or the Inland Empire, the data is loud enough to act on.

80% less HVAC energy. That's what peer-reviewed research in Energy and Buildings (2025) found when schools switched from natural to mechanical ventilation — while still hitting IAQ targets.

Moreno-Rangel and co-authors modeled four ventilation strategies across 11 Mediterranean-climate zones. Mechanical ventilation won every zone. Layered CO2 demand control on top and saved another 8%.

Natural ventilation has a nice ring to it. Open windows, breezy classroom, save on fan power. The energy bill tells a different story. When classrooms run on operable windows, HVAC ends up conditioning way more outside air than needed, and comfort swings with the weather.

For California shops bidding school district retrofits, this is a pitch you can actually defend. Peer-reviewed numbers beat vendor datasheets at a facilities meeting every time.

The CO2 demand-control detail matters. Instead of fixed outside-air rates, the system reads room CO2 and ramps fresh air to actual occupancy. Empty room at 2 p.m. during staff meeting? No over-ventilation. Full room during third period? Air delivered. Sensors are cheap.

Spec list: VFDs on supply fans, commissioned economizers, classroom CO2 sensors tied to demand control, a DDC that enforces setpoints. Miss one and the savings walk.

Rotary pressure exchanger lifts transcritical CO2 COP 18.2%.

Saeed et al., Applied Thermal Engineering 2024. Tested a transcritical CO2 system pairing a rotary gas pressure exchanger with two low-lift ejectors. Up to 60 bar of free pressure lift.

Addresses the worst part of CO2's operating envelope. Relevant to warm-climate supermarket retrofits under CARB rules.

Two-stage R-290 heat pump: 18.3% more heating capacity at -8.3°C.

Purdue built a variable-speed two-stage propane (R-290) liquid-to-liquid heat pump with independent compressor control. Published in Applied Thermal Engineering 2025.

Indirect architecture keeps flammable charge outdoors. Roadmap for where California A3 code paths may land.

A2L vs A3: What the CFD Actually Showed

Wang et al., International Journal of Refrigeration (2024). CFD simulation of R-1234yf (A2L) and R-290 (A3) leaks in an automotive AC cabin.

Results:

R-1234yf: stayed below its 6.2% LFL in the passenger compartment.

R-290: exceeded its 2.1% LFL at every monitoring point.

Only evaporator-outlet leaks were modeled — not a full sweep. But the A2L-vs-A3 contrast lines up with the service procedures already diverging between the two classes, especially in small enclosures.

R-454B posts 38% higher EER than R-410A in rack cooling tests.

A 2025 Applied Thermal Engineering paper ran 24 realistic in-rack tests. R-454B hit ~5.8 EER at 9 kW, comparable PUE, substantially lower TEWI. GWP 466 vs 2088.

Efficiency story is now on R-454B's side. Service practice needs A2L-rated leak detection and updated brazing.

MOF-coated desiccants beat cooling-coil dehumidification by 65 to 85%.

A 2025 Journal of Building Engineering paper compared conventional cooling coils, silica gel, zeolite, and MIL-100(Fe) MOF desiccant exchangers. MOF won big at sensible/latent ratio 0.5.

Coastal California light commercial and data centers are the obvious early markets. Latent-load-heavy climates get the most out of this technology.

A typical mid-sized HVAC shop quietly walks away from about $53,779 a year in after-hours revenue. Average job value of $450, a 2.5x emergency multiplier, ten missed calls a week. The money doesn't show up on any P&L because it never enters the business.

Jane Blanchard, who leads brand and marketing at ServiceForge, said the assumption that those calls go to the biggest franchise is wrong. "It ultimately goes to whoever picks up. You don't have to be the biggest business in town to win that call — but you do need to be the one that picks up."

AI isn't the fix. One in three callers hangs up when they reach AI on a business line. At 11 p.m. on a no-heat call, people want a human. Blanchard's frame: AI is a workflow tool — appointment reminders, transcriptions, summaries — not a relationship tool.

The missed-call windows aren't the 3 a.m. ones owners assume. They're 6–9 a.m. (people calling before work), lunch hour, and 5–9 p.m. Owners don't call those "after-hours." Customers do, because the phone isn't getting picked up.

Web forms have a 30-minute shelf life. Homeowners submit to three or four shops at once. Whoever calls back first, qualifies, and books wins.

The starting move: pull one week of call data, count what came in outside your hours. The number tends to do the rest of the work.

Field MPC on dual-fuel: 27% bill reduction, furnace never fired.

A 2024 Applied Energy study ran model-predictive control on a small commercial heat pump plus gas furnace setup for two months. Time-of-use rates plus gas prices meant the heat pump won every hour. 23% of load shifted off-peak.

The controller was inexpensive edge hardware. Not research-grade. Deployable on real jobs.

Measured VRF SCOP of 5.35 in a real university building.

A 2025 Applied Sciences paper logged 67% less primary energy and 79% lower emissions per square meter versus a gas-boiler baseline. Grid was 83% renewable. Methodology uses real seasonal data, not rated conditions.

Zoning and heat recovery drive the gains. Useful when bidding commercial retrofits.

The biggest efficiency win left on GSHPs is not hardware. It is controls.

A 2024 Energy and Buildings systematic review covered ground loop designs, working fluids, and hybrid configurations. The authors flagged controls as the largest remaining research gap. Poor scheduling of compressors, circulators, and aux heat can cut theoretical COP 10 to 20%.

Hybrid systems with solar thermal recharge and seasonal thermal storage showed the strongest cold-climate gains. Relevant for mountain and high-desert California installs.