Let me save you a procurement meeting: the inverter with the highest peak-efficiency number on its datasheet is usually not the most efficient system you can actually build. Not because the number is fake. Because it's measured under conditions that don't exist on a roof in July.
I work on the quality and brand compliance side at deye. I review spec sheets, installation manuals, and incoming batches before they reach our integrators—roughly 400 SKUs and documentation sets a year. In our Q1 2024 audit we pulled 46 datasheets from across the market and tried to map their efficiency claims against how the equipment actually gets installed. The pattern was consistent: peak efficiency clustered into a narrow band, and the differences at the very top said almost nothing about annual yield.
So here's the argument I keep making, and keep losing, in spec meetings.
Peak efficiency is a lab condition, not a project condition
Most modern hybrid inverters, deye's included, land between roughly 97% and 98% peak efficiency. That number is real and it's tested. It's also measured at one DC input voltage, one ambient temperature, and one load point that the manufacturer picks. Move any of those three and the curve moves with it.
What you should be reading instead is the shape of the curve—especially the ends. Converters are least efficient at light load and at the top of their rated window. A unit that hits 97.8% at 60% load and falls to 94% at 15% load will underperform a flatter unit on a system that spends its mornings and evenings in low-light conditions. That's most residential systems.
When an integrator sends me a deye 12kw hybrid inverter datasheet and asks whether 97.6% beats a competitor's 98.1%, my answer is always the same: show me both curves at 25°C and at 45°C, then show me your string layout. Every inverter derates at high ambient temperature. A half-point gap at the peak evaporates the first afternoon the unit sits in a hot garage at 46°C.
Do the math on what that half point is even worth. A 12 kW system running 1,400 full-load hours a year, at $0.15/kWh, gains about $3.80 annually from a 0.2-point efficiency improvement. Three dollars and eighty cents. That's the prize for winning the datasheet argument.
Worth fighting over? No. Worth understanding? Yes, because it tells you where the real losses are.
The cheapest efficiency upgrade isn't in the inverter
Here's the part that surprises people. The single highest-return efficiency change on most of the systems I review isn't a better inverter. It's one gauge size up on the DC cable.
A 2% voltage drop on a 12 kW array at full production is about 240 W of heat in the conductors. That's twenty times the annual value of a 0.2-point peak-efficiency improvement, and it costs a few hundred dollars in copper on a typical residential run. The reason it gets skipped isn't cost. It's that nobody puts cable sizing in the same paragraph as inverter selection, so it never makes it into the comparison spreadsheet.
Grounding and equipotential bonding belong in the same conversation, for a different reason. NEC 2023 Article 690.47 sets the grounding requirements for PV systems, and I've watched more projects get delayed by an undersized equipment grounding conductor than by an efficiency shortfall. That's not a performance issue—until it is. A system that fails inspection isn't efficient. It's a liability sitting on a roof, and the rework cost lands on whoever signed the drawing.
Which brings me to the thing that actually wrecks efficiency numbers in the field.
Where the efficiency argument really dies: the battery
Last spring an installer forwarded me a link to a dumfume 12v 300ah lithium lifepo4 battery and asked whether it could carry a 6 kW critical-load panel through a hybrid inverter.
On paper it looked plausible. 300 Ah at 12.8 V nominal is about 3.8 kWh. The problem isn't capacity—it's the discharge rate. A 6 kW load on a 12.8 V bank pulls roughly 470 A. The continuous discharge rating on most batteries in that class sits well below that, and the BMS will either current-limit or disconnect outright. Neither outcome is something you want to explain to a homeowner at 9 p.m. during an outage.
The battery wasn't defective. It was the wrong battery for the load profile, and the spec sheet made it easy to miss because capacity is the number printed in the largest font.
Capacity tells you how long. Continuous discharge rating tells you whether. Those are different questions, and only one of them causes callbacks.
I've made this mistake myself, in a smaller way. In 2022 I approved a low-voltage battery package to save about $600 per unit against the 48 V alternative. Nine months later we were eating a retrofit on eleven units because the BMS was throttling hard in cold weather and the customers read it as a system fault. Net loss after labor and freight: north of $4,000. The 'cheaper' choice looked smart right up until it didn't.
Cycle life claims deserve the same skepticism. The 6,000-cycle figure in a brochure usually assumes 80% depth of discharge at 25°C. Put the same battery in an unconditioned enclosure in Arizona and you're running a different test.
You can't manage an efficiency you can't see
The best analogy I have for system monitoring comes from outside our industry. A vital sign monitoring system in a hospital doesn't measure 'health.' It measures heart rate, SpO2, blood pressure, and respiration separately, continuously, with alarms at defined thresholds. If it handed clinicians one aggregate wellness score once a day, they'd unplug it.
Solar monitoring works the same way, or it should. Aggregate production numbers hide the thing you're trying to find. Module-level or string-level data is what turns a vague 'the system seems low' into a specific diagnosis.
This is where the microinverter conversation usually starts, and I think it starts in the wrong place. When someone asks me what does a microinverter do, the honest answer is two things: it does maximum power point tracking per module, and it reports per module. The first reduces mismatch losses. The second gives you resolution. But it also puts one more electronic assembly per panel in the sun, adds a failure point per module instead of per string, and raises the $/W. I'm not arguing microinverters are wrong—I'm arguing the efficiency case for them is narrower than the marketing suggests, and the diagnostics case is broader.
We didn't have a formal commissioning verification step until 2023. It cost us when a batch of units shipped with monitoring gateways configured to the wrong time zone, and three sites logged six weeks of production under the wrong tariff window. Nobody noticed because nobody was reading the data against a baseline. The third time a variant of that problem showed up, I finally wrote a 14-point commissioning checklist. Should have done it after the first.
What I actually ask for now
When a vendor or an internal team brings me an efficiency comparison, I want three things, in this order:
- Weighted efficiency—CEC or European—not peak.
- The full curve, with the derating point marked and the test ambient temperature stated.
- The continuous discharge rating and BMS behavior of whatever battery is in the same drawing.
If a supplier can't produce those three, the peak number they're leading with is the least interesting claim in their document.
'But we need datasheets to compare'
Yes. And I'm not arguing you ignore them. I'm arguing you stop letting one number decide a six-figure project. Datasheets are how you shortlist. They were never meant to be how you decide.
The fair objection here is that weighted efficiency and curve data are harder to collect than a single headline figure, and that a lot of vendors don't publish them. That's true, and it's also the point. If a manufacturer won't publish the derating curve, you've learned something useful about how much engineering is behind the brochure.
It took me about four years and somewhere north of 600 datasheets to understand that efficiency is a property of the system, not of a component. The inverter is the loudest part of the install. It's rarely the part that decides whether the project pays back.
Buy the system. Read the curve. Size the cable. Check the battery's continuous rating, not its headline capacity. And when the next meeting starts with someone waving a 98.2% figure, ask what load point it was measured at.
The silence that follows is usually the most useful efficiency data you'll get all week.