Battery Storage

The 18% Gap: Why SMA Systems Get Blamed for Design Errors

2026-11-17 · Kofi Asante

The Call I Get Every Quarter

I got the call on a Tuesday afternoon, which is usually when these things surface. A 480 kW commercial rooftop outside Las Vegas, commissioned in late 2023, was producing about 18% under its modeled output—or rather, 18% under the P50 estimate, which is the number the owner actually cares about because it's the number in the financing documents.

The installer's first message was direct: "We think the SMA inverters are underperforming."

I want to be fair to them, because that's a reasonable first hypothesis. SMA has a big installed base, the Sunny Tripower line is what everyone specifies on commercial rooftops, and when output doesn't match the model, the inverter is the easiest thing to point at. It's visible, it logs faults, and it has a brand name on the front that a client can Google.

But in the roughly 40 rush diagnostics I've handled over the past two years—mostly commercial and utility-scale arrays in Nevada, Arizona, and Southern California—the inverter itself has been the root cause maybe three times. The other 37 were design, commissioning, or monitoring issues that just happened to show up first on the inverter's display.

This article is about the other 37. Specifically, about why solar panel installer training keeps producing the same string sizing mistakes, why those mistakes cost more than most people think, and what a real design-review step looks like when it's done properly.

The Problem Isn't the Inverter. It's the String.

Here's the pattern I keep finding. The array is fine. The modules are fine. The inverter is fine—it's actually doing exactly what it's supposed to do, which is trip offline when DC input voltage exceeds its window.

On that Las Vegas rooftop, the fault log showed repeated "DC input voltage too high" events clustered in the early morning hours during the previous winter. That's a dead giveaway for anyone who's spent time on cold-weather string design. On a module with a Voc temperature coefficient of about -0.3%/°C, a string that reads 410V at STC will read closer to 455V when the cell temperature drops to -10°C. Add ten modules to that string and you're at a number that pushes the upper MPPT limit, then eventually the max DC voltage limit, and the inverter shuts down to protect itself.

The installer hadn't done the cold-temperature voltage calculation. Not because they were lazy—because the design tool they used had defaulted to a minimum temperature that was 15°C warmer than what actually shows up on a Nevada rooftop in January.

"What most people don't realize is that string sizing tools are only as good as the temperature data you feed them. The defaults are usually set for a mild climate that exists in very few places."

If I remember correctly, the original design had 14 modules per string on a 1000V-rated Sunny Tripower. At STC that's around 574V—comfortable. At -10°C with the actual module coefficient, it's north of 630V, which is still under the 1000V limit but close enough to the MPPT upper bound that the inverter starts clipping in the morning. The output loss compounds over a winter. By spring, the owner is looking at an 18% annualized gap and the installer is looking at a warranty claim.

The Deeper Problem: Training That Teaches the Tool, Not the Physics

Here's where I have to be careful, because this is the part that gets installers defensive.

Most solar panel installer training programs—the ones that run two to five days and end with a certificate—teach the design software. They walk you through the SMA Sunny Design interface, show you where to enter module count, and let the tool spit out a string configuration. That's useful. It's also not the same thing as understanding why the tool is making the recommendation it's making.

What the tool does is apply a formula. What the installer needs to know is when the formula's inputs are wrong. And the inputs are almost always wrong in one of three places: the minimum design temperature, the module's actual Voc coefficient (which varies more between manufacturers than the datasheet suggests), or the expected soiling and shading losses that affect the MPPT operating point.

I'm not 100% sure why some installers consistently get this right while others consistently miss it. My best guess is that it comes down to whether someone on the team has actually done a field measurement of cold-morning string voltage, versus just trusting the simulation. The ones who've been burned once tend to build a check into their process forever after.

The ones who haven't been burned yet usually haven't because their climate is forgiving. Nevada is not forgiving. Neither is the high desert in New Mexico, or the mountains outside Flagstaff. The temperature swing on a Nevada storage-plus-solar site can run from 40°C in July afternoon to near freezing by 4 a.m. in January. That's a thermal cycle the string design has to survive thousands of times over the system's life.

Which brings up a useful comparison. Your customer expects this array to last 25 years. Compared to the 4.6 billion years the planets in our solar system have been around, 25 years is nothing—but it's still roughly 9,000 daily thermal cycles, and every one of them is a small stress test on the string design. If the design has no margin at the cold end, you're not just losing output in year one. You're accelerating connector and module-level degradation for the entire warranty period.

What This Actually Costs

Let me put numbers on the Las Vegas job, because vague cost warnings don't change behavior.

The 18% output gap on a 480 kW array, at Nevada commercial rates and the site's PPA structure, worked out to roughly $14,000 in lost revenue in year one alone. The diagnostic visit and re-stringing labor was about $9,000. The rework required pulling modules on two rows—about 60 modules total—which meant the roof warranty paperwork had to be re-filed with the membrane manufacturer, adding another $2,000 in administrative cost.

But the number that actually mattered to the installer wasn't any of those. It was the phone call from the asset owner's CFO asking whether the installer "knew what they were doing." That's the cost that doesn't show up in a change order. The installer had done 11 projects for that owner over four years. After the string sizing incident, they did two. The remaining nine went to a competitor who charged 6% more per watt but had a documented design-review step.

When I switched my own diagnostic practice from "audit the inverter logs" to "audit the design inputs first," my callback rate on these jobs dropped by roughly a third. The pattern is that consistent.

What a Real Fix Looks Like

The solution here is short, because the problem was long.

First, run the cold-temperature voltage calculation by hand for every string. Not the tool's default. The actual site minimum design temperature from ASHRAE or a local weather station, applied to the actual module's temperature coefficient from its datasheet—not the typical value. It takes 20 minutes and it catches the failure mode that causes most of these calls.

Second, build a design-review gate that is separate from the designer. We didn't have a formal review step at my previous firm until 2022, when a similar-sized job cost us a $40,000 change order. The reviewer's only job is to check three numbers: string Voc at minimum temperature, string Vmp at maximum temperature, and the inverter's MPPT window against both. That's it. It's a 15-minute check and it has caught every one of the last nine near-misses we've had.

Third, use the monitoring platform the way it's meant to be used. SMA's Sunny Portal will flag DC input overvoltage events if you set the alert threshold correctly. Most installers leave the default thresholds, which are set for a mild climate. Adjust them for the site, and the platform becomes an early-warning system instead of a post-mortem tool.

None of this requires new equipment. It requires treating string sizing as a design decision with consequences, rather than a step the software handles.

The inverter is rarely the villain. But it's always the first thing the client sees, which means it's always the first thing that damages the brand perception. A string sizing error that gets caught in review is a boring Tuesday. A string sizing error that ships is a phone call from a CFO, a lost account, and a story that follows your company for years.

Leave a Reply