Someone has quoted you a number. Maybe it's 20%, maybe it's "up to a quarter more from the same roof," and it arrived before anybody looked at a single month of your production data. This article is not about which optimization levers exist - our solar system optimization complete guide walks through those. It's about the narrower and more expensive question: how do you find out whether the number you have been handed is true for your array, before the invoice?
TL;DR: A yield claim is only checkable if you know what it's a percentage of. Get the baseline in writing, pull thirteen months of your own metered production, model the same array in PVGIS, and treat the gap between those two as the ceiling on what any retrofit can recover. Under light-to-medium shading, independent measurement put module-level optimizers about one point ahead of a plain string inverter, against the +20% a vendor case study reports for a deliberately awkward roof. Both numbers are real. Only one of them is a forecast for a typical house.
What Does a Twenty Percent Yield Claim Actually Mean?
Nothing at all, until somebody says what it's a percentage of. That sounds pedantic. It's the single distinction that decides whether you can ever hold anyone to the figure.
There are three different baselines hiding behind the same sentence, and they are not interchangeable. A gain against nameplate compares your roof to a laboratory rating no installed system has ever hit. A gain against a simulation compares it to a model the vendor ran, with inputs the vendor chose, which you usually never see. A gain against your own metered production over a stated period is the only one that can be verified after the work is done, because it is the only one where both numbers come from your meter.
Ask which one it is. The answer is diagnostic in itself: a firm that has done this before will have the distinction ready, and a firm that hasn't will treat the question as an obstacle.
How Do You Build a Baseline Before Anyone Quotes You?
Thirteen months of your own data, and there's a reason for the odd number. Twelve gives you one full seasonal cycle. The thirteenth gives you an overlapping month, so you can put January against January rather than against an average, which is what separates a genuine fault from an ordinary dull winter.
Export monthly kWh from your inverter portal and write it in a spreadsheet next to a PVGIS estimate for the same array. The EC JRC PVGIS database takes your location, tilt, azimuth and system size and returns an annual estimate built from satellite irradiance, and it costs nothing. The gap between that estimate and your meter is the number that matters. It is the ceiling. No retrofit can recover more than the energy you are actually losing, and a claim that exceeds your own measured shortfall is arithmetically impossible before anyone argues about hardware.
For scale: NREL's PV Fleet Performance Data Initiative, covering 2,200+ sites and 19,000+ inverters across 37 US states, found inverter availability alone causes 2.3% average annual energy loss, rising to roughly 8% in the first six months (NREL, 2024). Across 500+ commercial systems totalling 11 GW, kWh Analytics found assets missing P50 estimates by 5 - 10% (Solar Generation Index, 2022). A 6 kW home at 8,000 kWh/year losing 10% is down 800 kWh, about $160 a year. That is your budget for fixing it, and it's worth knowing the figure before you hear a price.
Which Part of the Claim Does Independent Measurement Support?
This is where the two kinds of evidence pull apart, and it's worth seeing the numbers side by side.
Allenspach and colleagues at ZHAW simulated light-to-medium chimney shading and measured a plain string inverter returning about 90.7% of unshaded annual energy, against about 92.1% with per-panel optimizers (Solar RRL 7(8), 2200596, 2023). That's roughly one percentage point of annual yield. The same work found commercial DC/DC optimizers running about 2 percentage points below their published efficiency, a conversion loss you pay every sunny hour whether anything is shaded or not.
| What is being compared | Measured result | What it is evidence for |
|---|---|---|
| String inverter, light chimney shade | ~90.7% of unshaded annual energy | The baseline case |
| Per-panel optimizers, same shade | ~92.1% of unshaded annual energy | About +1 point, typical light shading |
| Optimizer conversion efficiency | ~2 points below datasheet | A permanent cost, not a one-off |
| Vendor retrofit case study, complex roof | +20% claimed | A ceiling for an awkward array |
Read that table the right way round. None of those figures is dishonest. The point is that they answer different questions, and a quote that cites the last row while your roof looks like the first two is answering a question about somebody else's house.
What Does a Vendor Case Study Leave Out?
Its own selection. A case study is written after the result is known, which means the roof was chosen because the number came out well. That isn't fraud and I don't think it should be treated as such - it's a fair demonstration that a ceiling exists. It is simply useless as an estimate of what a median roof will do, and it gets quoted as though it were one.
So read every published gain for three things the prose usually skips. What was the array like before? A retrofit on two orientations with a chimney has far more to recover than a clean unshaded south pitch, and the starting condition is what produced the headline. Over what period was the gain measured? A summer month flatters module-level electronics, because that's when shadows move most across a roof. And what does the comparison hold constant? A retrofit that replaced ageing panels at the same time isn't measuring the optimizers at all.
Here's a test I'd apply to any case study before believing it applies to me: does the write-up describe a roof I would recognise as mine? If the photographs show three dormers and a mature oak, and my array is one clean rectangle, the document is evidence about dormers.
Who Should Ignore All of This?
Some people genuinely should buy the hardware without a thirteen-month audit, and pretending otherwise would be its own kind of dishonesty.
If your roof carries two or more orientations, heavy all-day obstruction, or panels past eight to ten years that have visibly diverged, the case is strong enough that waiting a year costs more than it saves. The SolarEdge P370 power optimizer retrofits string systems at 99.5% efficiency with IP68 weatherproofing and adds per-panel monitoring without an inverter swap; above its 370 W rating you are into the S-series, and the two families cannot share a string. The power optimizer vs microinverter guide covers which architecture suits which roof.
The audit matters most in the ambiguous middle: a system that feels a bit down, on a roof with nothing obviously wrong with it. That's precisely where a confident percentage is most persuasive and least earned.
How Do You Turn the Claim Into a Payback Number?
Three multiplications, and they take about five minutes.
Take your metered annual kWh, multiply by the claimed percentage, and you have the extra energy. Multiply that by whatever a kWh is actually worth to you - your export rate if the surplus goes to the grid, your import rate if it displaces a purchase, and those are usually very different numbers. Divide the quoted price by the result and you have payback in years.
Run it on the honest input as well as the optimistic one. On our 6 kW, 8,000 kWh example, a claimed 20% is 1,600 kWh; the ZHAW-style one point is 80 kWh. At $0.20 per kWh that is $320 a year against $16 a year. A $1,200 retrofit pays back in under four years on the first figure and in seventy-five on the second. The decision isn't close in either direction, which is exactly why the baseline question is worth being tiresome about. Our solar panel payback period analysis has the full method, and the solar panel efficiency calculator guide helps confirm the shortfall is real before you chase it.
One more thing I'd put in writing. If a vendor is confident in the figure, ask for it as a performance guarantee naming the measurement window, the data source and the remedy. A percentage without those three is unenforceable, and the request costs you nothing to make.
Summary
A solar yield claim is only worth as much as the baseline behind it, so establish yours before you accept a quote: thirteen months of metered production against a PVGIS estimate for the same array, with the gap between them treated as the hard ceiling on anything a retrofit can return. Independent measurement puts per-panel optimizers about one point ahead under light shading and charges roughly two points of conversion loss for the privilege, while vendor case studies showing +20% describe deliberately awkward roofs and should be read as ceilings rather than forecasts. Convert whatever number you are given into an annual cash figure and a payback period before comparing it to the price, because the same claim can mean $320 a year or $16. Buy the hardware without waiting only where the roof has obvious, heavy, all-day obstruction or multiple orientations. If you are starting from the export itself and the columns mean nothing yet, our guide to reading solar inverter data explains what each figure is telling you.