Reading a commercial solar production model: what P50, P90, and degradation actually mean for your NPV
A commercial solar proposal is a 25-year cash flow forecast built on three variables the owner rarely sees called out. P50 versus P90, the degradation curve, and the availability assumption. What each one is doing in the model, and where honest proposals differ from optimistic ones.
The proposal in your inbox has a first-year production number on it. Somewhere on page four, in smaller type, there is a 25-year cumulative kWh figure. A savings table sits underneath it. The IRR and payback numbers on the cover page are derived from that table.
Three inputs upstream of that table decide whether the cover page is a fair representation of what the system will actually deliver, or a marketing artifact. The three inputs are almost never explained. They are: the confidence interval on annual production, the annual degradation curve, and the assumed system availability. Two developers can quote the same 250 kW system, model the same first-year kWh, and produce 25-year NPV numbers that differ by more than 100,000 dollars. The delta lives entirely in these three lines.
What a production model actually is
A commercial PV production model is a Monte Carlo simulation. The engineer feeds a site-specific model, typically PVsyst or SAM, with a typical meteorological year (TMY) dataset, panel and inverter specs, a shading analysis, a soiling assumption, and a wiring loss stack. The model runs the site against many possible weather years and returns a distribution of annual production outcomes.
That distribution is the entire point. The single first-year kWh number on a proposal is a point on that distribution. Which point matters.
P50 versus P90: the confidence interval
P50 is the median. In a given year, there is a 50 percent chance actual production lands at or above the P50 number, and a 50 percent chance it lands below. P50 is what most residential and lower-tier commercial proposals quote by default, because it is the highest defensible number.
P90 is the 90 percent confidence value. In a given year, there is a 90 percent probability actual production lands at or above the P90 number. It is a more conservative figure, roughly 6 to 10 percent below P50 depending on site variability and dataset quality.
P99 is used by tax equity investors and lenders on utility-scale projects. It is not standard in the C&I range but occasionally shows up on financed deals. Approximately 12 to 15 percent below P50.
Here is why this matters. If a developer quotes a P50 first-year production of 400,000 kWh, and the site's true P90 is 368,000 kWh, the proposal is quietly assuming an 8 percent production premium the owner has a 50 percent chance of not receiving in any given year. Over 25 years, that variance is real dollars. Signing a savings pro forma built on P50 numbers means signing a document that will overperform in half the years and underperform in the other half. If a lender is sizing debt against those same numbers, a bad weather year can push debt service coverage below covenants.
What to ask. Every commercial proposal should disclose whether the modeled kWh is P50 or P90. If the answer is P50, ask for the P90 number in the same units. A developer who cannot produce it on request is not modeling the site with the software they should be using.
The degradation curve
Silicon PV modules produce less each year than the year before. The physical mechanism is well understood: light-induced degradation in the first year, followed by a steady linear decline driven by encapsulant discoloration, potential-induced degradation, and cell-level micro-cracking accumulated through thermal cycling.
The degradation number a proposal uses is the second decisive variable in the 25-year model, and it is the one where marketing has drifted the farthest from field data.
Two conventions are common in C&I proposals:
- 0.5 percent per year linear degradation. This is the number modern Tier 1 monocrystalline modules quote on their datasheet warranty. It is a manufacturer marketing figure derived from accelerated laboratory aging, not from installed field performance.
- 0.7 to 0.8 percent per year linear degradation. This is what large-scale field studies from NREL, LBNL, and DNV have found across commercial installations over the last decade. It is the more defensible planning number.
The gap looks small on a spreadsheet. It is not small over 25 years.
On a 250 kW system with a P90 Year 1 production of 368,000 kWh, the cumulative 25-year kWh under each curve:
At a Colorado C&I effective retail rate of roughly 0.11 dollars per kWh blended, a 324,000 kWh cumulative shortfall over 25 years is roughly 35,600 dollars of foregone savings on a single 250 kW project. That number scales linearly with system size and roughly linearly with retail rate. On a 1 MW system, the same modeling gap is more than 140,000 dollars.
What to ask. What degradation rate is assumed, and is it derived from the module datasheet warranty or from field data? A serious developer will use 0.7 percent as a planning number and treat the datasheet 0.5 percent as an upside case, not the reverse.
The availability assumption
Even at P90 production and a 0.7 percent degradation curve, a production model still overstates lifetime output if it assumes the system is generating 100 percent of the time. It is not. Inverters fail, string monitoring flags anomalies that take days to resolve, communications drop, and unplanned outages happen. If the site does not have proactive monitoring and a responsive O&M contract, those interruptions add up.
Field data on well-maintained C&I systems in the United States generally shows availability in the 98.0 to 99.0 percent range. Systems without active monitoring, or with reactive-only O&M, run 95 to 97 percent.
The one percent gap between a well-monitored system and a neglected one is another two to three percent of lifetime kWh. On the same 250 kW system, that is another 200,000 kWh over 25 years, or roughly 22,000 dollars in a Colorado C&I rate environment.
What to ask. What availability figure is baked into the model? Is there a signed O&M contract with performance obligations, and does that contract carry a production guarantee that pays the owner cash if annual output falls below a specified threshold? Availability numbers without contractual backing are aspirations.
The stacked effect
The three variables compound. Here is what happens on the same reference project when each variable is set to the honest planning number rather than the marketing number:
Same panels. Same inverter. Same roof. Same interconnection. 121,000 dollars of divergence over 25 years, entirely inside three lines of the model.
This is not an argument that the honest model is pessimistic. It is the argument that the honest model is what the system will actually deliver against typical assumptions. A proposal built on the left column will underperform its own pro forma in most years. A proposal built on the right column will meet or slightly exceed it in most years, which is what an owner should want from a 25-year asset.
How this shows up in a real conversation with a developer
A commercial owner comparing two bids can back-solve for these variables even if they are not disclosed on the proposal.
- Ask for the modeled first-year kWh at both P50 and P90 in writing. The delta between them is the site's weather variance and gives you a read on the quality of the meteorological dataset the developer used.
- Ask for the assumed annual degradation rate and the source. A developer who cites the module datasheet is marketing. A developer who cites NREL Technical Report 51664 or the DNV C&I field study is engineering.
- Ask what availability figure is in the model. If the answer is 100 percent, or if the developer looks surprised by the question, they are not modeling the site the way an operator would model it.
- Ask whether the O&M contract carries a production guarantee that pays cash below a defined kWh threshold, and what that threshold is as a percentage of the modeled P90. A guarantee at 95 percent of P90 is meaningful. A guarantee at 90 percent of P50 is theatre.
- Ask which software produced the model. PVsyst and SAM are the two credible options for C&I. Aurora and Helioscope are competent for shading and layout but their production engines are less rigorous. A proposal built off a PVWatts online calculator is a residential-grade tool being used for a commercial asset.
- Ask for the loss stack. A production model has a chain of derate factors: soiling, wiring, mismatch, inverter efficiency, transformer, availability. If the loss stack is not itemized, the model is a black box.
If two proposals come back with materially different P90 numbers on the same roof, the differences almost always trace back to inputs a developer chose, not to physics. Shading assumptions, soiling frequency, thermal derate, and inverter clipping ratio are all knobs. Turning them optimistically inflates every downstream number.
Bottom line
The IRR on the cover of a commercial solar proposal is only as credible as three lines buried in the model. P50 versus P90 sets the range of years that will disappoint you. The degradation curve compounds silently across 25 years. The availability assumption decides whether your Year 15 kWh actually shows up. Getting all three on the record before signing costs an owner nothing and separates developers who model with engineering discipline from those who model to a target IRR.
For a 25-year asset attached to a property that in most cases outlives the ownership entity that bought it, that is the diligence worth doing.
If you would like Solara Pro to review a production model you have received, we are happy to run a second-set-of-eyes pass on the P90, the degradation curve, and the availability assumption, and tell you plainly what we would change.
Solara Pro is a commercial solar developer based in Broomfield, Colorado. This briefing is general guidance for owners reading production models. Every site-specific analysis should be verified against a stamped engineering model built for the actual roof, load, and interconnection. This is not tax, legal, or engineering advice.