Who Pays When a Solar System Underproduces?

A homeowner calls in month thirteen. The proposal said 14,200 kWh a year. The monitoring app says 12,600. They want to know what you are going to do about it.

The instinct is to work out who is at fault. That is the wrong first move, and it is expensive, because the most likely answer to "who pays" is nobody — and the second most likely is the manufacturer — and you cannot tell which without doing the diagnostic work first. Sales organisations and installers absorb shortfalls every year that were never theirs, because the conversation started at liability instead of at cause.

So: what the estimate actually promised, how to find out whether there is a real gap, and where accountability lands once you know.

One cause is common enough to name separately: a production model built assuming a tree would be gone, with nobody recording who was removing it or by when. Who is responsible for tree trimming on a solar project covers where that lands.

An estimate is not a guarantee, and the difference is the whole subject

Nearly every residential PV solar sale is made on a production estimate — a simulation of expected annual output, produced by design software from the system specification and a long-term weather dataset for the location.

Two properties of that number matter enormously and are almost never explained at the kitchen table:

It is modelled on typical weather, not on next year's weather. The datasets behind these models describe a representative long-run year. An actual year is never that year. Irradiance varies, and a single year landing several percent below the long-term average is entirely ordinary.

It is described in the contract as an estimate. Residential agreements generally state that actual production will vary and that no specific output is guaranteed. That language usually does its job for normal variance.

A production guarantee is a completely different instrument: a contractual commitment to a minimum output over a defined period, with a stated remedy for a shortfall. It exists only if somebody wrote it down.

The third instrument, which is not the second one either

Module performance warranties get pulled into these conversations and do not belong there. A performance warranty guarantees that an individual module will still produce a specified share of its nameplate rating in a given year, measured against a published degradation curve.

That is a per-component test against a curve. It is not a system-output test against a proposal. An array can miss its estimate by fifteen percent with every module on the roof sitting comfortably inside warranty, because the two instruments measure unrelated things. Conflating them produces a claim that gets denied and a homeowner who is now angrier than when they started.

The same layering problem shows up across the whole warranty stack — which warranty answers which failure is worth being clear about before you need it.

Where the gap between the estimate and the outcome came from what was said at the kitchen table rather than from the model, the question changes character entirely: who is liable when a solar salesperson misrepresents savings.

First, establish that the shortfall is real

A meaningful proportion of reported underproduction is not underproduction. Before anything else:

Is the data trustworthy? Monitoring communication gaps produce missing intervals that read as zero. A figure taken from inverter reporting is not the same as a figure from the production meter. Clipping — an inverter operating at its AC limit while the array could have produced more — is normal design behaviour on a deliberately oversized DC array and is routinely misread as loss.

Is the comparison a full year? Seasonal distribution is uneven, and a shortfall measured over any partial period tells you very little. Twelve consecutive months, or wait.

What was the weather? This is the single most important correction and the one most often skipped. The honest test is the model rerun against the irradiance that actually occurred at that location in that period. A great many "underproducing" systems are performing exactly as designed in a below-average solar year.

Was the system available? Downtime is not underperformance. Inverter faults, utility outages and grid events all remove production hours. Availability and efficiency are separate questions and separate obligations.

Then, establish what was actually built

The model described a specific system. Systems change between proposal and commissioning, and the changes are frequently legitimate — but the estimate does not update itself.

Check the as-built against the modelled:

• Module count and model. A substitution or a reduced count changes expected output directly. If equipment changed, the relevant comparison is a corrected model, and who absorbs the cost of a substitution is a question with its own answer.
• Tilt, azimuth and layout. A plane dropped at design review, or an array shifted for fire setbacks, changes the yield.
• Inverter model and configuration. Different AC capacity, different clipping behaviour.
• Export limits. Where the utility restricted the system, output can be capped by interconnection terms rather than by hardware — and utilities limit system size more often than sales conversations assume.

A system that does not match its model is not underproducing. It is producing correctly for what it is, against an estimate that was never revised. That revision is a closeout obligation, and it is one of the quieter items in PV solar project closeout.

Where accountability lands, by cause

Once you know why, the allocation is reasonably clear. This is the map:

Cause	Who it lands on	Typical remedy
Below-average weather	Nobody	None. This is normal variance
Monitoring or measurement artifact	Nobody — there is no shortfall	Fix the reporting
Component below its performance curve	Manufacturer	Warranty claim, usually replacement
Workmanship — miswired string, wrong export limit, self-shading racking	Installer	Workmanship warranty; correct the installation
As-built does not match as-sold	Follows the change: substitution, design revision, or the party that caused it	Corrected model, or make-good on the difference
Estimate built on wrong site inputs	Whoever owned the survey and the model	Contract and consumer-protection exposure
Written production guarantee missed	Whoever gave the guarantee	Whatever the guarantee specifies
Downtime and availability	Whoever holds the service obligation	Restore the system; guarantee terms may apply
Tree growth or new shading after install	Homeowner	Generally excluded from guarantees

Three of those rows deserve more than a line.

Wrong site inputs are the one with real teeth

Every other cause on that list is a performance question. This one is a representation question, and it is the reason estimate discipline matters.

If the model was run on the wrong azimuth, or against a roof plane that does not exist, or with a mature tree left out of the shade analysis, the objection is not that the simulation was imperfect. It is that the proposal described a different house. Disclaimers about variance address variance; they do not address a document that misstated the site. The exposure here runs to contract and to state consumer-protection law, and it is meaningfully larger than a warranty claim.

This is also where the site survey stops being a scheduling formality. The survey is the evidentiary basis for the number in the proposal.

Workmanship causes are cheap to fix and easy to miss

A string wired into the wrong input, an inverter export limit left at a commissioning value, a second array row shading the first at low sun angles — these produce persistent, unremarkable-looking shortfalls that no amount of weather normalisation explains. They are also fully correctable, which makes them the best possible outcome of an underproduction investigation. Rule them out early rather than late.

The homeowner-caused row is real, and needs saying gently

Trees grow. Neighbours build. Panels soil, and in some climates snow matters. Most guarantees exclude changed site conditions explicitly, because the alternative is guaranteeing output against a site nobody controls. The conversation is easier when the exclusion was explained at signing rather than produced in year four.

The structural problem for sales-only organisations

Here is the seam this whole subject sits on.

The production estimate is generated by the party closing the sale. The party who has to answer for it — often years later, usually by phone, always to an unhappy homeowner — may be the installer, the service provider, or whoever now holds the account. The estimate creates a long-tail obligation, and it is routinely created by somebody who will not be holding it.

That is not a criticism of sales organisations. It is a description of how the division of labour actually works, and it is fixable with two things:

Preserve the estimate as evidence. The model inputs, the software and version, the shading data, the date, and the signed disclosure documents. Not a screenshot. An estimate you cannot reconstruct is an estimate you cannot defend, and the default in an undocumented dispute is that the customer's recollection stands.

Allocate estimate risk in writing. Between the sales organisation and whoever installs and services, somebody should own the consequences of the number. In most agreements nobody does, which means it lands on whoever the homeowner can reach.

Almost all of this is easier when the reference and the first months of data were filed at closeout rather than reconstructed under pressure, which is the practical case for verifying production against the design before the last draw rather than after the first complaint.

A worked sequence

A homeowner reports a shortfall. In order, and no skipping:

1. Pull twelve months of interval data and check for gaps, outages and meter-versus-inverter discrepancies. Roughly a third of reports end here.
2. Weather-normalise. Rerun the model on actual irradiance for the period. A gap that closes to within a few percent is normal variance, and the answer is a clear explanation rather than a remedy.
3. Reconcile as-built against as-modelled. Module count, model, layout, inverter, export limit. If they differ, correct the model before continuing.
4. Check availability. Separate downtime from efficiency. Downtime points at service, not design.
5. Check the components against their curves. If a module is genuinely below its performance warranty, it is a manufacturer claim.
6. Inspect for workmanship causes. Strings, configuration, self-shading, soiling.
7. Only now ask what was promised in writing — estimate, guarantee, or neither — and apply the map above.

Steps 1 to 6 are diagnostic and cost a few hours. Step 7 is where money moves. Doing them in the other order is how organisations pay for weather.

The bottom line

Underproduction is a diagnostic problem that people treat as a liability problem, which is why it costs more than it should. Most gaps are weather, measurement, or a system that was legitimately built differently from the one modelled — and none of those are anyone's debt. The gaps that are someone's are identifiable: a component below its curve, an installation error, a system that does not match what was sold, an estimate built on the wrong site, or a guarantee somebody actually wrote.

Three habits prevent nearly all of the avoidable cost: keep the estimate defensible, reconcile the as-built when anything changes, and make sure monitoring works before you close the project.

If you would rather the design, the install and the service after it sat with one accountable party instead of three, get in touch. Coverage is confirmed per service area rather than promised as blanket availability.