Compare the assumptions, not the price

Two commercial solar proposals can quote nearly identical hardware and differ by years on payback. The difference is almost always in assumptions that were never printed. Normalise those and the comparison becomes real.

Put two proposals side by side and the eye goes to three numbers: the capacity, the price and the payback. All three are outputs. None of them can be compared until the inputs behind them match.

The instinct to compare them anyway is strong, because they are the only three numbers presented in a form that looks comparable. That presentation is the problem rather than the arithmetic.

The capacity is an output of a design, and two designers looking at the same roof will fill it differently depending on what clearances they left and whether they surveyed it or traced it.

The price is an output of a scope, and scopes differ far more than prices do.

The payback is an output of a model, and a model is a stack of assumptions about yield, consumption, tariffs and certificates, any one of which can move the answer by years.

The work of comparing proposals is therefore not arithmetic. It is normalisation: making the two documents describe the same job under the same assumptions, and then seeing what is left.

It is worth saying what this is not. It is not an argument that the lower price is suspicious or that the higher price is thorough. Both happen, and neither can be established from the documents until the inputs match.

There is a shortcut worth knowing. Where a proposal is unwilling to state its assumptions, the comparison is already finished, because nothing in it can be checked.

What follows is the order to do that in, and the four places the difference usually turns out to be hiding.

Before any number is compared, establish that both proposals cover the same work. This single step resolves most apparent price gaps.

The reason scopes differ so widely is that several of the items below cannot be priced accurately until somebody has been on site. A business that has surveyed the building prices them. A business that has not, excludes them.

The first row is the one that ends projects. A board with no spare ways and a superseded protection standard is a rebuild, not a connection, and it is invisible from a desk. Another post owns that ground: see "Why the switchboard decides your solar programme".

The rule to apply is simple. An item excluded from both proposals is a shared unknown, and the buyer carries it. An item excluded from one is not a saving, it is a difference in scope, and it should be priced before the comparison is made.

Ask each party to quote the same exclusion list. Most will, and the ones that will not have answered a different question.

One more exclusion worth listing separately, because it does not appear in the table and it should. Making good. Who patches a penetration, who reinstates a flashing, and who is responsible if the roof leaks in the first winter after the work.

Annual generation is the first modelled figure, and it is the one most often presented without its workings.

Four things sit underneath it. The orientation and tilt actually proposed, not a generic value. The shading model, and whether it was built from a survey of the roof or from an aerial image. The soiling and loss assumptions. And the degradation curve applied across the modelled life.

Differences in any of these are legitimate. Differences that are not disclosed are not.

Degradation deserves particular attention because it compounds. A model applying a lower annual decline produces a materially better twenty-year total from identical hardware, and the assumption sits in a footnote if it is stated at all.

The useful question is a ratio rather than a total: how many kilowatt hours per kilowatt of installed capacity per year does the model expect. That number is comparable between proposals in a way a total is not, because it removes the capacity difference.

Then ask what shading was modelled. A plant room, a parapet, a neighbouring building or a stack removes generating hours from part of the array, and a model built from an aerial trace will not have seen any of it.

Where the two proposals disagree on yield by a meaningful margin for the same roof, one of them has modelled a different roof.

Ask, finally, which weather dataset the model used. Generation models run on historical irradiance data for a location, and two datasets for the same site will disagree slightly even when everything else matches.

Generation is only half the model. What the generated kilowatt hours are worth is the other half, and it moves the answer more.

The critical assumption is the self-consumption ratio: how much of the generation the site uses itself rather than exporting. Self-consumed generation is worth the retail rate the site avoids paying. Exported generation is worth whatever the feed-in arrangement says. Those two numbers are not close.

It is also the assumption most likely to be optimistic by default, because the modelling tool has to put something in the field and the flattering value is rarely the one that gets questioned.

A proposal that assumes a site will consume most of its own generation, on a site that will consume barely half of it, has not quoted optimistically. It has modelled a different business.

The ratio is derivable from the site's own interval data, which is why that data belongs in the brief rather than in the response. Give both parties the same year of interval data and the assumption stops being a variable. Another post covers how the data is read: see "Size it from the load, not the roof".

Then check the tariff itself. Which rate the avoided kilowatt hours were valued at, whether demand charges were modelled at all, and what escalation rate was applied across the modelled life. An escalation assumption compounding over twenty years is doing more work in the final number than most people realise.

Demand charges are the assumption most often left out entirely. On a site billed on maximum demand, an array that lowers energy consumption without touching the peak leaves a large part of the bill untouched, and a model that ignores the distinction will overstate the saving: see "Demand charges are a different bill".

Certificate value appears in almost every commercial proposal, often as a fixed deduction, and it is the figure least suited to being fixed.

Small-scale certificates are created at installation and their value depends on the capacity, the installation date and the market price on the day¹. A number quoted in a proposal is an estimate made against conditions that will have moved by the time the system is built.

That does not make it dishonest. It makes it a forecast, and it should be labelled as one. Another post covers the mechanism in full: see "Why nobody can quote you an exact figure".

Check also whether the quoted figure is a deduction already applied to the price or an amount the buyer is expected to receive later. Those are different commercial positions and they are frequently presented in the same format.

For comparison purposes, the safest approach is to strip the certificate line out of both proposals and compare what is left. If one proposal only wins with a favourable certificate assumption, that is worth knowing.

Where a scheme has a published estimating tool, use it rather than the proposal's figure for a sanity check².

All of the above reduces to a single request, and it is the one worth making at the start rather than at the end.

Ask each party for a one-page assumption register: yield per kilowatt, shading basis, self-consumption ratio, tariff and escalation, certificate treatment, degradation, and the full exclusion list.

It should fit on one page. If it runs to five, the proposal has been padded rather than explained, and the assumptions are still not comparable.

A business that can produce it in a day has modelled the site. A business that cannot has modelled a template.

The register also has a second life. Once the system is commissioned, it is the document the installed result gets measured against, which turns a sales artefact into an accountability one.

And it makes the final decision defensible internally, which matters more than it sounds. A recommendation that names its assumptions can be approved by people who were not in the meetings, and that is usually the difference between a project that proceeds and one that keeps getting deferred.

One last note on how to run the process. Send the same brief to every party, including the interval data, the roof age, the switchboard details if you have them, and the exclusion list you want priced. Proposals written against different briefs cannot be normalised afterwards, however carefully they are read.