The Numbers to Question in Any Feasibility Model

The Numbers to Question in Any Feasibility Model

General information for South Australian landowners and small developers only. This is not financial, planning, construction, valuation or tax advice, and nothing here comments on the viability of any particular project or model. Route questions about construction pricing, specifications and buildability to a quantity surveyor or your builder; questions about what a completed dwelling or allotment is likely to sell for to an independent valuer and experienced local agents; questions about what a site can accommodate and which assessment pathway a proposal would attract to a planning consultant; questions about how a development would be taxed to a registered tax agent; and questions about finance structure and serviceability to your own accountant or finance broker. Market conditions and statutory arrangements change over time, so confirm the current position before relying on any model — including your own.

The model that looked precise

A feasibility model arrives looking finished. The columns align, the subtotals reconcile, the bottom line sits in bold at the foot of the page. Formatting does something quiet and dangerous here: it transfers confidence. A number that began life as somebody's guess, typed into a cell on an ordinary afternoon, emerges from the spreadsheet indistinguishable from a number that was quoted, dated and signed. By the time the model reaches you, the guess has been formatted into a fact.

This is worth saying plainly because of where feasibility models actually fail. The arithmetic can certainly be wrong — a formula can point at the wrong cell, a subtotal can quietly leave a row out, a piece of logic can be built on a misreading — and any model you intend to rely on should have been checked for exactly that. But the inputs upstream of the formulas deserve at least as much scrutiny as the arithmetic — particularly an input that nobody flagged as soft, because flagging it would have made the page look less finished.

This piece is a companion to what a feasibility model can and cannot decide. That article asks which of your questions a model can settle and which it cannot; this one asks something prior — where the model's inputs came from in the first place. The two questions are separate, and both need answering before a bottom line deserves your trust. If you want the full walkthrough of how a study is assembled line by line, that is set out in how a feasibility study for an Adelaide site actually works. The scope here is narrower: four inputs that deserve interrogation in any model put in front of you, and a single test that catches most of the trouble.

Question the yield

The yield — how many dwellings or allotments the model assumes the site will carry — is the input everything else leans on. Revenue multiplies off it. Construction cost scales with it. Change it and the whole page changes with it. So the first question is not whether the yield looks reasonable. It is: who derived this, and against which constraints?

There is a wide gap between a yield formed by a planning consultant who has read the title, the zoning, the overlays, the frontage, the easements and the servicing position, and a yield produced by dividing the site area by an assumed lot size. Both produce a tidy count. Only one of them has been tested against this site's actual criteria, and it is the better informed of the two for that reason. A model rarely tells you which kind it holds, so you have to ask — and if the answer is that the yield came from a selling agent's material, or from what a nearby project achieved, treat it as an aspiration until someone qualified has tested it against this land.

Two boundary points sit underneath the question. A concept scheme, however carefully drawn, remains subject to planning consent: the relevant authority determines what is approved, and your planning consultant advises you on what the site is likely to support — those are different roles, and no feasibility model collapses them into one. And where a model's yield leans on an anticipated rezoning, remember that a rezoning does not itself grant development approval; what a change of that kind would actually open up for this land is a question for your planning consultant or solicitor, and any application still has to be made and determined. A yield that quietly assumes both steps have already succeeded is not an input. It is a wish with a border around it.

Question the build rate

The build rate looks like one number, but it is three claims wearing a single cell: whose quote is this, at what date, and for what specification?

Whose quote matters because rates have provenance. A rate a builder priced against actual drawings for this site is one kind of evidence. A rate remembered from a previous project, or inherited from whoever built the template, is another kind entirely — and the cell displays them identically. At what date matters because construction pricing moves, and a quote is a snapshot of the market on the day it was given. A model built around pricing obtained before the design settled is carrying a number whose shelf life may already have expired, and nothing on the page will say so. For what specification matters because a rate is meaningless without knowing what it buys: which standard of finish, and whether it includes the things this particular site will demand — retaining, fill, demolition, service connections, the driveway that has to negotiate the fall of the land. Two rates can differ substantially and both be correct, because they are pricing different buildings.

These are questions for a quantity surveyor or a builder, not for the spreadsheet's author to settle by assertion. What the spreadsheet can do — and should — is record, beside the rate, its source, its date and its specification basis. If the model cannot tell you those three things, that silence is itself the finding.

Question the approval duration

Time appears in a feasibility model even when no row names it. Every period the project spends waiting for a decision, the land is being held, and holding costs accrue whether or not the model prints them — rates, and, depending on how the land is held and funded, interest, insurance and the alternatives forgone. So the duration assumption deserves the same interrogation as any cost line: whose experience produced it, across how comparable a set of cases?

A common risk worth checking here is the point estimate. A model that carries a single tidy duration is asserting that approval timeframes are a fact you can look up, when they are actually a spread you have to estimate — an argument made in full in how long development approval really takes in South Australia. Applications that look alike on the surface can carry different assessment pathways, different referral triggers and different notification outcomes, and their durations scatter accordingly. In research we co-authored on South Australian planning applications, the pattern that mattered was that comparable cases cluster into ranges with identifiable tails — which is exactly the shape a feasibility model should hold. A model that carries one duration scenario is not cautious or optimistic; it is unexamined.

The practical question to put to any model, then, is not "is this duration right" — nobody can certify that in advance — but "does the model survive the slower scenario". If the bottom line only works when the approval lands at the hopeful end of the range, you have learned something the bold total was not going to volunteer.

Question the contingency

The contingency row is where models confess their character. Ask one question of it: is this a priced buffer, or a hope?

A priced buffer is built against identified risks. Ground conditions that remain unknown until the geotechnical work is done. Service connections whose cost cannot be confirmed until the authority responds. The drift between the date the pricing was obtained and the date a contract will actually be signed. Each of those is a nameable exposure, and a contingency assembled from them can be defended line by line. A hope, by contrast, is a figure typed in because the template had a row for it — chosen by convention, sized by habit, connected to nothing about this site.

Two further tells are worth knowing. First, watch what happens to the contingency under pressure: a model whose author trims the buffer until the bottom line clears the hurdle has already answered your question about its reliability. Second, do not let the contingency and the margin blur into each other. The contingency is there to absorb what goes wrong; the margin is the reward for the risk of the whole undertaking, and what margin a project actually needs is its own question, taken up in how developers think about margin on cost. A model that spends its contingency to fatten its margin, or leans on its margin to excuse a hollow contingency, has confused the two jobs — and will discover the difference at the worst possible moment.

Provenance is the test, not precision

Pull the four questions together and they are one question asked four times: where did this come from, and how sure are we?

That is the test of a feasibility model. Not its precision — precision is cheap, and the most dangerous models are the most precise-looking ones. The test is provenance: whether every material line can answer for itself. Who supplied this number. On what date. Against what evidence. And, where the honest answer is "nobody yet — this is a placeholder", whether the model says so, visibly, instead of dressing the placeholder in the same font as the quotes.

This is also why the lines a template generated deserve the question most, not least. A template's fixed rows carry defaults that nobody involved in your project chose, sourced from circumstances that are not yours, at dates nobody recorded. The problem is not that a template's numbers are necessarily wrong. It is that a fixed row gives you no way to find out — you cannot trace what you cannot open. A model whose every line can be edited, sourced and dated beats a template with locked rows for the same reason a witness who can be cross-examined beats one who cannot: not because the testimony is friendlier, but because it can be tested.

The habit that follows is simple to state. Annotate every material assumption with its source and date. Mark the unverified as unverified. Then re-run the bottom line with each questioned input pushed toward its unfavourable edge, one at a time, and watch which movements the result survives. A model's real output is not a single number. It is a map of which inputs the number depends on — and that map tells you which questions to take to your own advisers next.

Putting the questions to work

If you would rather start from a model built to be interrogated, this is what our investment return analysis report is for. Every cost line in it is editable and carries its stated basis, so your own advisers can trace, challenge and replace any input; where market evidence is involved, our part is to compile and coordinate it so your own advisers can check its sources, while any conclusion about market value is a matter for a certified practising valuer you engage — the report is not a valuation and does not stand in for one; and the whole document is an evidence-based assessment from historical cases, not a guarantee of any outcome. Any concept scheme it contains remains subject to planning consent. Where options are examined, they are compared side by side rather than auto-ranked, and any recommendation is signed by a named professional who stands behind it. Cyberate PM's role is to coordinate the inputs — from quantity surveyors, valuers, planners and your own tax and legal advisers — not to substitute for their opinions. You can read what the report covers at the investment return analysis report. Where the questions above need answering across the whole site rather than inside one model, that is property development feasibility. And if it would help to talk through whether your project warrants one, start a conversation with us — a conversation is a practical way to sort which of the questions above already have answers, and which the model in front of you has been quietly leaving open.