Week 7 Session: The Number Someone Wants

A real business case that came out below one, an evaluation that says exactly where the pressure went, and an hour spent finding out what you would have done

19 April 2027

None of the evidence today is invented. The project is real, the numbers are the proponent’s own, and the criticisms were written by the agency whose job it is to make them. The negotiation you run in the second half is a teaching simulation built on that record — the case is sourced, the roles are not.

The case#

The Great Western Highway runs across the Great Dividing Range through the Blue Mountains. The 34 km between Katoomba and Lithgow is the last unimproved stretch. In 2022 the NSW Government put a business case for upgrading the eastern and western sections, and Infrastructure Australia evaluated it, as it must for proposals seeking $250 million or more of Commonwealth funding.

From that evaluation, dated 8 September 2022:

“The project is expected to generate a cost to society, with a NPV of -$579.5 million.”

The proponent’s benefit–cost ratio was 0.57. Including wider economic benefits and land use benefits, 0.68. A ratio below one means the costs exceed the benefits as counted.

The project was funded anyway: $2,030 million from the Australian Government in April 2022 and $2,500 million from the NSW Government in March 2022 — both commitments made before the evaluation was published.

Sit with that for a minute before going on. Nobody involved was hiding anything. The analysis said the project did not pay for itself, the analysis was published, and the money had already been committed. Whatever pressure operated here, it did not operate by concealment.

Where the pressure actually went#

Infrastructure Australia’s evaluation is specific about where the appraisal leans. Four findings, quoted:

On the largest single benefit:

“the largest economic benefit (28% of total benefits) is derived from avoided routine maintenance of the existing roads and the basis of this contains some assumptions and methodological issues which could overstate the quantum of benefits in the Cost Benefit Analysis (CBA).”

On the contingency allowances applied to the two sides of the comparison:

“The business case also assumes higher contingencies in the base case O&M (of 20% - P50 and 30% - P90) compared to the project case O&M (East & West sections: 10% - P50, 15% - P90).”

“Overall, we consider the base case estimates of O&M costs to be disproportionate relative to the O&M cost estimates in the project case. This will have the effect of overestimating the benefits and the BCR of the project.”

And on what happened when the first method gave the wrong answer:

“The cost-benefit analysis (CBA) of the four shortlisted options of the whole program identified that the ‘Do Minimum’ option had the highest NPV and BCR. A multi-criteria analysis (MCA) was undertaken after the CBA and this significantly changed the prioritisation of options from that indicated by the CBA. The weightings and scores within the MCA implied extremely high valuations on some identified benefits, far higher than calculated for those items by the CBA. This suggests an optimism bias in the design of the MCA results that favours the project option.”

Study the contingency finding hardest. The doing-nothing case was allowed a generous margin for costs that might blow out; the build case was allowed a tighter one. Each figure is arguable on its own. Set side by side they push in one direction, and it is the direction that favours the project option — which is Infrastructure Australia’s own characterisation of what the appraisal does.

That is the thing to learn to see, and note carefully what it is not. Nothing here establishes that any individual set out to reach a chosen answer, and you should not claim it does. What the evaluation documents is a pattern of assumptions that lean the same way — an asymmetry consistent with optimism bias, not a fabrication, and not a proven intention.

Run it#

Groups of four. Three of you hold the appraisal; one takes the proponent’s interest, and that person’s brief is simple: the ratio needs to clear 1.0.

Rules, because the exercise is worthless without them. The proponent may only ask for particular assumptions to be revisited, and must give a reason each time. They may never ask for the answer to be changed. The pressures available are the ones this real appraisal actually shows — the maintenance-savings benefit, the contingency margins on either side, the 42% apportionment of program benefits to these two sections while the Central section stays unfunded, the wider economic benefits, and the choice of appraisal method itself.

Everybody else answers each request with exactly one of three moves, and the group writes it down:

• Accept — the request brought new evidence, or exposed a flaw in your reasoning, that materially changes the assumption. Record which of the two it was.
• Reject — the request supplied no evidence and no reasoning that changes the assumption. Record what was missing. Wanting a different answer is not a reason, and neither is asking twice.
• Widen — the request revealed more uncertainty than your number implied, without justifying a move in either direction. Widen the range, leave the centre where it is.

Keep the log as you go. Do not tidy it afterwards.

The audit#

Stop the negotiation with ten minutes left and read your own log back.

For every accept, name the evidence or the reasoning flaw that justified it. If the honest entry in that column is “they pushed twice”, mark it. Nobody is graded on how few of those they have; the skill being trained is seeing them at all, and they are far easier to see in a log than in the moment.

Then compare what your group did with what the real business case did. Between 0.57 and 0.68 sits a decision to count additional categories of benefit — a legitimate analytical choice, and one that could only move the ratio upward, and it still did not get them to one.

Finally, look at what Infrastructure Australia recommends to the whole sector. Its appraisal guidance says optimism bias “refers to the underestimation of the likelihood of an adverse event”, that research “indicates that optimism bias is a significant problem in infrastructure”, and lists what to do about it, including: ensure that underlying assumptions are transparent; compare estimates with past comparable projects; conduct independent peer reviews.

Every one of those is a defence that works only if it is in place before the pressure arrives. Check which ones your group had.

Leaving#

Write one sentence, for yourself: the assumption in your own week 3 ledger that you would find hardest to hold if someone with authority wanted it moved.

Nothing is due. Next week the room estimates, and the room turns out to have a bias of its own.