You are wrong in a predictable direction.

That number was built from five separate pieces. Measured work priced off a bill by one team. Preliminaries built from a resourced programme by another. Escalation from a market forecast. Contingency from a percentage rule. Overhead and profit from a company rate.

Five independent build-ups, done in different weeks by different people, adding up to within a seventh of one percent of a round million.

That does not happen. Not by arithmetic.

The round number came first

What actually happened is the thing that always happens. Somebody formed a view early about what this job had to come in at — from the client's budget, from what the last one went for, from a conversation at a site visit — and every subsequent number was built with that figure already in the room.

Nobody instructs it. Nobody writes it down. The estimator prices the work honestly, arrives at $1,043,000, and then goes back through looking for the places where he has been generous, because he knows what the number needs to be. He finds them, because there are always places. The waste allowance comes down a point. The gang rate assumes the good gang. The programme assumes the crane is available when you want it.

Every single adjustment is defensible on its own. That is exactly what makes it invisible.

This has a name. The first figure that lands in a room becomes the reference point everything else is measured against, and people adjust away from it far less than the evidence should move them. In construction the anchor is almost never a calculation. It is a budget, a previous job, or a number the client said out loud.

Nobody in this story is lying

There are two different failures here and it matters which one you are dealing with, because they have different fixes.

The first is honest. Ask anybody how long a task will take and they will imagine it going the way it is supposed to go — the drawings arrive, the gang turns up, the concrete is on time — and estimate that. They are not stupid. They know jobs go wrong. But when they picture this job, they picture it working, because that is what planning a thing feels like from the inside. Then they add ten percent for luck, which is nowhere near enough, because the historical overrun on work like this is not ten percent.

The second is not honest, exactly, but it is not lying either, and it is far more powerful.

An estimate is not only a prediction. It is also a bid for something — approval, a contract, a place on the tender list. And numbers that are optimistic get approved. A contractor who prices this job at $1,140,000 because that is what he thinks it will cost does not win the job and does not get to be right. He gets to be unemployed.

So estimates are not scattered randomly around the truth. They are pulled toward whatever number gets the outcome the estimator needs. Not by fraud. By selection.

The winner's curse

Which brings us to the part that is specific to this industry, and is not a psychological weakness at all. It is arithmetic, and it is unavoidable.

Six contractors priced this job. All six are experienced. All six did honest take-offs. Because the job contains real uncertainty — the ground, the market, the productivity — their six numbers came out spread across a range: $1,000,000, $1,032,000, $1,068,000, $1,094,000, $1,131,000, $1,180,000.

The average of those six is $1,084,167. That is the industry's collective best view of what this job costs.

We tendered $1,000,000, and we won.

We did not win because we are eight percent better than five other competent firms. We won because, on this particular job, our assumptions happened to be the most optimistic of the six. The tender process does not select the most efficient contractor. It selects the one whose errors ran furthest in one direction.

Every job you win, you win by being the most optimistic person in the room. That is not a warning about your company. It is a structural property of competitive tendering, and it applies to every contractor who has ever won anything.

The view from outside

There is one fix, and it is almost insultingly simple. Stop asking what this job will cost.

Ask instead: of the jobs we have done that look like this one, what did they finish at compared to what we tendered?

That question needs no judgement about the rock, the crane or the gang. It does not care how good your team is or how well you have planned. It asks only what happened last time, twenty times over. It is called taking the outside view, and the technique built on it — reference class forecasting — came out of work by Bent Flyvbjerg on why large projects overrun as reliably as they do.

Pull twenty comparable jobs. Piling and substructure, similar value, similar client type. Divide each final account by each tender sum and line them up.

The median is 1.11. The eightieth percentile is 1.19. One job out of twenty came in under its tender. One.

Apply the median to this contract and the expected outturn is $1,110,000.

Now hold that next to something Cost & Cash produced twenty-three weeks ago, from a completely different direction — from the cash forecast, the productivity trend and the steel price. That article put the final cost at $1,120,000.

Two methods that share no inputs, landing within one percent of each other. The detailed forecast and the crude historical ratio agree, and they both disagree with the tender.

Practical insight

Find the last ten jobs your company completed. You need two numbers for each: the tender sum and the final cost. Not the final account — the cost. Somebody in commercial has both.

Divide one by the other, sort the ten, take the middle.

That single number is worth more than every risk workshop your company has ever run, because it contains every risk that actually materialised, including all the ones nobody thought to write down.

Then compare it to the contingency you are carrying today. On this job the answer is 4.8% held against an 11% median. If your ratio is anything like the industry's, you will find the same thing, and it will take you an afternoon rather than a career to find out.

Key takeaways

✔ Five independent build-ups landing $1,388 under a round million is not arithmetic. The target existed before the estimate.
✔ Anchoring works through defensible individual adjustments, which is exactly why nobody notices it happening.
✔ The planning fallacy is honest: asked about this job, people picture it working, because that is what planning feels like from the inside.
✔ Strategic misrepresentation is not fraud. Optimistic numbers get approved and win work, so estimates are selected for optimism rather than accuracy.
✔ The winner's curse is arithmetic, not weakness. Six honest bids spread across a range, and the job goes to the most optimistic one every time.
✔ The outside view needs no knowledge of your project — only what happened to twenty projects that looked like it.
✔ Median outturn on twenty comparable jobs was 1.11 of tender. One in twenty came in under. The contingency here is 4.8%.

What's coming next

That is the end of judgement. We have fourteen risks, anchored scales, a grid we now know the limits of, and a historical ratio that says the whole tender is eleven percent light.

What we do not have is a way of combining any of it. You cannot add a median to a heat map. You cannot ask a reference class how much money to hold this month rather than next.

For that we need to stop treating each risk as a single number and start treating it as a range — which means being clear about something we have been sloppy with all track. There are two entirely different reasons you do not know a number, they behave differently, and only one of them gets smaller when you go and find out more.

Next week: the rock and the gang rate look like the same kind of uncertainty. They are not, and treating them the same is why contingency gets spent on the wrong things.