The Estimate Was Made by the People Who Needed to Win

A forecast is not only a technical product. It is also an artefact of whoever produced it, and of what they needed it to say.

Most organisations respond to a large overrun by improving their estimating. They buy a better model, mandate a range instead of a point, require a second opinion, add a contingency policy. All of it is sensible, and almost none of it addresses the mechanism that produced the number.

Here is the uncomfortable version. In the great majority of cases, the people who produce the estimate are the people who need the proposal to proceed. They may be a bid team whose year depends on winning. They may be an internal sponsor who has spent six months building support. They may be an engineering group that wants the work. They are rarely dishonest, and that is precisely the difficulty — the effect operates without anyone lying, and it survives every process improvement that leaves the incentive in place.

The Economist, reporting on project performance in 2005, located the problem exactly there: "Problems arise most frequently when initiation gets separated from execution." Securing the work, it went on, pushes bidders toward optimistic assumptions about costs and revenues — and it attributed the mechanism to what Max Bazerman of Harvard Business School calls self-serving bias — the same phenomenon he uses to explain why good accountants do bad audits. [SOURCE DETAILS REQUIRED]

The Strategic Context

The audit analogy is the part worth sitting with. Auditors are trained, licensed, professionally obligated and personally liable. They are also paid by the party whose accounts they examine. The profession's response to that structure was not to exhort auditors to try harder; it was decades of structural intervention — rotation, independence rules, restrictions on adjacent services. The structure was treated as the problem, because it was.

An enterprise's capital estimates sit in a comparable structure and are almost never treated the same way. The bid team produces the number that decides whether the bid proceeds. The sponsor produces the benefits case that decides whether their initiative is funded. The vendor produces the effort estimate that determines the size of their contract. In each case the producer holds a position on the outcome, and in each case the recipient treats the output as a technical fact.

The 2005 report named two amplifiers, and both remain live. The first is the public sector, where, in its assessment, after-the-event accountability to the project's paymaster — the taxpayer — is less rigorous, so the corrective feedback that might discipline forecasting is weaker. The second is the prestige project, where bidders are "chasing glory almost as much as commercial gain", and where part of the value being pursued does not appear anywhere in the financial case. The report's examples were a national stadium reconstruction whose developer admitted mounting losses on a £750 million project after the price of steel doubled in a year, and a trans-national oil pipeline that opened several months late and five to ten per cent over its four-billion-dollar budget. [FACT CHECK REQUIRED]

What Gets Blamed Instead

Three explanations usually arrive before the structural one, and each is partially true, which is why each is durable.

"The estimate was immature." Often accurate, and it points somewhere useful. Kul Uppal, writing in Cost Engineering in 2002, identified two root causes behind the majority of rework across a body of engineering and construction projects: poor definition of project requirements before the estimate was prepared, and "failure to recognize invalid assumptions behind these project requirements". [SOURCE DETAILS REQUIRED] Note the second one carefully. It is not that the assumptions were wrong; it is that nobody recognised they were assumptions. That is a question about who was in the room, not about the maturity of the technique.

"The scope changed." Frequently true, and irrelevant to the original number, which was wrong on the day it was written for reasons unrelated to what happened afterwards.

"We should have used a range." A range is better than a point, and organisations that adopt ranges routinely find it collapses back to a single figure somewhere between the analyst and the approval paper. [Related article: From Estimate to Commitment]

None is a lie. All three locate the fault in method or in events, and leave undisturbed the question of who produced the number and what they needed it to say.

Reframing the Issue

The reframing is small and it changes what a leader does next. Stop treating an estimate as a measurement and start treating it as testimony — a statement made by an identifiable party who has a position on the matter.

Testimony is not worthless — courts, boards and regulators rely on it constantly. But nobody competent receives it without asking who is speaking and what they stand to gain, and nobody competent treats the answer as an accusation. It is the first question.

Applied to a capital paper, the question becomes: for each material number here, who produced it, and what outcome did that party want? Not "is this person trustworthy" — almost always yes — but "is this person positioned to be wrong in a particular direction". The answer is usually available, usually uncomfortable, and usually absent from the paper. [Related article: What Must Be True: The Assumptions Register as a Strategy Instrument]

The Same Mechanism, Inside and Outside

The structure is easiest to see in a bid, but it is not confined to one. Consider, hypothetically, an energy retailer building the internal case for a smart meter rollout. The programme team preparing the case has been advocating the rollout for two years. Their benefits model assumes a rate of customer self-service adoption, an installation productivity figure drawn from a favourable pilot, and a meter failure rate supplied by the preferred vendor. Every input is defensible in isolation. Every input also sits at the optimistic end of its plausible range, and no single choice is large enough to argue about. The aggregate will not survive contact with the field, and it was produced entirely in good faith.

Now the external version. A commercial building services contractor, hypothetically, is bidding a large office fit-out. The estimator has priced the mechanical works with an allowance for access to occupied floors, based on the client's stated intention to decant two levels at a time. The bid manager knows that allowance is the difference between winning and losing. Nobody instructs the estimator to change it. The tender simply goes out with the client's stated intention treated as a fact rather than as an assumption held by a party who has not yet had to act on it.

In both cases the defect is invisible at the level of any individual number, and obvious at the level of the whole. That is the signature of the mechanism, and it is why line-by-line scrutiny rarely finds it.

The Two Interventions That Actually Change the Room

The corrective is not exhortation. It is changing the composition of the group that sets the number, and the available evidence describes two ways to do that.

Put the people who will deliver it into the room where it is promised. Siemens, having launched a worldwide initiative to improve its project management, introduced project managers into the company's sales teams specifically to temper their more extravagant promises. The report is unusually honest about the cost: it is "a move that requires a careful balance between reining them in and killing the deal". [FACT CHECK REQUIRED] That sentence is the whole of the trade-off. A delivery voice in a sales conversation will lose some deals that a purely commercial team would have won, and some of those deals would have been profitable. The enterprise is choosing a distribution, not eliminating a defect — fewer wins, and fewer of the wins that destroy value. The choice is defensible; it is not free, and any organisation adopting it while promising no effect on win rate has not adopted it.

Give the challenge to people with no stake in the answer. Nick Lavingia, writing in the same journal a year later, describes the peer review in terms that make its purpose unmistakable: its goal is "to constructively challenge the project team's assumptions, alternatives considered, decision logic and path forward", and participants "should include peers who are not associated with the current project". [SOURCE DETAILS REQUIRED] The second half is the mechanism. A review staffed from the programme, the division or the supply chain is a technical check; a review staffed from outside the outcome is an independence device. Most organisations run the first and describe it as the second.

Neither intervention requires new analytical capability. Both require someone senior to accept a slower and less agreeable approval process, which is why both are commonly announced and quietly diluted.

Decision Framework

The instrument is a single addition to the approval paper, and it is deliberately small enough to survive contact with a busy governance calendar.

The estimate provenance line. For each material number in the paper, four fields:

Where the third field reads "nobody", the paper says so. It does not resolve the gap; it discloses it, and lets the approver price it. That is the entire design intent. A provenance line that triggers a remediation process will be gamed within two cycles; one that simply makes the structure visible tends to survive, because it costs the producer nothing to complete honestly.

Three tests then read the paper.

The beneficiary test. Does anyone in the estimating chain gain from approval? Almost always yes. The test is not diagnostic on its own — it is a prompt to look for the next two.

The refusal test. Has anyone in this process ever returned a number that killed a proposal, and what happened to them afterwards? This is the strongest single indicator available, and it is answerable from memory in most organisations. Where nobody can name an instance, the challenge function is decorative regardless of how it is described in the governance framework.

The prestige test. Is any part of the value here reputational — a first, a flagship, a category entry, a relationship with a client the enterprise wants to be seen serving? If so, is that value stated in the case, or is it being carried silently by an optimistic financial number doing work it was not designed to do? Reputational value is legitimate and often substantial. It should be argued on its own terms rather than smuggled in through the cost line.

From Strategy to Execution

Immediate. Add the provenance line to the next three approval papers, without changing anything else. Do not announce a policy; the point of the first three is to find out what the fields say when nobody has prepared for them.

Medium-term. Constitute a small challenge capability whose members are not drawn from the initiatives they review, and — this is the part that decides whether it works — give it standing rather than a process step. Standing means the review's objections are recorded in the approval paper whether or not they are accepted, and the approver signs having seen them. A challenge function whose findings can be resolved before the paper is written is not a challenge function. [Related article: Enough Technical Depth to Test the Answer]

Long-term. Decide, at board level, what the enterprise's posture is on the deal it is willing to lose. Every structural intervention here reduces the win rate somewhere. An organisation that will not name that price will keep the intervention in its governance framework and out of its behaviour, and will continue to be surprised by numbers that were always going to be wrong in the same direction. The alternative posture is also legitimate: accept the bias, price it explicitly through a standing adjustment, and stop describing the raw figure as a forecast.

Signals to Monitor

• Whether the variance between approved and outturn figures is symmetric. Persistent one-directional error is a structural signal, not a competence signal.
• The gap between the analyst's range and the figure in the approval paper — and where in the chain the range disappeared.
• How often an independent challenge changes a number materially, as against how often it changes wording.
• Whether the enterprise can name a proposal stopped by its own review process in the past two years.
• Bid win rates rising while margin realisation falls, which is the signature of the promise moving ahead of the delivery.
• Reputational or strategic value appearing in the narrative of a case but nowhere in its numbers.

Questions for the Leadership Team

1. For our three largest approved investments, who produced the central financial number, and what did that party want?
2. Can anyone here name an occasion when our review process stopped a proposal — and what happened to the person who raised the objection?
3. Where a delivery voice sits in a commercial conversation, have we accepted the cost the arrangement carries, or have we asked for it at no cost to win rate?
4. Is our challenge function staffed from inside the work it examines?
5. Are our forecast errors symmetric, and if not, what standing adjustment have we made?
6. Which of our current proposals carries reputational value that is not written down anywhere?

Closing Perspective

Nothing in this argument requires believing that estimators are unreliable. It requires only accepting that a number produced by an interested party is testimony, and that testimony is received differently from measurement.

The organisations that handle this well have not found a better model. They have changed who is in the room when the number is set, accepted the cost of doing so, and stopped asking a single figure to carry both a technical judgement and a commercial ambition. Everything else — the ranges, the contingency policies, the second opinions — operates downstream of that choice, and cannot substitute for it. [Related article: From Producer to Orchestrator] [Related article: The Estimating Loop Nobody Closes] [Related article: Your Scoring Model Decided Before the Bids Arrived] [Related article: Volunteers Are Not a Sample] [Related article: The Front End Owns the Outcome] [Related article: What Is the Risk of Not Doing It?] [Related article: How Does Work Get Into Your Enterprise Without a Decision?]

About EraNorth Insights
EraNorth Insights publishes practical analysis on strategy, projects, operations, transformation and decision intelligence for professional and organisational use. About EraNorth.

Copy link Print