Every week, business owners are asked to accept numbers they did not produce: a supplier’s recovery plan for a late job, a manager’s forecast for a new product, a contractor’s estimate for an expansion, a salesperson’s view of how many quotes will convert. Each number arrives looking like a measurement. Each was actually produced by someone, and that someone often has a stake in the answer.
When forecasts turn out to be badly wrong, the usual response is to improve the method: better spreadsheets, ranges instead of single figures, a contingency policy. These help, but they often miss the main cause. In most cases, the people who prepared the estimate are the people who needed the proposal to go ahead. They are rarely dishonest. That is exactly the problem: the effect works without anyone lying, and it survives every improvement in method that leaves the incentive unchanged.
This article explains how to treat an estimate as testimony from someone with a position, how to test a forecast against what has actually been achieved, how to separate facts from assumptions and hopes, and a few simple habits that make numbers more trustworthy without slowing every decision down.
An estimate is testimony
A useful shift is to stop treating an estimate as a measurement and start treating it as testimony: a statement made by an identifiable person who has a view on the outcome. Testimony is not worthless. Courts, boards and lenders rely on it constantly. But nobody sensible receives it without asking who is speaking and what they stand to gain, and asking is not an accusation.
For each important number in a proposal, ask:
- Who produced it?
- What outcome did they want?
- Who checked it from outside the proposal?
- What could the checker change?
Where nobody independent has checked a figure, say so. Simply recording it lets the decision-maker weigh the number accordingly.
The pattern has been noticed for a long time. A 2005 report in The Economist on project performance observed that problems arise most often when the people who win work are separated from the people who deliver it, because winning pushes bidders towards optimistic assumptions. It linked this to what Harvard’s Max Bazerman calls self-serving bias, the same tendency he uses to explain why capable auditors can sign off poor accounts.
What gets blamed instead
When a forecast fails, three explanations usually arrive first, and each is partly true:
- “The estimate was too early.” Often fair. Research by Kul Uppal, published in Cost Engineering in 2002, identified poorly defined requirements and failure to recognise invalid assumptions behind them as major causes of rework in engineering and construction projects. Note the second point: the problem was not only that assumptions were wrong, but that nobody recognised they were assumptions.
- “The scope changed.” Often true, but it does not explain an original number that was wrong on the day it was written.
- “We should have used a range.” Better than a single figure, but ranges have a habit of collapsing back to one number somewhere between the analyst and the approval.
None of these is a lie. All of them place the fault in method or events, and leave aside who produced the number and what they needed it to say.
Each input defensible, the total impossible
The effect is hard to spot line by line. Consider a business case for a new service. The team assumes a customer take-up rate from the best month of a pilot, a staff productivity figure from the most experienced person and a cost figure supplied by the preferred vendor. Each input is defensible on its own. Each sits at the optimistic end of its range. No single choice is worth arguing about, but together they produce a total that will not survive real conditions, and every input was chosen in good faith.
That is why line-by-line checking rarely finds the problem. The useful question is about the whole: if every input is slightly optimistic, what does the total look like?
Check the rate the claim depends on
Most forecasts depend on a rate: productivity, conversion, uptime, take-up, defect rate, hours per unit. A simple, powerful test is to compare the rate the claim requires with the rate the business has actually achieved.
Writing in PM Network in 1998, John Sahlin described a project review where a contractor’s performance had been running at around 70% efficiency, yet the recovery plan claimed the project would finish at 92% overall. Given how much of the budget had already been used at 70%, reaching 92% would have required the remaining work to be done at a rate far beyond anything achieved, and possibly beyond what was possible at all. The plan was not ambitious. It was arithmetically unavailable.
You do not need to be able to build the model to ask this question. You only need to find the rate the claim rests on and ask whether it has ever been achieved.
Four questions for any forecast
- What exactly is being claimed? One sentence, with a number and a date. If the presenter cannot state it, the report describes activity rather than an outcome.
- What rate does the claim require, and what rate have we achieved? This is the key test. Most unsound claims fail here.
- What would have to change to make that rate achievable? A credible recovery names a specific change: more people, less scope, a removed obstacle. An incredible one names effort, focus or commitment.
- If this is wrong, when will we find out, and what will it have cost by then? The answer tells you how hard to test the claim now.
A useful supporting rule: every forward-looking claim should state the rate it depends on, next to the rate actually achieved. One line on a page: “This forecast assumes ___; our achieved rate over the last two months was ___.”
Separate facts, estimates, assumptions and preferences
Confidence is a human signal, not evidence. In any significant proposal, ask the author to label what is:
- observed: facts from records;
- calculated: derived from observed facts;
- assumed: believed but not yet tested;
- preferred: a judgement about what the business should want.
This is not bureaucracy. It shows where further evidence could change the decision and where disagreement is really about values. It also helps separate the quality of a decision from its outcome. A good decision can meet bad luck, and a poor decision can be rescued by good luck. Judge decisions by whether the reasoning and evidence were sound at the time, not only by how they turned out.
Put the person who delivers into the room where it is promised
Two practical changes alter the incentive rather than just the method.
Involve the people who will deliver. The 2005 Economist report described Siemens placing project managers in sales teams to temper over-optimistic promises, and acknowledged the difficult balance between restraining salespeople and losing the deal. That trade-off is real. A delivery voice in a sales conversation will lose some work that would otherwise have been won. The business is choosing fewer wins and fewer of the wins that lose money. In a small business, that might mean the person who runs the job reviews the quote before it goes out.
Have someone without a stake challenge the numbers. Writing in Cost Engineering in 2003, Nick Lavingia described peer reviews intended to challenge a project team’s assumptions, alternatives and reasoning, carried out by peers not associated with the project. That independence is the point. A review by people who want the project to proceed is a technical check, not a challenge.
Three quick tests of the estimating culture
- The beneficiary test: does anyone who prepared the estimate gain from approval? Almost always yes. That is a prompt, not a verdict.
- The refusal test: has anyone ever produced a number that stopped a proposal, and what happened to them? If nobody can remember an example, challenge is not really happening.
- The reputation test: is part of the attraction prestige, such as a landmark customer or a first for the business? If so, state that value openly rather than letting an optimistic financial figure carry it silently.
A worked example
This is an illustration. A fit-out business has subcontracted the joinery for a large office job to a specialist. The joinery package was budgeted at 1,000 hours. After 600 hours, the work completed is worth 420 budgeted hours. The subcontractor’s monthly report says productivity has been poor but “a recovery plan will bring the package in on budget”.
The owner applies the four questions:
- The claim: the package will be finished within 1,000 hours.
- The required rate: 580 budgeted hours of work remain, with 400 hours left in the budget. The remaining work would have to be done at 1.45 times the planned rate. The subcontractor has achieved 0.7 times the planned rate so far, and its best month was 0.95.
- What would have to change: the plan names “extra focus” and “experienced staff”. It names no specific change in method, scope or crew.
- When we will find out: within about six weeks, by which time the fit-out’s handover date would be at risk.
At the achieved rate, the remaining work needs about 829 hours, a total of about 1,429 hours. Even at the subcontractor’s best month, it needs about 611 hours, a total of about 1,211 hours. The recovery plan is not credible.
The owner asks for a revised forecast stating its assumed rate, agrees an additional crew for the critical rooms, re-sequences the work and resets the handover plan with the client early, rather than discovering the problem in six weeks.
The owner also looks inward. The original 1,000-hour figure came from the estimator who won the job, using a productivity rate from the business’s best previous project. The business now requires every estimate for packages above a set value to state the productivity rate it assumes, alongside the average rate achieved on the last five similar jobs, and to be reviewed by the operations manager, who will have to deliver it.
How this applies to a small Australian business
Small businesses rely heavily on estimates from suppliers, contractors, staff and themselves. Practical steps:
- Ask who produced each important number and what they wanted.
- Find the rate each forecast depends on and compare it with what has been achieved.
- Require a specific change behind any recovery plan.
- Label facts, estimates, assumptions and preferences.
- Have the person who will deliver review quotes and estimates before they go out.
- Ask someone without a stake to challenge significant numbers.
- Judge decisions by their reasoning, not only their outcomes.
The estimating project costs from cost drivers, plans that detect rather than predict and reports that change decisions articles cover related practices.
Signals worth watching
- Recovery plans that rely on effort rather than specific changes.
- Forecasts with no stated assumptions.
- Every input in a proposal sitting at the optimistic end.
- Estimates prepared and reviewed only by people who want approval.
- Nobody ever having stopped a proposal with a number.
- Overruns always explained by scope changes or bad luck.
Common mistakes
- Treating estimates as measurements.
- Checking line by line but never the total.
- Accepting a recovery rate that has never been achieved.
- Mixing facts, assumptions and preferences.
- Judging decisions only by outcomes.
- Improving the method while leaving the incentive untouched.
Frequently asked questions
Isn’t questioning estimates a sign of distrust? No more than asking for a second quote is. Asking who produced a number and how it was tested is normal diligence, and good estimators welcome it.
What if I don’t understand the technical detail? You do not need to. Find the rate the claim depends on and compare it with the record. That question can be asked of any forecast.
How do we stop staff inflating estimates to protect themselves instead? Compare estimates with actual results over time, for both over- and under-estimates, and discuss the pattern openly. The aim is accuracy, not caution.
Should we always adjust optimistic forecasts down? Where you have a record showing consistent optimism, a standing adjustment based on that record is reasonable. Better still is fixing the process that produces the optimism.
Who should review significant estimates in a small business? The person who will have to deliver the work, and occasionally someone outside the business, such as an adviser or an experienced peer.
Questions to ask
- Who produced the most important number in our current proposal, and what did they want?
- What rate does our main forecast depend on, and have we ever achieved it?
- Which recovery plans rely on effort rather than specific change?
- Has anyone ever stopped a proposal with a number here?
- Do our proposals separate facts from assumptions and preferences?
- Who without a stake checks our significant estimates?
Bringing it together
Numbers come from people, and people often need a particular answer. Treat estimates as testimony, ask who produced them and who checked them, find the rate each forecast depends on and compare it with what has actually been achieved, and require specific changes behind any recovery plan. Separate facts from assumptions and preferences, involve the people who will deliver, and invite challenge from someone without a stake. You do not need to build the model to test it. You only need to know where to point.
Source: KEVOS notes, drawing on reporting in The Economist (2005) on project performance, K. Uppal (2002) and N. Lavingia (2003) in Cost Engineering, and J. P. Sahlin (1998) in PM Network. Examples and figures in this article are illustrations.
