Some estimates can be built from records: the business has done the job many times, and past costs and durations are a good guide. Others cannot. A first job of a new kind, a new product, a new market, an unfamiliar site or technology: there is little or no data, and the estimate has to come from people’s judgement. That is normal, and judgement can be good. The trouble is how it is usually gathered.
The common method is to ask the most senior or confident person, or to discuss it in a meeting until a number emerges. Both are vulnerable to predictable errors. The first number spoken becomes an anchor that pulls every later estimate towards it. People defer to seniority or to whoever sounds most certain. Ranges are too narrow, because people underestimate how much they do not know. Optimism creeps in, especially when the estimate affects a sale someone wants to win. The resulting number looks like a team view but is often one person’s guess, slightly adjusted.
This article explains the biases that distort expert estimates, simple ways to structure judgement so it is more reliable, how to run a basic Delphi process, why ranges beat single numbers, and how to record the basis of an estimate so it can be reviewed and improved. It is general information for owners and managers who quote, plan or invest in work without much history to rely on.
Why unstructured judgement goes wrong
- Anchoring. The first figure mentioned, from a customer, a boss or an early guess, shapes everything after it, even when people know it was arbitrary.
- Deference. Junior people adjust towards senior ones, and quieter people towards confident ones. Useful knowledge stays unspoken.
- Overconfidence. Estimated ranges are usually much narrower than the real spread of outcomes.
- Optimism. People imagine the work going as planned and underweight the many ways it can be delayed.
- Motivated estimates. When an estimate affects winning a job or approval for a project, it tends to drift towards the answer that helps.
None of these means experts are poor estimators. It means the process of gathering their views needs some structure.
Five ways to structure judgement
1. Break the question down
A large uncertain number is easier to estimate as smaller parts. Instead of “how long will the fit-out take?”, estimate the parts: site preparation, each area, access constraints, finishing, defects. Some parts may have data even when the whole does not. The estimating project costs from cost drivers article covers building estimates from what actually causes cost.
2. Look for an outside view
Before relying on your team’s view of this particular job, ask what usually happens with jobs like it. Other businesses, suppliers, industry associations or your own records of loosely similar work can give a starting range. Then adjust for what is genuinely different. Starting from how similar work usually turns out is a strong guard against optimism.
3. Gather independent estimates before discussion
Ask each person to write down an estimate, with a range and their reasons, before anyone speaks. Then compare and discuss, focusing on the reasons behind different numbers. Then let each person revise privately. This simple “estimate, talk, estimate” sequence keeps the value of discussion while reducing anchoring and deference.
4. Ask for ranges, not points
Ask for a low, a most likely and a high figure, or ask “what is the figure you are 90% sure we will not exceed, and 90% sure we will exceed?”. A range captures uncertainty that a single number hides, and gives the business something to plan contingency against. The how confident is that finish date article covers combining three-point estimates into an overall forecast.
5. Record the reasons
The reasons behind an estimate are often more valuable than the number. They show which assumptions matter most, which can be checked, and which would change the estimate if they turned out wrong.
Choose the right people
The quality of judgement depends on who gives it. Include people with hands-on experience of the work, not only those who manage or sell it. Include at least one person likely to be sceptical, and, where possible, someone from outside the business who has done similar work. Ask people to say if they have an interest in the answer, such as a salesperson whose commission depends on the quote. Interests do not disqualify anyone, but they should be visible.
A simple Delphi process
The Delphi method is a structured way of combining several experts’ judgements over a few rounds, with anonymity to reduce deference. A small business version:
- Choose a panel of four to eight people with relevant knowledge, including some from outside the immediate team if possible, such as a subcontractor or supplier who has done similar work.
- Ask a precise question, with the units, scope and assumptions spelled out.
- Round one: each person gives an estimate, a range and their main reasons, independently and anonymously, for example on a form collected by a neutral facilitator.
- Summarise and share: the facilitator shares the median, the spread and the main reasons given, without names.
- Round two: each person reconsiders and gives a revised estimate, explaining any change or why they are holding their view.
- Repeat once more if views are still moving significantly, then stop.
- Report the result as a range with the median, and note any remaining disagreement and its reasons.
Two cautions. The aim is not forced agreement: a persistent minority view with a good reason is valuable information and should be recorded. And a single survey or a one-off meeting is not a Delphi; the method’s value comes from anonymous, reasoned feedback across rounds.
Record the basis of the estimate
A basis of estimate is a short record that travels with the number:
| Item | What to record |
|---|---|
| Scope | Exactly what the estimate covers and excludes |
| Method | Decomposition, outside view, Delphi, expert opinion or data |
| Sources | People consulted, past jobs, supplier quotes, published data |
| Key assumptions | The few assumptions that drive the result |
| Range and confidence | Low, most likely and high, and how confident the team is |
| Date and owner | When it was made and who is responsible for updating it |
With a basis, an estimate can be reviewed, challenged and updated when assumptions change. Without one, it becomes a number nobody can explain six months later. The from estimate to commitment article covers turning a range into a commitment deliberately.
Keep the estimate separate from the price
An estimate is a forecast of what the work will take. A price or a commitment is a business decision that also reflects margin, competition and appetite for risk. Mixing the two produces estimates shaped by what the customer will pay. Make the estimate first, honestly, as a range. Then decide the price or commitment separately and record why it sits where it does. If the business chooses to quote below the realistic estimate to win work, that should be a visible decision, not a hidden adjustment to the estimate.
Calibrate your team over time
Calibration means checking whether your ranges are as reliable as you think. If the team gives ranges it is 90% sure of, actual outcomes should fall outside them roughly one time in ten. Keep a simple record of estimates and outcomes. If outcomes fall outside the ranges far more often, the team is overconfident and should widen its ranges. People become better estimators surprisingly quickly when they get regular feedback on how their estimates turned out.
Update as evidence arrives
Judgement is a starting point, not a final answer. As the work begins and real data arrives, compare it with the estimate and revise. Early results from a pilot, the first stage of a job or the first weeks of sales are often more informative than any amount of further discussion. Keep track of how the business’s past estimates compared with outcomes. Over time, that record tells you whether your team tends to be optimistic, by how much, and for which kinds of work.
A worked example
This is an illustration. A joinery business is asked to quote for its first fit-out inside a heritage-listed building: custom shelving and counters in an old bank building being converted to a café. The business has fitted out many modern premises, but never a heritage building, where walls are uneven, fixings may need approval, and access is limited to certain hours. The question is how many installation hours to allow.
The owner sets up a quick Delphi. The panel is two senior installers, the project manager, the owner and a subcontracted plasterer who has worked in heritage buildings. The office manager acts as facilitator.
Round one. Each person independently writes an estimate of installation hours, with a range and reasons. The estimates are 220, 260, 300, 340 and 480 hours. The reasons include uneven walls needing scribing, restricted access hours, uncertainty about whether fixings into original walls will be allowed, and the plasterer’s experience that heritage jobs usually need extra protection and clean-up.
Feedback. The facilitator shares the median of 300 hours, the range and the reasons, without names.
Round two. Revised estimates are 280, 300, 320, 360 and 420 hours. The median moves to 320. The plasterer keeps the highest estimate, explaining that heritage approvals for fixings often take weeks and can force design changes. That reason is recorded.
Basis of estimate. The owner records the scope, the method, the panel, the key assumptions (fixings approved as designed, access five days a week, walls within a stated tolerance) and the range of 280 to 420 hours, with 320 as the most likely.
The quote. The business prices installation at 340 hours, slightly above the median given the newness of the work, and states clearly that delays or redesign caused by heritage approvals are excluded and will be priced as variations.
Installation takes about 345 hours. One fixing detail needs redesign after a heritage review, and it is priced as a variation as agreed. The business records the outcome against the estimate, so the next heritage job starts with real data.
How this applies to a small Australian business
- Break uncertain estimates into parts.
- Start from how similar work usually turns out, then adjust.
- Collect independent estimates before discussing.
- Ask for ranges and the reasons behind them.
- Use a simple Delphi for important first-of-a-kind estimates.
- Record a basis of estimate with assumptions and sources.
- State key assumptions and exclusions in quotes.
- Compare estimates with outcomes to learn your team’s tendencies.
- Decide prices separately from estimates, and record why.
Signals worth watching
- Estimates that match the first number anyone mentioned.
- The most senior person’s estimate always prevailing.
- Single-figure estimates with no range.
- Nobody able to explain how an estimate was made, or by whom.
- Consistent overruns on new kinds of work.
- Estimates that drift towards what wins the job.
- Outcomes regularly falling outside stated ranges.
Common mistakes
- Discussing before estimating independently.
- Treating a confident voice as evidence.
- Ranges that are far too narrow.
- Calling a single survey a Delphi.
- Losing the reasons behind the numbers.
- Never comparing estimates with outcomes.
- Adjusting the estimate to fit the price you want to quote.
Frequently asked questions
How many people do we need for a Delphi? Four to eight is usually enough for a small business. Diversity of experience matters more than numbers.
Does anonymity really matter in a small team? It helps even when people could guess who said what, because writing estimates privately before seeing others’ views reduces anchoring and deference.
How long does this take? A simple two-round Delphi can be done in a day or two by email or short forms.
What if the panel cannot agree? Report the range and the reasons for disagreement. Persistent disagreement often points to the assumption that most needs checking.
Can we use this for sales forecasts too? Yes. The same methods work for forecasting demand for a new product, the take-up of a new service or the response to a price change.
Should we use the median or the most likely value? For planning, use the range. For a single figure, the median of a well-run panel is a sensible starting point, adjusted for the business’s appetite for risk.
Questions to ask
- What part of this estimate do we have data for, and what is pure judgement?
- How do jobs like this usually turn out, elsewhere or in our records?
- Did everyone estimate independently before discussing?
- What range would surprise us, low and high?
- Which assumptions would change the estimate most?
- How did our last estimate of this kind compare with the outcome?
- Who on the panel has an interest in the answer?
Bringing it together
When data is thin, estimates rest on judgement, and judgement is easily distorted by anchoring, deference, optimism and overconfidence. Structure it: break the question down, start from how similar work usually turns out, gather independent estimates with reasons before discussing, and ask for ranges. For important first-of-a-kind decisions, a simple Delphi over two or three anonymous rounds combines several experts’ views well. Record a basis of estimate, state assumptions in quotes, keep estimates separate from prices, and compare estimates with outcomes so the business’s judgement improves over time.
Source: KEVOS notes, drawing on earlier KEVOS handbooks on duration and cost estimating worksheets, bases of estimates and the Delphi method in practice. Examples and figures in this article are illustrations. This article is general information.
