What it means
A Delphi exercise runs in rounds. A facilitator poses a question, collects individual answers privately, circulates an anonymised summary such as the median and the range, and then invites participants to revise, usually asking anyone holding an extreme view to explain their reasoning.
The anonymity is the whole point. In an open meeting, estimates converge on whatever the most confident or most senior person says first, whereas anonymous rounds let a quiet specialist with genuine insight move the group without having to win an argument.
Businesses use it for questions that resist modelling: how quickly a new technology will be adopted, what a market will be worth in five years, how long an untried project will take, or how likely a particular risk is. It also appears in risk assessment, capital planning and long-range demand forecasting where historical data simply does not exist.
The usual measure of progress is convergence. If the spread of estimates narrows across rounds while the central estimate stays reasonably stable, the panel has genuinely settled, and if the spread stays wide the honest conclusion is that the question is uncertain rather than that more rounds are needed.
The method has real limits. It is slow, typically taking two to four weeks for three rounds, the result depends entirely on the quality of the panel, and pushing too hard for consensus can produce a comfortable middle number that nobody actually believes.
In practice
Real-world examples.
Example
A medical device company needs a five-year demand forecast for a product category that does not yet exist commercially. It runs three Delphi rounds with twelve clinicians, regulators and distributors, and uses the converged median as the base case in its investment appraisal.
Example
An insurer estimating the frequency of a rare operational loss event convenes an anonymous expert panel because internal loss data contains only two incidents in a decade. The exercise produces a range rather than a point estimate, which is fed straight into the capital model.
Example
A city transport authority uses a Delphi panel to estimate how long a complex tunnelling project will take, deliberately keeping responses anonymous so junior engineers are not swayed by the programme director's public optimism. The panel's median estimate is nine months longer than the official plan, and the schedule is revised.
Formula
Calculation
There is no single algebraic formula. The standard measures per round are the median estimate and the spread, calculated as: spread = highest estimate - lowest estimate, with convergence judged by the fall in spread between rounds.
Worked example. A company asks seven industry experts to estimate next year's revenue for a new product line, in millions of dollars.
Round 1 responses: $8m, $10m, $11m, $12m, $14m, $18m, $25m. The sum is $98m, so the mean is $98m / 7 = $14m. The median, the fourth value in order, is $12m, and the spread is $25m - $8m = $17m.
Participants receive the anonymised summary and the written reasoning behind the $8m and $25m answers, then revise.
Round 2 responses: $10m, $11m, $11m, $12m, $12m, $13m, $15m. The sum is $84m, so the mean is $84m / 7 = $12m. The median is still $12m and the spread is now $15m - $10m = $5m.
The mean has moved from $14m to $12m as the outlier pulled back, the median has held steady at $12m, and the spread has fallen from $17m to $5m, a reduction of $12m or 70.6%. That pattern, a stable centre with a narrowing range, is what convergence looks like, and the panel can reasonably stop at $12m.Case study
Seen in the real world.
Copperline Robotics is an invented company used here for an illustrative purpose only. Its planning meetings for a new product forecast always ended the same way: the chief executive gave a number, everyone agreed with a small variation around it, and the plan was built on that figure. The last three launches had missed the forecast by an average of 40%.
The head of strategy ran a Delphi exercise instead, using fifteen participants drawn from sales, engineering, service and two external distributors, with responses collected privately. Round one produced estimates ranging from 4,000 to 30,000 units. After two more rounds and circulated reasoning, the range narrowed to 7,000 to 13,000 with a median of 9,000, less than half the number the leadership team had assumed.
Copperline planned production capacity around the Delphi median with an option to expand. Actual first-year sales came in at 9,800 units, and the company avoided both the inventory write-down and the cash strain that a plan for 20,000 units would have produced. The illustrative lesson is that removing hierarchy from a forecast can be worth more than adding another spreadsheet.
Watch out
Common mistakes.
- Letting participants see who said what. Once names are attached, the seniority effect the method exists to remove comes straight back.
- Running rounds until everyone agrees. Forced consensus produces a comfortable average rather than an honest estimate, and a persistent wide spread is itself a useful finding.
- Assembling a panel of people who all share the same background. Convergence among similar thinkers is easy to achieve and tells you very little.
Questions
People also ask.
How many experts and rounds does a Delphi exercise need?
Panels of roughly eight to twenty participants and two or three rounds are typical, since additional rounds usually shift the median very little.
Is the Delphi method suitable when good historical data exists?
Usually not, because statistical forecasting will be faster and more reliable, and Delphi earns its keep where data is absent or the future is structurally different from the past.
How do you know when to stop?
Stop when the spread of estimates stops narrowing meaningfully between rounds, which indicates the panel has settled rather than merely tired.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%Related