Cost estimation techniques: what each method requires

A number is asked for before the project is defined. The reflex is to look for the right method, as if the choice belonged to the estimator. It rarely does: what can be produced at that point is already set by what is known, and the useful question is what may legitimately be said about the figure once it exists.

What an estimate is, and what it is not

An estimate is a quantitative assessment of the likely amount or outcome of a variable, such as project costs, resources, effort and durations (PMI, Practice Standard for Project Estimating, 2nd edition, 2019, glossary). It is not a measurement, and it is not a forecast that happens to be imprecise: it is a value produced on purpose from incomplete material. The Dictionnaire de management de projet (AFITEP, 3rd edition, AFNOR, 1996) describes the operation as giving the most probable value and duration of a task "from insufficient and often more or less false data", and attaches to it an obligation that the rest of this article rests on: those values are necessarily accompanied by their degree of confidence.

One consequence follows, and it is a referential that states it rather than us: "every point estimate value will likely prove to be wrong" (AACE International, Recommended Practice 18R-97, revision of 7 August 2020).

Cost and effort are not the same thing

That same 1996 dictionary carries the articulation in a single line: cost is the expression of a charge in monetary units, within a defined economic base, and is not to be confused with price. Its entry for charge, glossed as work load, defines it as a quantity of work to be performed, expressed in terms of the resource assigned to an action, usually in man-months or labour hours. English-language standards count the same quantity and call it effort. Cost sits downstream of it: it prices work that has first been counted.

Effort does not divide by headcount

Most cost and duration estimates rest on estimating the amount of effort an activity requires within the project scope, and the step from effort to duration is where a common shortcut fails. The 2019 practice standard states it and illustrates it: an activity requiring eight labour units may be performed by one resource over eight calendar days, or by two resources of similar performance over four, and "this relationship may not be linear". The PMBOK Guide, 6th Edition (2017) names three reasons the linearity breaks: the law of diminishing returns, the number of resources, since doubling them does not always halve the time, and motivation, with student syndrome and Parkinson's law. Effort is estimated; duration is derived from it, and not by division.

The methods, and what each one requires

There is no canonical list, and the differences are not cosmetic. Six methods in the GAO Cost Estimating and Assessment Guide in its 2009 edition, twenty-three techniques in three categories in the 2019 practice standard, five tools per process in the PMBOK Guide, 6th Edition, two generic categories and five sector-specific methods at AACE International, four applied item by item in the Washington State Department of Transportation manual, six in PRINCE2 7 (AXELOS, 2023). An article that announces "the estimating methods" owes its reader the house it takes its list from. This one takes its architecture from the PMI Practice Standard for Project Estimating, 2nd edition, 2019, and its conditions of use from Table 11 of the GAO guide in its 2009 edition.

That standard sorts techniques into three categories: quantitative, with analogous, parametric and bottom-up estimating; relative, with affinity grouping and planning poker; and qualitative, with expert judgment, observations, interviews and surveys. What follows covers the quantitative family and expert judgment. The relative family belongs to agile contexts and is left aside here.

Analogous estimating

Also known as top-down estimating. It prices a new item against a known one.

What it requires is little. It needs minimal project detail and is usually faster, easier and less expensive to implement. Table 11 of the GAO guide in its 2009 edition lists the same profile from the other side: few data required, based on actual data, reasonably quick, good audit trail.

What it produces is an order of magnitude, and a cross-check on estimates produced another way.

Its conditions of use, in the referentials' own words: it provides "no detailed basis for justifying decisions or estimates"; "an analogy relies on a single data point"; adjustments are subjective, accuracy depends on the similarity of the items, the effect of a design change is difficult to assess, and the method is blind to cost drivers.

The most directly reusable rule of the section is the reasonable person test set out by the GAO guide in its 2009 edition: the sources of the analogy and any adjustment must be logical, credible and acceptable to a reasonable person. Its own illustration is precise. An estimator who assumes that a new component will be 20 percent more complex, and cannot explain why, has produced an unacceptable adjustment factor. The complexity has to be related to the system's parameters.

Parametric estimating

Based on historical information from very similar projects, but taking scale differences into account. Its condition is suspensive, and the 2019 practice standard writes it that way: parametric estimates are more accurate and more reliable than analogous ones, "but only if a statistical relationship exists between the variables used to calculate the estimate".

What that relationship is, concretely, comes in three simple shapes in the GAO guide in its 2009 edition. A rate uses a parameter to predict cost through a multiplicative relationship, and its units are always dollars per something. A factor uses the cost of another element with a multiplier, often expressed as a percentage. A ratio is a function of another parameter, and is often used to estimate effort.

The validity condition that bites concerns the data behind the relationship: the attributes of the programme being estimated have to fall within, or at least not far outside, the dataset the cost estimating relationship was built on. The guide's own example is a one-million-line software program estimated with a relationship built on programs of 10,000 to 250,000 lines.

Its conditions of use, from Table 11: it lacks detail, it requires investment in the model, it meets cultural barriers, and it demands an understanding of how the model behaves.

Bottom-up estimating

Also called deterministic or detailed estimating, and recognised to produce the most accurate and most reliable estimate, which is why its result also carries the name definitive estimate.

What it requires is a list, and the list is the whole condition: a detailed work breakdown structure, an activity list, and a comprehensive directory of project resources. The work breakdown structure is taken as acquired here. In the GAO guide in its 2009 edition the method appears under the name engineering build-up, with the matching profile: it requires detailed design, and it is slow, laborious and cumbersome.

Its own limitation is systemic rather than accidental. Because it works from the activity list, it may overlook many of the system-level costs, integration, configuration management and quality assurance among them.

Expert judgment, which is not on the same plane

The 2019 practice standard places expert judgment in the qualitative category, not alongside the three quantitative families. The GAO guide in its 2009 edition defines it as expert opinion, also known as engineering judgment, commonly applied to fill gaps in a relatively detailed work breakdown structure when one or more experts are the only qualified source of information.

Two institutional positions face each other here, and this article does not settle them. The GAO guide in its 2009 edition reserves it for a plausibility check: "because of its subjectivity and lack of supporting documentation, expert opinion should be used sparingly and only as a sanity check". The 2019 practice standard makes it a category in its own right and carries it twice in its table of techniques. Two publications, two doctrines.

Three-point estimating, which is not a fourth method

It is not a method but a treatment, applicable to any quantitative technique. That is the point of this section, and it is what a flat list of five methods gets wrong.

The 2019 practice standard publishes two distinct formulas and names them both. The triangular distribution gives E = (O + ML + P) / 3. The PERT formula gives E = (O + 4ML + P) / 6, and the standard says why: it weights the most likely value by four times its value to reinforce its significance.

The same standard publishes a worked example: optimistic 66,000, most likely 100,000, pessimistic 210,000, PERT result 113,000. Applying to those three published values the triangular formula the standard prints on the previous page gives 125,333. That second figure is derived here rather than published there, and it is worth deriving: on three identical numbers, the weighting moves the result by more than twelve thousand, and downwards. That is what weighting the most likely value by four does in practice.

The criterion for reaching for it at all comes in one line: the wider the range between the pessimistic and the optimistic values, the greater the value in using the technique.

What decides which method can be used

How well a project is defined is a matter of degree rather than a yes or no, and it firms up progressively. That point is established in its own right elsewhere in this hub and is taken as acquired here.

Four houses then say the same thing in four vocabularies, and none of them cites the others. The checklist in the GAO guide in its 2009 edition orders the methods by phase of the life cycle: analogy early in the life cycle; expert opinion very early, if an estimate could be derived no other way; the build-up method later, in acquisition, when the scope of work was well defined and a complete work breakdown structure could be determined; parametrics where a database of sufficient size, quality and homogeneity was available; and extrapolation from actual cost data at the start of production. AACE International indexes its estimate classes on a percentage of project definition reached (Recommended Practice 17R-97, revision of 6 March 2019). The AFITEP dictionary of 1996 indexes its four classes on the study deliverable available, from a preliminary study to a project already under way. And the Washington State Department of Transportation, whose manual is one owner's internal doctrine rather than a general rule, writes that estimating methodologies are applied to individual bid items and may change as the project develops (Cost Estimating Manual for Projects, M 30-34.04, February 2026).

They do not agree on which upstream factor commands, and that is worth saying once. AACE International and the GAO guide point to the maturity of the definition, the 2019 practice standard points to the availability of data, and Fish, in the second edition of 1923, pointed to the purpose of the estimate, the data available and the time available. What they agree on is the shape: what decides sits upstream of the estimator's preference.

The choice is not made once for a whole project either. Any combination of methods may be found in any given class of estimate (AACE International, Recommended Practice 17R-97, revision of 6 March 2019); the AFITEP classes of 1996 combine methods item by item inside a single class; and the Washington State manual changes method bid item by bid item. Reading the axis project by project is the natural beginner's mistake, and it is the one thing the figure above is drawn to prevent.

Accuracy is a consequence, not a choice

Six referentials published between 1996 and 2026 converge on one requirement, and none of them argues for it. The AFITEP dictionary of 1996 makes the degree of confidence obligatory. The 2019 practice standard says not to rely on a single cost, value or schedule estimate, but to use several techniques or models, compare the results and determine the reasons for any large variations. The GAO guide in its 2009 edition says that high-quality cost estimates usually fall within a range of possible costs, the point estimate lying between the best and the worst case, and adds that "a point estimate, by itself, provides no information about the underlying uncertainty other than that it is the value chosen as most likely". The Washington State manual makes it a rule of public communication: an estimate is a range, not a single number. PRINCE2 7 (AXELOS, 2023) recommends including a level of confidence with every estimate. And AACE International writes that an estimate reflects a range of potential outcomes, each value in that range carrying a probability of occurrence (Recommended Practice 18R-97, revision of 7 August 2020).

One rule out of this is immediately applicable, and the 2019 practice standard states it as a defect to avoid: early estimates involve broad assumptions, yet estimating techniques often return very specific numbers, which gives an impression of accuracy that was never reached. Round to the next higher increment, and give a range.

Where that range turns into money is another subject. The range is a property of the estimate; the contingency and the reserve are lines of a budget, and they belong to cost management rather than to estimating.

What the number is for, on a contract project

Once produced, the figure does two things at once. On a public works contract the owner's estimate is prepared for the final contract review ahead of advertisement, and "it is used to obligate construction funds and to evaluate contractors' bids" (Washington State Department of Transportation, Cost Estimating Manual for Projects, M 30-34.04, February 2026, again as one owner's internal doctrine and not as a general rule).

This is where a vocabulary distinction has to hold. The AFITEP dictionary of 1996 is explicit that cost is not to be confused with price, and a contract text shows what that difference does. In the December 2000 edition of NF P 03-001, the French standard form for private building works, the contract price on a lump sum contract is the amount stated by the contractor in its letter of engagement or its tender and accepted by the owner, while on a remeasurement contract it is the unit prices of the schedule of rates. And the owner owes no additional payment where modified works have cost the contractor more than the works originally provided for. The cost can move without the price moving.

The calculation that produced the price is not what binds either. In that same edition, the only priced document ranked in the contractual order of precedence is the accepted letter of engagement or tender, and it ranks first. The detailed price breakdown is annexed without contractual value, while remaining usable to establish interim valuations and to value additions and omissions, unless the particular conditions of the contract provide otherwise.

The same exercise therefore yields two documents rather than one. The detailed estimate of any phase is kept confidential until the end of bid opening, while the blank schedule goes out to the bidders.

Hence a traceability requirement, on which three houses converge. NF ISO 21502:2021 recommends that estimates be justifiable. The PMBOK Guide, 6th Edition (2017) asks the supporting documentation to give a clear and complete understanding of how the estimate was derived, and lists what that takes: the basis of the estimate, the assumptions, the constraints, the range of possible estimates and the confidence level. The Washington State manual puts it in seven words: "An estimate without documentation is not an estimate."

Where it comes from

The methods were not born in a referential, and the formula that gets copied most did not come out of a statistical argument.

The first recorded measurement of the gap is from 1840. In a paper on British canals, Jones found an average ratio of actual cost to estimate of 2.79 (Jones, "Engineers and Their Estimates", Journal of the Franklin Institute, vol. XXV, 1840, reported by Hollmann, "The Early History of Cost Engineering", 2016 AACE International Technical Paper TCM.2104, which is the piece read here).

The taxonomy is older than the referentials that publish it. In the second edition of 1923 of Engineering Economics, Fish sets out five abbreviated procedures and orders them: unit cost of product, costs of parts already known, estimating by ratios, estimating by analogy, and overall unit costs. Facing them stands the detailed procedure, with its quantity take-offs, its unit prices and its indirect costs apportioned. What the PMI published in 2019 was named and ordered in 1923. The same chapter also gives another determinant of the choice, in three terms: the purpose of the estimate, the available data, and the time available for making it.

The name of the best-known formula is regularly got wrong. PERT stands for Program Evaluation and Review Technique (Malcolm, Roseboom, Clark and Fazar, "Application of a Technique for Research and Development Program Evaluation", Operations Research 7(5), 1959). The middle word is Evaluation.

Its origin is documented by one of that paper's own authors. In 1962, answering analysts who asked where the 1959 assertions came from, Clark wrote a letter to the editor of the same journal (Clark, "Letter to the Editor, The PERT Model for the Distribution of an Activity Time", Operations Research 10(3), 1962). Three things come out of it. Taking one sixth of the range as the standard deviation is a convention that was selected, introduced with "suppose we select". The expected value was obtained by empirical numerical manipulation, the exact calculation being judged ponderous relative to the reliability of the results. And the beta distribution was a convenience: it is the one that first came to the author's mind, and he writes that it is not suggested that the beta or any other distribution is appropriate.

The reason is a collection constraint rather than a statistical one. Estimates had to be produced periodically, formally and at low cost, for thousands of activities.

Public audit entered the subject ten years later. In 1972 the Comptroller General of the United States published Theory and Practice of Cost Estimating for Major Acquisitions (B-163058), which found that estimates of the cost to develop and produce weapon systems were frequently understated, and that without realism and objectivity in the estimating process bias and over-optimism creep in, pushing estimates too low. The GAO guide in its 2009 edition reprints that finding and says these characteristics are still valid today.

In short

Two moves come out of all this, and neither of them is a method.

Before reaching for a technique, look at what is actually on hand: how far definition has gone on the item being priced, and whether there is comparable data, a usable relationship, or a breakdown detailed enough to build on. The technique follows from that answer instead of preceding it.

And never hand over a figure on its own. What makes it usable by somebody else is the spread around it and the trail underneath: what was assumed, what was used, how much confidence it carries. A bare number is one that the next reader will treat as a commitment.

Stop guessing. See the real impact.

Frequently asked questions

Q.Is a cost estimate the same thing as a quoted price?

No. An estimate is an internal assessment of a probable value, built from incomplete data and given with a degree of confidence. A quote is a figure a supplier undertakes to hold and a buyer accepts. The first describes what is known; the second commits.

Q.Where does the uncertainty go once the estimate is done?

It changes owner. The range belongs to the estimate and says how far the figure may move. Once a budget is built, that uncertainty is carried by named envelopes, contingency and management reserve, which are objects of the budget rather than of the estimate.

Q.How much time should be spent estimating?

In proportion to what an error would cost. Fish, in the second edition of 1923, set the rule out: a large percentage error on a small item affects the total no more than a small percentage error on a large one, and time should be apportioned accordingly.

References

• AACE International - AACE International Recommended Practice No. 18R-97 - Cost Estimate Classification System - As Applied in Engineering, Procurement, and Construction for the Process Industries - Revision of 7 August 2020
• AACE International - AACE International Recommended Practice No. 17R-97 - Cost Estimate Classification System - Revision of 6 March 2019
• AACE International - John K. Hollmann - The Early History of Cost Engineering - 2016
• AFNOR - NF P 03-001 - Marchés privés, cahiers types, CCAG applicable aux travaux de bâtiment faisant l'objet de marchés privés - Édition de décembre 2000
• AFNOR - NF ISO 21502:2021 - Recommandations sur le management de projet - Juin 2021
• AFNOR, AFITEP - Dictionnaire de management de projet français-anglais-espagnol - 3e édition, 1996
• AXELOS - PRINCE2 7 - Managing Successful Projects - 2023
• Comptroller General of the United States - Theory and Practice of Cost Estimating for Major Acquisitions (B-163058) - 1972
• Franklin Institute - T. Jones - Engineers and Their Estimates - vol. XXV, 1840
• GAO - GAO-09-3SP Cost Estimating and Assessment Guide: Best Practices for Developing and Managing Capital Program Costs - March 2009
• INFORMS - D. G. Malcolm, J. H. Roseboom, C. E. Clark, W. Fazar - Application of a Technique for Research and Development Program Evaluation - vol. 7, no. 5, 1959
• INFORMS - Charles E. Clark - Letter to the Editor - The PERT Model for the Distribution of an Activity Time - 1962
• John Charles Lounsbury Fish - Engineering Economics: First Principles - 2nd edition, 2nd printing, 1923
• PMI - PMI Practice Standard for Project Estimating - 2nd edition, 2019
• PMI - A Guide to the Project Management Body of Knowledge (PMBOK Guide) - 6th Edition - 2017
• WSDOT - Cost Estimating Manual for Projects - M 30-34.04, February 2026