The five AACE classes describe how well a project is defined. What they quietly set is how deep the work beneath your rates has to go — and the point at which carrying a benchmark stops being a defensible choice.
AACE International’s cost estimate classification system sorts capital project estimates into five classes according to the maturity of the project definition available when the estimate was prepared. Class 5 sits at the beginning, Class 1 at the end. The generic recommended practice, 17R-97, was first published in 1997 and exists to replace a vocabulary that had become useless: “order of magnitude”, “budget”, “definitive” and “detailed” all meant something different in every organisation that used them.
The system rests on one determination. Definition maturity is the primary characteristic; end usage, estimating methodology, effort and expected accuracy are all secondary characteristics that follow from it. And the assessment is made against the status of specific deliverables in a maturity matrix — has the geotechnical investigation been done, is the alignment fixed, are the structural arrangements settled — not against a single figure for percentage design complete.
That is a stricter idea than it first appears, and it is the reason this page is written from the build-up upwards rather than from the definition downwards. A class is a claim about what you knew, and every claim about what you knew implies a claim about how the rates beneath it were produced. The methodology column in the classification table is not advice. It is the observable evidence that the class you have written on the cover is the class you actually delivered.
TX1:Trinity lets a project start as parametric formulas and mature item by item into full resource build-ups, in the same file, without a rebuild at every gate.
Start a free 14-day trial17R-97 is deliberately industry-independent. It defines the five classes and the principle behind them, and organisations without a sector-specific practice available should apply it directly. Everything else in the family adapts that skeleton by supplying deliverable maturity matrices for a particular sector, because the deliverables that signal maturity in a refinery have nothing in common with the ones that signal it on a rail corridor.
18R-97 was the first of those industry adaptations, covering the process industries — oil, gas and petrochemical — and its matrices are built around process flow diagrams, piping and instrumentation diagrams, equipment lists, plot plans and piping take-offs. It is also the source of the accuracy ranges most often quoted, which matters when those figures are reproduced for a sector they were never measured on.
For Australian transport infrastructure the relevant practice is 98R-18, published in 2020 for road and rail, with matrices written around corridor alignments, pavement design, bridge and tunnel details and traffic management. Others follow the same pattern: 56R-08 for building and general construction, 97R-18 for pipelines, 96R-18 for power transmission line infrastructure. Queensland’s PCEM phases align with AACE classification principles, which is why the two frameworks are usually cited side by side on Australian projects and why an estimate built to the PCEM structure can be classified without rework.
The generic practice. Defines the five classes, the primary characteristic and the secondary ones. Use it where no sector practice exists.
Process industries, and the first industry-specific practice. Source of the accuracy ranges most commonly quoted in the general literature.
Road and rail transport infrastructure, published 2020. The matrix Australian civil estimators should be assessing against.
Each class is described by five attributes: definition maturity, typical end usage, estimating methodology, the effort required, and the expected accuracy range. The maturity bands overlap on purpose, because different industries and project types cross the thresholds at different points of engineering completion. The values below are the process-industry figures from 18R-97; sector practices such as 98R-18 set their own.
| Estimate Class | Definition Maturity | Typical End Usage | Estimating Methodology | Typical Accuracy Range |
|---|---|---|---|---|
| Class 5 | 0% to 2% | Concept screening, feasibility | Capacity factored, parametric models, judgement | −50% to +100% |
| Class 4 | 1% to 15% | Feasibility studies, concept evaluation | Equipment factored, parametric models | −30% to +50% |
| Class 3 | 10% to 40% | Budget authorisation, funding approval | Semi-detailed unit costs with line items | −20% to +30% |
| Class 2 | 30% to 75% | Control baseline, bid evaluation | Detailed unit costs with forced quantities | −15% to +20% |
| Class 1 | 65% to 100% | Check estimates, bid validation, change orders | Detailed unit costs with detailed material take-offs | −10% to +15% |
The one thing to take from this page: estimate quality is set by the project definition available to the estimator, not by the effort spent, the software used or the time taken — so the honest move when definition is thin is to classify accordingly, not to work harder and claim a class the information cannot support.
Read the methodology column as an instruction rather than a description and the classes separate into two groups. Class 5 and Class 4 are factored: capacity or equipment factors, parametric relationships and judgement applied to a scope that is barely drawn. Benchmark rates are not a compromise at this end, they are the correct tool, because there is nothing under the scope for a build-up to attach to. Class 3, Class 2 and Class 1 are unit-cost estimates: semi-detailed with line items, then detailed with forced quantities, then detailed with full material take-offs.
Class 3 is the hinge, and it is also the class that carries the heaviest decision. Its end usage is budget authorisation and funding approval — the investment decision. From Class 3 onward the estimate must be decomposed into real items with real quantities, and a rate that came out of a benchmark table cannot respond to the construction method, crew or site constraint that the reviewer will ask about. By Class 2 the methodology is explicitly detailed unit costs, and by Class 1 it is detailed take-offs behind every rate. There is no version of those two classes that a pasted schedule of rates can honestly produce.
So the depth of the build-up is not a matter of estimating philosophy. It is set by the class you have been asked for, and the class is set by the definition that exists. An estimate labelled Class 2 whose rates are all carried forward from a benchmark library is not a Class 2 estimate with a minor documentation gap. It is a Class 4 estimate wearing a Class 2 label, and the error that matters is not the rate — it is the accuracy range everyone downstream is now relying on.
Turned around, the classification table becomes a specification for the work beneath the rates — which is the form in which it is actually useful to an estimator deciding where to spend the next fortnight.
| Class | What the rates can honestly be | What the estimate must show |
|---|---|---|
| Class 5 | Factored and parametric throughout. Benchmarks are the correct method, not a shortcut. | The factor logic, the analogue project and why it is comparable. |
| Class 4 | Parametric, with a handful of build-ups where one package dominates the cost. | Which elements are factored and which are priced, and the basis of each. |
| Class 3 | Line items with quantities; build-ups for the packages carrying the cost, benchmarks on the tail. | The method assumed, the production evidence for the major items, and the split of built versus carried rates. |
| Class 2 | Detailed unit costs across the schedule, built from resources and crews. | Resource-level build-ups, the construction methodology they assume, and a priced indirect structure. |
| Class 1 | Detailed unit costs behind measured take-off, at tender or check-estimate depth. | Take-off traceable to drawings, quotations for major supply, and every rate openable to its components. |
Mixed maturity is the normal case. Disciplines rarely advance together; the pavement may be settled while the structures are still at options stage. Where definition is uneven, the estimate should take its class from the weakest area that materially affects cost, not from the average. Averaging is how a project ends up reporting a Class 3 range over a Class 4 unknown, and the unknown is usually the expensive one.
The ranges are typical, not promised. AACE is explicit that published accuracy ranges are values observed across many completed projects, generally expressed at an 80 per cent confidence interval, and that the range for any particular estimate must be established by risk analysis against that project — the complexity, the novelty, the ground, the market. The ranges also assume contingency has been applied; without it the true spread is wider. AACE’s contingency practices, 40R-08 and 44R-08, are the companion documents, and the distribution they produce is what a P50 or P90 figure is drawn from.
The asymmetry is deliberate. Every range runs further up than down, because capital projects overrun more often and by more than they underrun, and the effect is strongest where definition is thinnest. An estimator who presents a symmetric range at low maturity has quietly removed the most reliable finding in the whole field.
You cannot buy a class with effort. More hours on a Class 4 scope produces a more carefully argued Class 4 estimate. If the geotechnical investigation has not been done, no amount of first-principles build-up will tell you what is in the ground — though it will tell you, precisely, what the answer depends on, which is the most valuable thing an early estimate can deliver.
The practical difficulty with classification is that most teams rebuild the estimate at every gate, because the tool that suited a factored Class 4 cannot hold a resource-level Class 2. TX1:Trinity is built to make that one continuous file. Early work is expressed as formulas over project parametrics — #name variables, scoped parametrics, assignment with set(#VAR, expr), and more than seventy functions over a dependency graph with cycle detection — which is exactly the machinery a capacity-factored or equipment-factored estimate needs.
As definition arrives, items are replaced one at a time with resource build-ups: typed resources for labour, plant, material, subcontract and other, assembled into group resources whose published rate is the sum of their component rows. Nothing else in the project has to be rewritten to accommodate that, so the estimate can hold factored items and fully built items side by side — which is the real state of every Class 3 estimate ever produced, and worth being able to show honestly rather than disguise.
Because it is a local-first Windows application over an EF Core and SQLite store, each gate submission can be kept as a .tx1 package — compressed, checksum-validated, AES-256 optional — so the Class 4 you issued and the Class 2 you issued eighteen months later are both recoverable and comparable. Code sets let the same costs report through several classification trees at once, which is what makes a maturity assessment auditable rather than asserted.
It is a framework published by AACE International that sorts capital project cost estimates into five classes according to how mature the project definition was when the estimate was prepared. Class 5 is the least defined and Class 1 the most. The generic recommended practice is 17R-97, first published in 1997, and it replaced older labels such as order of magnitude, budget and definitive, which meant different things to different organisations. The point of the system is to give owners, designers, contractors and financiers one vocabulary for estimate reliability.
The maturity of the project definition, and nothing else. End usage, methodology, effort and expected accuracy are all secondary characteristics that follow from it. The assessment is made against the status of specific deliverables in an industry maturity matrix, not against a single percentage of design completion. This is why a class cannot be raised by spending more hours: without the deliverables, the information the higher class depends on simply does not exist yet.
RP 98R-18, the classification system as applied in engineering, procurement and construction for the road and rail transportation infrastructure industries. Published in 2020, it provides deliverable maturity matrices built around transport deliverables such as corridor alignments, pavement design, bridge and tunnel details and traffic management. For process plant work the reference is 18R-97, for buildings 56R-08, for pipelines 97R-18 and for transmission line infrastructure 96R-18.
Partly, but not entirely. Class 3 methodology is described as semi-detailed unit costs with line items, so the schedule has to be decomposed into real items with real quantities. Benchmark rates can still carry the minor and secondary items, but the packages that dominate the cost need build-ups that respond to the actual method, crew and site constraints. If every rate in a Class 3 estimate came from a benchmark table, the estimate has the appearance of Class 3 and the information content of Class 4.
No. AACE is explicit that the ranges in the classification tables are typical values observed across many completed projects, offered as a guide rather than as a prediction for any single estimate. The accuracy range for a particular estimate has to be established by risk analysis against that project, considering its complexity, novelty, site conditions and market exposure. The published ranges also assume appropriate contingency has already been applied, and they are deliberately asymmetric because overrun is more likely than underrun.
Start the estimate parametric, mature it into resource build-ups as the design lands, and keep every gate submission as a package you can reopen.