Industries / Construction and property / EPC contractors
Front-end design and estimating
Before an owner decides to build, engineers define the plant well enough to price it and judge whether it is worth building; this phase is called front-end loading, and its last step is usually FEED (front-end engineering design).
7
Milestones achieved, from 1947 to 2011.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1947
Costs scaled from older plants
Early in a project, a plant's cost is often estimated from a similar plant of another size. In 1947 Williams published in Chemical Engineering the rule that cost grows with capacity raised to a power, usually about 0.6, so a plant twice as large costs about 1.5 times as much. In 1948 H. J. Lang proposed multiplying the cost of a plant's purchased equipment by one factor for fluid plants, another for solids plants and a third for mixed ones. Both are still taught and used. In a 2016 test on three biorefinery designs, Tsagkari, Couturier, Kokossis and Dubois found that the worst of several such shortcut methods missed by more than 180%.
Lessona rule of thumb gives a starting number, and its error depends on how closely the new plant resembles the plants behind the rule.
Credit
Williams; H. J. Lang
1963
One index for plant costs
Chemical Engineering magazine has published its Plant Cost Index (CEPCI) every month since 1963, with 1957–59 set at 100, so that estimators can move a past plant's cost to current prices. It is a weighted sum of four sub-indexes, built mostly from US producer price indexes: equipment (about 51%), construction labor (29%), engineering and supervision (16%) and buildings (5%), by Dimitri Mignard's 2014 account; the formula was overhauled in 1982 and 2002. The magazine says the index has been written into the cost-escalation clauses of construction contracts.
LessonMignard links its worldwide use to a lack of local process-plant indexes elsewhere, so a US index, adjusted by a location factor, prices plants in other countries.
Credit
Chemical Engineering magazine
1981
First plants overrun their estimates
A pioneer plant is the first commercial plant to use a new process. Edward Merrow, Kenneth Phillips and Christopher Myers of RAND, working for the US Department of Energy, analyzed detailed records of 44 process plants built by 34 oil, chemical and mineral companies, and asked why inaccurate cost estimates are typical of such plants and why their performance falls short. Merrow went on to found Independent Project Analysis (IPA) in 1987, which benchmarks owners' projects and says its databases now hold more than 20,000 projects.
Lessonthe estimate for a new process carries risks that its single number does not show.
Credit
Edward Merrow, Kenneth Phillips and Christopher Myers, RAND
1996
A score for project definition
The Construction Industry Institute (CII) published its Project Definition Rating Index (PDRI) for industrial projects in June 1996. It is a checklist of 70 scope-definition elements in a score sheet, weighted with input from 54 experienced project managers and estimators, which gives a project one score for how well it is defined before detailed design. On a set of real projects, the scores explained part of the spread in project success, with a coefficient of determination (R²) of 0.40. CII said the industry had until then "lacked non-proprietary tools" for this stage. It is now in version 5.0.
Lessonhow well the scope is defined predicts outcomes, but explains well under half of them.
Credit
CII Front End Planning Research Team
1997
Estimates graded by definition
AACE International's Recommended Practice 18R-97 sorts estimates for process plants into five classes by how much of the project is defined, not by how much effort went into the numbers. A Class 5 estimate, made at 0–2% definition by scaling from capacity or by analogy, is expected, once contingency is added, to land between 20% and 50% below and 30% and 100% above the final cost for 80% of projects. A Class 1 estimate, at 65–100% definition, falls between 3% and 10% low and 3% and 15% high. The August 2020 revision adds that for weak project systems or risky projects the high end can be two to three times the table's figure.
Lessonan estimate's accuracy is set by what has been decided, not by how carefully it was added up.
Credit
AACE International
2007
Open estimates for power plants
The US National Energy Technology Laboratory (NETL) published its "Cost and Performance Baseline for Fossil Energy Plants" in May 2007. The May 2025 revision of Volume 1 gives block flow diagrams, stream tables, equipment lists and capital costs by account for 20 coal and gas power plant designs, counting its appendix cases, with and without CO2 capture. NETL rates them as AACE Class 4 estimates: 15% low to 30% high for coal plants and 15% low to 25% high for gas combined-cycle plants. It warns that its method "does not fully account for the unique cost premiums" of first plants with capture.
Lessoneven a public reference estimate says that the first plants of a new design will cost more than its range.
Credit
NETL, US Department of Energy; Marc Turner and co-authors (2025)
2011
Most megaprojects miss
Edward Merrow's book Industrial Megaprojects (Wiley, May 2011) drew on IPA's database of more than 300 projects of over $1 billion, averaging $2.6 billion in 2010 dollars. IPA says large overruns, delays, poor operability and safety incidents mark "well over half" of them, and traces the failure mostly to the early phase: the business strategy and the basic technical data package. In a 2012 article Merrow reported that only 22% of oil and gas megaprojects were successful; the rest averaged 33% real cost overrun and 30% schedule slip, and 64% had serious, lasting production problems in their first two years. IPA scores project definition with its FEL Index, which it calls "highly predictive" of outcomes.
Lessonthe path to a failed megaproject is usually laid down before the decision to build.
Credit
Edward Merrow, IPA
Watch for
Feasibility models in hours
Roebling, formerly Synonym, says its AI-based platform lets owners and engineers "model, refine, and validate project concepts in hours and days instead of months", and lists feasibility work through FEL-2 for fermentation and biomanufacturing companies.
Achieved whena client or reviewer publishes a Roebling estimate beside a conventional estimate of the same scope, or beside the cost later approved.
Claimed by
Roebling, February 2026
Open
01
Agents price NETL's plants
A public test of AI agents as cost estimators. Each agent gets the design basis, block flow diagram, stream table and major equipment list of one of the 20 designs in NETL's May 2025 baseline, with the case name removed, and must return the total plant cost by NETL's account structure. Scores compare each answer with NETL's own estimate. A second, harder track gives the block flow diagram only, as for a Class 5 estimate.
The briefCredit · unclaimed02
An open plant cost index
A free, documented monthly index of process-plant construction costs, built from public price series (US producer price indexes and labor-cost series, and the matching Eurostat series for Europe), with its weights, code and data published, so that students and small firms can bring old estimates up to date without a license.
The briefCredit · unclaimed
Open 01 · the brief
Agents price NETL's plants
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of AI agents as cost estimators. Each agent gets the design basis, block flow diagram, stream table and major equipment list of one of the 20 designs in NETL's May 2025 baseline, with the case name removed, and must return the total plant cost by NETL's account structure. Scores compare each answer with NETL's own estimate. A second, harder track gives the block flow diagram only, as for a Class 5 estimate.
Why it is still open
Agent tests in process engineering cover other steps. CeProBench (March 2026) scores knowledge, flowsheet and parameter tasks; ChemCost (May 2026) scores agents on the purchase cost of a reaction's chemicals from 230,775 supplier quotes; a January 2026 workflow sets up process simulations. None scores a plant's capital cost against a published estimate. Firms such as Roebling, under Watch for, sell AI-assisted estimates without a public test.
Built with what exists
NETL's 2025 Volume 1 report (public, on OSTI) as the answer key; any agent framework with a spreadsheet or Python tool; open-weights and commercial models. Run every agent several times per case, since a model that passes once may fail on a repeat (see the Note "All twenty times"). A human baseline using Lang factors on the same inputs.
First step
Copy the case B31A equipment list and total plant cost by account from the 2025 report into a machine-readable answer key, and run one agent on it.
Done looks like
All 20 cases, at least three agents, five runs each, and a published table of every run's error against NETL. An agent passes a case only if all five runs fall inside NETL's stated range for it (15% low to 30% high for coal, 15% low to 25% high for gas). Results reported separately for cases that appeared in earlier NETL editions, to show whether models are recalling old figures.
Open 02 · the brief
An open plant cost index
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A free, documented monthly index of process-plant construction costs, built from public price series (US producer price indexes and labor-cost series, and the matching Eurostat series for Europe), with its weights, code and data published, so that students and small firms can bring old estimates up to date without a license.
Why it is still open
The CEPCI is a paid subscription, and its publisher forbids any third party to use its data "for any educational, commercial or other use" without consent. Mignard (2014) notes the lack of local process-plant indexes outside the US. The US Bureau of Reclamation publishes free construction cost trends in 35 categories for the types of projects Reclamation itself builds, adding water treatment plants and plastic pipelines from October 2025. Searches of GitHub and the web in November 2026 found no open plant index.
Built with what exists
the US Bureau of Labor Statistics' public data API; Eurostat's producer price data; Python and pandas; a scheduled GitHub Actions job to update it monthly; weights taken from the split of equipment, materials and labor in NETL's published cost tables.
First step
Find the public producer price series for the product groups Chemical Engineering names among its 41 inputs, such as carbon steel plate and fans and blowers, and chart them from 2000.
Done looks like
A monthly series from January 2000 for the US and the EU, with code, weights and data public and rebuilt from scratch by a second person; a published comparison with the Bureau of Reclamation's trends and with yearly CEPCI values already quoted in open papers, with every year's gap shown.