PEICI method note

PEICI publishes the cost–capacity parameters behind screening-level capital estimates for industrial energy equipment. This page is the method: what the numbers are, how they were derived, what they can carry and what they cannot.

It is free, never gated and versioned. If you are deciding whether to trust a figure from this site, everything you need to make that decision should be on this page — including the parts that do not flatter the data.

The Sizing Set

Five published parameters turn an index into a calculator.

  • C_ref — reference cost. What the equipment cost at the reference size, in USD of a stated cost year.
  • S_ref — reference size. The capacity that cost applies to, in the equipment’s native unit.
  • n — scaling exponent. How cost moves with size. There are two of them, which is the next section.
  • ICM — installed-cost multiple. Installed cost divided by equipment cost, at the reference size.
  • LF — Input Price Relativity (IPR). Relative price of equipment, installation labour and bulk materials by region, base region North America = 1.000. The symbol stays LF in the published data and the schema; the name is what a reader sees.

The formula

r               = S / S_ref

equipment_cost  = C_ref × r^n_eqp × LF
installed_cost  = C_ref × ICM_ref × r^n_inst × LF

ICM_at_S        = ICM_ref × r^(n_inst − n_eqp)

Worked, in native units, on the package steam boiler at 20,000 lb/hr against a reference of 50,000 lb/hr:

r          = 20,000 / 50,000            = 0.4
equipment  = $236,276 × 0.4^0.6318     = $132,434
ICM_at_S   = 2.213 × 0.4^(0.4248−0.6318) = 2.675
installed  = $132,434 × 2.675          = $354,288

Every number in that example is published on the heat factor table. Worked examples on this page stay in native units and do not follow the site’s unit toggle — they are here to be checked against the published data, and the published data is native.

Two exponents, not one exponent and a fixed multiple

This is the finding the whole product rests on. Installed cost and equipment cost scale at different rates: the installed exponent runs about 0.12 lower than the equipment one. So the installed-cost multiple is not a constant — it falls as plant gets bigger.

That was measured in 36 of the 37 equipment types where both cost bases were published, with a median of about 2.12× at the small end of each range and 1.33× at the large end. The single exception is the 9-stage centrifugal compressor at 1900 psia discharge, where it rises slightly, 1.07× to 1.11×. It is named because a finding with a named exception is a finding; one without is a claim.

The consequence for anyone pricing plant: on a package steam boiler the multiple is 2.213 at 50,000 lb/hr and 2.675 at 20,000 lb/hr. Same equipment, same method, 21% more installation cost per dollar of kit, purely because the plant is smaller. A calculator applying one fixed multiplier discards that and under-prices small plant — which is most of what our clients build.

How the exponents were fitted

Ordinary least squares on log-transformed cost and capacity — log C = log a + n log S — so n is the slope on a log-log plot.

  • 346 raw cost–capacity observations collected, of which 308 are used by the published fits
  • 88 fits, across 27 equipment types and 47 equipment/subtype combinations
  • 77 meet the quality bar; 11 do not, and all of them are published

Every fit rests on 3 to 6 points. That is enough for a defensible exponent and not enough for a tight prediction band.

The full table, including every fit that failed, is at the exponent register.

The quality bar

A fit meets the bar when both hold: the capacity range spans at least 10×, and R² is at least 0.90. 77 of 88 fits meet it; 7 fail on range, 4 on R², none on both.

5 of the failures have an R² of 1.000 across three points spanning a factor of 2.5 to 6.7. A perfect fit over too narrow a range is not a scaling law — it is a straight line through three points that happen to be collinear. That is why the range condition exists, and why a high R² alone is not enough to publish an exponent.

The 10× is not an arbitrary number either. It is the same decade over which cost–capacity scaling is generally defensible, and it is the same threshold at which the calculator starts warning that a result is indicative only. A bar that matches the warning is coherent; two unexplained tens would not be.

The installed-cost multiple, and two numbers on this site that differ

The multiple published in the register is computed from matched raw pairs — installed cost divided by equipment cost at the same observed capacity — not from the ratio of the two fitted curves.

The calculator’s ICM_at_S, though, is derived from the two fitted exponents, because it has to return a value at any capacity and not only at the handful that were observed. So this site publishes two installed-cost multiples for the same equipment at the same size, and on the cooling tower — the steepest gradient in the set — they differ:

Units and conversions

The site offers a metric/imperial toggle. These are the constants behind it. Every one is exact by definition rather than measured, which is why the toggle can never introduce an error.

Native unitPaired unit1 native =Exact because
lb/hrkg/h0.453592371 lb = 0.45359237 kg, by definition
MMBtu/hrMW0.29307107021 Btu(IT) = 1055.05585262 J, by definition
gallonsm³0.0037854117841 US gal = 3.785411784 L, by definition
gpmm³/h0.22712470704as above, × 60
acfmm³/h1.699010795521 ft³ = 0.028316846592 m³, by definition, × 60
MW_coolington284.345136…1 ton of refrigeration = 12,000 Btu(IT)/h, by definition
kWhp1.341022…1 hp (mechanical) = 745.699872 W, by definition

MVA, MW, MWth, MWac and MWh are already SI and have no imperial pair. MVA in particular has none and never will — it is not MW either, because the two differ by power factor.

The exponents are dimensionless. Cost is fitted against the ratio S / S_ref, and a ratio of two capacities is the same number whichever consistent unit both are expressed in. Converting lb/hr to kg/h multiplies both by 0.45359237 and the factor cancels. So n_eqp, n_inst and ICM are identical in metric and imperial, and switching unit system cannot change a cost.

There is a second kind of conversion that can be wrong, and the difference matters:

  • Definitional — lb/hr to kg/h, MMBtu/hr to MW. Pure unit algebra. Needs nothing, cannot be wrong, is not adjustable. Shown as Exact.
  • Duty-based — lb/hr to MW thermal, gpm to MW cooling. Needs an enthalpy, an efficiency or a temperature range. Can be wrong, and you can change it. Shown as Assumed, with the parameters on screen.

Collapsing the two would fail in both directions: exact arithmetic made to look uncertain, or an assumed duty basis made to look exact. The second is how somebody sizes plant against an enthalpy nobody chose.

Currency

Every published figure is USD. GBP and EUR are display conversions at a fixed published rate, frozen for the life of the factor set version and never a live lookup — because a rate that moves between two page loads means one enquiry gives two answers, and an estimate that cannot be reproduced is not a citation asset.

Currency conversion is not a regional cost adjustment. Input Price Relativity does that job, and all five regions now carry a published value. Converting a North American cost to pounds still leaves it a North American cost — to price a European project, select the European region and change the input, not the symbol.

Accuracy class

Pemspring screening uncertainty envelope, informed by AACE Class 5 practice — order of magnitude, concept screening, 0–2% project definition. The numbers are unchanged: −20%/−50% low, +30%/+100% high. The name changed because once enough Pemspring estimates have been matched against outturn, the envelope can be calibrated from our own results — and an envelope named after somebody else’s standard cannot be narrowed on your own evidence without looking like a downgrade.

Best caseWorst case
Low-20%-50%
High+30%+100%

Per AACE International Recommended Practice 18R-97. Shown asymmetric because it is asymmetric. Averaging it into a single ± figure would misrepresent the risk: the downside and upside of a screening estimate are not the same size, and presenting them as if they were is how a range becomes a number in somebody's business case.

Input Price Relativity and Delivery Environment Uplift

Input Price Relativity is an input-price factor. It captures the relative price of equipment, installation labour and bulk materials, plus a productivity allowance. It excludes delivery-environment cost entirely — remoteness, logistics, camp, local-content rules, contractor market conditions, country risk and schedule extension. Those push the other way and none of them is quantified here. Outside a mature industrial region, treat the published factor as a lower bound on the true cost difference.

Input Price Relativity by region, and its derivation
RegionIPRBasket rangeGradePLIPFDutyRecomputed
NA1.000by definitionA110%1.000000
EU0.9570.872–0.975A0.7461.15242%0.957070
APAC0.9310.819–1.014B0.56751.25956%0.931010
SAM0.9310.874–0.938C0.4911.305410%0.930563
MEA0.8870.792–0.964D0.3971.36188%0.886592

IPR = 0.594(1 + duty) + 0.244(PLI × PF) + 0.162(0.5 + 0.5 PLI). The weights come from PEICI’s own installed-cost-multiple register rather than a textbook Lang factor: the median ICM is 1.68× at the geometric mid of the fitted ranges, so equipment is 1 ÷ 1.68 = 59.40f installed cost, and installation splits 60/40 between labour and bulk materials.

PF = 1 + 0.6(1 − PLI), tagged Assumed. It is one transparent constant rather than five per-region guesses, and it is the weakest parameter here — it will be replaced with Pemspring’s own project hours as soon as the project register supports it.

The PLI was checked against measured construction labour costs. Eurostat euro-area construction labour is EUR 34.40/hour (2025); BLS ECEC US construction total compensation is USD 48.58/hour (March 2025). At EUR/USD 1.05 that is a ratio of 0.744 against a euro-area GDP PLI of 0.746 – agreement within 0.3%. The PLI is a sound proxy for construction labour cost.

These values are not validated against any commercial location-factor product. They are derived from published price data and are reproducible from the parameters on the method page.

A region with no published factor is disabled in the calculator with “not yet published” rather than defaulted to 1.00, and the difference matters: 1.00 would mean no adjustment is needed, and what would actually be true is that no data exists. Defaulting would silently conflate the two. All five regions currently carry a published value, so nothing is disabled today — the rule governs what happens when a region is added faster than its price basket.

Published grassroots location factors assume equipment is roughly 20–25% of installed cost, because they price a whole plant including offsites, engineering and indirects. PEICI’s scope is equipment plus its direct installation, and equipment is 59% of that. Equipment is internationally traded from the same global vendors at broadly the same price, so a majority of a PEICI number does not move with location, and the factors compress towards 1.00 as a consequence. Applying a published grassroots location factor to a PEICI number would over-correct it.

Known limits

  • Within-region spread exceeds between-region spread. APAC ranges 0.82 to 1.01 and MEA 0.79 to 0.96 across their own baskets, while the five regional points span only 0.89 to 1.00. A single factor for a region this coarse is a screening device, not a site figure.
  • Delivery-environment cost is excluded entirely. See scope.
  • NA is fixed at 1.000 as the base. That base is really the US Gulf Coast: Mexico's own PLI is 0.549 and its computed factor 0.925, so a Mexican project priced at 1.000 is over-priced by roughly 8%.
  • The factor is size-dependent and is published as a single number. Equipment share of installed cost runs from 47% on small plant to 75% on large, so location bites harder on small plant. At the extremes the factor moves by roughly 0.05 either side of the published value.
  • PLI is a whole-economy price level, not a construction-sector one. Validated for the euro area only.
  • 2023 price data, 2026 publication.

What IPR is not: Delivery Environment Uplift

Delivery-environment cost — remote logistics, local-content rules, site access and laydown, contractor-market premium, camp and security, abnormal working, schedule extension — is excluded from IPR entirely and carried as a separate, project-specific input. It is never published as a regional constant and never derived from geography: selecting a region sets IPR and says nothing about the delivery environment. Its default is not assessed, which is a different claim from zero, and any estimate carrying an unassessed uplift is labelled a partial screening estimate rather than a total.

Validity, refusal and multiple trains

  • Outside the validity range, the calculator refuses. A power law fitted over one range does not hold outside it, so the tool returns nothing rather than a number that looks authoritative and is not.
  • Beyond one decade from the reference size, it warns and calculates. Same decade as the quality bar, and for the same reason.
  • Above the largest single unit, it prices multiple trains. Capacity is added by installing more machines, so cost goes linear. Trains are sized equally, because equal trains are what gets built and an undersized last one misprices the set.

One caveat on the last of those: a single-unit ceiling is currently populated on one record only. That is a gap in the data, not a claim that the others have no ceiling — it is on the v0.2 list.

Cost basis and re-anchoring

Most seeded values come from a 1998 US cost basis (DOE/NETL-2002/1169), with transformers at 2011 and solar at 2023. Every seeded entry is flagged as needing re-anchoring, and the reason is worth stating plainly rather than leaving in a field name.

The shapes are current. The levels are not. An exponent describes how cost moves with size, and that relationship is stable over decades. A reference cost describes what something cost in a particular year, and that is not. Re-anchor the reference cost against a present-day quotation before using any figure from this site in a business case.

Limits of this version

  • Single source, single cost year for most of the register.
  • Three to six points per fit — a defensible exponent, not a tight prediction band.
  • Units are process-industry units. Converting to megawatts needs a stated duty basis per equipment type, and those are Assumed.
  • No coverage yet of heat pumps, electric boilers, transformers beyond a published exponent, or solar and storage regressions.
  • Input Price Relativity is a screening device, not a site figure. Within-region spread exceeds between-region spread: the APAC basket alone runs 0.819–1.014 while all five regional points span 0.887–1.000. It also excludes delivery-environment cost entirely, which is carried separately as DEU and defaults to not assessed.
  • Heat pumps and electric boilers are blocked pending written reuse confirmation from the Danish Energy Agency, whose Technology Catalogue is the only public source with capacity-varying cost data for them. We would rather publish an empty row with the reason than substitute a proxy and let somebody quote it.

Revision policy and version history

Revisions are published, never silently corrected. Every restatement goes to the revision log with the old value, the new value and the reason. This method note is versioned too, and its own history is below.

  • 1.0 — first publication of the method note.
  • Validation order corrected. The train count is now determined before the range check, so the range test applies to the per-unit capacity rather than the total. Under the previous order, multiple-train pricing was unreachable on every record in the set.
  • Register counts corrected. An earlier note described 90 fits with 38 meeting the quality bar. The data holds 88 fits, of which 77 meet it; the 38 was the qualifying equipment-cost subset, now 39. No exponent and no quality flag changed — the values were right and the description of them was wrong.
  • Location factors published for all five regions, and renamed. An earlier note said factors outside North America were not published. Four are now published — EU 0.957, APAC 0.931, SAM 0.931, MEA 0.887 — each with its basket range and a coverage grade, derived from World Bank price levels and validated against Eurostat and BLS construction labour costs. The adjustment is now called Input Price Relativity, because the old name said where the number applied and the new one says what it is. No value changed in the process, and the field names in the published data and the schema are unchanged.
  • Delivery Environment Uplift added as a separate input. IPR always excluded remoteness, logistics, local content, contractor market and country risk — that exclusion was stated in prose and is now a named variable with its own default of not assessed. An estimate carrying an unassessed uplift is labelled a partial screening estimate rather than a total. Nothing about the published IPR values changed; what changed is that the caveat is no longer doing work the data model should do.
  • Accuracy band renamed to the Pemspring screening uncertainty envelope, informed by AACE Class 5 practice. The band is unchanged: −20%/−50% and +30%/+100%.

Both of those are here rather than quietly fixed because a method note whose version history begins at “1.0, everything fine” is less trustworthy than one that shows its working.

Sources and how to cite

Every input, its licence and the attribution it requires are listed at sources. The fitted exponents, with their R², sample sizes and capacity ranges, are at the exponent register.

Cite as: Pemspring Energy Solutions, PEICI method note, with the version and date shown above. Published values are in USD and in each equipment record’s native units.

Discuss whether this fits

Tell us the problem, the site and the decision you need to reach. If this is not the right service we will say so and point you at the one that is.

United Kingdom and United States enquiries are routed to the same team. We will tell you plainly where we do and do not have local delivery capability.