ForensicBIM Valuation Methodology — How the Numbers Are Built

Every monetary figure in a ForensicBIM report comes from established valuation and accounting concepts, applied to measured model content and checked against independent experts. This explainer names those concepts and the standards and guides they come from, so finance teams, auditors and asset managers can check our reasoning against sources they already trust.

It is written for three readers:

• Asset owners and finance teams who want to know whether a figure can support a budget or a balance-sheet decision.
• Auditors and reviewers who need the conceptual basis and the published references behind each figure.
• BIM and information managers who want to see how model quality becomes money.

This explainer covers what each figure means, which recognised framework it follows and how we test it. It does not publish our internal coefficients or code. Those are proprietary and calibrated on our own reference corpus. Readers who need to verify a particular figure can ask us for a reviewer session.

We value the information asset: the digital model data delivered with a building, bridge, tunnel, road or other built asset. We do not value the physical asset. A model is an identifiable, non-monetary asset without physical substance, so it is treated as an intangible asset, in the same family as software and databases.

The unit of valuation is one IFC file (ISO 16739). Every figure is derived from what that file measurably contains:

• Geometry: distinct shapes, surface area, detail and the IFC classes used.
• Information: properties, classifications, materials and quantities that carry real meaning. Empty, placeholder and disguised-missing values count for nothing.
• Structure: how elements are connected, contained and typed.
• Quality: whether the data can be trusted and reused, from the ForensicBIM quality audit.

The method is generic. It reads the asset type from the file content itself, not the file name. A bridge, a tunnel or a building gets the logic that fits it, and assets without rooms or storeys are not penalised for lacking them.

Two questions sit behind every report, and they answer different needs:

ForensicBIM uses no new valuation theory. Every figure applies one of five well-established bodies of work to model data. The table maps each report figure to the concept and the standard behind it.

1. The three valuation approaches

The International Valuation Standards (IVS 105) recognise three approaches: market, income and cost. IVS 210 applies them to intangible assets. We use all three because each answers a different question. For model data the market approach is the weakest, because few models are sold on their own. The cost and income approaches carry most of the weight.

2. Cost approach: what it takes to have the data again

IVS 105 defines the cost approach as the current cost to reproduce or replace an asset, less deterioration and obsolescence. It names two methods:

• Reproduction cost: an exact replica. For a model this is the effort to rebuild the same IFC content when the end result is already known: geometry, information, structure and export.
• Replacement cost: an asset of equal utility. This is closer to re-acquiring the data from scratch, without the original to copy, which takes longer.

IVS 105 also describes the summation method: value the components separately and add them up. Our effort estimate follows this logic. It sizes geometry, information and structure as separate work pools and adds them.

The cost approach then deducts for obsolescence. For model data, obsolescence means content that cannot be trusted or reused: unclassified proxies, placeholder values, duplicate identifiers, broken spatial structure. The ForensicBIM quality audit measures these and the valuation applies them.

3. Parametric effort estimating

We cannot time an expert rebuilding every model. Instead, effort is predicted from measured size drivers, calibrated on expert estimates. This is parametric estimating, the method the US GAO describes for programme cost estimates and the one behind software models such as COCOMO. The GAO Cost Estimating and Assessment Guide sets the tests a parametric model must pass: documented drivers, a calibration data set, stated accuracy and a known valid range.

4. Income approach: value in use

IAS 36 defines value in use as the present value of the future cash flows expected from an asset. For model data the cash flows are avoided costs: operate and maintain work the owner does not have to do because the data exists and is reliable.

The evidence that these savings are real and large comes from NIST GCR 04-867 (Gallaher et al., 2004). It put the cost of inadequate interoperability in US capital facilities at $15.8 billion a year in 2002. About two thirds fell on owners and operators, mostly during operation and maintenance. Research on BIM for facility management (Becerik-Gerber et al., 2012; Teicholz, 2013) identifies where the savings come from: finding information, locating and identifying components, fault finding and planning maintenance.

Discounting over a service life follows life-cycle costing practice: ISO 15686-5, RICS life-cycle costing guidance and, for US federal owners, OMB Circular A-94.

5. Deprival value: value to the owner

Deprival value asks what the owner would lose if deprived of the asset. Bonbright set out the idea in The Valuation of Property (1937); Baxter named it in 1975. The rule is simple:

• An owner would never lose more than the cost of getting the data again, so value is capped at re-acquisition cost.
• If re-acquiring is not worth it, the owner loses only the benefit the data would have delivered: its value in use. Classical theory uses the recoverable amount here; we use value in use because model data is rarely sold, so a transfer value is too thin to rely on.

Deprival value is the lower of the two. It is the most defensible single number for insurance, claims and budgeting, because it cannot exceed either what the data costs or what it earns.

6. Information as an asset

The idea that data has asset value of its own is now mainstream. Moody and Walsh (1999) set out how information differs from physical assets: it is not used up, it gains value with use and loses it when it is inaccurate or out of date. The Bennett Institute and the Open Data Institute (2020) review how data value is measured. Both support two features of our method: missing or unreliable data counts for nothing, and data loses value if nobody maintains it.

7. Recognition on the balance sheet

An owner who received model data as part of design or project costs may be able to recognise it as an intangible asset. The relevant standards are:

• IFRS: IAS 38 Intangible Assets for recognition and amortisation; IAS 36 for impairment.
• Dutch GAAP: RJ 210 Immateriële vaste activa; RJ 121 for impairment.
• US public bodies: GASB Statement 51 for intangible assets; GASB Statement 42 for impairment of capital assets.

All three require an identifiable asset, future economic benefits or service potential, and a cost that can be measured reliably. Our carrying amount shows what recognition would look like. It starts at cost, is amortised straight-line over the useful life and is written down when the recoverable amount falls below it.

Every report figure is built from the IFC file in two independent routes: what the data costs to have again, and what it saves in use. They meet in the value ladder.

The cost route and the income route share only the audit. A model can therefore be expensive to rebuild and still be worth little in use, or the reverse.

Step by step

1. Read the file: The audit reads the IFC file as delivered. Nothing is entered by hand and nothing depends on the file name.
2. Audit quality: About 45 checks score validity, structure, geometry, information and consistency. Unreliable content, such as placeholder values, unclassified proxies or duplicate identifiers, is treated as obsolete and counts for nothing.
3. Measure content: The audit counts distinct shapes rather than copies, measures modelled surface area, and counts informative facts and relations.
4. Estimate reproduction effort: Effort is estimated separately for the geometry, information and structure pools and then added up, following the summation method of IVS 105. The estimate is parametric: calibrated on expert judgement, with a stated range.
5. Price the effort: Hours times a blended hourly rate give the reproduction cost, always shown as a range.
6. Identify what is maintained: Each element gets an operate and maintain role from its IFC class, for example equipment, network runs or fabric. For infrastructure, fabric is sized by measured area, length or volume rather than by element count.
7. Estimate yearly savings: Five mechanisms are counted per element group: searching and checking information, repeat visits, fault finding and isolation, re-measuring for maintenance plans, and surveys before alterations. Only data that is present and reliable earns a saving.
8. Discount over the useful life: The yearly savings are brought back to a present value over the useful life you choose. This is the value in use.
9. Build the ladder: The two routes are combined with standard valuation rules into recoverable amount, deprival value, value at stake and the carrying amount if recognised.
10. Project forward: The report shows how value develops over the useful life in two cases: data kept up to date, and data left alone and going stale.

The effort model behind reproduction cost is calibrated on independent expert estimates and tested on models it has not seen. Out of sample, 70% of expert consensus values fall inside the range we report. That is the confidence level printed on the Valuation tab.

The expert round

• 121 estimates from 7 independent BIM experts, each with a stated confidence level.
• 43 calibration models: architectural, structural, MEP and infrastructure, from single-trade supplier files to multi-storey buildings and bridges.
• Each expert split their estimate into the three work pools: geometry, information and structure.

How the estimates are combined

1. Clean the data: Obvious inconsistencies are resolved by rule, for example a stated total that does not match the sum of its pools.
2. Correct for personal bias: Some experts are systematically fast or slow. Each expert's bias is estimated from the models they share with others and removed. This follows standard practice in expert elicitation (Cooke, 1991; Jørgensen, 2004).
3. Form a consensus: For each model and pool, the confidence-weighted geometric mean of the corrected estimates is calculated. The geometric mean is used because effort estimates are skewed: one very high guess should not dominate.
4. Fit and test: The effort model is fitted to the consensus with confidence as weight. Candidate models are compared by leave-one-out cross-validation (Stone, 1974): each model is predicted by a fit that never saw it.

Accuracy metrics

A typical error of ×1.45 means a model valued at 70 hours would usually lie between 48 and 102 hours by expert judgement. That is why every reproduction cost is shown as a range, not a point.

Example anchors

The last two rows are deliberate: we publish misses as well as hits. The MEP case rests on a single low-confidence estimate, so it carries little weight in the fit.

Valid range and warnings

The model is only trusted inside the range it was calibrated on: up to about 10,000 distinct shapes, 40,000 elements and 350,000 m² of modelled surface. A model beyond that is flagged as extrapolated. Missing inputs, such as no measured surface area or no relation data, are flagged and lower the stated confidence.

Quality scores

The quality and forensic scores that feed the valuation are benchmarked on a frozen release of 4,138 unique models. Releases are built offline and versioned, so a client can always tell which benchmark their report used. Models are compared only with the same facility type: bridges with bridges, never with buildings.

This example uses round, illustrative numbers for a mid-size building model. It is not a real file and the rates are not our defaults; it shows how the figures relate to each other.

Inputs (illustrative):

• Reproduction effort from the calibrated model: 70 h, range 48–102 h.
• Blended hourly rate: €100.
• Yearly operate and maintain saving from the data as delivered: €4,000.
• Useful life 20 years, discount rate 4%, cost increase 2% a year.
• Re-acquisition cost (new survey and remodelling, no original to copy): €18,000.
• Transfer value (what another party would pay): €2,000–4,000.

The present value is the sum of each year's saving brought back to today:

Here S is the yearly saving, g the cost increase rate (2%), r the discount rate (4%), and N the useful life in years. This is the standard present-value formula of IAS 36 and ISO 15686-5, not a ForensicBIM invention.

What the example shows

• Value in use far exceeds cost: That is normal for good data: it is cheap to produce relative to what it saves over decades. It is also why an owner should budget to keep it current.
• No impairment: The recoverable amount stays well above the carrying amount, so no write-down is needed.
• Neglect costs money: If the data is not maintained and its usefulness falls by about 6% a year, the same 20 years yield about €37,900 instead of €65,700.
• A poor model reverses the picture: If the audit finds the data mostly unusable, the yearly saving and so the value in use drop towards zero. The carrying amount is then written down, even though the reproduction cost has not changed.

Our figures are well-founded estimates with stated ranges, not appraisals or audited values. Read them with these limits in mind.

Known sources of uncertainty

Useful life and accounting limits

You can set the useful life from 5 to 30 years for buildings and up to 50 years for infrastructure. Accounting rules may be stricter. RJ 210 presumes a maximum of 20 years for intangible assets unless a longer life can be justified. IAS 38 and GASB 51 have no fixed cap but require the life to be supportable. Choose the life with your auditor.

Sensitivity

Value in use is most sensitive to the useful life and the discount rate. In the worked example, cutting the life from 20 to 10 years lowers value in use by about 45%, from €65,700 to €36,000. Reproduction cost is most sensitive to the hourly rate, which scales it one for one.

These are the sources a reviewer can consult to check each concept independently. Practical guides come first; standards and research follow.

Practical guides and manuals

Valuation standards

Accounting standards

Life-cycle costing, BIM & research citations

Frequently asked questions

Why don't you publish the formulas?
The coefficients are calibrated on our own expert rounds and reference corpus, and they are our intellectual property. The concepts, standards, validation method and accuracy are published here. A reviewer who needs to verify a specific figure can request a confidential reviewer session.

Why is value in use so much higher than reproduction cost?
Reproduction cost is a one-off effort. Value in use is the saving the data delivers every year for decades. For good data the second is usually much larger, which is the case for maintaining it.

Can a model have zero value?
Yes. If the audit finds that the data cannot be relied on, its value in use can be zero, even when rebuilding it would cost money. A recognised carrying amount is then written down.

Does a bigger file mean a higher value?
Not by itself. Value follows distinct content and usable information, not file size or element count. Copy-pasted geometry and empty properties add bytes, not value.

Can I change the assumptions?
Yes. Useful life is set on the Overview tab and drives the Valuation and Potential tabs. Cost at recognition can be edited on the Valuation tab.

Why are bridges not compared with buildings?
Different assets have different modelling norms. Benchmarks only compare like with like, and say so when there are too few comparable models.

Is this an audit opinion?
No. It is an evidence-based estimate to support a decision by the owner and their auditor.

Glossary