How to Run a Royalty Benchmarking Analysis

What the CUT Method Does, and When to Use It

The comparable uncontrolled transaction method, called the comparable uncontrolled price method in the OECD Guidelines, prices a controlled transaction by reference to the price charged in a comparable transaction between independent parties. Applied to intangibles, it compares the royalty in a controlled license against the royalties in uncontrolled license agreements covering sufficiently similar intangibles on sufficiently similar terms. Where reliable comparables exist, it is regarded as the most direct and reliable way to apply the arm’s-length principle, because it rests on an actual market price rather than a constructed one.

The method’s strength is also its constraint. Its reliability depends entirely on the comparability of the agreements identified. A royalty rate observed in a third-party license is only evidence of the arm’s-length rate for the controlled license to the extent the two transactions share the economically relevant characteristics that drive a royalty. The work of a CUT study is therefore not the arithmetic at the end; it is the disciplined screening that determines which agreements belong in the set.

This guide addresses running royalty rates, meaning royalties expressed as a percentage of a sales base, which are the most common form for licensed intangibles and the form most readily benchmarked from disclosed agreements. Lump-sum and mixed structures raise additional conversion questions that are out of scope here.

A note on entitlement before rate. Benchmarking answers the question of what rate is arm’s length. It does not answer the prior question of which entity is entitled to receive the royalty at all, which turns on the analysis of functions, assets, and risks set out in the DEMPE analysis. Run that analysis first; a perfectly benchmarked rate paid to the wrong entity is not a defensible position.

The Workflow at a Glance

A CUT royalty study moves through four analytical stages, each documented so the path from raw database population to final rate is reproducible, and closes with the write-up that makes the result defensible.

The stages are sequential, but the screening is iterative in practice: a qualitative review that rejects too many agreements for one reason often sends the analyst back to widen or refine the quantitative filters. What matters is that the final documented path is clean and the rejection reasons are explicit.

The steps below turn the framework into a defensible study, from characterizing the tested license through the two screens and the financial analysis to the documentation that lets a reviewer retrace the whole path.

Everything downstream is calibrated to the tested transaction, so it must be characterized precisely before any searching begins. Accurate delineation of the controlled license under the Chapter I framework is the starting point. The specification should capture at least the following.

• The intangible itself: its nature (patent, know-how, trademark, software, or a bundle), its field of use, and its stage of development.
• The rights granted: exclusive or non-exclusive, the geographic territory, the term, sublicensing rights, and rights to enhancements, updates, or improvements.
• The royalty base: the sales measure on which a running royalty would be charged (net sales, gross sales, or another base), since a rate is only meaningful against its base.
• The functional profile: which party performs which functions, bears which risks, and contributes which other assets, because these shape what an independent licensee would pay.
• The industry and market: the sector and the economic conditions of the market in which the license operates.

This specification becomes the benchmark against which every candidate agreement is judged. The more precisely the tested transaction is described, the cleaner the accept-or-reject decisions in Step 3.

Quantitative screening reduces a large agreement database to a manageable candidate pool using objective, reproducible filters. The major commercial databases of third-party license agreements, populated largely from public sources such as securities filings, support searching on structured fields and full text. The study should record the database used, the date of the search, and every filter applied.

The screening is best run as an ordered cascade of filters, with the count of remaining agreements recorded after each step. That running tally, the funnel from the full database down to the primary set, is part of the documentation: it shows the search was systematic and lets a reviewer see how each criterion narrowed the population. A typical cascade for a brand or trademark royalty search runs in this order.

• Industry or sector. Restrict to the sectors relevant to the tested intangible, using the database’s industry codes or classification scheme. Record the exact codes used.
• Type of licensed property. Restrict to agreements licensing the relevant kind of intangible, for instance marketing intangibles such as trademarks and trade names, while removing unrelated categories such as manufacturing know-how, patents, or process technology where those do not match the tested transaction.
• Consideration and royalty basis. Keep only agreements with a variable running royalty, expressed as a percentage of a sales base of zero or greater, removing lump-sum and other structures that cannot be compared on a percentage basis without conversion.
• Full-text keywords. Apply full-text filters reflecting the technology, product, or field of use. Wildcard stems are useful here: a stem such as “brand” can be set to capture “brand,” “brands,” and “branded,” which widens recall without manual enumeration.
• Normalized base. Restrict to agreements whose royalty is expressed against a base comparable to the tested transaction’s base, typically net sales, so the rates are commensurable before any are compared.
• Date. Restrict to agreements with effective or license dates within a defined window, recent enough to reflect comparable economic conditions.
• Related-party removal. Remove agreements between related parties, since only uncontrolled transactions are valid comparables.

The output of this stage is a list of potentially comparable agreements, the primary set. It is expected to be over-inclusive: quantitative filters cannot judge true comparability, only surface candidates that the qualitative stage will accept or reject.

Where a single database yields too few comparables after screening, for instance in a narrow sector where third-party brand licenses are scarce, the study can run a second, parallel search in a different database, or widen the first search into adjacent sectors with similar licensing characteristics, and then combine the accepted agreements from both. Running two sequential searches and pooling the results is a recognized way to build a more robust final set, provided the comparability standard is held constant across both.

Qualitative screening is the heart of a CUT study. Each candidate agreement from the primary set is read and assessed against the tested transaction, then accepted or rejected for a documented reason. This is where comparability is established, and where a study is defended or lost on audit. The discipline is to apply the comparability factors consistently and to record a specific rejection reason for every agreement excluded.

The comparability factors

The OECD Guidelines identify special factors affecting comparability for transactions involving intangibles. Applied to a royalty search, these are the criteria against which each candidate is judged.

Applying the factors

For each candidate, the analyst compares the agreement against the tested transaction on each factor and decides whether the differences are absent, acceptable, or disqualifying. Three outcomes are possible.

The agreement is accepted where it is comparable on the economically relevant factors, with any differences either immaterial or able to be addressed by a reliable adjustment. The agreement is rejected where a material difference cannot be reliably adjusted for, with the specific reason recorded, for instance “reject: licenses a trademark, tested transaction licenses manufacturing know-how,” or “reject: non-exclusive, tested license is exclusive worldwide,” or “reject: royalty base is gross sales, not comparable to tested net-sales base.” The agreement is adjusted where a quantifiable difference can be corrected, in which case the adjustment and its basis are documented.

The standard to hold to is that a rejection reason should be specific enough that a reviewer could agree or disagree with it on its face. Vague exclusions (“not comparable”) are the weakness an examiner probes first.

A disciplined study records the qualitative review as a funnel, using a small set of standardized rejection reasons and reporting how many agreements each reason eliminated and how many remained. A consistent taxonomy keeps the review reproducible and makes the rejected-agreements appendix easy to defend. A workable set of standard reasons is the following.

Applying the same taxonomy across the whole primary set, and recording the eliminated and remaining counts at each step, turns the qualitative screen into a documented, auditable funnel rather than a set of one-off judgments.

The commensurate-with-income overlay

Under the US section 482 regulations, the bar for a comparable license is higher than the general OECD standard. The commensurate-with-income principle requires that controlled royalties track the income attributable to the licensed intangible, and the regulations require that a CUT involve the same or comparable intangible used in connection with similar products or processes, with similar profit potential. In practice this means an analyst building a study that may be tested under US rules should weight the profit-potential and same-intangible factors heavily, since a comparable that clears the OECD threshold may still fall short of the stricter US requirement. Where the tested transaction has US exposure, screen to the stricter standard.

Agreements with multiple royalty rates

A single license agreement frequently discloses more than one royalty rate, and how those rates are collapsed into one observation for the set is a judgment that affects the range. There is no universal rule. The correct treatment depends on why the agreement carries multiple rates, and the governing principle is comparability: the single figure entered for an agreement should represent the rate an independent licensee would pay for the same thing the tested transaction covers, against the same base. Three cases recur.

Tiered rates on the same intangible. Where the rates are volume or sales tiers for one intangible (for instance a higher rate on an initial band of sales that steps down as volume grows), the agreement prices the same right throughout; only the rate changes with volume. The most defensible figure to enter is the effective rate the licensee pays, which is the total royalty across the tiers divided by total sales, because it reflects what the licensee bears given where its volume falls across the bands. A simple average of the stated tier rates ignores how much sales volume sits in each band and can therefore misstate the effective burden. In practice, many published benchmarking studies adopt the simpler convention of taking the arithmetic average of an agreement’s tiered rates, on the basis that all observations are then weighted equally and the underlying sales distribution is often unavailable. That convention is common and defensible as a documented house method, but where the sales distribution can be supported, the sales-weighted effective rate is the more accurate figure, and the choice between them should be stated.

To make the difference concrete, suppose an agreement charges 8 percent on the first 10 million of net sales, 5 percent on the next 15 million, and 2 percent above 25 million, and the licensee’s expected net sales are 30 million. A simple average of the three stated rates is 5.0 percent. The sales-weighted calculation tells a different story: 0.80 million of royalty on the first band, 0.75 million on the second, and 0.10 million on the third, totaling 1.65 million on 30 million of sales, for an effective rate of 5.5 percent. The half-point gap is the difference between the two conventions, and the reason to prefer weighting where the distribution is known.

Different intangibles, fields of use, or territories. Where one agreement prices different things (a patent at one rate and a trademark at another, or different rates by territory or field of use), blending the rates mixes non-comparable items. The cleaner treatment is to split the agreement into separate components and keep only the component that matches the tested transaction, discarding the rest. A patent benchmarking study takes the patent rate and drops the trademark rate. The components would be combined only where the tested transaction is itself the same bundle, in which case they are weighted to mirror the tested transaction rather than averaged.

Rates that vary over the term. Where the rate changes across the life of the license (for instance a lower rate in early years rising later, or a minimum royalty alongside a running rate), the figure to enter is a representative steady-state rate or a present-value-based effective rate, not a simple average of the stated rates.

Whatever method is used, the agreement’s multiple rates and the basis for collapsing them must be recorded in the study. An examiner who pulls the underlying agreement will see the multiple rates and ask how the single figure was derived, and an undocumented average over a tiered structure is a soft spot in an otherwise sound analysis.

The final set of accepted agreements yields a set of observed royalty rates. From these, the analyst computes the arm’s-length range. Because even a well-screened set retains some comparability imperfection, the convention is to use the interquartile range, the band from the 25th to the 75th percentile, to reduce the influence of the least comparable observations at the extremes. A controlled royalty that falls within the interquartile range is generally regarded as arm’s length; one that falls outside it invites adjustment, often to the median.

Reading the result

Observations are weighted equally in the computation: each accepted agreement contributes one rate to the set, and where an agreement carries multiple rates that have been collapsed to a single figure under the rules above, that single figure is the one observation it contributes, so that no agreement is over-represented in the range.

A few further cautions apply to the financial analysis. The range is only as reliable as the final set that produced it; a tight interquartile range built from four loosely comparable agreements is weaker evidence than a slightly wider one built from a dozen closely comparable agreements. The number of observations matters, and a very small final set should prompt a return to widen the search rather than a false confidence in a narrow band. Finally, the base must be held constant: every rate in the set must be expressed against a base comparable to the tested transaction’s base, or converted to one, before the percentiles mean anything.

A CUT study is only as defensible as its documentation, because the value of the method lies in the reproducibility of the screening path. The study file should allow a reviewer to retrace every step from the database population to the final rate. At a minimum it should record the specification of the tested transaction; the database used and the date of search; every quantitative filter and the exact codes and keywords applied; the full primary set; the accept, reject, or adjust decision for each agreement with its specific reason; the final set; the computation of the range; and the conclusion on the controlled rate. The rejection reasons are the part an examiner scrutinizes most closely, so they should be specific and consistent across the set.

It is good practice to present the results in two appendices: one listing the accepted agreements, each with its parties, a short synopsis of the licensed rights, and the selected royalty rate; and a second listing the rejected agreements with the standardized reason for each. The accepted-agreements appendix is what supports the range; the rejected-agreements appendix is what demonstrates the screen was applied evenly.

One subtlety belongs in the documentation. The same licensor and licensee can appear in more than one accepted agreement and still count as separate observations, where the agreements license distinct intangibles or distinct rights. Identical parties do not make duplicate data points if the underlying IP differs; the duplicate-agreement rejection reason applies only to true duplicates of the same agreement. Where the same pair appears more than once, the study should note briefly why each entry is a distinct comparable.

A consolidated worklist for the full CUT study. Each item should be completed and evidenced in the documentation file.