A whole-of-market-evidence approach to FRAND determination

In June 2026, the European Patent Office in collaboration with our team at BRELA Research in Economics and Legal Analytics published a study on the methodologies that courts use to determine FRAND rates for SEP licenses. One of the key findings of the study is that courts in different jurisdictions have increasingly embraced comparable licenses analysis as the predominant methodology for the purpose of determining FRAND rates, or for assessing whether SEP holders’ licensing offers comply with their FRAND obligations.

The popularity of comparable licenses is understandable, given that commercial transactions in the licensing market are uniquely well-suited to reveal the value of a patented technology. Courts have largely adopted the framework that the FRAND rate for a SEP license is a rate that a willing licensor and a willing licensee would have agreed upon – and comparable licenses voluntarily concluded between similarly situated parties are the most straightforward empirical approach to operationalize this hypothetical negotiation framework.

Nevertheless, as often in economic analysis, the devil is in the details. A comparable licenses analysis usually consists of three elements: selection, unpacking, and adjustments. First, one has to select the comparable licenses to be used in the analysis. To be used in a comparable license analysis, existing licenses should be actually comparable to the license at hand. Also, in order to reveal a FRAND rate, comparable licenses should themselves be FRAND. Second, the license needs to be unpacked – a complex analysis dealing with lump-sum payments, cross-licensing provisions, releases for past infringements, various types of discounts, and licensing agreements covering multiple technologies. Sometimes, there are too many uncertainties involved in this unpacking; therefore, the possibility to reliably infer an effective rate from an agreement is an important selection criterion as well. Finally, if the “comparable” licenses are not comparable to the license at issue in all important respects, these differences need to be accounted for through adjustments. Adjustments to the determined rate may also be necessary if the comparable licenses feature rates that appear too low or too high to be FRAND.

Each of these steps is complicated and often contested. Therefore, despite the broad consensus that now exists regarding the fundamental meaning of FRAND and the predominant role of comparable licenses for FRAND determinations, parties’ views and even different courts’ determinations of FRAND rates may still diverge substantially.

The difficulties and subjectivities involved in selecting and unpacking comparable licenses are well-known and have often been highlighted by commentators and practitioners. An aspect that has received less attention, however, is that the existing practice of comparable license analysis often relies on non-market information. In particular, courts routinely rely on patent counts for selecting comparable licenses, scaling between licenses to different portfolios, unpacking cross-licensing agreements, and adjusting for changes in portfolio strength over time. Therefore, while the premise behind the use of comparable licenses is that they reveal the market’s valuation of a patented technology, the way in which comparable licenses are selected and unpacked in practice is not as anchored in the reality of the SEP licensing market as the theoretical premise of comparable license analyses may suggest.

In this article, we propose a different, more internally consistent approach to the comparable licenses analysis, an approach that we call the whole-of-market-evidence approach. The idea behind the approach is simple: if one uses comparable licenses as an indicator for the market’s valuation of a license, one should ideally rely on market evidence for each of the steps of the analysis (selection, unpacking, and adjustments). This approach would often necessitate data on a larger number of licenses – including licenses that involve neither of the parties to the litigation. The whole-of-market-evidence approach is thus often more an aspiration than a reality, given the difficulty of observing the rates and other terms of third-party licenses. Nevertheless, we discuss how publicly available data about the broader SEP licensing market can be used to assist fact finders with the selection, unpacking, and adjustment of specific comparable licenses.

Existing approaches to selecting comparable licenses

Existing approaches to selecting comparable licenses often entail a mix of empirical and normative considerations and use licensing market data along with numerous other sources of information.

A typical negotiation between a patent owner (prospective licensor) and implementer (prospective licensee) entails two sets of (potential) comparable licenses: the prospective licensor may have existing licenses with other implementers, and the prospective licensee may have existing licenses with other patent owners. In many real-life situations, the prospective licensor may also need a license for some of the prospective licensee’s patents, and both parties’ existing agreements may also be cross-licenses. The presence of cross-licensing complicates the analysis but does not fundamentally alter the main point. Therefore, let us consider a simpler situation of a negotiation for a unidirectional license between a licensor (A) and implementer (I), where the licensor has existing licenses with other implementers J, K, and L; whereas the implementer has existing licenses with other patent owners B, C, and D.

Most courts would consider that the natural starting point for a FRAND determination in this case are the existing licenses between patent owner A and the other implementers J, K, and L. Sometimes, a patent owner has licensed its patents on very similar terms (similar per unit or ad valorem rates) to different implementers; but more often than not, the rates and other licensing terms differ between different agreements. Also, a majority of agreements involve lump-sum payments, and lump-sum payments cannot be simply extrapolated from one license to another if different licensees make different use of the technology. Thus, the licenses with implementers J, K, and L may yield different effective rates, when compared with the focal license between A and I, raising the problem of selection.

In the EPO study, we observed that courts may use inclusive or selective approaches (see figure below). In an inclusive approach, the three licenses between A and implementers J, K, and L may be considered as comparable licenses, and define a possible range of market rates for the portfolio. But if the effective rates differ between these agreements, these differences probably reflect differences between these licensees, or differences between the circumstances of the agreements, so that a rate that was acceptable for the license between A and J may not have been acceptable for the license between A and L. Alternatively, the circumstances of the agreements between A and J and A and L were indeed similar, so that there is no objective justification for the difference between the effective rates of the agreements. In either case, the licenses between A and J and between A and L cannot simultaneously be good comparables for the license between A and I, or at least one of these two agreements itself is not FRAND. Therefore, while the existing agreements with J, K, and L identify a range of rates at which A’s portfolio has been licensed, they do not identify the range of the FRAND rate for the license between A and I.

In a selective approach, the fact finder would first select the most comparable license (one court referred to this as the “awesome comparable”)^[1] among the licenses between A and implementers J, K, and L. If none of these licenses is sufficiently comparable or reliable, one may turn to licenses between I and other patent owners B, C and D. Nevertheless, each licensee is different, and the circumstances of each agreement are different as well. One license may be most similar to the focal license between A and I in some respects, but another license is the most comparable license in other respects. Courts have advanced a large variety of selection factors to justify that one license is more comparable than another, and the importance ranking of these factors differs from one judgment to the other. Where one court finds that licensees must be headquartered in the same country to be truly comparable, another court may find that licensees should be of similar size or make similar products; for others, it may be more important that the selected comparable agreement was reached in the absence of litigation, is recent, or has important non-royalty terms also found in the focal agreement.

Many courts turn to normative arguments to make the selection: factors are disregarded if – under FRAND – they should not matter for the value of the license. If the court finds, e.g., that non-discrimination requires offering similar rates to companies of different size, the size of the licensee is not used as a factor to select a comparable license – even if the evidence indicates that larger licensees have in fact negotiated licenses with lower rates. This creates a bit of a paradox: the court uses a comparable license, which supposedly reflects the market value of a portfolio; but it takes issue with how the licenses are actually priced in the market. It is, of course, possible that existing licenses are indeed not on FRAND terms, but one cannot simply rely on some aspects of these agreements while disregarding others as non-FRAND. If, for example, a court finds that rates differ between the agreements between A and J, K, and L in ways that lack objective justification, neither of these licenses provides direct evidence of the rate at which the patents of A would have been licensed if A had indeed applied uniform rates. Instead, one would need to carry out a complex counterfactual analysis to determine the rates that would have emerged in this counterfactual scenario – an analysis that is just as uncertain and complex as the hypothetical negotiation framework that the comparable licenses are supposed to assist with.

In the alternative, a court may select comparable licenses based on purely empirical considerations (i.e., considering licenses to be comparable if the evidence suggests that negotiations in the existing SEP licensing market would result in similar prices for these two licenses; independently of whether the price should be the same for the different licenses based on normative considerations). Nevertheless, parties are typically unable to provide courts with the relevant data to support such an analysis. For example, if A has licensed its patents to J at lower rates than to K, one may wonder whether implementer I would be more similar to J or to K. However, there are many different ways of comparing the different companies, and there is typically no data in the record on whether implementer I typically pays rates that are more similar to those paid by J or those paid by K. The agreements that implementer I has with B, C, and D do not help, unless one also has information about agreements between these other patent owners and implementers J, K, and L (thus, agreements involving neither A nor I, the two parties of the dispute).

Existing approaches to unpacking comparable licenses

Similar uncertainties surround the second step of the analysis: the unpacking of the selected comparable licenses. Even a carefully chosen comparable rarely states, on its face, the figure a court actually needs. The aim of unpacking is to derive from an existing agreement an effective rate – whether per unit or ad valorem – that can be compared across agreements and transposed onto the focal license (the license under determination). Most SEP licensing agreements, however, are not structured as simple, unidirectional running-royalty licenses, and every departure from that baseline requires assumptions to convert the agreement’s terms into an effective rate. The difficulties tend to cluster around three recurring features of real-world agreements: lump-sum payments, cross-licenses, and licenses covering multiple technologies.

The most common complication is the prevalence of lump-sum payments. Converting a lump sum into an effective rate requires dividing the payment by a measure of the licensee’s use of the technology over the term of the agreement, typically the licensee’s sales of licensed products. This immediately raises the question of which sales figures to use. Parties negotiate lump sums based on expectations about future sales, and these expectations may or may not have materialized. An effective rate computed from the sales projections available to the parties at the time of the agreement can therefore differ substantially from an effective rate computed from the licensee’s actual sales. Moreover, the parties’ contemporaneous projections are often not in the record, so that experts substitute analyst forecasts or their own reconstructions. Furthermore, a lump-sum payment is not economically equivalent to a running royalty with the same expected value. By accepting a fixed payment, the licensee assumes the risk that its sales fall short of expectations, and captures the benefit if sales exceed them; the licensor, for its part, obtains certainty of payment. Whether and how to account for this difference in risk allocation when comparing lump-sum agreements with running-royalty agreements is itself a contested question.^[2]

The deepest difficulties, however, arise from cross-licensing. Where the comparable agreement is a cross-license, the payment flowing between the parties reflects the difference between the values of the two licensed portfolios, not the gross value of either. To derive an effective rate for the patent owner’s portfolio, one must therefore place a value on the consideration that the patent owner received in kind – that is, on the license to the counterparty’s portfolio. In practice, this valuation is almost invariably carried out based on relative patent counts: the counterparty’s portfolio is assumed to account for a share of the total value of the transaction corresponding to its share of the relevant patents. Here, the tension noted above becomes particularly apparent. The premise of a comparable licenses analysis is that patent counts and similar proxies are insufficient indicators of value, and that market transactions are needed to reveal what a portfolio is actually worth. Yet the unpacking of cross-licenses re-introduces patent counts into the middle of the analysis. If relative patent counts were a reliable measure of relative portfolio value, one could dispense with comparable licenses altogether and determine rates top-down; if they are not, the effective rates derived from unpacked cross-licenses inherit the weaknesses of the patent counting on which the unpacking relies. 

Each of these unpacking steps requires assumptions; each assumption can be made in several individually defensible ways; and the choices compound across steps.^[3] As noted above, the possibility to reliably unpack an agreement is itself an important selection criterion – although courts have generally held that the mere fact that a license requires unpacking does not, by itself, exclude it from a comparable licenses analysis.^[4]

Relying on ease of unpacking as a selection criterion, moreover, creates a problem of its own. If only simple agreements – unidirectional licenses with running royalties, without releases, covering a single technology – can be reliably unpacked, then the set of usable comparables is tilted towards a particular type of transaction and, often, a particular type of licensee. There is no reason to assume that the agreements that are easiest to unpack are also the agreements that are most representative of the market.^[5]

Existing approaches to adjusting comparable licenses

Even the most carefully selected and unpacked comparable license usually differs from the license under determination in some respects, and existing practice accounts for these residual differences through adjustments. Also, courts may use comparable licenses, even though they consider the rates for those licenses were themselves not FRAND. Also, in such a case, a court will use adjustments to account for non-FRAND factors.

For instance, some courts have used licenses to other patent owners’ portfolios and derived a rate for the portfolio at issue on the basis of the relative number of patents in the different portfolios (“scaling”), even though there were notable qualitative differences between the portfolios. In other cases, comparable licenses to the portfolio itself were used, even though the court believed that the value of the portfolio had changed over time. In yet other cases, comparable licenses were used, even though the court accepted that because of the licensing environment at the time of the agreement, the patent owner or the implementer had not been able to obtain an agreement at a FRAND rate. All these situations have one thing in common: the FRAND determination rests on a comparable license, which provides the closest approximation of a FRAND rate for the license at issue, but it is still “off” by a certain, unknown amount. The court feels confident that it knows the direction of the needed adjustment, but there is no objective indication of the size. 

Compared with selection and unpacking, adjustments have received far less scrutiny. In the EPO study, we observed that courts discuss adjustments much less frequently than the other two steps. Yet the choice of adjustment factors and of their magnitudes is no less consequential – and no less contested: like the assumptions made in unpacking, adjustments can be made in several individually defensible ways, and seemingly small differences can translate into substantially different rates. Also, similarly to existing practices of selecting and unpacking comparable licenses agreements, adjustments frequently rely on non-market data, such as patent counting to adjust for changes in portfolio strength over time. Finally, adjustments may also be required where the rates of the comparable agreements themselves appear too low or too high to be FRAND – a determination that cannot be made based on the selected comparable agreements alone, and that calls for a benchmark external to them.

The whole-of-market-evidence approach as a potential solution

The common thread running through the difficulties described above is that each step of the existing practice – selection, unpacking, and adjustments – imports non-market information into an analysis whose very rationale is reliance on market evidence. The whole-of-market-evidence approach does not eliminate any of the three steps; each remains necessary. Rather, it disciplines each step with evidence drawn from the broader SEP licensing market.

At the selection stage, data on a larger set of licenses allows empirical regularities to substitute for ad hoc selection factors. Whether the size of the licensee, the presence or absence of litigation, the recency of the agreement, or the licensee’s product mix is associated with systematically different rates is not a question that needs to be answered normatively; it is an empirical question, which can be answered if one observes a sufficient number of agreements across the market. In the example introduced above, the question whether implementer I as a result of a market-based negotiation would be likely to pay rates closer to those paid by J or those paid by K can, in principle, be answered by observing the agreements that I, J, and K have concluded with other patent owners – precisely the third-party evidence that is typically missing from the record.

At the unpacking stage, market evidence offers an alternative to patent-count-based conventions. The clearest illustration is the valuation of cross-licenses: rather than valuing the counterparty’s portfolio by its share of patent counts, one can value it at the rates that this counterparty has itself achieved in its own unidirectional licensing agreements – that is, at the price the market has actually paid for that portfolio. Similarly, market-wide evidence on the relationship between lump-sum payments and licensees’ realized sales, across many agreements, can inform the assumptions used to convert individual lump sums into effective rates, and can indicate whether lump-sum agreements systematically embed a risk discount relative to running-royalty agreements.

At the adjustment stage, finally, market-wide evidence provides an alternative to rescaling individual agreements by patent counts. If rates for comparable portfolios have generally risen or fallen between the date of the comparable agreement and the date of the license under determination, this observed market-wide trend can be used to update the comparable rate without recourse to assumptions about the evolution of patent numbers. Likewise, observed differences between the rates of agreements with different geographic scopes can anchor territorial adjustments in actual market practice. And where the rates of the selected comparables themselves appear too low or too high to be FRAND, the distribution of rates across the broader market provides the external benchmark against which such outliers can be identified.

Market evidence beyond the courtroom

It is worth emphasizing, finally, that the value of public market evidence is not confined to the courtroom. Most FRAND rates are not determined by judges; they are agreed in bilateral negotiations, where neither party has access to the discovery apparatus of litigation, and where each must nevertheless form a view of what FRAND requires. This is the rationale behind the FRANDLY platform. By assembling the public record of the SEP licensing market – annual and quarterly reports, press releases, and legal documents covering licensing agreements and more than 600 identified SEP licenses since 2010 – into reproducible benchmarks, the platform aims to make the whole-of-market-evidence approach usable also at earlier stages of a licensing negotiation. Reliable public benchmarks serve both sides of the negotiation. They give patent owners a defensible, data-grounded basis for their offers, and implementers an evidentiary basis for their counteroffers – reducing, for both, the risk that their positions are later characterized as falling outside FRAND, or their conduct as unwilling.

About BRELA

BRELA (Research in Economics and Legal Analytics) provides evidence-based research and expert analysis at the intersection of technology, law, and policy. The firm specializes in the empirical analysis of technology standards, patents, SEP licensing, and FRAND determinations, turning large technical and legal datasets into clear, decision-relevant insights for corporations, law firms, standards organizations, and public-sector institutions, including the European Commission and the European Patent Office.

Beyond data-driven studies and consultancy, BRELA provides independent, methodologically rigorous expert testimony in complex SEP and FRAND matters. BRELA’s Director, Dr. Justus Baron, recently acted as expert witness in the UK FRAND proceedings Samsung v. ZTE, where the Court described his evidence as well-reasoned, carefully presented, and unanswered by any opposing expert. 

To discuss how BRELA can support your standards, IP, or policy questions, contact Dr. Justus Baron at justus.baron@brela-research.com.

About FRANDLY

FRANDLY is a licensing intelligence solution for SEPs that combines two complementary offerings. The platform delivers granular SEP licensing data, including royalty benchmarks, market analytics, and supporting documentation, while our research reports provide in-depth economic analysis of key SEP licensing questions, such as ‘The Value of 5G’.

Together, these give professionals and legal experts the evidentiary basis to make informed decisions across SEP negotiations, litigation, valuation, and policy development. Our data and analyses support reasoned, defensible determinations of FRAND licensing terms.

[1] InterDigital v. Lenovo [2023] EWHC 1583 (Pat), at 105. ↩

[2] Lucent Technologies, Inc. v. Gateway, Inc. (Fed. Cir.) addresses the significant risk-allocation differences between running-royalty and lump-sum licenses, and on the need for an evidentiary basis before converting between the two structures. ↩

[3] Each step of the comparables analysis rests on significant assumptions that are a recurring source of disagreement between expert witnesses and between first-instance and appeal courts. Concrete illustrations from the study: in Optis v. Apple, the first-instance court ([2023] EWHC 1095 (Ch)) built an inclusive analysis averaging across values derived from 14 agreements, which the Court of Appeal ([2025] EWCA Civ 552) overturned; in InterDigital v. Lenovo, the appeal narrowed the accepted comparables from 27 licenses to 2. ↩

[4] Courts have frequently unpacked cross-licenses rather than discarding them; and in InterDigital v. Lenovo, the court rejected the patent owner’s proposal to rely only on running-royalty licenses because of unpacking uncertainties, instead finding the lump-sum agreements with major implementers to be the best evidence of a FRAND rate. ↩

[5] Excluding all agreements with a lump-sum component would omit a significant and distinct portion of the licensing market and skew the resulting picture of market prices towards agreements with smaller licensees, since large implementers show a strong preference for lump-sum structures. ↩