There is a conceptual problem right at the heart of the consultations on proposed steering conduct requirements (CRs) for Apple and Google by the Competition and Markets Authority (CMA) in the UK. On the one hand, the CMA presents the new interventions as standard rules to ensure that developers can communicate with users, present alternative offers and direct users to external purchasing mechanisms on fair and reasonable terms. On the other hand, the proposals go materially beyond removing contractual anti-steering clauses and venture into a whole new area of prescriptive behavioural rules and, most remarkable, price regulation. The CMA may be taking aim at steering, but it certainly ends up in a different territory altogether.
The difference is significant. For instance, in both consultations, Chapter 3, sets out detailed requirements governing how an external purchasing route must be presented, accessed and used relative to the platforms’ in-app purchasing systems (paragraph 6); the permissible content, language, visual design and prominence of, and user interaction with, a single interstitial screen (paragraph 7, read together with the definition of “Neutral Language” in paragraph 2(h)); and prohibited forms of additional friction, including withholding pricing or promotional information, requiring additional access steps, and presenting deterrent warnings or disclosures (paragraph 8). They also limit the circumstances in which a steering fee may be charged for a non-recurring transaction to transactions occurring within seven days of the user reaching the alternative purchasing mechanism (paragraph 9(b)(i)), while misapplying that limit to recurring subscription payments, subject to a prohibition on charging for automatic renewals of subscriptions entered into before the redirection occurred (paragraph 9(b)(ii)).
Although similar mechanics appear in Japan and Brazil, it does not establish that the CMA’s requirements are necessary or proportionate. Apple’s Japanese implementation requires its in-app purchasing system to be offered at the same time as, and at least as prominently as, an alternative payment option, together with a disclosure sheet and a seven-day attribution period. These are features of Apple’s implementation, not requirements prescribed in that form by Japanese law. The Mobile Software Competition Act (MSCA) read together with the JFTC Guidelines, instead apply a case-specific test: they assess whether fees or design choices make link-out transactions practically difficult, prohibit reduced link visibility and promotional pop-ups, and permit warnings only where they are factual, non-discriminatory and no more extensive than necessary (JFTC Guidelines, pp. 60-63, Hypothetical Scenarios 87 and 89-95).
Brazil is a closer precedent for some of the detailed mechanics. The Administrative Council for Economic Defense (CADE) settlement requires Apple’s in-app purchasing system to be displayed alongside an active external link, gives developers control over design and language subject to equitable presentation, and limits Apple’s notices to neutral, objective information without unnecessary friction (CADE Opinion, paragraphs 169-174 and Schedule I, sections 3.1-3.6.1, pp. 42-46). It also permits a 15 per cent commission on sales completed within seven days of link activation (paragraphs 235-239 and Schedule II, section 4.1, pp. 61-62). Even so, it does not validate the CMA’s wider framework: it sets a specified rate rather than requiring cost- and value-based fee methodologies, lasts for three years, and allows reassessment where its objectives are not achieved (paragraphs 246 and 261-273, pp. 65 and 67-69). It therefore provides precedent for particular steering safeguards, not for the CMA’s broader system of interface prescription and continuing price supervision.
More fundamentally, the CMA proposes to regulate both the level and structure of steering fees, amounting in everything but name to a price regulation. Rather than setting an exact numerical rate, the proposed CRs would require each platform to demonstrate that every fee is fair and reasonable by reference to specified pricing principles and would prohibit fees exceeding the level determined in accordance with those principles. The cost- and value-based principles constrain the fee’s level, while the administrative-simplicity principle constrains its structure and operation (Chapter 3, paragraphs 9-13; and Chapter 4: Apple, paragraph 4.77; Google, paragraph 4.78).
The resulting framework is therefore extraordinarily complex and comes with contradictory justifications and exemptions. The proposed CRs identify recoverable and excluded costs, prescribe the treatment of “common costs” and a “reasonable return”, and require the value contributed by both the platform and developers to be considered. They also require records capable of substantiating the calculation. The accompanying monitoring framework contemplates implementation plans, reports every three months, data on transactions, uptake and charges, user-journey information and, where necessary, behavioural audits. Compliance would therefore require substantial new accounting, reporting and administrative systems of the very kind that the administrative-simplicity principle is ostensibly intended to avoid (Chapter 4: Apple, paragraphs 4.148-4.153; Google, paragraphs 4.147-4.152).
This raises several important policy concerns with the proposal, including the introduction of continuing supervision over fee methodologies, interface choices, and aspects of commercial design. While safeguards may be needed to prevent platforms from undermining the practical effectiveness of steering through excessive charges or artificial friction, the depth and breadth of the proposed requirements raise fundamental questions of proportionality, administrability, and institutional competence. Nor is it clear that the CMA’s proposed requirements are targeted at concerns specific to steering practices, or that they could be implemented without impairing important sources of user value, legitimate protections, and incentives to invest in platform development. It is not clear from the proposals whether the CMA has fully assessed the potential adverse consequences.
Moreover, under the CMA’s own framework, a conduct requirement must be effective, no more onerous than necessary, the least onerous of the equally effective alternatives identified by the CMA, and must not create disadvantages disproportionate to its aim (Appendix A, paragraph A.4(a)(i)-(iv); Chapter 5: Apple paragraph 5.2; Google paragraph 5.2). This has not been established by the CMA in the proposals. It says that a high fee could be used to defeat steering and establishes a need for an effective anti-circumvention rule. However, this does not establish that the CMA fee methodology is justified nor that it must apply from the first day. That additional step requires evidence and consideration of less intrusive designs capable of achieving the same objective. The CMA does not provide that, despite there being designs from other parts of the world that establish steering rules but in a far less complex, intrusive, and disproportionate way.
Against that proportionality framework, the consultations give rise to five principal concerns:
- The first concern is that the CMA imports concepts used in utility and access-price regulation without sufficiently explaining why those concepts should be transferred to an integrated and evolving app-store ecosystem or, more fundamentally, why app-store distribution should be treated as a stable and separable access service comparable to an airport, fixed telecommunications network, energy grid, or rail infrastructure.
- The second concern is that long-run incremental cost (LRIC) is materially under-specified. It cannot be meaningfully calculated without defining the relevant increment and counterfactual, and those choices materially affect the resulting cost estimate.
- The third concern is that the value-based principle requires an inherently uncertain allocation of ecosystem value between the platform and developers, adjusted for market power and the respective contributions of each side, without providing a clear methodology for making that allocation.
- The fourth concern is that the proposals do not adequately assess the dynamic and distributional effects on investment and cross-subsidies, or the implications for security, developer tools and smaller developers. Nor are the claimed benefits of lower steering fees robustly quantified.
- And lastly, the fifth concern is that the requirements provide no sufficiently clear prospective safe harbour. Each platform must formulate its fee and demonstrate to the CMA’s satisfaction that its level, structure and operation comply with the prescribed principles. This leaves material uncertainty as to which fee arrangements will be regarded as compliant unless and until the CMA has assessed them.
There is also a final point to explore. The CMA is not the first competition or regulatory authority that has proposed or established steering rules. No other authority, however, has pursued the same type of regulation as the CMA. While rules and enforcement elsewhere have been associated with problems too, they have been more specific to the defined concerns. The experience and evidence from other markets should be used by the CMA to tailor better and more proportional conduct requirements.
Concern 1: The Utility-Regulation Analogy Underlying the Cost-Based Principle
By the standards of anti-steering rules applied around the world, the CMA clearly errs on the extreme side. What is the model of thinking behind the proposed conduct requirements? The concern is that the CMA relies on methods developed for regulated access to relatively stable infrastructure without first establishing that app-store services can be isolated and priced in the same way.
Both proposals essentially follow the same structure. Each applies the cost-based, value-based and administrative-simplicity principles in paragraph 10 of the proposed CR. Each also refers to four regulatory comparators: the CAA’s Heathrow price control, Ofcom’s regulation of wholesale fixed telecommunications, Ofgem’s RIIO framework, and the Office of Rail and Road’s approach to Network Rail access charges (Chapter 4: Apple paragraphs 4.104-4.108 and footnote 128; Google paragraphs 4.105-4.108 and footnote 133).
The CMA, however, nowhere explains why those indirect comparators should determine the regulation of app-store steering, while the more directly relevant experience of jurisdictions already regulating alternative payment routes is given little methodological weight. Nor does it establish that app-store services can be isolated, attributed and priced in the same way, or explain why this model is preferable to the more targeted approaches used in other app-store markets. The platforms may recover forward-looking LRIC, an appropriate allocation of common cost and a reasonable return, but only for services used to a ‘material and direct extent’ in facilitating or supporting steered transactions (Chapter 3: Apple and Google paragraphs 10(a) and 12).
The cited regulatory comparators are therefore an awkward selection. In reality, platform markets are very different from stable infrastructure markets. The comparators all concern legacy infrastructure sectors with long histories of state ownership and economic regulation, factors that have profoundly shaped their market structures and contributed to high levels of concentration. Moreover, they generally begin with a service or asset that can be identified and costed with some stability, partly because innovation is not as systemic and market changing as in platforms. By contrast, a digital purchase through an app store is integrated in a system that combines distribution, discovery, rule-setting, software infrastructure and protections for users. Those functions are technically and commercially interdependent.
The CMA starts from the “provision of steered transactions” and suggests that the incremental costs may be limited to the technology required to offer external links. This is just not correct. That approach defines the regulated activity far too narrowly and transfer techniques developed for different kinds of infrastructure without first establishing their economic fit. The CMA describes these approaches as being used for “more utility-like services” and says that they “could offer relevant parallels” for steering. But it does not undertake a comparison of the economic characteristics that would make those parallels reliable and relevant, such as the relevant increment, cost causation, demand uncertainty, the treatment of innovation or the interdependence of services (Chapter 4: Apple and Google paragraph 4.106).
Revenue associated with digital transactions may support a wider package of functions: app review and distribution, search and discovery, APIs and development tools, account and subscription management, security and privacy controls, device integration, and access to an established user base. This does not mean that every such function should be charged to a steered transaction. The proposed CR itself requires a material and direct connection. It does mean, however, that the CMA should explain why the selected accounting and valuation techniques can separate and price the relevant services reliably in an ecosystem whose components and business models continue to change (Chapter 3: Apple and Google paragraphs 10-13). The CMA seeks to bridge that gap by permitting an allocation of common costs and a reasonable rate of return, and by relying on evidence of the platforms’ profitability. But those considerations concern the permissible recovery and likely level of the fee; they do not establish that the relevant services and costs can first be separated and attributed reliably (Chapter 4: Apple paragraphs 4.108-4.113; Google paragraphs 4.108-4.111).
The platforms’ treatment of physical goods and services also reinforces the concern that the CMA has not established the necessity of its chosen regulatory model. Apps selling physical goods and services, including ride-hailing and grocery delivery, may generally use developer-selected payment arrangements and direct users outside the platform’s billing system without the app-store commission applied to digital content (Chapter 1: Apple paragraph 1.7 and footnote 11, and Chapter 4: paragraphs 4.20 and 4.64; Google paragraph 1.7 and footnote 11, and paragraphs 4.20 and 4.65).
But that comparison does not answer the central pricing question. A commission on a steered digital transaction may remunerate not only the incremental cost of enabling steering, but also the platform and app-store services that the developer continues to receive, including distribution, discovery, software infrastructure and access to users. The consultations largely collapse those distinct elements into the cost of supporting the external purchasing route. Indeed, the CMA itself relies on the physical-goods comparison in concluding that the incremental cost of permitting steering should be very low (Chapter 5: Apple paragraph 5.15; Google paragraph 5.13). That may be relevant to the cost of the link itself, but it does not establish what, if anything, may reasonably be charged for the wider services that continue to support the developer and the transaction.
Nor does the CMA seem to be unaware of the underlying problem. Apple argued that its commission remunerates the combined value of distribution, discovery, developer tools, trust and safety, and payment services, and that restricting its ability to charge would permit developers to free-ride while weakening its incentives to invest. It submitted that cost-based regulation is unsuitable for a platform characterised by high innovation, unpredictable demand and jointly supplied services whose costs are difficult to allocate; that it would exclude Apple’s contribution to joint value creation; and that applying it only to steered transactions would discriminate against developers using in-app purchasing. Apple also stated that it does not ordinarily allocate costs systematically to particular products or services and could not therefore provide reliable estimates of the relevant App Store expenditure (Chapter 4: Apple paragraphs 4.89-4.93 and 4.109).
Google made a related but distinct argument. It maintained that its fees remunerate differentiated services, including trust, discovery, delivery and continuing support and that cost-based pricing would prevent an app store from being rewarded for investment, penalise more efficient operators with lower costs, and weaken incentives for risk-taking and innovation. It also argued that such an approach was inconsistent with the legal framework for fair pricing. Google could not provide a UK cost figure for the Play Store because Play’s business is organised globally (Chapter 4: Google paragraphs 4.90-4.93 and 4.109).
Nevertheless, the CMA acknowledges these difficulties. As to the accounting problem, however, it responds that the platforms’ current inability to identify the relevant costs is not a reason to reject the principle and that they should instead be required to create the necessary accounting records (Chapter 4: Apple paragraphs 4.92-4.93, 4.109 and 4.114-4.116; Google paragraphs 4.92-4.93, 4.109 and 4.112-4.114). But accounting is not magic and cannot make the reality fit with the narrow model that the CMA has outlined of platform costs.
Concern 2: The Cost Principle Turns On Unresolved Attribution Choices
The second concern is rather operational. Even if a cost-based principle is accepted, the consultations do not define the increment or the allocation rules that would produce the permitted fee. The concern is therefore not with cost recovery in principle, but with the substantive policy choices hidden inside the calculation.
An incremental-cost exercise cannot begin until the regulator identifies the activity to be added or removed in the counterfactual. Possible increments include one additional steered transaction, one developer’s participation, all steering developers, the UK storefront, a distribution function, or a broader platform activity. The choice determines which costs are incremental, which are common and how much cost can be attributed to steering. It is therefore a substantive policy choice, not merely a technical accounting input.
The consultations reveal this difficulty. In the CMA’s LRIC illustration, the incremental activity may be framed as the technology that enables developers to provide external links (Chapter 4: Apple paragraph 4.107; Google paragraph 4.107). The final proposal is not a pure LRIC cap: it also permits an appropriate allocation of common cost and a reasonable return. Even so, the illustration matters because the additional amount depends on judgments about which shared functions have the required connection with steered transactions. The direct engineering cost of permitting an additional link may be modest, while the economically contested issue is the share, if any, of the surrounding platform services that should follow the transaction reached through that link.
The exclusions in Chapter 3, paragraph 12(b) of in both consultations further shape the result. The cost base must omit the platform’s own downstream digital content; general discovery, marketing and developer tools not used by steering developers to a ‘material and direct extent’; and in-app payment security or privacy features that are not supplied or required for the steered transaction. Each exclusion is capable of being justified by reference to causation. Their combined effect, however, makes the material-and-direct threshold central to the outcome. Paragraph 12(a)(iii) requires common costs to be allocated through “objective, causally relevant and proportionate cost drivers”, but it does not identify those drivers or explain how the benefits and costs of jointly supplied services are to be divided (paragraphs 12(a)(iii) and 12(b)).
The same problem remains when common costs are considered. Complexities become confusing, and the proposals end up split between different objectives. For instance, Apple told the CMA that it does not ordinarily allocate costs systematically to individual products or services and could not provide accurate and reliable estimates of its investment in the relevant App Store changes. It was also unable to provide the annual cost of operating the UK App Store (Apple consultation chapter 4: paragraphs 4.93 and 4.109). The proposed CR would therefore require Apple to create and maintain records capable of supporting the calculation (Chapter 3: paragraph 12(d); Chapter 4: paragraph 4.116). At the same time, the CMA expects the resulting fee to be “significantly lower” than Apple’s current charges because it regards the App Store as a low-cost, high-volume platform business and points to Apple’s profitability (Chapter 4: paragraphs 4.110-4.113). What is the principal motivation for that conclusion? And ‘significantly lower’ for whom? The evidence seems to come from general profitability, but such general metrics cannot decide the allocation question that the proposed records are intended to answer.
Google similarly stated that it could not provide an annual UK’s cost for operating Play because the service is organised globally (Google consultation, chapter 4: paragraphs 4.109 and 4.114 and footnote 139). The proposed method must nevertheless allocate shared activities and investment, including review, security, updates, developer APIs, Android infrastructure and arrangements with device manufacturers, among Android, Play services, steered transactions and other activity. Better records could cure a data gap. But they do not determine the conceptual basis on which globally shared and jointly productive costs should be divided. The CMA’s expectation of a materially lower cost-based fee, through reduced profitability, comes from an estimated average return on capital employed for Alphabet of about 38 per cent over ten years, compared with an estimated cost of capital of 10-15 per cent. It finds that Play Store margins exceeded those of Google Services and Google overall (Chapter 4: paragraphs 4.110-4.111).
This may be correct or incorrect, but it does not help to address the specific issue. Profitability does not define the increment, choose the cost driver or convert itself into a lawful fee. Those additional steps require separate justification. What happens is that the cost principle cannot simply operate as a neutral calculation. The CMA first has to identify the regulated increment, the relevant cost drivers and the rules for allocating the costs of shared services. Those choices can materially determine the outcome, so they should be justified as part of the conduct requirement rather than left to bilateral accounting exercises after the rule takes effect.
Concern 3: The Value Principle Does Not Identify Its Own Allocation Rule
How are the platforms supposed to work out and the CMA supposed to assess, what share of the relevant value belongs to the platform, while removing the effects of market power and recognising the value contributed by developers, when there is no clear benchmark for doing so? The value-based principle permits a fee above the cost-based measure where the platform’s offer to steering developers is especially differentiated, valuable or the product of continuing innovation. The calculation must also remove the effect of substantial and entrenched market power and recognise the value that developers contribute to the wider mobile platform (Apple and Google consultations: Chapter 3, paragraphs 10(b) and 13(c)-(e); Chapter 4: Apple, paragraphs 4.120 and 4.122-4.124; Google, paragraphs 4.118 and 4.120-4.122).
The framework, however, treats market power as a one-sided issue. It discounts the platform’s contribution for the effects of its market power while treating the developer’s contribution as an offset, without considering whether particular developers themselves possess substantial downstream market power or countervailing bargaining strength. This matters because the developers most likely to steer are large, established services with strong brands and direct customer relationships. Spotify, for example, sells Premium subscriptions outside the iOS app, illustrating that at least some developers can monetise iOS users without using Apple’s in-app system. The consultations do not explain how these material differences between developers should affect the allocation of value.
The CMA itself acknowledges that adjusting for market power and accounting for the two-way exchange of value would be complex and open to a degree of subjectivity (Chapter 4: Apple paragraph 4.126; Google paragraph 4.124), but none of this explains why a framework that the CMA accepts is so indeterminate should nevertheless be used to regulate steering fees. Paragraph 13 of Chapter 3 in each consultation adds further conditions, but these largely amount to broad standards rather than an allocation rule. They do not explain how the different elements of value are to be measured, weighed, or converted into a fee.
The consultations invoke standard-essential-patent (SEPs) licensing and FRAND analysis as possible analogies, including comparable licences and hypothetical negotiation, to estimate value without rewarding bargaining power created by lock-in (Apple consultation, footnote 137; Google consultation, footnote 140). This is an awkward analogy. The institutional setting is materially different. A FRAND obligation usually traces back to a commitment made through a standard-setting process; SMS status is imposed later under statute. The CMA would therefore not be applying an earlier undertaking to a defined asset. It would be reconstructing an acceptable division of value after the ecosystem, user base and commercial dependencies have formed. That difference does not rule out a value inquiry, but it removes one of the anchors that gives the FRAND exercise its frame of reference.
The established valuation techniques do not remove that uncertainty. A licence used as a comparator may reflect the threat of litigation, the availability of injunctions and the bargaining characteristics of a particular portfolio. Top-down and bottom-up models depend on assumptions about contribution, the relevant base and the appropriate aggregate return. Those problems arise even when the subject is a defined patent portfolio governed by an express licensing commitment. Even the UK IPO has noted that there is no single methodology for calculating a FRAND rate, and that comparable licences are often difficult to assess because their terms are confidential, and that different methods of determining aggregate royalties can produce widely varying outcomes.
Here the proposed calculation must account for distribution, discovery, trust, security, developer participation, access to users and ecosystem investment, while excluding value attributed to entrenched market power. The consultations do not identify a market transaction that reveals that allocation. They also reject the platforms’ proposed comparators because those comparators do not simultaneously involve sufficiently similar services, effective competitive constraint and recognition of the value supplied by both sides (Chapter 4: Apple paragraphs 4.127-4.129; Google paragraphs 4.125-4.128).
There is also circularity in the proportionality analysis. The CMA may properly distinguish a transfer of revenue from a net loss to society (Chapter 5: Apple paragraphs 5.17-5.18; Google paragraph 5.15). But it cannot assume, before applying the proposed methodology, that the relevant part of the existing fee is entirely a product of market power, use that assumption to predict a lower fee, and then give the lost revenue little weight because it is said to be a market-power transfer. The amount attributable to market power is one of the matters the value exercise is supposed to determine. The CMA rejects Apple comparators unless they involve sufficiently similar services, effective competitive constraint and recognition of value supplied by both sides. The same evidential discipline should apply to the CMA’s own estimate.
Concern 4: Dynamic and Distributional Effects, And an Unproven Benefit Case
The CMA’s case rests on the expectation that lower fees will produce substantial benefits while treating the principal costs – lower platform revenue, compliance expenditure, altered investment incentives and security or consumer-protection trade-offs – as small or avoidable. The consultations do not quantify those effects with comparable rigour or explain how they will be distributed across large and small developers, users and the wider ecosystem.
The benefit case is stated in broad and potentially significant terms. The proposals suggest that lower commissions may lead to lower prices, greater developer investment, new entry, more innovative business models and increased payments to downstream creators or smaller developers. They also state that, even with relatively low uptake and modest price reductions, tens of millions of pounds each year could be retained by developers or passed on to UK end-users (Chapter 5: Apple paragraphs 5.35-5.45; Google paragraphs 5.32-5.42). Those are theoretically plausible channels of benefit. The difficulty is that the key quantities, likely uptake, the size of any fee reduction, pass-through, substitution between payment routes and differences across developer categories are not yet observed. Nor does it conform to observed user reality in other jurisdictions where platform fees have been cut: actual fees for big and popular apps, do not seem to move with the platform fees charged on the developer.
The corresponding cost analysis gets even narrower. Under Chapter 5 of both consultations, the CMA treats lower platform revenue largely as a transfer associated with market power and considers compliance costs unlikely to be substantial and concludes that residual privacy and security risks will be very small or immaterial (Apple paragraphs 5.13-5.34; Google paragraphs 5.13-5.31). The concern that remains is what the intervention could do to expected returns on platform investment, the funding of common services and cross-subsidies, spending on security and developer tools, the treatment of smaller developers, and the structure of charges elsewhere in the ecosystem. Those effects may be real even if part of the existing fee reflects market power. Much of the CMA’s analysis proceeds on the basis that developers may choose not to steer and will do so only where the benefits exceed the costs (Apple paragraphs 5.18 and 5.33; Google paragraphs 5.16 and 5.30). That does not, in itself, resolve the distributional question of which developers can use the remedy effectively and which may remain dependent on the existing in-app purchasing system.
Interface, Security and Consumer-Protection Effects
The effective-steering provisions are broad. Both platforms would be prohibited from preventing, restricting or limiting steering, directly or indirectly, through contractual terms, technical design requirements, app-review processes, enforcement practices or other measures. The definition of an End-user Redirection Mechanism includes functionality that enables, or is ‘reasonably likely to encourage’, an end-user to access an alternative destination (Chapter 3: paragraph 2(i) and paragraphs 3-5). A nominal permission to steer can be ineffective if the surrounding conditions materially suppress its use. At the same time, the breadth of the drafting means that legitimate technical, privacy, security and commercial standards may fall within the initial scope of the prohibition and depend on the availability of a justification.
The CR also contains a security-related exception. A platform may restrict a redirection mechanism where the restriction is “strictly necessary” for an “objectively justifiable” purpose of preventing malware, fraud, unlawful content or content harmful to children (Chapter 3: paragraph 3(b)). Though the consultations discuss privacy and security at length and rely on several mitigations, including continuing app review, the availability of the integrated payment system alongside steering, a single interstitial screen, protections for children and the fact that the CRs requires link-outs for transactions rather than app downloads, the CMA provisionally concludes that it has received no robust evidence of observed privacy or security harm from steering in other jurisdictions (Chapter 4: Apple paragraphs 4.17-4.74 and chapter 5: 5.19-5.31; Google paragraphs 4.17-4.75 and 5.17-5.28).
The narrower concern lies in the operative drafting. In Chapter 3, the exception in paragraph 3(b) is confined to the listed purposes and subject to a strict necessity test, while the equal-treatment and no-additional-friction obligations in paragraphs 6-8 extend to ordinary design and information choices. But what does this mean in practice? The proposed CRs do not explain how privacy, authentication, informed consent, refunds, subscription administration, and other protections not expressly listed in paragraph 3(b) will be assessed. Consequently, a measure intended to address a genuine risk could initially be characterised as prohibited friction and only subsequently assessed through enforcement.
Interface neutrality is also not realistically available. Showing the integrated payment option beside the developer’s route may favour the platform because users already know it; removing it may instead favour the developer. A disclosure screen adds a step, but its absence can leave users unclear about the counterparty, destination, refund route and protections that apply. Stored credentials, ordering, branding, wording, placement and the number of interactions can all affect completion. The CRs therefore require the CMA not merely to prohibit deception or deliberate obstruction, but to decide which differences between two technically and legally distinct purchasing routes are permissible.
Experience with choice mechanisms in adjacent digital markets cautions against assuming that a mandated presentation will by itself change established behaviour. European browser and search choice measures have reported limited market-share effects, while changes to search defaults indicate that many users return to familiar services. These are not direct studies of app-store purchases and should not be treated as such. Their narrower relevance is that visibility alone may not overcome familiarity, switching costs and user habit. The CMA should therefore define the particular friction it seeks to prevent and evaluate the remedy using observed completion, abandonment and comprehension rather than treating visual similarity as an independent objective.
The CMA’s own evidence confirms that relatively small changes in a purchasing journey can have substantial commercial effects. Developers reported that moving subscriptions to the web could reduce take-home revenue by about 6 per cent; that conversion for steered transactions could be approximately 30-60 per cent, compared with rates consistently above 60 per cent for in-app purchase; and that a single interstitial could reduce conversion by estimates ranging from about 5 per cent to more than 60 per cent (Chapter 4: Apple paragraphs 4.61-4.63 and paragraph 4.97; Google paragraphs 4.62-4.63 and paragraph 4.97). The CMA should test its own design choices rather than assume that the selected architecture will be neutral for competition.
Design obligations can also redistribute security and consumer-protection responsibilities. In an integrated purchase, the platform may participate in authenticating payment, administering subscriptions, processing refunds, detecting fraud and providing a route for disputes. An external transaction can place some or all of those functions with the developer, a payment provider or another intermediary. The relevant question thereafter is which entity controls payment credentials, recurring charges, cancellation, refunds, user information and redress in the particular journey. Measures directed to demonstrable security, privacy, and consumer-information risks should be permitted where they are transparent, objectively justified and proportionate. Platform-published fraud or malware figures can illustrate the asserted scale of the risk but remain interesting evidence. The CMA should compare benefits and disadvantages on a common evidential basis and identify how the effects fall across different groups. A broad prediction of lower prices and innovation is not enough to establish that the proposed fee and interface regime will avoid disproportionate dynamic or distributional costs.
Concern 5: The Framework Provides No Prospective Safe Harbour
Each platform must formulate a fee and then satisfy the CMA that all three principles have been applied correctly. The consultations preserve discretion but offer no prospective safe harbour and no decision rule that a platform can use with confidence before charging. In particular, the documents say that cost-based and value-based calculations can inform and cross-check one another, but they do not state which result governs when the two differ materially (Chapter 4: Apple paragraphs 4.139-4.141; Google paragraphs 4.138-4.140).
A platform may therefore understand that its present arrangement is unlikely to be accepted yet remain uncertain about the lawful cost drivers, common-cost allocation, valuation assumptions and relationship between the two calculations. Given that the consultations do not say that the CMA’s acceptance of an implementation plan would bind it or protect the platform from later enforcement, compliance would in practice be settled through continuing regulatory engagement rather than through a rule whose application can be predicted in advance.
The CMA also relies on its ability to introduce greater prescription at a later stage. It states that, where a higher-level conduct requirement proves ineffective or a specific corrective measure becomes necessary, it has the legal power to impose more detailed requirements (Chapter 4: Apple paragraph 4.78; Google paragraph 4.79). The CMA has therefore already accepted the legitimacy of staged intervention. The proportionality question is not whether further detail may ever be required, but how far the initial requirement should go before evidence of circumvention or ineffectiveness has emerged. That question is also not answered by the CMA’s conclusion that it has identified no equally effective, less onerous design (Chapter 5: Apple and Google paragraphs 5.8-5.10). The consultations simply do not assess a lighter two-stage alternative under which the platforms would first be required to permit effective steering, with scrutiny of fee levels and structures triggered where evidence showed that remuneration was frustrating the remedy.
More broadly, predictable, proportionate and evidence-led enforcement reduces the risk that competition rules develop into continuing regulatory supervision of product design and pricing architecture. That concern does not determine the precise form of the UK remedy, but it is directly relevant to whether the CMA’s proposed first stage is the least onerous equally effective means of securing effective steering. The uncertainty is amplified by the accompanying monitoring framework, which envisages implementation plans, compliance reports every three months, extensive transaction and pricing information, and possible behavioural audits (Chapter 4: Apple paragraphs 4.149-4.153; Google, paragraphs 4.148-4.152). Where the substantive fee standard is itself indeterminate, those mechanisms make continuing regulatory engagement part of the compliance architecture rather than merely a check against a standard that can be applied prospectively.
The experience of the EU’s Digital Markets Act (DMA) further illustrates why the line between anti-circumvention and price supervision matters. On 23 April 2025, the European Commission imposed a EUR 500 million fine on Apple for non-compliance with the DMA’s anti-steering obligation. Apple appealed, revised its commercial terms and remained subject to further compliance assessment. The chronology supports two conclusions. A broadly worded steering right may require active enforcement against circumvention, but a formal decision does not necessarily settle the permissible structure of remuneration. Repeated changes to the applicable terms can also impede developers’ commercial planning and reinforce the risk that fee supervision becomes an open-ended process of regulatory renegotiation. The CMA should therefore identify in advance the circumstances in which more intrusive scrutiny of fee levels or structures will be triggered. Clear escalation criteria would preserve the CMA’s ability to respond to circumvention while reducing uncertainty about the operation of the initial conduct requirement.
A Less Onerous Alternative: A Conduct-First Sequence
The CMA is not the first authority to adopt or propose steering rules. Other regimes have also encountered enforcement and design problems, but the comparators considered here do not use the same combination of ex ante cost allocation, ecosystem valuation and continuing supervision proposed by the CMA. Comparative experience should therefore be used to tailor a narrower and more proportionate UK requirement.
Japan provides a closer comparator
As noted earlier, Japan’s MSCA, is arguably a more appropriate comparator for the CMA, than the utility regimes cited by the CMA because it specifically regulates designated providers of smartphone operating systems (OS) and app stores, among other forms of specified smartphone software, and addresses restrictions on alternative payment arrangements. Article 8 MSCA, read with the JFTC Guidelines, asks whether platform conduct creates a high likelihood of making the use of an alternative payment-management service difficult in practice, even where such use is not formally prohibited.
The examples include imposing excessive financial burdens, displaying the alternative through smaller or less prominent buttons or text, and repeatedly promoting the platform’s own payment service. The assessment of a financial burden is contextual and may consider the aggregate charges attached to the competing routes and whether an efficient alternative-payment provider could continue operating (JFTC Guidelines, pages 50–52, Hypothetical Scenarios 72–75).
More directly for steering, the Guidelines treat as potentially prohibited a fee that creates a high likelihood of making link-out transactions practically difficult, worsening the visibility of link-out buttons or text, and displaying a pop-up promoting the platform’s own payment service when a developer provides a link-out (JFTC Guidelines, pages 60–61, Hypothetical Scenarios 89–92). The level of a link-out fee is assessed case by case, taking account of the benefits developers obtain from the app store and the fees imposed by efficient alternative app stores. A blanket prohibition on pricing information or payment-page links, however, is not justified where the risk can be addressed through verification, correction and restrictions directed at developers likely to mislead users (JFTC Guidelines, pages 61–63, Hypothetical Scenarios 93–95).
The Guidelines also place security and consumer protection inside the substantive assessment. Requirements concerning payment-information handling, effective refund and cancellation procedures, parental controls and objective cybersecurity standards may be justified where they are reasonable and limited to what is necessary. By contrast, a general prohibition on alternative payments is not justified merely by invoking security where objective standards, review and less restrictive measures can address the risk (JFTC Guidelines, pages 52-54, Hypothetical Scenarios 76-80).
Japan’s framework therefore offers a less onerous alternative for the CMA. Rather than determining the permissible level of every charge in advance, the CMA could specify the forms of discrimination and circumvention that are prohibited, monitor their practical effects, and intervene where a fee or interface design makes an alternative commercially unusable. Japan still requires transparency about fees: designated providers are expected to explain why a financial burden is reasonable in light of the benefits obtained from the app store and must report the calculation method and reasons for imposing it. But that is materially different from prescribing a cost-based and value-based methodology for every fee from the outset. The Japanese model is narrower and more targeted than the DMA, combining ex ante obligations with detailed guidance and case-specific assessment. This comparison concerns the design of the framework rather than established enforcement outcomes.
Conclusion
Ultimately, all five concerns point to the same conclusion: the CMA should avoid immediately adopting a comprehensive fee methodology as a way to make steering effective. A staged rule would address the identified steering concern from the outset while reserving detailed price supervision for evidence of circumvention. Looking at the Japanese framework, this matters under the CMA’s proportionality framework because a conduct requirement must be no more onerous than necessary and, where multiple equally effective options exist, the CMA must choose the least onerous; it must also avoid disadvantages disproportionate to its aim (Chapter 5: Apple and Google paragraph 5.2).
The timing further reinforces the need for caution over the form and sequencing of intervention. Google’s revised global terms entered into force in the UK on 30 June 2026, the date on which the consultations were published, and the CMA states that it will assess their likely impact during the next phase of its work. The consultations also recognise that several relevant overseas and domestic proceedings remain subject to appeals, remedial processes or continuing compliance assessment (Chapter 1: Apple paragraphs 1.15–1.17; Google paragraphs 1.14–1.17). These considerations make it difficult to conclude, on the present record, that the proposed fee methodology is the least onerous equally effective means of securing effective steering. The analysis therefore supports retaining a conduct-first approach before introducing comprehensive fee regulation.
You can also watch Fredrik Erixon and Dyuti Pandya discuss this topic on YouTube.