In competition law, how to define markets in the context of two-sided platforms is a hot topic. Drawing on the judgment of the UK Competition Appeal Tribunal in BGL/Comparethemarket v. Competition and Markets Authority, we address some of the main questions and misunderstandings about multi-sided market definition.
By Andreea Antuca, Gunnar Niels & Helen Ralston-Smith
Market definition in the context of two-sided platforms has been subject to debate in competition policy for some time. The judgment of the UK Competition Appeal Tribunal (“CAT”) in BGL/Comparethemarket v. Competition and Markets Authority provides useful additional guidance, but also leaves questions unanswered.
This case arose following the Competition and Markets Authority’s (“CMA”) infringement decision against BGL for the use of wide most-favored-nation (“MFNs”) clauses (also known as price parity clauses) in the distribution of home insurance by its price comparison website (“PCW”) Comparethemarket. In the UK, PCWs emerged in the early 2000s, quickly becoming a popular way for consumers to search for, and purchase, insurance products.
The CMA recognized that these PCWs are, in general, a “force for good,” stimulating competition between suppliers by engaging consumers and assisting in search and comparison, thereby countering consumer inertia (the majority of home insurance policies in the UK continue to be renewed each year, without the customer looking to switch). However, concerns arose about PCWs’ use of MFNs — in particular, wide MFNs, which require suppliers to offer the PCW their best price and not offer a lower price on another PCW or online channel.
In 2015, the CMA had prohibited the use of wide MFNs by PCWs in the private motor insurance market. In 2017, it opened a Chapter 1 and Article 101 TFEU investigation into Comparethemarket’s use of wide MFNs in home insurance, which resulted in an infringement decision and fine of £17.9m in November 2020. CompareTheMarket was the most popular PCW for home insurance products, with approximately 50% of sales through this channel; however, this fell to approximately 15% when other channels (including renewals) were taken into account, rendering PCWs a small part of the overall market. Understanding whether home insurance sales through these other channels formed part of the relevant market was important in assessing the competitive effects.
Common questions in the general platform market definition debate considered in this case include:
- How many markets should be defined? A single market encapsulating competitive constraints on both sides of the PCW platform (insurers and consumers) or one for each side?
- How should competitive constraints faced by the PCW platform be assessed?
- How to apply the SSNIP (small but significant and non-transitory increase in price) test when the platform’s consumer prices are zero
In cases involving two-sided platforms, the answers to these questions have not always been consistent. In MasterCard, the European Commission defined several interrelated markets, considering that the definition of one single market would ignore the different levels of competitive interactions in payment card platforms: the platform (competing with other payment methods); the acquiring banks on one side of the platform (competing for the business of merchants): and the issuing banks on the other side (competing for cardholders). This approach was generally followed in subsequent cases involving payment card schemes across Europe. Yet, in Amex, the U.S. Supreme Court took a different view: it found that if both sides of customers (merchants and cardholders) are required to participate at the same time for a transaction to occur then one single market must be defined. The Commission’s merger control decisions (e.g. in Microsoft/LinkedIn) seem to have shifted towards a position of defining one single market for transaction platforms, and the UK CMA has also adopted this approach, for example in cases involving online restaurant platforms.
In terms of how to define the relevant markets, indirect network effects between the two sides of the platform may make demand substitution more complex to assess, but this complexity is not insurmountable. In European competition policy, we have observed a degree of consensus on the usefulness of the SSNIP test, at least as a thought framework, which from an economics perspective is appropriate. This was recently confirmed in the Commission’s draft updated Market Definition Notice (issued for consultation in November 2022). The value of the SSNIP test in exploring demand substitution was also confirmed by the CAT in Comparethemarket. In earlier European cases involving two-sided platforms — e.g. the Swedish and Italian competition investigations into hotel booking platform Booking.com — the focus was often on a comparison of product characteristics, and potential competitors that did not look like Booking.com were rejected. In comparison, in Comparethemarket, the CAT embraced the use of the SSNIP test to understand demand-side substitutability (as had the CMA and the BGL itself).
II. WHEN IS A PLATFORM TWO-SIDED?
The label “two-sided platform” is commonly used for all sorts of digital platforms, not just in competition policy but also in the business community. But when is an intermediary a truly two-sided platform and what does this mean for market definition?
Two-sidedness arises when a platform serves two groups of consumers, and the platform’s value to one group increases with the use of the platform by the other group — commonly referred to as indirect or two-sided network effects.
In the case of PCWs such network effects are clear: the value of the PCW to an insurer increases with the number of consumers browsing on the PCW, as this raises the chance that the insurer makes a sale; similarly, the value of the PCW to consumers increases with the number of insurers available on the PCW. In this case, the indirect network effects work in both directions: from consumers to insurers and vice versa. Sometimes, however, indirect network effects go in only one direction — e.g. with social media platforms, the value of the platform to advertisers increases with the number of consumers using the platform, but it is less clear that the value of the platform to consumers increases with the number of advertisers (it may in fact decrease).
Figure 1 Types of network effects
As illustrated in Figure 1, network effects can also be single-sided (between the same type of user), in which case they are often referred to as “direct” network effects — examples include users on a social network site and traders in a financial market.
Network effects may not be the only economic characteristic exhibited by a platform. When defining markets it is important to be explicit about all characteristics to avoid over-generalizing. Moreover, the two-sided network effects may not always be strong, in which case the traditional approach to market definition may require only minimal adjustment.
Supermarkets, for example, also serve two user group that benefit from one another — producers and consumers — as do airports, restaurants, music festivals, business conferences, but in comparison to digital platforms, the two-sided network effects in these situations tend to be relatively weak. There are only so many brands a supermarket can stack on its shelves and consumers who can fit in a store. In such situations, other characteristics can be more important to factor into the approach to market definition — for example, the role played by supermarkets in vertical supply chains, and the economies of scale, density and scope that can arise.
One trap to avoid with two-sided platforms is an assumption that competitors must all share the same product characteristics as the provider of the focal product. What matters is substitutability in the eyes of users, not the product characteristics as such. The CAT judgment makes an important finding in this respect. It highlights that PCWs are simply the latest addition to the many ways consumers and insurers transact with one another, and that to overlook the potential competitive constraint of traditional intermediaries and direct channels is an error:
Ultimately, the Decision fails to see what it calls Price Comparison Services in their true and proper context, which is as an insurance intermediary. […] These days, intermediaries come in all shapes and sizes: thus, the seller of a product will often seek to sell insurance also. The price comparison website is a (relatively late) addition to the list of indirect channels or Interfaces by way of which insurance products can be sold. […] It is a failure to understand this dimension that is a hallmark of the Decision.
III. HOW TO TEST BOTH SIDES PROPERLY: ONE SSNIP OR TWO?
When defining the relevant market for a two-sided platform, does one need to test one or two markets, and with one or two SSNIP tests?
Among the three parties involved — the CMA, BGL and the CAT — the Comparethemarket case discussed all possible combinations. The CMA proposed one overall market and one SSNIP test; BGL proposed one market encompassing both sides and two SSNIP tests that need to be considered together; while the CAT defined two separate markets with two different SSNIP tests. This range of options reflects the ongoing debate in the economic literature that has yet to reach a consensus on how many separate markets should be defined when dealing with multi-sided markets (perhaps because there is no single correct method). However, what ultimately matters is whether the exercise of market definition properly identifies the existing constraints from other channels (e.g. direct insurer websites, brokers, renewals) on the hypothetical monopolist platform, which competes on both sides simultaneously.
While all parties agreed that both sides need to be properly tested, the CMA took the view that the same constraints apply to both sides of the market, and hence that one SSNIP test is sufficient to ascertain the substitutability of both insurers and consumers. In its view, an increase in the commission fee paid by insurers to PCWs would directly test whether insurers would switch to other channels in sufficient numbers to constrain the hypothetical monopolist. Up to this point, all parties agreed that this is an appropriate test for the insurers’ side, or, as the CAT put it, “a vanilla” application of the SSNIP test. On this side of the market, the CAT agreed with the CMA’s conclusion that insurers are unlikely to be a sufficient constraint on the hypothetical monopolist and will continue using the PCW even if the fees increase — i.e. the CAT defined a narrow market on the insurers side (BGL had shown evidence that disputed this finding).
However, the CMA went on to argue that since the insurers could pass on the fee increase in their insurance retail price, the same thought experiment will also test for a degradation in the conditions on the consumer side. Both BGL and the CAT criticized the CMA approach to testing both sides of the market with a single SSNIP test for doing only half the test required to define the market. This is because the hypothetical increase in retail price faced by consumers through the CMA’s pass-on assumption would have been too small to test a SSNIP properly, as this pass-on of commissions represented an increase of only 1.8–3.5% in the retail price of insurance.
So, how can one test the consumer side properly when no price is charged? The starting point for defining a relevant market is to identify the relevant price and non-price dimensions of competition. The conceptual framework of the SSNIP can be applied to each, although the P in SSNIP refers to price. Below we consider the advantages and disadvantages of the tests considered on the consumer side in the Comparethemarket case.
A. SSNIP on Price Dimensions
The traditional SSNIP test is applied to the price dimension. In Comparethemarket, the challenge was that none of the PCWs, which together formed the hypothetical monopolist, charged their users to use the website. The PCWs’ business model is to charge commissions to insurers when policies are sold. Thus, applying the usual 5–10% SSNIP to a zero price on the consumer side would not lead to a change in market conditions because 10% of zero is still zero. While the economic literature has recognized that the absence of a price on which to apply the SSNIP test adds complexity, this challenge can be overcome by adapting the standard SSNIP test.
In particular, while the PCWs did not charge users who searched for home insurance quotes, they did seek to make their site attractive to users by encouraging insurers to offer attractive retail prices — i.e. the PCW users did pay once they purchased insurance from a provider they found via the PCW. As such, BGL proposed to focus the SSNIP test on the consumer side on this price-related dimension. This was in line with the dimension tested indirectly by the CMA under the assumption that the increase in commission on the insurer side will be passed on by the insurers to consumers, but with the notable change that it would have applied a material increase as required by the “significant” condition in the SSNIP test. Since PCWs users are generally price-sensitive, this is an appropriate dimension on which to test how they would react if prices increase once the hypothetical monopolist takes actions that worsen the retail price on that side of the market.
The CAT considered that such a retail price test would be inadequate because (i) it is difficult to see how consumers would be aware of the price increase and (ii) it only captures the substitutability of consumers who actually end up buying an insurance policy, not of those who simply browse the PCW. In the CAT’s view, these latter consumers are also relevant to the market that is being tested, and it therefore proposed a “special” SSNIP test that introduced a positive price for access on the consumer side which could be:
an absolute charge for using the service (e.g., £2 per comparison or an annual subscription of £5 for unlimited use) or an element in the Premium charged to such consumers only that specifically itemizes the cost of the service (e.g. 5% of the Premium is attributable to the price comparison service)
The CAT approach is not without controversy. For many services, the SSNIP test is applied to the sale price even though there are consumers who browse but do not purchase. For example, the usual SSNIP test for a supermarket considers the impact on the hypothetical monopolist’s profits if consumers switch away in response to an increase in the displayed price of the products it offers without imposing changes on how consumers access the supermarket. In focusing on the browsing consumers, the CAT’s test seems to overlook the main constraint imposed by consumers; namely the forgone commission fee the PCW receives when a browsing consumer makes a purchase.
To illustrate this, consider a scenario where all consumers who purchase on a PCW are rational and know that the saving they make via this intermediary is higher than the access charge. In this case, they will continue to use the PCW and generate a commission for the hypothetical monopolist. Even if the browsing consumers switch to other channels, the hypothetical monopolist will still find it profitable to increase the price based on the purchasers. This highlights that the emphasis on browsing consumers is misplaced and may be influenced by the CAT’s decision to separate the consumer side from the insurer side, where the commission is paid upon a sale.
Furthermore, both forms of the “special” SSNIP test proposed by the CAT have their shortcomings.
The option of introducing a direct charge for making a query or a subscription price to the PCW is not entirely novel and has been previously considered as a SSNIP alternative in markets with zero prices. However, it is important to approach this carefully because it fundamentally changes the business model for the hypothetical monopolist and moves away from market reality. Especially in two-sided markets where not charging one side of the market is common, moving from paying nothing to a small positive price is a rather radical change, and may not be an optimal choice for a hypothetical monopolist that is maximizing its profit on both sides at the same time.
Moreover, behavioral economics highlights that goods with zero prices are perceived differently by consumers, as they have a higher intrinsic value — there is something special about “free.” Thus, charging £2 instead of nothing has a more negative impact on consumers’ decisions than increasing the price from £2 to £4. This consumer reaction risks exaggerating the effect of the price introduction, and could lead to frequently finding wide markets on the consumer side when a zero price is involved. Caution is needed to ensure that the positive charge is not leading to a change beyond what is required by the “small but significant” part of the SSNIP test.
The second form of the SSNIP test proposed by the CAT — a 5 percent surcharge in the premium price of insurance applied to all users of a PCW — is not that different from the SSNIP test on the retail price which we discussed above. Both approaches test the price sensitivity of consumers who make a purchase. While the CAT seems to indicate that the surcharge would apply to all consumers browsing a PCW regardless of the channel they ultimately use to buy their insurance, this is likely to be unfeasible in practice. To illustrate this, consider that this surcharge will need to be applied after visiting a PCW to three different groups of consumers: (i) PCW users who buy insurance from an insurer they discovered on the PCW; (ii) PCW users who then decide to buy insurance directly from an insurer they did not discover on the PCW; and (iii) PCW users who then decide to buy insurance by accepting their renewal offer. In the absence of a mechanism to track users across channels and the ability to charge them for the use of the PCW (especially when the user purchases from an insurer that has no connection with the PCW), it is highly unlikely that the hypothetical monopolist could enforce a surcharge on those PCW users, and it is more likely to enforce the additional 5% surcharge only on its own website. This practical difficulty in the thought experiment proposed by the CAT means that the 5 percent surcharge test would most likely materialize in the same manner as a 5 percent increase in the retail prices displayed on the PCW.
B. SSNIP on Non-price Dimensions
The price dimensions are not the only options to test for market definition. The Commission’s draft Market Definition Notice acknowledges that “non-price elements are particularly relevant for the assessment of substitution” when a zero price is present on the consumer side, and considers, as an alternative, assessment of “the switching behavior of customers in response to a small but significant non-transitory decrease of quality (‘SSNDQ’).” It is generally accepted that testing non-price dimensions can be more difficult, but there are examples where it has been done. For example, the Commission used an SSNDQ test in Google Android to assess whether manufacturers, users and application developers would switch away from app stores for Android to app stores for other licensable smart mobile operating systems in the event of an SSNDQ.
In Comparethemarket, the most relevant qualitative dimensions of competition on the consumer side included the level of marketing and advertising undertaken by the PCW to attract users and the usefulness of the comparison service. For example, an important way through which PCWs attract consumers to their websites are online advertising or catchy TV adverts featuring well-known actors and singers, or mascots such as meerkats. While the usefulness of a website is difficult to assess, a change in marketing and advertising spend is one dimension that can more easily be tested in market definition. A decrease in this dimension by the hypothetical monopolist would have an impact on the visibility of the PCW channel compared to other distribution channels, and can highlight where consumers would divert.
This type of analysis has been considered by the CMA, for example in the Hunter Douglas/247 Home Furnishings merger. The CMA assessed the closeness of competition between the merging parties using the ranking of competitors on Google search paid and organic search results, which is a manifestation of the marketing and advertising spend of a firm.
Even though similar information on the ranking of competitors from other channels was available in Comparethemarket, the CAT overall took the view that non-price dimensions are too difficult to use and did not discuss further the possibility of applying a SSNIP on the consumer side along these dimensions. We consider that the information in this case did lend itself to an SSNDQ on marketing and advertising.
The rise in online advertising as an important channel to reach consumers means that there is reliable information that can be used to understand the impact of marketing on consumers’ behavior and a firm’s profitability, which renders feasible the use of an SSNDQ in market definition. For example, online search engines that provide advertising services, share relevant metrics such as the list of competitors listed in response to a query such as “home insurance” with those that take part in their advertising auctions. Importantly, as with the SSNIP test itself, one must focus on the hypothetical monopolist, and therefore the competition between the firms that are considered part of the hypothetical monopolist (here: the rival PCWs) needs to be disregarded. This means that one needs to examine what happens if all PCWs in home insurance reduce their marketing and advertising budget at the same time by the same percentage. Intuitively, one would expect that a reduction in the number of times the hypothetical monopolist bids to show an advert is an opportunity for a competitor channel to attract that consumer to switch. The CAT did not engage with this type of test in the Comparethemarket case.
IV. CONCLUSION: TEST BOTH SIDES, WHETHER IN ONE OR TWO MARKETS
The Comparethemarket case proved a battleground for different approaches to two-sided market definition. One of the most important messages to emerge from the case is that in two-sided platforms there needs to be a proper test for competitive constraints on both sides, even if prices are zero on one side. Such platforms compete on both sides simultaneously. They set prices and other competitive parameters so as to attract users on both sides, taking into account the externalities between the two sides. A hypothetical platform monopolist would do the same. For this reason it makes sense to define a single platform market, assessing the constraints on both sides. However, the same aim can in principle be achieved by defining two markets, one for each side (as the CAT preferred), provided that the network effects between the sides are taken into account.
A challenging question that may arise under either approach is what to do if the answer to the SSNIP question differs between the two sides: a SSNIP is profitable on one side of the platform, but not on the other side. This would indicate that the platform has a degree of market power on one side, but still faces strong competitive constraints on the other side. If the business practice or agreement in question has effects on both sides, the competitive constraints on both sides are relevant. The CAT did not explicitly address this question, although it ultimately concluded that there was no evidence of harmful effects from Comparethemarket’s wide MFNs on either side.
Many platforms have a business model in which one side is charged a zero price, in order to attract users and thus make the platform more valuable to users on the other side. In such situations it is still meaningful, and feasible, to assess market definition on non-price competitive dimensions, as also recognized in the Commission’s draft Market Definition Notice. The CAT’s approach to introduce a small price where previously there was none is also a feasible test, although one that must be applied with caution, since going from “free” to a non-zero price could be interpreted by consumers as something more than a “small but significant increase in price.”
 Economists at Oxera Consulting LLP. Gunnar Niels and Helen Ralston-Smith acted as economic experts for Comparethemarket in the BGL/Comparethemarket v. CMA case, and Andreea Antuca also advised on the matter. The opinions expressed in this article are our own.
 See, for example, European Commission (2021), “Support study accompanying the evaluation of the Commission Notice on the definition of relevant market for the purposes of Community competition law,” Final study, section 3, and Niels G. and Ralston, H. (2021), “Two-sided market definition: some common misunderstandings,” European Competition Journal, 17:1, 118-133.
 Competition Appeal Tribunal judgment (2022), Case No: 1380/1/12/21 BGL (Holdings) Limited & Others v. Competition and Markets Authority, 8 August.
 CMA (2021), “Case 50505 Price comparison website: use of most favoured nation clauses,” Non-confidential infringement decision.
 CMA (2017), “Digital comparison tools market study,” Final report, 26 September, para. 2.
 CMA (2014), “Private motor insurance market investigation,” Final report, 24 September; CMA (2015), “The Private Motor Insurance Market Investigation Order 2015,” 18 March, para. 4.1
 CMA (2021), “Case 50505 Price comparison website: use of most favoured nation clauses,” Non-confidential infringement decision.
 CJEU, Case C-382/12P, EU:C:2014/2201, 11 September 2014.
 Ohio et al. v. American Express Co. et al., 138.Ct. 2274 (2018).
 European Commission, Case M.8124, Microsoft/LinkedIn, 6 December 2016.
 CMA (2017), “Just Eat and Hungryhouse: A report on the anticipated acquisition by JUST EAT plc of Hungryhouse Holdings Limited,” 16 November.
 European Commission (2022), “Commission Notice on the definition of the relevant market for the purposes of Union competition law,” 8 November.
 Konkurrensverket, Decision 15/04/2015, Ref. no. 596/2013; Autorità Garante della Concorrenza e del Mercato, Provvedimento n. 25422, I779 – Mercato dei servizi turistici-prenotazioni alberghiere online, Bollettino n. 14, 27 Apr 2015.
 Competition Appeal Tribunal judgment (2022), Case No: 1380/1/12/21 BGL (Holdings) Limited & Others v. Competition and Markets Authority, 8 August, para 128.
 See Niels, G. (2019), “Transaction versus non-transaction platforms: A false dichotomy in two-sided market definition,” Journal of Competition Law & Economics, 15:2–3, 327–357; European Commission (2021), “Support study accompanying the evaluation of the Commission Notice on the definition of relevant market for the purposes of Community competition law,” 1 June, p.41, section 3.1.1.
 Competition Appeal Tribunal judgment (2022), Case No: 1380/1/12/21 BGL (Holdings) Limited & Others v. Competition and Markets Authority, 8 August, para. 138.
 Competition Appeal Tribunal judgment (2022), Case No: 1380/1/12/21 BGL (Holdings) Limited & Others v. Competition and Markets Authority, 8 August, paras 95(7) and 102(1).
 Competition Appeal Tribunal judgment (2022), Case No: 1380/1/12/21 BGL (Holdings) Limited & Others v. Competition and Markets Authority, 8 August, para. 139(2)(i).
 Competition Appeal Tribunal judgment (2022), Case No: 1380/1/12/21 BGL (Holdings) Limited & Others v. Competition and Markets Authority, 8 August, paras 139(2)(ii) and 144(8).
 Competition Appeal Tribunal judgment (2022), Case No: 1380/1/12/21 BGL (Holdings) Limited & Others v. Competition and Markets Authority, 8 August, para. 144(9).
 Shampanier, K., Mazar, N. and Ariely, D. (2007), “Zero as a special price: The true value of free products,” Marketing Science, 26, 742-757.
 European Commission (2022), “Commission Notice on the definition of the relevant market for the purposes of Union competition law,” 8 November, paras 97–98 and footnote 47.
 European Commission (2018), Case AT.40099, Google Android, 18 July, paras 284–305.
 CMA (2020), Case Hunter Douglas N.V. / 247 Home Furnishings Ltd merger inquiry, Final report, 14 September, paras 8.21–8.53.