Introduction
Data has been described as the world’s most valuable commodity.[1] Yet despite its importance, especially in the field Artificial Intelligence (AI), transfers of data are rarely discussed within the context of Subsidy Control. In this article, Hana Hammouda and Sattam Al-Mugheiry explore the economics of data and consider the circumstances in which the Subsidy Control Act 2022 may apply to public authority transfers of data.[2] They have considered this matter in detail in relation to State aid but this article focuses solely on the implications for the UK subsidy control regime.[3]
In addition to more traditional marketing applications, the value of data will be increasingly significant in the context of AI development as new data sets become more scarce and therefore valuable.[4]
It will be argued that, depending upon the particular circumstances, the transfer or sharing of data with enterprises can amount to a subsidy in UK law. Different approaches to valuing data will be explored, highlighting the risk of public authorities inadvertently giving away valuable public assets and granting unlawful subsidies that distort competition. Public authorities are uniquely positioned to hold and create large valuable troves of data in fields such as healthcare.
The Value and Significance of Data in the 21st Century
Not withstanding that the ownership of data in the UK is rarely straight forward[5] and its processing is governed by complicated laws,[6] this does not necessarily impact its value, nor does it stop the routine collection and processing at scale by public authorities.
Collecting and trading data creates and shifts information across companies and individuals. The value of data can be unlocked when it can be accessed by firms or researchers who are well-placed to innovate and generate insights and knowledge from it. When personal data is exchanged by companies and impacts consumer privacy, the data market can create harm that reduces consumer welfare, or it can facilitate innovation and be used in ways which increase consumer welfare.
The Transfer of Data as a Subsidy in UK Law
Public authorities hold vast amounts of health, travel, educational and financial data. Where a public authority makes a transfer of this data to an enterprise, we argue that the transfer can constitute “financial assistance” as defined by the Subsidy Control Act 2022[7] (SCA). In order to do so will look to the SCA and the application four-limb subsidy test set out in the statutory guidance[8] to the SCA.
UK subsidy control law focuses on financial assistance given by public authorities to enterprises.[9] Financial assistance[10] requires the below market cost provision of a good or service[11] which could easily occur in the complex and rapidly changing area of data.
Section 2 of the Subsidy Control Act 2022 defines a subsidy in UK law:
(1) In this Act, “subsidy” means financial assistance which—
(a) is given, directly or indirectly, from public resources by a public authority,
(b) confers an economic advantage on one or more enterprises,
(c) is specific, that is, is such that it benefits one or more enterprises over one or more other enterprises with respect to the production of goods or the provision of services, and
(d) has, or is capable of having, an effect on—
(i) competition or investment within the United Kingdom,
(ii) trade between the United Kingdom and a country or territory outside the United Kingdom, or
(iii) investment as between the United Kingdom and a country or territory outside the United Kingdom.
(2) For the purposes of this Act, the means by which financial assistance may be given include—
(a) a direct transfer of funds (such as grants or loans);
(b) a contingent transfer of funds (such as guarantees);
(c) the forgoing of revenue that is otherwise due;
(d) the provision of goods or services;
(e) the purchase of goods or services.
To assess whether a transfer of data is a subsidy, in accordance with the SCA definition, we need to assess the economic advantage that transfer gives to the recipient. The statutory guidance to the Subsidy Control Act[12] states that financial assistance is “a wide concept”[13] and as such a data transfer, if the data is of established financial value to the recipient, can meet the definition of financial assistance.
Applying the Four-Limb Subsidy Control Test
Financial assistance becomes a subsidy when it meets all four limbs of the test contained in the Subsidy Control Act 2022. Assuming the data is assessed as valuable by one of the methods considered below, we can move swiftly past the first limb as we are specifically considering a transfer from public resources controlled by a public authority.
The second limb of the test, that the financial assistance confers an economic advantage on one or more enterprises, is also likely to be met, as we are concerned with a transfer to companies. The provision of the data on an open basis to all who want it will not be problematic. It is selective provision of the data that will carry the highest risk of subsidy. The application of the second limb will be potentially more problematic in the case of entities that carry out more than one function such as charities or research institutes that sell goods or services (i.e. when they also engage in economic activity). In those cases, there will be potential routes to preventing a subsidy or making it compliant.
The third limb of the test, that financial assistance be specific – benefitting one or more enterprises over others, may be more complicated. The third limb focuses on the financial assistance being to specific beneficiaries, this can include a particular sector, industry etc. In the case that only one or two beneficiaries are involved, this should be easy to identify. Where a public authority shares data with an entire sector or group, such a granting of data may not be immediately noticed as being specific.
The fourth limb of the test, relating to the impact of the financial assistance on trade and investment will be specific to the area and directly impacted by the value of the data. This limb requires assessing the market(s) in which the beneficiaries operate, and the nature of domestic and international competition as well as trade flows. It is submitted that the more valuable the data and the wider the range of applications, the more likely it is to have an effect of competition or investment within the UK or internationally.
Valuing Data and a Consequent Subsidy
It has been reported that Sainsbury’s and Tesco make £300m a year[14] from selling information on individual shopping habits collected through loyalty card schemes. Deals include a partnership between Sainsbury’s Nectar card and Channel 4 to tailor adverts to specific groups. These clear examples show a significant financial value can be placed on some types of data.
These real-world examples provide a useful benchmark for considering the market value of data, but do not shed light on how that value has been determined, raising the question of how to approach the calculation of the value of data.
The web of value to different stakeholders means that estimating the market value of different types of data is a complex task. These challenges can be overcome, as illustrated in the examples discussed below.
Consumer Led Approaches to Valuing Data
One way to do this is to explore the value of data to the different stakeholders who have the potential to derive value from it.
Consumers from whom the data is derived may value their privacy and would be willing to pay to avoid its use. In contrast, consumers may obtain value from data if it results in targeted marketing or discounts which improves their consumer outcomes and or allows them to optimise their decision making, maximising their utility.
Valuing data from the perspective of the consumer can be achieved through quasi-experimental techniques where a firm has offered consumers the choice to either (i) provide their data and receive a service for free; or (ii) pay a fee to not share their data. The price paid represents a lower bound for the value of data for an individual consumer. More commonly, the value of data from the consumer perspective has been estimated using stated preference through willingness-to-pay and willingness-to-accept type experiments. In these experiments, consumers are asked how much they would need to be willing to pay or receive to avoid disclosing or to disclose their data. These figures can be aggregated to provide an overall value.
Business Led Approaches to Valuing Data
Holders of data can derive value through utilising it to make operational decisions, with the aim of maximising profitability. Using the Nectar card example, Sainsbury’s can derive value from the ability to create detailed profiles of customers’ preferences and spending habits, predict if shoppers can afford to spend more and offer targeted deals, and optimise decisions on inventory management, staffing and marketing. The value of data from the holder’s perspective in this example is the company’s expectation of additional profit (via reductions in cost or increase in sales) that can be derived through use of the data.
Other firms could also derive value through utilising the same data, or by combining it with other data to yield further insights, with the aim of maximising their profitability. For example, Channel 4 (through its partnerships with Nectar, Boots Advantage Card, and Tesco’s Clubcard) will offer advertisers access to consumer insights, providing them with the ability to target viewers with bespoke ads based on their purchasing behaviour. Predictive analytics and AI could further unlock value. Examples of where other companies can derive value through use of data might be where life insurance firms could increase the accuracy of their risk models when setting premiums based on purchasing behaviour, and banks could increase the accuracy of their credit risk models when setting interest rates on borrowing based on purchasing behaviour.
Again, the value from the perspective of an individual company can be estimated based on its willingness to pay which may be based on its expectation of additional profit, and it could be argued that a rational commercial operator would be prepared to pay up to that amount (factoring in a reasonable profit). As data is non-rivalrous,[15] sellers could in theory sell to multiple companies. However, the value of the data may be linked to whether it is also made available to other competitors. Unique use of such data is likely to create more opportunities for profit than data that is made universally available, or made available to several market players, as it can form part of a firm’s IP, thereby increasing the potential value.
Estimating the impact of data on profitability can be achieved through use of randomised control trials. One example of this is in digital advertising, where information garnered from cookies could be withheld from a random subset of publishers. The difference in revenue between the group of publishers with access to the information present in cookies, and the group without, provides an estimate of the increase in profits which results from using the data.
Public Sector Focused Valuation of Data
Governments can also derive value from data to improve its policymaking with a view to maximising social welfare. Governments capture and share large amounts of data on a range of topics but may not always be best placed to utilise that data, or may wish to share that data with researchers, academics, and private firms to unlock economic value.
Managing Subsidy Risk
In the context of subsidy control rules, public bodies should be careful not to give away data for free[16] or too cheaply, to one or more enterprises. Otherwise, they will likely be granting a commercial advantage in the form of a subsidy.
The risk seems likely to arise when, while making agreements relating to data, a public authority fails to consider the market value of the data, inadvertently granting a subsidy in the course of permitting access or sharing data. This may arise when data is shared for one purpose but technically allows another to be pursued of far greater value, for instance, permitting use of data for the purposes of training an AI model.[17] Technology companies are likely to have considered far more carefully the potential value to be obtained from contracts and data in this area than hard pressed non-specialist public authorities.
In the case of a transfer of data of value, public authorities will need to take appropriate action to make the subsidy compliant with the Subsidy Control Act 2022.
Avoiding a Subsidy
One alternative to looking for a compliant route to a subsidy would be to model the transfer on a Commercial Market Operator (CMO). Such a transaction should be evaluated against what is (or is believed could be) undertaken by a private sector ‘commercial’ entity, and not any public policy objectives (such as wider economic or social impacts).[18] In this case the appropriate amount would need to be determined and paid for by the recipient of the data.
Alternatively, data could be sold via a competitive process, such as an auction, to allocate data and to generate public revenue. Auctions could allocate data by enabling market forces to designate who receives the data and overcomes the pricing problem through the bidding mechanisms (which let potential buyers reveal their individual values of an item).
To the extent that governments make data available to all at no cost, this is would likely not be considered a subsidy. A similar alternative route for a public authority might be to provide the data available for free to those that request it. If either of these courses was followed, the provisions of the data would not be selective and hence avoid giving a subsidy. These options may be more complex depending on the data, as other legal duties may come into play,[19] but would allow for public authorities to gain any wider benefits they aimed to achieve by distributing the data.[20]
Conclusion
In the age of AI, it is clear that data has value. Providing access to that data for a subset of individuals or organisations for free or at a cost below what a private sector ‘commercial’ entity might sell it for (i.e. where the transfer of data is not compliant with the CMO principle), is likely to be a subsidy. Achieving compliance with the CMO principle and avoiding any subsidy would involve identifying the market value of the transfer and ensuring that amount (or greater) is paid by the recipient.
If the provision of data is deemed to be a subsidy, then the transfer arrangement could still be made lawful by adherence to the seven subsidy control principles,[21] which recognise that well-designed subsidies can bring about benefits for society, by spurring businesses to undertake activity which would not happen otherwise. There are a range of ways in which the use of data can drive innovation and lead to positive externalities which may create an incentive for providing the data on a subsidised basis, although the value of those externalities can be just as hard to quantify as valuing the data.
In any transfer of data care should be taken in the transfer or allowing of use of data not to inadvertently grant a subsidy in the course of other activities.
We anticipate that in future public authorities will be more conscious that transactions involving data might involve subsidy. Likewise, we would anticipate that as datasets become scarcer the willingness of a competitor to challenge a public sector data deal would increase.
[1] Kiran Bhageshpur, “Data Is The New Oil — And That’s A Good Thing” (Forbes, 15 November 2019) < https://www.forbes.com/sites/forbestechcouncil/2019/11/15/data-is-the-new-oil-and-thats-a-good-thing/> accessed 24 October 2024.
[2] We are particularly grateful to Tom Middleton of Grant Thorton for contributing his expertise and insights on this article.
[3] S Al-Mugheiry and H Hammouda, ‘State Aid and Data Transfers’ [2024] 23 European State Aid Law Quarterly 290 – 293.
[4] Tal Roded and Peter Slattery, “What drives progress in AI? Trends in Data” (FutureTech MiT, 19 March 2024) <https://futuretech.mit.edu/news/what-drives-progress-in-ai-trends-in-data#:~:text=Data%20is%20required%20to%20train%20and%20validate%20AI%20models.&text=AI%20models%20use%20training%20data,changed%20between%202010%20and%202023> accessed 24 October 2024.
[5] See for example: Mark A. Prinsley and Oliver Yaros, “Data ownership and contact tracing” (Mayer Brown, 21 May 2020) <https://www.mayerbrown.com/de/insights/publications/2020/05/data-ownership-and-contact-tracing> accessed 24 October 2024.
[6] Most significantly the Data Protection Act 2018.
[7] Subsidy Control Act 2022 (SCA 2022).
[8] Statutory Guidance – Published in November 2022 and updated in July and December 2023
<https://assets.publishing.service.gov.uk/media/658025b295bf65000d719140/uk_subsidy_control_regime_statutory_guidance.pdf> accessed 24 October 2024.
[9] Brought in to implement the UK’s obligations under the Trade and Cooperation Agreement between the European Union and the European Atomic Energy Community, of the one part, and the United Kingdom of Great Britain and Northern Ireland, of the other part [2021] OJ L 149/10.
[10] Used rather than “advantage” in Article 107(1) of the Treaty on the Functioning of the European Union [2012] OJ C 326.
[11] The transfer of data as State aid in EU law will be the subject of another article.
[12] Statutory Guidance (n 9).
[13] ibid para 2.5.
[14]Hannah Boland, ‘The loyalty card cult netting supermarkets millions’ (The Telegraph, 7 January 2024) <https://www.telegraph.co.uk/business/2024/01/07/supermarket-loyalty-clubcard-nectar-card-goldmine-data/#:~:text=Customer%20data%20is%20anonymised%20by,selling%20this%20in%2Dhouse%20data> accessed 24 October 2024.
[15] Non-rivalry means that consumption of a good by one person does not reduce the amount available for others.
[16] Unless it is granted free to all.
[17] Such a failure would mean that the transparency requirements under s 33 of the Subsidy Control Act 2022 were not met so the arrangements would be vulnerable to challenge.
[18] These should not be included in this assessment as this should be a market orientated approach.
[19] For example, the UK General Data Protection Regulation.
[20] Noting Managing Public Money concerns could come into play given this would be giving away an asset of value.
[21] Public authorities should consider the subsidy control principles A to G when designing a subsidy, and they must not give a subsidy unless they are of the view that it is consistent with the subsidy control principles.