11

Agricultural machines are equipped with digital data sensors and actuators. Sensors collect data on the mechanical movements and navigation position of the machine. Actuators use data inputs to activate mechanical movements and steer the machine. For example, a GPS signal sensor captures the precise location of a machine; actuators steer the machine on the basis of instructions received from a computer programme. This combination enables the collection of field-level data and implementation of agronomic advisory services, for example, for automated seeding, fertilising and chemicals inputs. [20]

12

While the market for agricultural machine sales is highly competitive, the market for data that drive aftermarket services is less so. Agricultural machinery manufacturers can design the machine in such a way that they have exclusive access to sensor data and actuators inputs. Once a farmer buys a particular machine, he is locked into the data channels controlled by the machine manufacturer. The manufacturer may use this monopolistic position in upstream data collection, or in access to the downstream implementation of data-driven agronomic services, to leverage his position in downstream services markets.  [21] The lack of interoperability between data formats and devices from different manufacturers reinforces this monopolistic market structure. Agricultural machine producers and service providers deliberately segment the data standards in order to increase switching costs for farmers, reduce competition in aftermarkets and apply monopolistic pricing in aftermarket services. [22] This was emphasised by KWS (an agricultural company) in the market investigation of the Bayer/Monsanto case:

“It is difficult to switch from one platform to another, since the industry is not able to agree on one common data protocol (joint data format), therefore there is high incentive for the farmer to decide on only one platform. Even though farmers keep the ownership of provided data and they can contractually request that their data are returned to them, from the technical point of view, such data are not compatible with another platform and can therefore not be easily transferred to another platform from a practical point of view.” [23]

13

According to the “Chicago Critique”, [24] there is no need to intervene in the aftermarket when farmers are rational. They will consider the combined costs and benefits in the primary and aftermarket before deciding on the purchase of a device. If farmers are myopic, however, they may struggle to combine cost and benefits in both markets. Lack of transparency at the time of purchase may be an obstacle to rational decision-making. For instance, with long machine lifetimes and a fast-evolving technology environment, mismatches may occur during the lifetime of the machine that are not predictable at the time of purchase.

14

This monopolistic lock-in position is not absolute. Data plug-ins and add-ons can circumvent the manufacturer’s monopoly on mechanical access. For example, Bosch has developed, in collaboration with some partners including Bayer’s Xarvio, Syngenta and AGCO, the Nevonex interface that seeks to overcome incompatibility problems between agricultural machines. [25] It consists of an interface that can take data input from various machines, brands and data formats, and send steering signals to a variety of machines, including retro-fitted mechanical devices on existing machines. [26] In agricultural machinery, there is the Isobus ISO technical standard initiative. [27] For example, Xarvio designed a data-driven sprayer that can be mounted on existing mechanical sprayers to give them data steering for precision spraying purposes. [28] These add-ons with open technical standards may help to overcome interoperability barriers between machines and data-driven agronomic advisory services. However, there are several limitations to these interoperability solutions. First, the Isobus standard suffers from “forking” into several proprietary versions that are not necessarily fully compatible. Second, interoperability does not ensure the transfer of historical farm data between machines and applications. [29] Third, it remains to be seen how successful interoperable data formats, such as the Isobus standard, will become in the market. Several economic models explain the ambiguity in incentives for firms to open up access to their exclusive data.

15

Incentives may vary according to firms’ market shares. Big manufacturers that benefit from a strong market position may be reluctant to give up their advantages. [30] Smaller manufacturers or companies that have no entrenched position in the agricultural machine market may prefer an open data standard. Smaller firms stand to gain more from interoperability than large firms with strong market positions. [31]

16

Another economic model that explains the ambiguity of incentives is the “co-opetition” (cooperation-competition) model. [32] Machine manufacturers can choose to open access to their machine data and make data formats compatible with common standards. This makes it easier for farmers to switch between agronomic service providers in the aftermarket. It increases competition between aftermarket service providers and decreases prices. That will make the machines more attractive to farmers and increase the market share of manufacturers who sell interoperable machines in the primary machine market. On the other hand, it may decrease aftersales revenue for manufacturers who are vertically integrated into aftermarket services. The net effect of all these factors is a complex empirical question that is hard to predict.

17

Large input producers have accumulated considerable knowledge in the use of inputs such as seeds, fertilisers, pesticides and other chemicals. Working closely with agricultural extension services, agronomic laboratories and big historical datasets, they have proprietary knowledge about the precise genetic composition and characteristics of seeds, and the biochemical interaction with chemical inputs, soil quality and weather factors. [33] Besides these large vertically integrated input providers, there are also smaller start-ups that are specialised in data-driven agricultural services only. They both require complementary data to generate tailor-made agronomic solutions. An individual farmer’s experience and data cannot match the insights obtained from large data pools and economies of scale and scope in the analysis of these fine-grained farm-specific data about actual input use and crop yields. Some data can be obtained from the market, and some need specific contractual relationships. For instance, coarse-grained land maps can be obtained from free satellite services (scale 10x10 meters) while more fine-grained mapping (up to 30x30 cm) is available for a price. [34] Farm specific data, such as detailed irrigation and soil data, need to be collected from the target farm. Combined with detailed weather forecasts, they enable the production of very granulated agronomic advisory services tailored to the needs of individual farmers and fields. [35] Service providers store historical data collected from farms. They have exclusive de facto control over these farm-specific data sets. [36] New market entrants will need to access these historical data to generate accurate prescriptions [37] when farmers desire to change service providers. This creates a data lock-in situation for farmers, which can jeopardise competition in the emerging markets of data-driven agronomic services. [38] Vertically integrated agronomic advisory services can be combined with agricultural inputs sales. This entails the risk of self-preferencing: the service provider can recommend its own upstream products even though they are not objectively the best or cheapest product to suggest. Self-preferencing may reduce competition in upstream inputs markets.

18

Integrated input producers can monetise their information advantage in the form of agronomic services. They can simply send the advice to the farmer and enable him to implement it manually on these fields. For instance, advice can be dispensed through apps on mobile devices. Combined with field navigation maps, the farmer can steer his machines as required. Alternatively, advisory services can be dispensed automatically by proprietary data interface devices that directly steer machines. Many companies have developed such interfaces. Monsanto’s subsidiary, the Climate Corporation, has introduced FieldView, a service that makes agronomic advice available to farmers and interacts with agricultural machines. [39] Bayer generated a similar service called FieldManager. [40] BASF uses the Maglis interface [41] and DowDuPont has various digital products that are sold in the US through its subsidiaries Pioneer [42] and Corteva Agriscience. [43] These companies can negotiate agreements with machine manufacturers to create a direct data access gate. For example, in the US, agricultural machine manufacturer John Deere came to an agreement with the Climate Company to let machines from the former interact with advisory services from the latter. [44]

19

Proprietary devices that dispense agronomic services can be used to leverage these integrated giants’ positions in the markets for data-driven agronomic services or inputs sales. Once farmers buy into a device that has special arrangements with a particular agronomic service provider, they are locked into the device and the agronomic aftermarket service, especially when it uses non-compatible and non-interoperable data formats and software design. This aftermarket lock-in enables firms to charge a monopolistic price for their advisory services and inputs sales. Firms can either choose a lock-in strategy to avoid farmers’ switching between data platforms, or choose an open system strategy to attract more farmers. Another strategy to circumvent exclusive cooperation between machine manufacturers and service providers is the use of machine add-ons that by-pass the manufacturers’ exclusive data channels. The above example from Nevonex illustrates the latter case.

20

Agronomic service providers face the same trade-off between competition and cooperation as machine manufacturers. They can perceive machines as complementary products because their customers may value agronomic services more when combined with the manufacturer’s machine compared to having the service alone. They may also perceive them as competitors if customers will pay less for the agronomic services when combined with the machine than when they buy the services separately. This is again a hard empirical question.

21

The machine-centred and the agronomic services-centred data-driven business models are complements, and there is some degree of convergence between the two. Machine manufacturers can either produce their own complementary agronomic advisory services or, alternatively, negotiate an agreement with other agronomic service providers to share these data channels for the purpose of dispensing agronomic services. The tension between competition and cooperation is always present. In line with the co-opetition model, [45] machine manufacturers seek collaboration agreements with inputs and advisory service providers, and vice versa. For example, John Deere, an agricultural machinery manufacturer, focuses on machine-based data collection while Bayer and Monsanto, agricultural input providers, focus on data-driven agronomic services in input markets. [46] There are many collaboration agreements between these companies. Monsanto (or Bayer, now), for example, has agreements with machine manufacturers John Deere, Agco and CNHI, through its subsidiary, the Climate Corporation, which specialises in data-driven agronomic services. [47] These collaboration agreements fall short of mergers, but they nevertheless involve coordination.

23

Statistical analysis requires large samples of data in order to extract insights. Economies of scale occur when adding more observations on the same variables in a sample increases the accuracy of the statistical predictions. For example, more observations on the response of crop yield to different types of fertiliser improve the prediction accuracy for fertiliser. More fine-grained soil maps allow for more precise application of fertiliser and chemicals in fields. [48] These improvements will, at some point, become subject to diminishing returns to scale or the number of observations.

24

Economies of scale constitute an argument in favour of data concentration in large databases, allowing firms to accumulate and combine data from many sources.

25

Contrary to ordinary goods that are rival and can only be used for one purpose at the time, data are non-rival. Many parties can use the same dataset at the same time for a variety of purposes. Economic efficiency gains occur when data collected by a firm can be re-used for other purposes. Economies of scope in re-use [49] can be realised either by the firm that collected the data and re-uses it in-house for other purposes or by sharing/selling the data to another firm that uses it for another purpose. For example, a tractor is a rival physical good and can only be used by one farmer at the time. If a tractor would be non-rival, all farmers could use the same tractor at the same time to work in different fields. The welfare gains would be enormous: it would suffice to invest in the production of a single tractor to cater to the needs of all farmers. This prospect can be achieved with data. For example, detailed farmland and soil survey maps can be used for precision farming applications for all types of inputs and services on that farm – but not on other farms. Collecting the data comes at a fixed cost. Re-use of the same non-rival data for another purpose entails quasi-zero marginal reproduction costs of an electronic data file. Economies of scope in re-use constitutes an argument in favour of wider access to data. Many applications of data in farming are re-use applications. Farmers share land and soil map data with agronomic services, or livestock health and production data with service providers.

26

Contrary to economies of scale, economies of scope in the re-use of data constitute an argument in favour of de-concentration of data, facilitating the distribution of data over many applications and allowing access by many firms for competing applications.

27

Data access may also come at a cost. Privacy and commercial confidentiality are important for the autonomy of private decision-making by firms and individuals and for extracting private value from these decisions. [50] When data are used by a competing firm to produce a substitute good or service, it may harm the interests of the original data collector. When data become widely available, it erodes the market value of the data for the original collector and may become a disincentive for continued investments for data collection.

28

An ideal data governance regime would thus seek an optimal combination of wider access and exclusive control, a balance between anti-competitive concentration and competitive decentralisation. [51] Data are not excludable by nature. They require technical and/or legal protection to ensure exclusive access for one party. That is what machine manufacturers and agronomic advisory services aim to achieve by channelling data-driven services through exclusively controlled devices.

29

When two datasets are complementary, more insights and economic value can be extracted from the analysis of the merged dataset, compared to applying data analysis to each of the separate data sets. [52] Economies of scope in data aggregation constitute another argument in favour of large data pools and concentration of data. They are a critical factor in the success of digital farming.

30

At the same time they trigger concerns from competition authorities, as shown by the market investigation of the European Commission’s Bayer/Monsanto merger decision:

“The more data (and the more specific data) you have, the more robust your algorithms will be and the more proven results you will have as references to your potential customers. (..) Covering more crop varieties, more climate areas, more soil types, etc allows you to expand your offering to other areas and cultures and because it is a complex system that constantly evolves, it is important to have different independent and broadly representative sources of information to build the necessary expertise.” [53]

31

The merger between Bayer, a chemicals producer, and Monsanto, mostly known for genetically modified seeds, was designed to generate more benefits from their combined agronomic research, including in the digital era where big data collection and the use of artificial intelligence to comb through these datasets would have become primary tools to advance research. The merger decision package sought to reduce economies of scale and scope in data aggregation in order to maintain competition. [54]

32

Economies of scale and scope are two distinct measures. In our fertiliser example, economies of scale are related to the number of observations on a particular fertiliser. Economies of scope occur when more variables are added to estimate the impact of fertilisers on yields, such as soil and weather conditions and the use of different chemicals.

33

Realising economies of scale and scope in data aggregation requires “big” datasets, usually across many farms, inputs, outputs and production conditions. Individual farmers cannot realise these benefits. It requires a third-party intermediary who collects and aggregates data from many sources in order to extract more insights from the pooled data compared to the insights that farmers could extract from their own datasets. [55]

34

Intermediaries are not necessarily large firms. Small start-up firms may be able to collect a sufficiently large data sample and reach a high level of economies of scale and scope in selected data domains. However, specialisation in one specific area is not sufficient to be competitive in a wide range of agronomic services markets that span many complementary and substitute products and crops. [56]

35

An important consequence of economies of scale and scope in aggregation is that the collective or social value of farm data is usually higher than the private value of data for individual farmers. [57]

36

The existence of a gap between private and social value implies an inherent market failure. Purely private data ownership rights may therefore not be an optimal allocation mechanism. Neither is purely public and common access to data because that would eliminate any market value for data. All benefits would be dissipated as user surplus and would dis-incentivise investment in data collection. This constitutes an argument in favour of regulatory intervention to overcome data market failures and put in place alternative data access regimes that seek to realize the social value of data while preserving competition in data-driven services markets.

37

Another source of market failure occurs when data aggregation generates negative externalities for farmers. Economies of scale and scope in data aggregation are subject to diminishing returns. [58] Once an agricultural service provider has collected a dataset that is sufficiently large to produce algorithms with a high level of prediction accuracy, the marginal value of collecting additional data from farms is low or zero. This depresses the market value of individual farm data [59] and puts farmers in a weak bargaining position with regard to their data. Even if they did not face data portability or interoperability obstacles, they would not be in a position to monetise the value of their data. Conversely, it explains to a certain extent why intermediary platforms cannot give farmers a meaningful remuneration for their data. They can only ensure their financial sustainability by charging for the data-driven services that they offer. [60]

38

Several digital economy studies [61] already highlighted how the data-driven platform economy is torn between efficiency and welfare gains from data aggregation in intermediary platforms and anti-competitive behaviour by these monopolistic data giants.

48

The US document distinguishes between farmers and service providers (called Agriculture Technology Providers or ATPs), and attributes original rights (ownership of data) to farmers:

“Farmers own information generated on their farming operations. However, it is the responsibility of the farmer to agree upon data use and sharing with the other stakeholders with an economic interest, such as the tenant, landowner, cooperative, owner of the precision agriculture system hardware, and/or ATP etc. The farmer contracting with the ATP is responsible for ensuring that only the data they own or have permission to use is included in the account with the ATP.”

49

The US text unambiguously attributes the ‘ownership of data’ to farmers [83] who generate data on their farming operations. Data ownership [84] is divorced from machine, device or land ownership, including from external parties that perform services on the farm. Ownership is attributed to the party that decides and manages the farming operations. The data are not considered as the product of device ownership, but rather of the farming operation: farmers’ efforts and practices. The text makes farmers responsible for any data sharing with other stakeholders with an economic interest by means of contracts, but this responsibility is ambiguous. There is no clarification in the text that implies rights for stakeholders or any mechanism to be used by them to access the data. This looks more like an advisory statement for farmers. The text uses ownership as the central legal concept when designing data rights. However, ownership rights are tradable/alienable. Although some parts of the text seem to provide inalienable consent rights to data re-use (see section 3.2. below), the data ownership design behind the US Principles limits the potential of the proposed rights with their alienable/transferable nature.

50

The EU Code has a more ambiguous wording. It distinguishes between data originators, providers and users. [85] It attributes data ownership rights to data generated during farming operations to data originators, i.e. the right to benefit from and/or be compensated for the use of data created as part of their activity. However, when data are produced by a service provider or external operator on the farm in the course of their activity, the operator might be considered as the data originator, not the farmer:

The originator (owner) - “the person or entity that can claim the exclusive right to licence access to the data and control its downstream use or re-use”  [86]

“The data originator of all the data generated during the operation is the one who has created/collected this data either by technical means (e.g. agricultural machinery, electronic data processing programs), by themselves or who has commissioned data providers for this purpose.”  [87]

“This Code recognises the data originator’s right, whether they are a farmer or another party…”  [88]

51

Clearly, the data originator may be different from the farmer, especially in automated data collection. It is not clear whether the operator would be the owner of the device, the device controller, or possibly the farmer who may have rented a device. Data collection may be conducted by a third party, commissioned by the farmer and with the aim to facilitate farmers’ decision-making. If data-collecting sensors in machines are not owned by farmers, the sensor/machine owners are the data providers, not the data originators:

"It can be assumed that the data originators are the farmers, also from data of sensors that are owned by the farmer.  If sensors are not owned by the farmers, the sensor owners are seen as data providers."  [89]

52

The notion of data provider refers to a natural or legal person who, under an agreement, delivers data to the data user and/or data originator. [90] Although these extra definitions (data user and data provider) create confusion regarding the attribution of original rights, the owner is data originator.

53

An additional provision in the Code adds to the confusion because it seemingly reverses the provision that data rights can be owned by other parties than the farmer:

“Rights regarding data produced on the farm or during farming operations are owned by the farmer and may be used extensively by them.”  [91]

54

The attribution of original rights emphasises data on “the farm or during farming operations”. This may be interpreted as indicating raw farm data only, not processed data. The text may be intentionally silent on processed data, leaving it out of the data charter and subject to the free market.

55

Both data charters have repetitive statements that indicate that related rights can be alienable and transferable, i.e. they can be traded through bilateral contracts. This is compatible with the concept of ownership. Bilateral negotiations imply that bargaining power will play an important role in deciding who ends up with the effective ownership right of the data. For example, powerful data aggregators who can extract more value from large data pools could end up acquiring the data. This may be beneficial for farmers if it allows them to obtain the highest value for their data. It may also be detrimental when farmers are locked-in machine data and data-driven services markets are foreclosed. For example, machine manufacturers or service providers may have exclusive control over access to the data that enables them to foreclose the market for downstream use of the data for the purpose of agricultural services. That leaves no other option to the farmer than to accept the proposed terms and conditions that may contain provisions to transfer ownership rights from farmers to machine producers or technology providers. Portability rights would open the door to circumvent lock-in situations, but it is unlikely to be enforced together with the alienable and exclusive data ownership understanding.

56

Despite the fact that the EU and US agricultural data charters were inspired by the GDPR, [92] the introduction of tradable data ownership rights shows how the charters represent a clear departure from the underlying principles of the GDPR, where rights to personal data are considered as fundamental and inalienable human rights. [93] Even if the data rights are allocated initially to the farmers, bargaining power determines which party eventually ends up with the rights to use the data. Consequently, data ownership as conceived in both the EU and US charters is not able to address lock-in concerns and broader data access problems in the sector. Inalienability might protect farmers from powerful service providers or machine producers in terms of controlling collected raw farm data in a sustainable way.  [94] Also, inalienable data rights do not necessarily exclude other stakeholders from using the data. For instance, while farmers can have rights to data portability when changing services, the same data can also be used for training algorithms by the previous technology provider, or the new tenant of a rented farm field might benefit from the historical data sets.

57

However, since farms are tradable assets, all the rights that they acquire should be by definition tradable as well. So, inalienable rights should stay with the farm, not with the farmer as an individual. Otherwise, it may create problems when farms are sold while rights remain with the farmer. Inalienable rights for farms not only protect the farmer while he owns the farm, but also allow him to sell the rights together with the farm. Alienable rights imply that the rights to data can be sold separately from rights to the legal entity of the farm. Big data firms might hoover up farm data rights without owning farms. That would handicap future owners of farms and diminish the value of farms.

58

Under the title “Collection, Access and Control” the US Principles state that;

“An ATP’s collection, access and use of farm data should be granted only with the affirmative and explicit consent of the farmer. This will be by contract agreements, whether signed or digital.”

59

This gives farmers an exclusive decision right over data access and re-use by third parties, as an attribute of their data ownership right. This clause aims to restrict onward data sharing by ATPs. At first sight, it gives farmers inalienable control rights vis à vis third parties. However, farmers can give their consent in the agreement to leave it up to the ATP to decide on sharing data with third parties. Unlike in the GDPR, there are no details regarding the modalities of the consent. Is a general consent statement valid, and is there a right to withdraw consent? This might result in ambiguities with regard to the alienability of the right to consent. The primacy of contracts implies that withdrawal and data retention rights can also be restricted by agreements, i.e. they are alienable. Repetitive statements throughout the text in favour of contractual superiority may be an indicator of the ATPs’ influence in designing the charter. [95]

60

The EU Code is again more ambiguous. Consent for data access and re-use may be given by the data originator or the operator. This boils down to the farmer’s consent only if the farmer and originator coincide. Contract dominates, however, which implies that consent and re-use rights are alienable, as in the US charter.

“The collection, access, storage and usage of the collected agricultural data can only occur once the data originator has granted their explicit, express and informed permission via a contractual arrangement.”  [96]  

“The data originator must give permission for their data to be used and shared with third parties, including circumstances in which decisions are made based on data."  [97]

“Right to determine who can access and use the data is attributed to this operator.”  [98]

“Parties … should establish a contract clearly setting the data collection and data sharing conditions…”  [99]

“Parties may not use, process or share data without the consent of the data originator.”  [100]

61

Like in the US Principles, details of the consent conditions are not mentioned, apart from a reference to GDPR based principles such as being explicit, express and informed. Therefore, it is unclear if a general consent statement in the contract is valid or whether there is a right to withdraw consent. This example demonstrates the limitations of the contract-based design in the charters. The validity of contractual relationships can be challenged by using the explicit, express and informed permission argument as most of the service providers’ terms and conditions are standard texts. So, the existing design is open to a number of problems in practice.

“Data originators must be given the possibility to opt-out of the contract and terminate or suspend the collection and usage of their data, provided that the contractual obligations have been met. This must be clearly stated in the contract…”  [101]

62

This statement signals that the consent rights could be inalienable and can be cancelled at any time by the data originator. Not surprisingly, this general statement is again subject to contractual provisions. Unless stated in the contract, there is no right to opt-out or terminate. [102] This problematic design is repeated all over the EU Code. It is difficult to see how these data principles could change anything if they would become law, compared to free and unregulated data markets, since contracts and market forces prevail over the principles.

63

Another rule mentions pseudonymisation or anonymisation of agricultural data in the EU Code:

“Data originator must give permission for their data to be used and shared with third parties… Information should only be given to third parties as aggregated, pseudonymized or anonymised data unless it is required to deliver the requested service and/or the conditions specified in the contract. Unless specified in the contract, the data user must take all precautions to avoid re-identification.”  [103]

64

The concepts of pseudonymisation and anonymisation are related to the privacy and identifiability of natural persons. They do not have any meaning for non-personal agricultural data in this regard. Data could be anonymised with respect to the identification of the farm. Farmers might not be happy about other parties accessing data on their farming practices because it may affect the asset value of their farm, their credit score, etc. This may imply that farm identification and physical location coordinates should be eliminated from the data. It would be more appropriate to link the clear aims and the re-use consent requirements such as to protect farms’ trade secrets, instead of using privacy law concepts.

65

Other principles in the US charter might also play a role in consent for access and re-use. For example, under “Transparency and consistency”, it states that;

“ATPs shall notify farmers about the purposes for which they collect and use farm data, third parties to which they disclose the data and the choices the ATP offers for limiting its use and disclosure.”

66

This formulation is not clear about the consequences of notification. Does the farmer have a right to object? The text is silent about this. The notification principle can become functional only if it is related to an inalienable consent right. Inalienability, in nature, is strictly related to binding legal rules that have clear enforcement mechanisms.

67

Under “Disclosure, Use and Sale Limitation” the US charter states that;

“An ATP will not sell and/or disclose non-aggregated farm data to a third party without first securing a legally binding commitment to be bound by the same terms and conditions as the ATP has with the farmer. Farmers must be notified if such a sale is going to take place and have the option to opt out or have their data removed prior to that sale. An ATP will not share or disclose original farm data with a third party in any manner that is inconsistent with the contract with the farmer. If the agreement with the third party is not the same as the agreement with the ATP, farmers must be presented with the third party’s terms for agreement or rejection.”

68

This statement implies an extension of contractual terms to third parties. Since there is no in rem legal framework for data that is enforceable towards third parties, the US charter attempts to use contracts as an enforcement tool for these principles.

69

Both charters also have rules that prevent unilateral contractual changes without the consent of farmers.

70

The US Principles include the following statements:

“An ATP’s principles, policies and practices should be transparent and fully consistent with the terms and conditions in their legal contracts. An ATP will not change the customer’s contract without his or her agreement. “

71

The EU Code has the following consent rule:

"Contracts must not be amended without prior consent of the data originator.  If data is to be sold or shared with a third party that is not initially mentioned in the contract, the data originator must be able to agree or refuse this, without financial or other repercussions.”  [104]

72

This protects farmers from unilateral actions. Service providers are obliged to maintain services under older terms and conditions (T&C). Provisions that forbid unilateral changes are indeed stating the obvious because both of the charters are designed to be enforced via contracts, and unilateral changes are not compatible with the mutual assent principle in contractual relations in any case.

73

Apart from this general statement in the texts, the EU Code provides a specific obligation to take consent for the new data-sharing situations that are not specified in the contract beforehand. The originator can refuse new sharing. The obligation for service providers is not to impose any response to this refusal that would result in negative consequences for farmers. Although it is not clear what the scope of this obligation or the meaning of the negative repercussions exactly is, it is a positive intention to protect the weaker party in case of refusal decision for the third-party access. The prohibition of financial or other repercussions plays an important role to compel service providers or machine producers not to limit machine functionality, for instance, if new T&C are rejected. The US text has no statement regarding the consequences of such an action.

74

The EU Code brings an interesting obligation for third parties’ access to data;

“The data user can only sell or disclose data to a third party if she/he has secured the same terms and conditions as specified in the contract between user and originator.”  [105]

75

However, it is unclear whether this is an additional obligation above the ones discussed before or an alternative one to share data, or whether the farmer and third party enter into a direct contractual relationship or this is only between companies, i.e. data user and third party. In the case of the latter situation, it is difficult to see how a farmer can enforce its rights against the third party as the farmer would not be a part of the contract between companies. The idea seems to protect farmers with the same contractual conditions, but the existing form of the text is ambiguous.

76

In general, there are various consent rules to collect, access or re-use of data, especially in the EU code. However, neither their scope and enforcement mechanisms nor their effects on the competitive dynamics in the sector are undisputedly clear. Also, contractual superiority emphasis throughout the text further limits the potential of the proposed rules and rights, if any. Apart from the discussed ambiguities and limitations of the texts, the core question, indeed, might be whether we really need consent-based rules and rights in the non-personal agricultural data setting from the competition policy perspective. On the one hand, consent rights for collection, access and re-use of data might increase farmers bargaining position if they are inalienable and binding. On the other hand, this may create another set of barriers against the free flow of data in the sector. This balance should be carefully considered when designing the sectoral consent rules. Instead of transplanting personal data concepts from the GDPR, rights need to be designed in accordance with sectoral conditions. Data protection law serves a more human rights-oriented policy preference that might not always be compatible with the needs of the sector, which is predominantly based on non-personal agricultural data.

77

In the US Principles, data portability is considered as a privacy-related issue, not a competition issue because there is no portability right for anonymous or unidentifiable data sets. Also, only primary data can be ported, not ‘aggregated’ data:

“Within the context of the agreement and retention policy, farmers should be able to retrieve their data for storage or use in other systems, with the exception of the data that has been made anonymous or aggregated and is no longer specifically identifiable. Non-anonymized or non-aggregated data should be easy for farmers to receive their data back at their discretion.”

78

As discussed above, anonymisation or identifiability are incompatible when discussing non-personal farm data as they are privacy law concepts. Protecting commercial confidentiality might have a certain rationale, but it is not related to the portability provision. When it comes to portability right, these nuances do not make sense. This indicates that the US Principles did not have a clear framework of failures to address them with their rules. It seems more like privacy-centric legal transplantation to an incompatible context, i.e. non-personal farm data.

79

The reference to “within the context of the agreement and retention policy” indicates that this is not an inalienable right. Farms can trade this right away in a contract, and market power will determine the eventual outcome of the negotiations. The US Principles actually do not change the status quo of the free B2B data market setting apart from its advisory statements in favour of contractual portability clauses.

80

In the EU Code, the rule regarding portability is as follows:

“Unless otherwise agreed in the contract, the data originator has the right to transmit this data to another data user. If agreed between the parties, the data originator shall have the right to have the data transmitted directly from one data user to another, where technically feasible.” [106]

81

This demonstrates the approach of the EU Code in favour of the primacy of contracts over principles again. However, this clause has a different design compared to other rules in the text. In situations where contracts remain silent on portability, the Code could be invoked to assume portability by default. This makes the scope of portability right somewhat broader in the EU Code compared to the US Principles. This is still an alienable design as this right can be removed by contractual clauses. However, the following sentence repeats the same right by saying that if agreed between the parties. This makes the approach of the text confusing because this jeopardises the possible legal interpretation of default portability right in the case of contractual silence. The text could have been clearer in this regard.

82

Portability is only possible in the EU Code “where technically feasible” – again an explicit transposition from Article 20 (2) GDPR. Technical feasibility might be problematic, especially when different and incompatible standards create an obstacle to the transfer. [107] Differentiation in standards is sometimes an intentional business strategy to prevent portability of data to competitors. [108] This may prevent farmers from enforcing their right even though the portability clause is not waived in the contract.

83

Indeed, the EU Code is aware of technical barriers to data portability and asks for transparency in this respect in the contract:

“The means through which they may migrate data pertaining to their farming operations to other service and the electronic data interchange standards and formats which are supported shall also be made clear.”  [109]

84

Even though there is an obligation on service providers to be clear about standards and interoperability, this falls short of a sector-wide standard and interoperability obligation. This rule is more about a transparency obligation for service providers. Therefore, even though the proposed rule in the EU Code becomes binding, this will not be effective in removing potential technical barriers to the free flow of data and will not ensure interoperability in the sector.

85

The portability rule is complemented by the following paragraph;

“This should be done without compromising restricted access to machine data or sensitive data (only relevant to the correct functioning of the machinery). This should be clearly specified in the contracts between farmers/contractors and device manufacturers.”  [110]

86

This statement caters to the wishes of machine manufacturers to protect their proprietary, sensitive and confidential data collected, stored and processed in machines, including data regarding the operations of the machine itself. Service providers or machine producers’ ‘proprietary data’ fall outside the data originators’ rights to portability.

87

The EU Code also mentions the data formats when receiving data from service providers;

“The data originator shall have the right to receive the data concerning their operation as specified in the contract, in a structured, frequently used and machine-readable format.”  [111]

88

The reference to “the right to receive the data concerning their operation” associates data rights with farming operations, not with the machine or device ownership. As noted before, this is important in a sector where renting machines and outsourcing of services is a common practice. It somewhat reduces the ambiguity in the attribution of original rights to data originators. It is obvious that stakeholders in the EU are aware of technical challenges to transfer data due to lack of standards and fragmentation in data formats, and this general principle could have been helpful to mitigate this problem to a certain extent, if it had been binding.

89

As a general consideration, although existing forms of portability related provisions in both the EU and US charters are not adequate to mitigate sectoral concerns deriving from farmers’ lock-in situations, one can expect that these voluntary portability rules might become a sector trend to implement more lenient data policies by service providers in terms of enabling the transfer of data to rivals when farmers desire to do so, especially for the EU Code as the US Principles are more focused on transplanting privacy principles rather than promoting competition in the market. However, it might not be realistic to expect that service providers will voluntarily renounce their exclusive control over the collected data. They have no incentive to weaken the advantages they derive from their exclusive data access. Still, the design of the EU Code, with a default portability right for farmers, [112] unless repealed by contract, could be an important step towards inalienable and binding rights for farmers, compared to the US Principles in which contractual clauses are always considered superior to the proposed rules and principles.

90

Our main focus in this section is to compare the attribution of original rights, data re-use/access rights and portability rights in these voluntary initiatives from a competition policy perspective. These rights could have major implications for data access related problems in the sector. However, it is also interesting to briefly evaluate other provisions in the charters, such as data retention/retrieval rights, purpose/storage limitation rules or prohibition of speculation and price discrimination. These may also affect the data access puzzle in this emerging sector. This sub-section discusses whether they are suitable for the non-personal machine ag-data setting and address the identified concerns.

106

Although there are some positive considerations, [122] attempts by the US and EU agricultural data charters to transpose some basic GDPR principles of personal data protection to non-personal machine-generated data run into several problems. For instance, notions of pseudonymization and anonymization or the right to be forgotten are related to the privacy of natural persons. They are not relevant for non-personal agricultural data. If the aim was to protect commercially sensitive data, it could have been stated more clearly without transplanting the GDPR concepts. In general, the absence of an obvious anchor for these rights in a natural person creates ambiguity with regard to the rights-holder: is it the farm or the farmer, or other parties?

107

These voluntary charters are naturally limited in terms of sector-wide validity and enforcement, except for the external auditing system in the US Principles. [123] An additional limitation in both data charters is the primacy of contracts over principles. Rights can be limited and alienated from farmers by contracts even though a company declares its participation in the charters. Markets and bargaining power in contractual negotiations will determine the outcome despite the proposed rules/principles. Even if the EU and US regulators would turn the voluntary principles proposed in the data charters into legally binding text, it is hard to see how they could correct B2B agricultural data market failures. [124]

108

These initiatives were not designed with a list of market failures and aftermarket competition concerns in mind. Instead, they transplanted rules designed to protect the privacy of individuals. One can, therefore, not expect these voluntary data governance initiatives in the US and EU to effectively address competition-related problems in this emerging sector.

112

There is a wide variety of third-party intermediaries that operate in the agricultural data market. Some of them behave in a “neutral” way: they are not vertically integrated with machine producers, inputs suppliers or agronomic services providers. As such, they have no stake in the sales of these products and no incentive to use the data to promote these sales. Of course, there are various shades of neutrality: some are more neutral than others. They range from not-for-profit to purely commercial data intermediaries. Their common characteristic is that they offer farmers some degree of control over the management of their data, sometimes combined with the promise that they can monetise farm data or appropriate a larger share of the benefits that data can generate. These intermediaries have been referred to as “Agri-Business Collaboration and Data Exchange Facility” (ABCDEF), [127] i.e. “neutral” B2B data platforms where farmers and agri-businesses can collaborate and exchange data in standardized formats. This could purportedly strengthen the position of farmers in the data market. The European Commission announced its support for the creation of “a common European agricultural data space to enhance the performance and competitiveness of the agricultural sector through the processing and analysis of production and other data, allowing for precise and tailored application of production approaches at farm level”. [128] This agricultural data space might also fall in the category of neutral intermediaries. However, no further details are known yet on this project.

113

The intermediary would act as a Farm Information Management System (FIMS), [129] comparable to Personal Information Management Systems (PIMS) [130] that have been suggested for personal data. They fulfil several roles: data storage, identity and permissions management, service and monetization management, standardized and secure data transfers through APIs, compliance management and accountability.

114

Large agri-business firms with vertically integrated data services are a step ahead of FIMS because they already have a large user base that they can leverage to generate network effects in data collection and better service production. It is not easy for FIMSs to overcome this disadvantage, unless they have a large and vertically integrated market side too. Some agricultural cooperatives may be in that situation as they sell agricultural inputs and rent machines. However, that makes them commercial stakeholders in at least one market and undermines their neutral third-party status. In France, for example, the InVivo agricultural group has started from its strong market position in agricultural products to add a data management and analytical dimension to its business. [131] There are many other examples of such intermediaries. [132] Some of these intermediaries have vertically integrated with data analytics firms. In the US, for example, GISC (Growers Information Services Cooperative) is an agricultural data cooperative that teamed up with IBM for data storage in the cloud to produce data analytics services that generate value-added on top of handling raw data. [133] Farmers pay for these services. This cooperative data business model retains some degree of neutrality with respect to products and services markets; it avoids self-preferencing in these markets. IBM has no stake in selling agricultural machinery or inputs, and it is neutral in this regard. As such, it may allocate a larger share of data-driven value-added to farmers. However, we do not have any information on the possibility for farmers to switch their data to other service providers than IBM. [134]

115

Other intermediaries have opted to stay neutral with regard to data analytics and use. They facilitate access and exchange of data but do not store or extract value from the data. For example, in the Netherlands, Join-Data is a not-for-profit agricultural data platform where farmers share their data with various agro-industry partners and companies that want to access data. [135] JoinData is set up by some large Dutch dairy and meat cooperatives involved in processing and distribution of agricultural inputs and livestock products. Some commercial firms are also members, including a bank and an IT services company that created the technical platform. [136] The platform manages data access authorisations for farms, but it does not store or analyse farm data. That is left to application providers. It is a mere passive and neutral data access & distribution platform, not an active data-driven agronomic services provider. It facilitates the transmission of data between farm machines and data users, including distributors of inputs and outputs, downstream industries, data-based agricultural service providers, with the authorisation of the farmer. JoinData membership terms & conditions do not say anything on ownership or access to data because it takes no responsibility for the handling of data. JoinData seeks to improve farmers’ trust by giving them more control over the use of their data at any time, by means of authorisations. Farmers do not pay for the service. Data users pay a fee for data communication, not for the data itself. It does not build data interfaces to facilitate data portability. Application providers have to build their own interfaces. It uses the AgroConnect data standard for data transmission and for its APIs. [137] Most members of AgroConnect are active in downstream data-driven services; some are manufacturers of machines and sensor devices. [138]

116

This model comes closest to a neutral third-party data intermediary. It gives farmers more control over who can access and use their data, reduces switching costs and avoids data lock-in for farmers. Farmers gain more subjective control over their data. That does not necessarily translate into capturing more value-added from agronomic services. That still depends on the farmer’s bargaining power with agro-service providers. It does not overcome the restrictions imposed by contracts between farmers and machines producers that may prevent them from accessing or porting their data, or may lock them into incompatible data formats. JoinData can only work when the original agreement with the data source (machine producer) allows it.

117

First, they require access to data sources. For example, the JoinData model works to the extent that data sources (machine producers) allow JoinData to manage the portability of their machine data. What would be their incentive to give away their exclusive access and allow other service providers to use their data? We can find some tentative answers to that question when looking at the membership list of JoinData’s data interoperability standard, AgroConnect. Members are mostly downstream agricultural services providers, [139] not upstream producers of data collection machines. The few exceptions are small machine and sensor producers that have very little to gain from maintaining data exclusivity. Their business model consists of selling machines and sensors, not selling data-driven analytics. We find the same pattern in membership of the more widely used Isobus interoperability standard for agricultural machines: only smaller machine manufacturers adhere to it while none of the larger ones do, except for a few of their machines in markets where they are not leaders. [140] This is in line with the predictions from economic theory. [141] When a platform is small, it can only gain from interoperability. Conversely, if the platform is large, gain from interoperability will be limited while its competitor will gain more. Consequently, dominant platforms’ incentives to accept interoperability will be low.

118

Second, they need to overcome several economic hurdles, similar to PIMS. [142] The parties (farmers, companies and platform operator) must find a sustainable data business model. That may be problematic. Farmers may not be willing to pay for storing and managing their raw data through FIMS, unless they receive well-defined monetary benefits in return. Some farmers may be motivated by the subjective feeling of more control over their data, independently of any monetary gains. Farmers may expect payment for the use of their raw data by agro-industry firms. This is unlikely to happen because the marginal value of individual farm data may be close to zero for a service provider as soon as it has reached a sufficiently large data pool where the marginal return to economies of scale and scope in aggregation come close to zero. That is why farmers usually have to pay a price for access to data-driven services, even if they deliver their own data to that service provider. New entrants in the data-driven services markets may subsidise data control services for farmers in order to attract more clients. This may be the case for JoinData. Eventually, however, full costs will have to be reflected on one or the other side of the market.

119

For data cooperatives, the only viable business model seems to require the production of data-based value-added services on top of the raw data delivered by farmers. This requires investment in data analytics as, for example, the case of the GISC in the US that collaborates with IBM to produce data-driven insights. Only large cooperatives with a sufficient volume of data collection can achieve the necessary economies of scale and scope in data aggregation to produce efficient data-driven services.

120

These economic considerations lead us to the conclusion that neutral third-party intermediaries are likely to remain outside the mainstream agricultural data market. It also raises questions about the potential benefits and limitations of voluntary interoperability and whether mandatory standards are necessary to overcome data-related competition bottlenecks in agricultural markets. We address this question in the next section.

121

In this section, we explore regulatory intervention as an alternative to overcome exclusive data access by device manufacturers and agronomic service providers. Data portability and/or interoperability is a necessary technical condition to unlock farm data. [143] However, it does not answer the question of who can use the portability or access rights and under which conditions. That is an important question because it affects the welfare of stakeholders in the agricultural production process. Policymakers can introduce mandatory portability to increase the joint welfare of farmers, agricultural industry and service providers, and consumers. The impact on these groups may not be evenly distributed, however, and can create equity and fairness concerns.

122

For personal data, the data subject as a natural person and originator of the data is the obvious rights holder and the basis for data protection rights in the GDPR. We argued in Section C that there is no equivalent for non-personal farm data, unless there is only one single data originator. When several parties contribute to an agricultural production process, they may all claim access rights to at least part of the data. Landowners may claim access to land use data from tenant farmers. Machine rental companies may compile usage data. Machine producers may collect data from all their machines. Agronomic service providers may collect data from all their client farms. Data analytics and other external service providers may claim use rights on the data that they process.

123

Some authors have suggested to distinguish between volunteered, observed and inferred data as a way to allocate data access rights. [144] Volunteered data have been willingly contributed by a user to service providers. For example, farms share their land and soil maps with rented seeding, fertilizer and harvest machines. Observed data are the result of interactions between users and the service provider. For example, combine harvesters collect data on the quantity of crops harvested. Fertilizer machines observe the type and quantity of chemicals used. Volunteered and observed are raw primary data. Inferred data are derived from raw data and produced by a data service provider by means of algorithms or other calculations and transformations. For example, raw land & soil maps and cropping pattern data are inputs for algorithms that recommend chemicals for crop protection. Combined with harvesting data, they can evaluate the productivity of a farm. Apart from the fact that the distinction between these three categories is not always clear, this categorisation does not resolve the question of who should get access to which type of data and under what conditions.

124

Currently, in the EU, portability right exists only for personal data in the GDPR. Even this right has significant limitations. [145] A very limited legal notion of portability right for non-personal data is mentioned in the Free Flow of Data Regulation, [146] only for cloud-based data services and on a voluntary basis through sectoral codes of conduct to be negotiated between industry stakeholders. In other words: it merely endorses the existing EU agricultural data charter. Other sectoral precedents for portable machine data exist, for example, in automotive, [147] energy [148] and payments services. [149]

125

The proposed Data Governance Act [150] includes regulation of data sharing services. Article 9 restricts the application of the regulation to three categories of data intermediaries: providers of bilateral or multilateral data exchange services, personal data sharing services and data cooperatives. Recital 22 explains that providers of data sharing services are specialised intermediaries that are independent from both data holders and data users. They assist both parties in their transactions of data assets. It covers services that intermediate between an indefinite number of data holders and users, not closed groups or an exchange platform that is exclusively linked to services provided by one data holder. It also excludes IoT data platforms that connect machines and devices, and services that generate value-added from the transformation and analysis of data without a direct relationship between data holders and users. We could not find any type of agricultural data service provider that would still fall within this very narrow definition of data sharing services. This leaves the category of data cooperatives as an alternative option. However, this category is not defined in the regulation. Even if there would be any agricultural data platforms that could be considered as data intermediary services under Article 9, the conditions that apply to these platforms under Article 11 are very general and do not go beyond what is already foreseen in the EU code of conduct that is investigated in detail in the section above.

126

The Digital Markets Act (DMA) [151] defines mandatory B2B data sharing obligations for (non-personal) commercial data. This applies only to very large “gatekeeper platforms” that provide “core platform services”. Agricultural data services are not covered by these DMA definitions. However, it is worth noting that DMA Article 6(h) introduces a real-time data portability right for business users on gatekeeper platforms. Article 6(i) mandates free access for business users to non-personal commercial data provided and generated by their activities on the platform. These articles introduce data access and portability rights for legal entities (i.e. businesses). Moreover, they go beyond the GDPR by abolishing any delays and mandating real-time access.

127

These clauses constitute a first step towards portability rights for non-personal commercial business data in the EU. While the DMA does not apply to agricultural data platforms, the European Commission announced its intention to prepare proposals for a Data Act in 2021. It would include general regulatory provisions for B2B sharing of non-personal and machine-generated data. [152] The details of this proposal are not known yet.

128

We can explore the conditions under which non-personal data portability rights could work for farms as business entities [153] and how this could increase competition in aftermarket services. In a simple one-to-one relationship between a farm and a machine or device producer, real-time portability and interoperability of machine data would separate primary machine markets from aftermarket services. It would enable farmers to select any aftermarket service provider of their choice. This would increase competition in aftermarkets. For example, a tractor or seeding machine could be steered by data-driven services from any provider. However, it would not prevent service providers from re-using the data for other purposes or sharing them with other businesses, unless re-use would be subject to consent from the farm to which the data pertain. One could think of a farm-centric portability and re-use right, limited to farms only and excluding other parties.

129

Exclusive rights for farms become complicated when more parties are involved in the agricultural production process. The farm’s central role in data collection may be eroded by competing data access claims from other parties. For example, machines can be owned by leasing firms, farmland can be owned by another party, farm data analytics and agronomic advisory services can be performed by a third party, etc. Leasing firms can claim access to machine data to monitor the use and performance of their machines; land owners may claim access to data on agricultural activities to monitor the quality of their land; and agronomic advisory services firms may claim rights over the service data that they produce. This leads to a debate on who gets access to which data under which conditions. Leasing firms may be granted access to mechanical machine data only, not to the quantity and quality of agricultural inputs and outputs. Land owners may, however, want to access data on the quantity and quality of inputs and outputs because that affects the quality and value of their land. Once these parties obtain a right to access these data, they may also claim the right to re-use the data without the consent of the farm.

130

Data access and re-use by other parties is valuable from a social welfare perspective. It enables these parties to improve the efficiency and economic value of their activities. At the same time, re-use may also impact the farm. Detailed land use data can affect the value of farmland and the creditworthiness of the farm. More reliable valuations are beneficial for society but not necessarily for the farm as a private undertaking. They may be used for price discrimination and speculation that are explicitly rejected in the EU code of conduct.

131

Like intellectual property rights, data need to have some degree of excludability in order to retain their market value. Making them widely available reduces their value to near-zero. [154] However, it may increase competition in downstream services markets, reduce prices and increase service quality. That may have positive welfare effects on farms via the services price channel. More competition in data-driven aftermarket services would help farmers to appropriate a larger share of data-driven productivity gains through lower prices and better service quality. [155]

132

The design of data access regimes is squeezed between two extremes. On the one hand, granting exclusive rights to farms when many parties contribute to the agricultural production process is not an optimal solution. On the other hand, generalised data portability and re-use for all is not ideal when negative externalities occur between parties. An intermediate solution that keeps some restrictions may be required to preserve the rights and welfare of some parties. All this indicates that a data access regime should be tailored to specific situations.

133

Smith warns us that the cost of intermediate data governance or data pooling regimes can be very high compared to the much lower costs of private ownership rights or fully open public domain regimes. [156] However, each of these cheaper regimes has its own costs. Public domain or full data sharing regimes may lead to underinvestment for lack of private incentives while exclusive private ownership data regimes lead to underutilisation of resources. Expensive data governance regimes should only be implemented if the benefits to society exceed the cost of governance.