Data sharing and monetization provides organizations with new sources of revenue and value creation. However, an accurate and scalable approach to data sharing and monetization for organizations is still lacking in practice. Due to the lack of efective mechanisms for control and enforcing governance as well as accurate valuation and pricing mechanisms, organizations are hesitant to share data. As a result, a large share of the economic value-creation potential of data is foregone. We propose a distributed-ledger-technology-based approach for decentralized data valuation incorporating federated machine learning to enable decentralized data-enabled learning and data valuation in a collaborative manner. We evaluate the proposed concept model with empirical evidence from expert interviews and single out the predictive maintenance context for future prototype development and testing.
Joint Workshops at 49th International Conference on Very Large Data Bases (VLDBW’23) — Workshop on Data Ecosystems (DEco’23), August 28 - September 1, 2023, Vancouver, Canada * Corresponding author. $ himmelsbach@uni-mannheim.de (T. Himmelsbach); mou@dbis.rwth-aachen.de (Y. Mou); decker@dbis.rwth-aachen.de (S. Decker); heinzl@uni-mannheim.de (A. Heinzl)
© 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License and consumer welfare [7, 8]. Decentralized platforms, CPWrEooUrckReshdoinpgs IhStpN:/c1e6u1r3-w-0s.o7r3g ACttEribUutRion W4.0oInrtekrnsahtioonpal (PCCroBYce4.0e).dings (CEUR-WS.org) fueled by Distributed-Ledger-Technology (DLT) and federated machine learning capabilities have the potential to contextual conditions, there are only very few objective overcome the drawbacks of centralized control, market measures and limited methodologies available to accupower, valuation, privacy, and security concerns by pro- rately determine the value of data. As such, the data qualviding a decentralized infrastructure that enables both, ity dimensions (e.g., completeness, accuracy, timeliness) secure data-enabled learning, and data valuation without or the relative position along the data value chain (e.g., an intermediary [9, 10, 11]. In our quest to facilitate a collecting, pre-processing, analyzing, using) help organisecure and scalable approach to data sharing and mon- zations to better gauge the value of their data [13, 14]. etization, we set out to answer the following research Extant literature proposes mainly four overarching, question: methodological data valuation concepts [13, 15, 1]: costoriented, market-price-oriented, risk-oriented, and usageRQ: How can Distributed-Ledger-Technology and federated oriented data valuation. Cost-oriented approaches build machine learning foster data monetization in decentralized their valuation on all costs that arise throughout the data data ecosystems? value chain such as data storage or data analytics. However, this approach is limited as the actual value created Due to the nascent nature of DLT and federated ma- by the data is entirely neglected. Market-price-oriented chine learning-based data monetization, we propose a concepts assume that data assets are traded on markets blockchain-based artefact including federated learning where their prices as an approximation of value are decapabilities and pre-evaluate it in the context of an Indus- termined. This approach is limited by the availability trial Internet of Things (IIoT) enabled predictive mainte- of comparable idiosyncratic use case-data combinations nance use case. In IIoT, the value of information exchange [4] and the assumed homogeneity among data buyers as well as the challenges thereof have been discussed in their willingness-to-pay [16]. Risk-oriented valuamanifold [12]. At the current stage, this research-in- tion approaches consider potential business risks (e.g., progress paper presents a conceptual model for an artifact measured as a function of probability and business cost that is to be developed and evaluated in the near future. of contingent outcomes) that may arise for a company The envisioned artefact draws on and extends the solu- from loss or misusage of data assets [13, 15]. Given that tion proposed by [1]. The presented conceptual model both, the probability and the business costs of adversarial addresses the two fundamental issues of centralized plat- scenarios are extremely dificult to quantify and forecast, forms —disproportional control and value capturing— the practicality of this approach is arguably low. Finally, as well as data privacy and data security by proposing usage-oriented valuation refers to the contribution that an approach for DLT-based data valuation and federated a data asset can generate to the company performance. data-enabled learning. The suggested approach combines [1] propose an approach for the usage-oriented data valthe scalable concept of compute-to-data (edge comput- uation suggesting that data valuation and pricing should ing) with a public, permissioned blockchain (Ethereum) comprise a combination of both, forward-looking exfor secure and transparent data access, valuation, and pected value anticipation and ex-post value measuring, monetization. We contribute to the growing body of In- which depends on the actual value contribution of a data formation Systems literature on data monetization and asset. data ecosystems, as well as Computer Science literature Data monetization provides organizations with an incenon federated machine learning in data ecosystems and tive to share their data with external parties and participrovide an innovative data monetization approach in pate in data ecosystems [4]. Data monetization describes form of a conceptual model to practitioners. the usage of data to achieve “quantifiable economic benefit” [ 17]. In a broader sense data monetization refers to both, indirect eforts aiming at improvement of business 2. Background processes and decision making, as well as external eforts aiming at outright selling data assets (via a data broker 2.1. Data Valuation and Data or independently) or data-based products and services Monetization [18, 19, 20]. While extant research examines the value creation potentials of data sharing manifold, the question of a fair valuation in data sharing arrangements is largely avoided [1].
Data valuation and pricing are key obstacles to data sharing and monetization. Data is a “non-rivalrous experience good” [4]. Non-rivalrous means that, once created, data can be exploited repeatedly by multiple parties without deteriorating in value [4]. Experience means that data must be used to realize its value [4]. Consequently, the value of the same data set varies significantly depending on the use case, context, and time. Even in adequate
Centralized digital platforms can be defined as “the extensible codebase of a software-based system that provides core functionality shared by apps that interoperate with move digital assets according to arbitrary pre-specified it, and the interfaces through which they interoperate” rules” [40]. Thus, smart contracts are algorithms that [21]. They enable multisided transactions and innova- comprise a-priori specified business logics (e.g., ownertions between diferent market players and create value ship, access-, and control rights), automatically execute through network efects [ 22, 23, 24]. A centralized data transactions accordingly, and record all transactions to platform refers to a technical environment for record- the blockchain [31]. In combination with blockchain ing, storing, analyzing, and presenting (big) data [25, 26]. infrastructure, smart contracts provide a “reliable, secure, Ecosystems are described by a “group of interacting firms and convenient approach to specifying an agreement, that depend on each other’s activities.” [27], e.g., devel- which is essential for data sharing” [1], enhancing the opers on Google Android depending on software updates transparency and traceability of transactions within the provided by the platform owner or data providers on a system. data platform such as Snowflake depending on certain standards and requirements for data storage. Tokenization refers to the “abstract representation of Decentralized data platform ecosystems can be under- physical assets in the form of blockchain tokens" [41]. stood as a subtype of data platform ecosystems. How- There are diferent token types, each with diferent charever, literature still lacks a generally adopted and recog- acteristics and taking central roles in the governance nized definition of decentralized data platform ecosys- and accessibility of decentralized DLT-based platform tems. In management and organizational theory decen- ecosystems. A high-level categorization distinguishes tralization is mostly referring to decision making and between three token types: Payment tokens, security authority [28, 29]. Correspondingly, in the platform tokens, and utility tokens [42]. In this paper, we focontext, decentralization regularly refers to governance, cus on two subtypes of utility tokens. Utility tokens are the mechanisms employed by platform owners aiming required for accessing the functionality of DLT token to orchestrate and influence ecosystem outcomes to fos- platforms. Without ownership of such tokens neither ter value co-creation [30]. On a more technical level the platform’s services can be used, nor any transactions decentralization in data platform ecosystems can also can be executed. The first utility token subtype, nonrefer to the data infrastructure [31]. Within this paper fungible token (NFT), is based on the ERC721 standard decentralized data platforms are understood as platforms [43]. Non-fungible means that while it can be transferred with decentralized data infrastructure and decentralized between participants within the ecosystem, the token governance building on DLT [32, 33]. With that, our un- is unique. Thus, NFTs certify ownership and tradeable derstanding of decentralized data platforms follows [34], rights to a digital asset. The second utility token subtype, who define a decentralized data marketplace as lacking fungible tokens, are classified by the ERC20 standard both a central authority and a central data repository. [43]. Fungible tokens are identical and interchangeable and represent access rights to digital assets. These acDLT serves as an umbrella term for multiparty systems cess rights can be traded and managed much like any operating in an environment without a central authority other good. In sum, smart contracts and tokenization or operator [35]. Blockchain technology is frequently provide an infrastructure for a decentralized data platregarded as a certain subgroup of DLT using a specific form ecosystem, enabling data sovereignty and trust and, decentralized data structure building on a chain of hash- thus, eliminating the need for a central platform owner. linked data blocks, representing transactions that are Further central aspects of decentralized data platform distributed and consistent among the network partici- ecosystems that enhance data sharing and monetization pants, the so-called nodes [35, 36]. Public blockchains are federated machine learning and digital twin capabiliallow all nodes to read the transactions logs while pri- ties. vate blockchains only permit the reading of transactions only to authorized nodes. Permissioned blockchains re- 2.3. Digital Twins and Federated Machine strict transaction validation, i.e., the participation in the consensus mechanisms, to chosen nodes while in per- Learning missionless blockchains all nodes validate transactions [35, 37, 38, 31]. Due to their distributed nature and the Digital twins refer to a virtual representation or digital peer-to-peer validation of transactions, DLT-based data replication of a physical object, system, or process [44]. platform ecosystems eliminate the need for a central plat- It is a digital counterpart that mirrors the characteristics, form owner [9, 39, 32]. behavior, and attributes of its real-world counterpart in real-time or near real-time and allows for diagnostics Central aspects of decentralized DLT-based data platform by using data captured from connected sensors. This ecosystems are smart contracts and tokens. Smart con- data can be further utilized to optimize the operation and tracts can be understood as “systems which automatically performance of the physical counterpart or predict future states [45]. Digital twins have been widely adopted in various fields, such as manufacturing, healthcare and transportation, due to their ability to ofer real-time monitoring and simulation, performance optimization, and fault prediction [46]. The emergence of Internet of Things (IoT) technology and machine learning has significantly accelerated the implementation and application of Digital twins. IoT allows real-time data collection from various sensors placed on the physical twin, and machine learning enables the processing of this vast amount of data.
Machine learning, as a subset of Artificial Intelligence (AI), has been central in enhancing the capabilities of digital twins. Machine learning provides the necessary algorithms and methods to analyze the data and create models capable of learning from this data, identifying patterns, and making predictions.
Currently, machine learning process models are majorly centralized. The process involved collecting data from various sources and aggregating it at a central point (e.g., server or cloud), where a machine learning model would then be trained [47]. However, this process raises several concerns [48]. Firstly, the transmission of data to a central repository results in security risks. Data can be intercepted during transmission, and the central repository itself can be a target for cyberattacks. Secondly, the aggregation of data at a central point raises privacy concerns and ethical issues. In many cases, the data used for machine learning contains sensitive information about individuals or organizations. Even if anonymized, the risk of re-identification through data linkage remains. Thirdly, the centralized approach requires significant computational resources and is less eficient. Large volumes of data have to be moved, requiring substantial bandwidth and storage. The latency associated with moving data to a central point can also slow down the learning process.
To address these limitations, the concept of federated (machine) learning was introduced [49]. Instead of requiring data centralization like other conventional approaches, federated (machine) learning describes a distributed machine learning approach that allows for training a global model on decentralized data sources while data remains on its original [11]. It is designed to address privacy concerns and data localization requirements, particularly in scenarios where data cannot be easily centralized due to privacy regulations or data ownership considerations.
The process typically involves the repetition of the following steps, namely, client selection, global model distribution, local model training, model verification, model aggregation and global model update. Federated Learning provides a methodology to build machine learning models using data located across diferent devices or servers while ensuring data privacy and reducing the requirements for data transmission. This strategy is particularly beneficial in scenarios where data privacy is critical, or where devices have limited connectivity or resources.
Figure 1 provides an economic-centered overview of the proposed conceptual model of DLT-based data monetization, focusing on the use case of predictive maintenance.
The exchange of information as base for innovative machine learning models that enhance the accuracy of predictive maintenance and prevent machine downtime has proven to be high [12], thus providing a suitable basis for our DLT-based data monetization concept model.
Currently, the value of datasets is estimated preacquisition, at the moment of sale based on cost, risk, market, or usage-oriented calculations. Post-acquisition value creation, not a-priori considered currently is not relfected in the data valuation, leading to high uncertainty for data sellers, especially regarding competitive data sets. The proposed reverse data monetization logic is a step-by-step, post-acquisition valuation and pricing approach based on data usage, actual costs, and generated business impact, aiming to consider future value creation in the determination of data value over the course of data usage. Data valuation and pricing is associated with the achieved business outcome by the data buyer after purchasing a data set. In the predictive maintenance use case, the business impact of a data set is determined by potential prevented machine downtime costs and production losses it can reduce. For instance, a dataset on rare frequencies of industrial pumps leading to breakdowns, ultimately leading to production stops of industry goods, would be of high value for competitors running similar pump systems in case machine downtimes could be prevented based on predictive maintenance precautions.
Potential machine downtime costs are calculated by the amount of produced items per hour with a certain profit per item equaling total costs of production losses. Furthermore, the value of a dataset is comprised of costs, such as data usage costs (curation, storage, monitoring, analysis) and (predictive) maintenance costs (employees, operating resources). Finally, a negotiable profit margin for the data seller completes the data valuation determination. The proposed reversed data monetization approach further comprises upfront and post-acquisition compensation. Upfront compensation is based on pre-acquisition valuation using traditional cost- and risk-oriented pricing models to mitigate any costs associated with the data buyer side. Post-acquisition compensation is determined by the actual costs and generated business impact along a pre-acquisition defined and negotiable time period. The value of data is determined at certain pre-acquisition deifned milestones over the course of data usage, actual costs, and generated business impact.
Figure 2 provides a technology-centric overview of the proposed DLT-based reverse data monetization concept model enabling a more accurate data valuation of data monetized between diferent actors in a digital ecosystem: The proposed model comprises a decentralized data fabric with built in services such as a digital twin visualization service, tokenization, federated machine learning capabilities, and an access and authority management, as well as a DLT-enabled data valuation meta space. In the following, the key components and functions of the proposed model are elaborated. trustful exchange of data within a local network of an organization as well as the exchange of meta data between the meta space and the local network of organizations.
The data fabric further comprises federated learning capabilities, enabling the data sellers to retain their data on-premise while allowing third parties access to a data set. Federated machine learning enables 3.2. Blockchain-enabled Data Valuation access to data assets distributed across multiple devices
Meta Space wcloituhdousterrveevreal[in11g].senAssitisvuechin,ftohremdaeticoennttroaliazecdendtartaal The Blockchain-based infrastructure provides the foun- storage and federated learning builds the foundation dation for a decentralized platform ecosystem and the for enhanced control and data sovereignty for the data reversed data monetization logic. The decentralized owner. Previously, the data seller had no control over platform ecosystem, we refer to as meta space, con- the data set, once it was sold to a third party. A further sists of two sub-dimensions: Data valuation space and built-in service of the data fabric, tokenization, enables Blockchain network. The Blockchain network comprises transparent and auditable access to data assets. The the key infrastructure elements of a public, permissioned ERC721 NFT certifies ownership and full rights to a blockchain (Ethereum), smart contracts, access control, digital asset [51, 52]. By acquiring an ERC20 data DIDs, and a data explorer. The data valuation space token that contains access rights to a certain data set, contains and visualizes data transactions through dig- a data buyer can access and use the data assets as ital twins along the data value chain from diferent actors pre-specified by smart contracts. Access can either mean within the ecosystem. It aims to provide a full traceability accessibility to a full data asset or parts of a data asset, of data transactions to enable the reverse data valuation or the deployment of approved algorithms by the data logic over the course of data usage. While the actual seller on the data asset. Furthermore, access to a data data sets are stored of-chain in a local organizational asset might be “perpetual”, “time-bound” or “one-time” network, the access control of meta data assets in form of [53]. By enabling traceability and control, tokenization tokens is stored on-chain [34]. The Blockchain network contributes to data sovereignty of the data owner. comprises smart contracts, that allow for a-priori speciifed business logics (e.g., ownership, access-, and control rights), automatically execute transactions accordingly, 4. Pre-Evaluation (e.g., micropayments for data transactions) and record all transactions as meta data to the blockchain [50, 31].
Therefore, the data infrastructure is designed to function without a central platform owner shifting the control and governance over the data assets to the data owners.
Moreover, the underlying blockchain provides the data owner with full traceability of the actual data usage of third parties, enabling a usage-based pricing approach.
To test and reach a better understanding of the suggested concept of DLT-based reverse data monetization, a first pre-evaluation with experienced industry experts was conducted. For this purpose, the data monetization logic was precisely described to receive feedback. The drawn conclusions and the feedback were incorporated in the proposed framework. Two key challenges need to be overcome. First, it might be dificult to persuade data 3.3. Data Fabric sellers to take part in this monetization scheme as risk To better understand the data value chain, the proposed is shifted to the data sellers. High costs may accrue for model comprises a data fabric, a decentralized framework, collecting and preparing the data prior to a monetization, that enables organizations to seamlessly and eficiently eforts which ask for some a-priori reward. High commanage, integrate, and distribute data across diverse and pensation in case of success might increase data seller’s distributed systems. The data fabric allows for a unified willingness to enter expenditure in advance, however a and consistent view of data, and facilitates data sharing, certain upfront payment might also become necessary. accessibility, and analysis by providing built in services Second, evaluating the step-by-step implementation as such as a digital twin toolbox. The visualization of busi- introduced within this paper one expert notes that it ness and production processes in a digital manner allows might be a challenge to achieve consensus between data for replication of the data value chain, including data buyer and data seller about when a particular milestone transactions, forming the basis for reverse data valuation. is reached and to what extent the data set was used for Connectors (E.g., Eclipse, Gaia-X) provide a secure and achieving this outcome. This paper aims at mitigating this second challenge by unambiguously defining the monetization milestones, ideally in conjunction with external third parties, representing it in smart contracts and tracking data usage using tokenization and a permissioned blockchain. Especially, a clear definition of milestones and corresponding to this triggering of the following actions like for example payments, implemented by smart contracts, are emphasized in that respect.
This paper presents a conceptual model for a prototype that is to be developed and further evaluated in the near future. The proposed concept combines the scalable concept of federated machine learning (compute-to-data) with a public, permissioned blockchain, a reverse valuation logic and addresses the two fundamental issues of centralized platforms, disproportional control, and value capturing, as well as key issues of data sharing, trust, data privacy, and ultimately, data valuation and pricing. The evaluation of the approach was only carried out in an initial step. The prototype is yet to be developed and tested in practice. Consequently, the technical specifications such as automation, security, and low transaction costs – while we recognize their importance for the prototype development [1] – were not central to the proposed conceptual logic in its current stage. Future work on this artefact will focus on an extensive evaluation with industry experts in order to complete and extend the proposed concept model. We recognize the dificulty of evaluating the actual value generated by using a data asset as the final value generation often cannot be measured and the value creation also depends on external factors. To mitigate that we aim to test our concept model in the predicitve maintenance use case, in which the actual data usage can efectively be measured ex-post. We aim to contribute to the growing body of Information Systems literature on data monetization and data ecosystems as well as literature in the field of Computer Science especially, federated machine learning and data ecosystems, and ofer an initial innovative data monetization approach that may reduce the hesitancy of ifrms to monetize their data and actively contribute to the value creation within their data ecosystems. ics: A research framework, Journal of management 323929.
information systems 35 (2018) 388–423. [30] A. Halckenhaeusser, J. Foerderer, A. Heinzl,
Plat[15] F. Bodendorf, K. Dehmel, J. Franke, Scientific ap- form governance mechanisms: An integrated
literaproaches and methodology to determine the value ture review and research directions, in: Proceedings
of data as an asset and use case in the automotive of the 28th European Conference on Information
industry (
net/10125/80023. [31] A. Zutshi, A. Grilo, T. Nodehi, The value proposi[16] S. Mehta, M. Dawande, G. Janakiraman, V. Mook- tion of blockchain technologies and its impact on erjee, How to sell a data set? pricing policies for digital platforms, Computers & Industrial Engidata monetization, Information Systems Research neering 155 (2021). URL: https://doi.org/10.1016/j. 32 (2021) 1281–1297. cie.2021.107187. [17] Gartner glossary, 2022. URL: https://www. [32] J. Pereira, M. M. Tavalaei, H. Ozalp, Blockchaingartner.com/en/information-technology/glossary/ based platforms: Decentralized infrastructures and data-monetization, accessed: 16 November 2022. its boundary conditions, Technology Forecasting [18] J. Baecker, M. Engert, M. Pfaf, H. Krcmar, Business & Social Change 146 (2019) 94–102. URL: https:// strategies for data monetization: Deriving insights doi.org/10.1016/j.techfore.2019.04.030. from practice., in: Wirtschaftsinformatik (Zentrale [33] C. Catalini, J. S. Gans, Some simple economics of Tracks), 2020, pp. 972–987. the blockchain, Communications of the ACM 63 [19] M. S. Najjar, W. J. Kettinger, Data monetization: (2020) 80–90.
Lessons from a retailer’s journey., MIS Quarterly [34] M. Travizano, C. Sarraute, M. Dolata, A. M. French, Executive 12 (2013). H. Treiblmaier, Wibson: A case study of a decen[20] B. H. Wixom, J. W. Ross, How to monetize your tralized, privacy-preserving data marketplace, in: data, MIT Sloan Management Review 58 (2017) H. Treiblmaier, T. Clohessy (Eds.), Blockchain and 9–13. URL: https://search.proquest.com/docview/ Distributed Ledger Technology Use Cases, Springer, 1885859901?accountid=14570. Cham, 2020, pp. 149–170. URL: http://dx.doi.org/10. [21] A. Tiwana, Platform Ecosystems: Aligning Ar- 1007/978-3-030-44337-5_8.
chitecture, Governance, and Strategy, Morgan [35] F. Glaser, F. Hawlitschek, B. Notheisen, Blockchain Kaufmann, 2014. URL: https://doi.org/10.1016/ as a platform, Business Transformation through C2012-0-06625-2. Blockchain: Volume I (2019) 121–143. [22] A. Hagiu, J. Wright, Multi-sided platforms, Interna- [36] M. Rauchs, A. Glidden, B. Gordon, G. C. Pieters, tional Journal of Industrial Organization 43 (2015) M. Recanatini, F. Rostand, K. Vagneur, B. Z. Zhang, 162–174. Distributed Ledger Technology Systems – A Con[23] M. A. Cusumano, D. B. Yofie, A. Gawer, The future ceptual Framework, Technical Report, Cambridge of platforms, MIT Sloan Management Review 61 Centre for Alternative Finance, 2018. (2020) 26–34. [37] G. W. Peters, E. Panayi, Understanding modern [24] Y. Yoo, O. Henfridsson, K. Lyytinen, The new or- banking ledgers through blockchain technologies: ganizing logic of digital innovation: An agenda Future of transaction processing and smart confor information systems research, Information tracts on the internet of money, in: Banking Beyond Systems Research 21 (2010) 724–735. URL: https: Banks and Money, 2016, pp. 239–278. //doi.org/10.1287/isre.1100.0322. [38] D. Trabucchi, A. Moretto, T. Buganza, A. MacCor[25] G. Mijsters, A Focus Area Maturity Model for Data mack, Disrupting the disruptors or enhancing as a Platform, Master’s thesis, 2019. them? how blockchain reshapes two-sided plat[26] P. Kramberg, A. Heinzl, Data platform ecosystems, forms, Journal of Production Innovation ManageHMD Praxis der Wirtschaftsinformatik 58 (2021) ment 37 (2020) 552–574. URL: https://doi.org/10. 477–493. 1111/jpim.12557. [27] M. G. Jacobides, C. Cennamo, A. Gawer, Towards a [39] S. Nakamoto, Bitcoin: A peer-to-peer electronic theory of ecosystems, Strategic management jour- cash system, Decentralized business review (2008). nal 39 (2018) 2255–2276. [40] V. Buterin, et al., A next-generation smart contract [28] V. Govindarajan, Decentralization, strategy, and and decentralized application platform, white paper efectiveness of strategic business units in multi- 3 (2014) 2–1. business organizations, Academy of Management [41] O. Ross, J. R. Jensen, T. Asheim, Assets under tokReview 11 (1986) 844–856. enization, Available at SSRN 3488344 (2019). [29] J. Zabojnik, Centralized and decentralized deci- [42] L. Oliveira, L. Zavolokina, I. Bauer, G. Schwabe, sion making in organizations, Journal of Labor To token or not to token: Tools for understanding Economics 20 (2002). URL: https://doi.org/10.1086/ blockchain tokens, in: International Conference on