Knowledge leaks in data-driven business models? Exploring different types of knowledge risks and protection measures

Fruhwirth, Michael; Pammer-Schindler, Viktoria; Thalmann, Stefan Article Knowledge leaks in data-driven business models? Exploring different types of knowledge risks and protection measures Schmalenbach Journal of Business Research (SBUR) Provided in Cooperation with: Schmalenbach-Gesellschaft für Betriebswirtschaft e.V. Suggested Citation: Fruhwirth, Michael; Pammer-Schindler, Viktoria; Thalmann, Stefan (2024) : Knowledge leaks in data-driven business models? Exploring different types of knowledge risks and protection measures, Schmalenbach Journal of Business Research (SBUR), ISSN 2366-6153, Springer, Heidelberg, Vol. 76, Iss. 3, pp. 357-396, https://doi.org/10.1007/s41471-024-00189-z This Version is available at: https://hdl.handle.net/10419/312594 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ ORIGINAL ARTICLE https://doi.org/10.1007/s41471-024-00189-z Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 Knowledge Leaks in Data-Driven Business Models? Exploring Different Types of Knowledge Risks and Protection Measures Michael Fruhwirth · Viktoria Pammer-Schindler · Stefan Thalmann Received: 16 February 2022 / Accepted: 17 June 2024 / Published online: 30 July 2024 © The Author(s) 2024 Abstract Data-driven business models imply the inter-organisational exchange of data or similar value objects. Data science methods enable organisations to discover patterns and eventually knowledge from data. Further, by training machine learning models, knowledge is materialised in those models. Thus, organisations might risk the exposure of competitive knowledge by sharing data-related value objects, such as data, models or predictions. Although knowledge risks have been studied in traditional business models, little research has been conducted in the direction of datadriven business models. In this explorative qualitative study, we conducted 28 expert interviews in three rounds (two exploratory and one evaluatory) and identified five types of risks along the three basic types of value objects: data, models and predictions. These risks depend on the context, i.e., when competitive knowledge could be discovered from shared value objects. We found that those risks can be mitigated by technology, contractual regulations, trusted relationships, and adjusting the business model design. In this study, we show that the risk of knowledge leakage is a relevant risk factor in data-driven business models. Overall, knowledge risks should be Michael Fruhwirth · Viktoria Pammer-Schindler Know-Center GmbH, Sandgasse 34/II, 8010 Graz, Austria Institute for Interactive Systems and Data Science, Faculty of Computer Science and Biomedical Engineering, Graz University of Technology, Sandgasse 36/III, 8010 Graz, Austria E-Mail: [email protected]; viktoria.pamm[email protected] Present Address: Michael Fruhwirth Silicon Austria Labs GmbH, Sandgasse 34/IV, 8010 Graz, Austria E-Mail: [email protected] Stefan Thalmann Business Analytics and Data Science-Center (BANDAS-Center), School of Business, Economics and Social Sciences, University of Graz, Universitätsstraße 15 Building F/III, 8010 Graz, Austria E-Mail: [email protected] K 358 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 considered already during business model design, and their management requires an interdisciplinary approach via a balanced assessment. The level of knowledge protection from a technology perspective highly depends on computer science innovations and thus is a moving target. As an outlook, we suggest that knowledge risk will become even more relevant with the extensive usage of machine learning and artificial intelligence in data-driven business models. Keywords Business model innovation · Data analytics · Data-driven business models · Knowledge risks · Risk management · Value objects 1 Introduction Developments in big data technologies and artificial intelligence (AI), as well as the availability of large data sets, hold the opportunity for the development of new products, services, and business models (Günther et al. 2017; Woerner and Wixom 2015), so-called data-driven business models (DDBMs) (Hartmann et al. 2016; Wiener et al. 2020). Such business models often imply the exchange of data and similar data-related value objects. Further, in such business models, sensitive information and competitive knowledge are materialised in data or models. At the same time, data science methods allow extracting information or knowledge from fine-granular, heterogeneous data, leading to potential risks when data is shared. Whereas before, knowledge needed to be represented in a much more explicit manner. Thus, it is challenging for organisations to evaluate what knowledge could be discovered from shared data sets (Zeiringer and Thalmann 2020). For instance, simply “looking at the data” (i.e., at the headers of a database or descriptive statistics over a single dataset) is not enough to assess which knowledge could be drawn from the data. Sharing data implies the risk—which we refer to as knowledge risks—that competitive knowledge could leak and spill over to other organisations. For example, we found such risks in a case study with an industrial company (Fruhwirth et al. 2019). In this case, novel knowledge of a real-world physical phenomenon (i.e., predicting the residual lifetime of a physical component) was generated from data and materialised in a model. Building new DDBMs around this model (i.e., offering the model) could imply the risk of leaking core knowledge, as one workshop with managers of this company showed. Further, the willingness to share data is often a prerequisite for a DDBM, but potential knowledge leakages negatively influence this willingness. Thus, DDBMs require balancing between sharing and protecting knowledge. Further, IP might be shared or could be re-engineered when offering machine learning (ML) models through an API (Application Programming Interface) (Hanzlik et al. 2021). Knowledge risks have been studied in strategic alliances (Hernandez et al. 2015; Jiang et al. 2016;Kaleetal.2000) and traditional business models (Al-Aali and Teece 2013). However, as shown above, DDBMs imply new types of risks, particularly that knowledge may spill over to competitors via sharing data and similar value objects. Although such risks exist, little has been written about how different types of offerings of DDBMs, or exchanged value objects in particular, relate to K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 359 knowledge risks. Therefore, we address the following research question in this paper: What knowledge risks are associated with sharing different types of data-related value objects in data-driven business models, and what are protection measures? To answer this research question, we interviewed 28 experts from industry and academia to explore cases of knowledge risks. We structured different types of risks, contextual factors and protection measures based on the three basic types of value objects: data, models and predictions. Based on our findings, we suggest three fields of action to mitigate knowledge risks in DDBMs: using technology, adjusting the business model design and establishing trustful relationships and contractual regulations. Managing knowledge risks in DDBMs requires a balanced view and interdisciplinary approach already during the design of a DDBM. 2 Background 2.1 Data-Driven Business Models Data-driven business models (DDBMs) have a conceptual focus on value creation from data (Guggenberger et al. 2020). A business model is a conceptual tool that allows a simplified description of how organisations create, deliver and capture value (Osterwalder and Pigneur 2010; Osterwalder et al. 2005; Teece 2010). Firms with a DDBM utilise data as a key resource for new business (Hartmann et al. 2016). They generate customer value through data analytics and machine learning (Schüritz et al. 2017b). Data analytics and machine learning techniques are used to discover insights from data (Kühne and Böhmann 2019). These insights are delivered as data analytics-based features, products, or services and support customers in their decision-making process (Schüritz et al. 2019) and enable the generation of new revenue streams (Schüritz et al. 2017a). Thus, data intermediation is the central value proposition (Dorfer 2016). Developing a DDBM requires business and technological capabilities (Stahl et al. 2023). Literature started to analyse and categorise DDBMs from different perspectives. Two common approaches are to differentiate based on the type of data sources used (e.g., internal existing or self-generated data vs externally acquired, customerprovided or free available data; see, e.g., Hartmann et al. 2016) or the type of analytics used (e.g., descriptive, diagnostic, predictive, vs prescriptive; see, e.g., Hunke et al. 2019). As data intermediation is the central value proposition (Dorfer 2016), it is also worthwhile to distinguish DDBMs based on the type of value proposition and offerings. For instance, Schüritz et al. (2019) differentiate between data, insights, and actions as offerings. Dehnert et al. (2021) further differentiate between data, information/knowledge, actions and non-data products and services in DDBMs. Hirt and Kühl (2018) describe Model-as-a-Service and Prediction-as-aService as two other types of offerings. These offerings can be differentiated by the type of exchange of value objects (Leski et al. 2021). A value object, as described in the e-3 value ontology, “is of value for one or more actors. Actors may value an object differently and subjectively, according to their own valuation preferences” (Gordijn and Akkermans 2003, K 360 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 Data-Driven Business Models Data-sharing Business Models Model-sharing Business Models Prediction-sharing Business Models Fig. 1 Subtypes of DDBMs based on exchanged value objects p. 120). Concerning DDBMs, such a value object can be data (e.g., Dehnert et al. 2021), models (e.g., Hirt and Kühl 2018)orpredictions or insights in general (e.g., Schüritz et al. 2019). By data, we understand a tradeable collection of “codified observation[s] fixed in a tangible medium” (Thomas et al. 2023, p. 256). Shared data can be in the form of specific data points, whole data sets (or data streams) or aggregated data (e.g., via descriptive statistics). By model, we understand a program or function that can identify patterns or provide predictions based on previously unseen input data. A model is a result of applying a machine learning algorithm to a set of (training) data. A model consists of its code and configuration. Hirt and Kühl (2018) differentiate between base models specific to one particular problem and transfer models that can be applied or transferred to a set of similar problems. The type of prediction encompasses identifying patterns, predicting events or attributes, or recommending actions based on incoming data applied to a learning model (Hirt and Kühl 2018). Predictions also represent target-specific insights that are shared to solve a specific (decision) problem of the customer, create customer benefit, and, in return, generate revenue. As Fig. 1illustrates, data-, model, and prediction-sharing business models can be understood as three subtypes of DDBMs. Differentiating DDBMs based on exchanged value objects is still under-represented in the DDBM literature, but a reasonable differentiation when it comes to knowledge risks: We assume that sharing different types of value objects leads to different types of risks. Examples of DDBMs that provide data as an exchanged value object are APIbased data-sharing business models in logistics (e.g., Möller et al. 2020). In such data-sharing business models (Schweihoff et al. 2023) or “data-as-a-service” business models (Chen et al. 2011), the business model owner grants other parties access to his own data set in exchange for compensation (Schweihoff et al. 2023; Vesselkov et al. 2019). One major obstacle to data sharing in organisations is the concern about exposing sensitive data and giving competitors a competitive advantage (Gelhaar and Otto 2020; Schweihoff et al. 2023). Thus, security aspects, such as usage restrictions K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 361 or cryptography, need to be implemented in such business models (Schweihoff et al. 2023). Examples of DDBMs that provide models as an exchanged value object are Language-Model-as-a-Service (Sun et al. 2022). In such a model-as-a-service business model, the user provides or uploads data to the service provider who builds (trains) a model based on this training data and his own human and/or machine intelligence (Hirt and Kühl 2018). Examples of DDBMs that provide predictions as an exchanged value object are prediction APIs (Santhosh et al. 2019). In a “prediction-as-a-service” or more general “analytics-as-a-service” business model, the provider applies a (machine learning) model to the input data provided by the customer to generate a prediction of events, recommendations or to identify patterns and finally to support decisions or automate actions for the customer (Hirt and Kühl 2018; Schüritz et al. 2019). We subsume these different terms under the term prediction for the context of this paper. 2.2 Knowledge Risks Emerging from Data Sharing DDBMs involve new types of risks. Large-scale data sharing can cause leakage of competitive knowledge and intellectual property (Zeiringer and Thalmann 2020; Zeng et al. 2012). This risk is called knowledge risk and comprises potential knowledge attrition, loss, leakage or spill-over of knowledge that could adversely affect the organization’s strategic advantage (Durst and Zieba 2017; Perrott 2007). Competitive knowledge of a firm can be discovered from shared data sets using advanced analytics methods (Ilvonen et al. 2018). Further, it is difficult for firms to evaluate which knowledge could be discovered by external actors from shared data (Zeiringer and Thalmann 2020). Known approaches for external acquisition of competitive knowledge that endanger a firm’s intellectual property are information leakage in supply chains (Zhang et al. 2012), industrial/data espionage (Thiel and Thiel 2015), or data breaches (Khan et al. 2021). An adversarial actor could also obtain valuable knowledge by reverse-engineering the firmware of a physical product to reconstruct an embedded algorithm (Thiel and Thiel 2015). For instance, it is technically possible to reverse-engineer black-box neural networks (e.g., Oh et al. 2019), or to steal machine learning models via API access (e.g., Tramèr et al. 2016). The described attacks can lead to unintended leakage or spill-over of knowledge, denoted as knowledge risk (Ilvonen et al. 2018; Zeiringer and Thalmann 2020). A knowledge risk is the “measure of the probability and severity of adverse effects of any activities engaging or related somehow to knowledge that can affect the functioning of an organisation on any level” (Durst and Zieba 2018, p. 2). Knowledge risks can be analysed by the factors that cause them and the preventive measures organisations can take (Durst and Zieba 2017). Managing knowledge risks in terms of knowledge protection is one core strategy of knowledge management (Loebbecke et al. 2016). It is crucial for organisations as knowledge is essential for competitive advantage (Jennex and Zyngier 2007). Therefore, knowledge protection prevents unwanted knowledge leakage to non-authorized people and organisations (Manhart and Thalmann 2015). Existing knowledge protection literature focuses on formal and explicit knowledge. It does not consider tacit knowledge in organisations (Manhart K 362 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 and Thalmann 2015) and the knowledge that can be discovered from data streams (Ilvonen et al. 2018). While explicit knowledge (e.g., materialised in data-related value objects) could quickly leave a company, tacit knowledge is more difficult to transfer and informal knowledge protection practices are needed (Thalmann et al. 2024). Finally, developing business models can be understood as a set “of concrete choices and the consequences of these choices” (Casadesus-Masanell and Ricart 2010, p. 198). Managers must balance expected risks and estimated returns when deciding between different business model design options (Casadesus-Masanell and Ricart 2010; Tesch et al. 2017). Such risks can threaten the profitability of the business model or even the firm’s value (Brillinger 2018), making it necessary to manage the risks. Risk management generally involves identifying, assessing and monitoring risks (Brillinger et al. 2020; Hallikas et al. 2004). Risks are usually evaluated by assessing the probability of a risk event and its impact on the business model (Hallikas et al. 2004; Brillinger et al. 2020). The problem with assessing nonfinancial risks, such as cyber security risks, is that little quantitative information is available, especially no reliable probability distributions (Franke 2020). Despite this, identifying and deciding how to deal with risks already in the business model design is crucial (Girotra and Netessine 2011). After identifying and being aware of risks, managers can adapt the business model design as a risk management measure (Brillinger et al. 2020). Our Conclusion from the Literature DDBMs can be differentiated based on the offering or, in particular, exchanged values. Based on the literature, we have stated that offerings in DDBM can be distinguished by three types of value objects: data, models and predictions. Further, knowledge protection literature recognises data sharing as a knowledge risk in general and that extracting knowledge from shared data is possible via data science methods. We already have the first evidence from previous research that exchanging data-related value objects can lead to knowledge risks (Fruhwirth et al. 2019). Nevertheless, the relationship between knowledge risks and exchanged data-related value objects in DDBMs has not been studied, and this connection has not been made by previous literature. 3 Research Method Our study aims to explore knowledge risks specific to DDBMs due to the specific nature of value objects. Given the novelty of the problem and lack of understanding of how and if knowledge risks occur in DDBMs, we applied an exploratory, qualitative research design that is appropriate for investigating why a certain phenomenon occurs (Yin 2009). The research design is qualitative, as we analysed interview data (see data collection section below), and exploratory, as we used a bottom-up data analysis method as informed inductive coding (see data analysis section below). K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 363 Table 1 Overview of our data collection process Interview Round 1 Interview Round 2 Interview Round 3 Interview participants 16 Interviews 7 Interviews 5 Interviews 7 Researchers (R1–R7), 9 Industry Experts (I1–I9) active in datadriven services, business model innovation and knowledge risks 3 Researchers (R8–R10), 4 Industry Experts I10–I13 active in datadriven services 5 Industry Experts (I14–I18) active in data-driven services Duration 35–75min 38–59min 40–59min Goal, main questions and content Focus on knowledge risks in DDBMs in general Focus on knowledge risks from sharing data-related value objects (data, models and predictions) Evaluation of results Presentation of 5 types of knowledge risks Main outcomes Knowledge risks differ if data, models or predictions are shared Identified five types of risks based on the three types of value objects Subtypes of risks for each type of shared value objects & contextual factors 3.1 Data Collection Due to the tacit and sensitive nature of the topic for organisations, we decided on expert interviews in three rounds as our primary data source (see Table 1), as interviews allow comprehensive discussions (Yin 2009). As interview partners, we selected 28 experts, 18 from industry (I1–I18) and 10 from research institutions (R1–R10) (see Table 3in Appendix A). We followed a purposive sampling strategy (Etikan 2016) and, in particular, an expert sampling strategy that is useful “when investigating new areas of research” and in particular when “there is currently a lack of observational evidence” (Etikan 2016, p. 3). As it was challenging to identify suitable cases (i.e., organisations) where knowledge risks have or could occur, as such information is not publicly available, we also selected consultants and researchers as informants who reported such cases. Academic experts reported on their experience and cases of knowledge risks in DDBMs based on their collaboration with industry (e.g., as part of research or consulting projects). We selected experts based on their knowledge and experience in developing DDBMs or supporting organisations in that process. For academic experts, we considered their recent publications on DDBM as an additional selection criterion. The selection of experts in the initial interview round was broader: we also selected experts in business model innovation and knowledge risks in general to explore the topic. We searched for experts in our immediate network and through an extended network on the LinkedIn platform (2nd order contacts). We conducted the interviews as face-to-face meetings or via digital communication software and audio-recorded them. Appendix A provides a detailed description of the experts who were interviewed. The scope of the first interview round was very broad, serving as a starting point to explore knowledge risks in DDBMs. After initial data analysis, we found that differentiating and analysing knowledge risks in DDBMs based on exchanged value objects is interesting and reasonable. Therefore, we conducted seven additional and K 364 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 more focused interviews with additional experts. In this second interview round, we presented and discussed the three data-related value objects (data, model, and predictions) and asked about cases and their relation to knowledge risks. In the first round, not all value objects were covered in each interview as the insights emerged over time. Further, we investigated motivations and practices in the design phase of a DDBM in detail, as we could now ask more focused questions in the second round. At the beginning of our semi-structured interview guideline, we presented working definitions of central concepts and an abstract problem definition, illustrated with a case example. The interview was divided into two parts: The first two-thirds of the interview focused on exploring the problem of knowledge risks in DDBM. The last one-third (only in interview round 1) focused on discussing requirements for ICT tools identifying and describing knowledge risks in DDBMs (Fruhwirth et al. 2021). We asked the interview partners for real examples from their context to concretise and ground the discussion as much as possible within their experience. The guideline was tested with a PhD student from the same subject (with practical experience) and methodological knowledge (training) regarding the guideline’s comprehensibility, question flow and structure. We adjusted our interview guideline for the second set of interviews through detailed questions (e.g., regarding protection measures) and a short presentation of our interim results. We presented each type of value object shortly and asked the experts how they perceived the knowledge risk related to each value object. To validate our results, we conducted a third interview round with five additional industry experts in data-driven services and data analytics. The interviews lasted between 40 and 59min. We again presented our problem definition, the concepts from the data analysis step after the two previous rounds and the five types of risks identified. Further, we provided one slide per type of risk with a short description and one example from the initial expert interviews. The three guiding questions for the evaluation interviews were: 1) Do you perceive these risks as relevant for your business? 2) Are there any other types of risks missing in that context? 3) Is the description of each risk reasonable for you?. Table 1summarizes our data collection process. 3.2 Data Analysis Interviews were fully transcribed and cleaned. Quotes used in this publication from interviews conducted in German were translated into English (marked with a “*”) and reviewed by a second researcher. We analysed this data following a qualitative content analysis approach via informed inductive coding (Mayring 2015)using MAXQDA V.11. For analysing the first round of interviews, the dimensions of analysis were themes that corresponded to the leading interview questions and developed a provisional coding scheme to structure the data. The major themes from the interview guideline have been “causes for knowledge risks”, “consequences of knowledge risks”, and examples. For the theme of the causes, we generated “influencing factors” and “mechanism” as our major categories. We distinguished between “type of knowledge K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 371 “On the other hand, when I talk about data that directly relates to the product, with which it is possible to draw conclusions about the architecture and technological specifics. Here, of course, the situation is different and the sensitivity of the information is higher.” (I17*, Manager Data Analytics Semiconductor Company) On the other hand, the manager also mentioned that sharing operational data from their production machines for maintenance or optimisation was perceived as less critical, as no conclusions on competitive knowledge are possible. The risk of knowledge leakage through data sharing depends on the context. If data is shared that relates to competitive knowledge, i.e., about their products or core processes, that allows an external party to make conclusions on the architecture or technology used, then it is perceived as critical. If the data relates to a more common context, such as the maintenance of machines, sharing data was perceived as less critical. Thus, what is competitive knowledge is very specific to the company and depends on its business model. One interviewee, therefore, pointed to the direction that internal balancing is necessary, i.e., at what stage is the retrieval of knowledge not acceptable for the company anymore? They need to take measures: “The internal discussions have to be held about when we have reached a level where drawing conclusions about the data or, for example, the vehicle’s configuration, the production, the development, is no longer acceptable for us, and we therefore have to do something else.” (I14*, Consultant Data-Driven Services) 4.1.4 Knowledge Protection Measures As we have seen above, knowledge risks in DDBMs are very contextual, i.e., if the shared data relates to competitive knowledge. One protection measure that our interviewees mentioned was to classify the data sources and to decide if this data can be shared or not, as one manager from the semiconductor industry mentioned: “And you have to have business processes in place. That’s what we have at our company in place, where you evaluate the data according to categories, from public to strictly confidential, for example.” (I17*, Manager Data Science Semiconductor Company) Another mechanism to tackle knowledge risks and enable data sharing is to involve a data platform. It mediates the data exchange between actors with technical measures implemented in the platform while preserving the provider’s knowledge. The automotive manager further mentioned here: “That’s why there are all these data-sharing platform initiatives, [enabling] data exchange under the premise of knowledge preservation. So, I can retain my knowledge but still share data. However this may work, it’s a task that probably needs to be solved so that it really takes off.” (I4*, Manager Data Analytics) K 372 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 The interviewee highlighted that knowledge protection concerns seem to be one of the main motivations for the rise of data platforms. However, he also acknowledges that protection concerns must be addressed properly before implementing a DDBM. There are also technical measures regarding secure technologies, like encrypting or decentralising data when performing data analytics and thus applying methods such as multi-party computation or homomorphic encryption. Another approach mentioned was to share only synthetic data, i.e., data generated by generative AI with similar properties necessary for sharing. Our interviewees frequently also mentioned using contracts such as NDAs (NonDisclosure Agreements) to tackle this risk. Nevertheless, they cannot prevent knowledge leakage when the contract is breached. Further, our interviewees frequently mentioned trusted relationships as a measure to mitigate knowledge risks. One practical approach mentioned was to begin sharing smaller and less critical data sets and to intensify the relationship over time. Firms and customers might be over-cautious and over-protective and, therefore, unwilling to share their data for fear of knowledge risks. This would imply that the DDBM is not implemented. This is especially the case as there is currently much awareness of data-related risks. Our interviewees reported the fear that others could benefit more from sharing and, therefore, as a consequence, decided not to share the data. This is perceived as a barrier for DDBMs, as one data science manager in the automotive industry mentioned: “Because all the companies in the [supply] chain are so afraid of losing knowhow, they don’t share the data. [...] This leads to the fact that it is sometimes difficult in the data environment for me to do business.” (I4*, Manager Data Analytics) Not realising a DDBM is the most extreme knowledge protection measure which is chosen if the perceived (vague) risks outweigh the perceived benefits of the DDBM. Therefore, our interviewees suggested balancing the expected benefits and possible risks: “And then there is also the question of the benefit: How much information can I gain when I give out data for further processing, versus the risk, what am I giving away?” (I18*, Managing Director Data Service Company) Thus, the risk can be reduced by running a data service or prediction model onpremise, i.e., locally at the customer’s premise, so that the data does not have to be shared. Another approach would be to use federated learning architectures,where the data stays local and only (transfer) models are shared or the weights of a neural network. A further knowledge protection measure is sharing models instead of data. Models are exchanged to protect the underlying data and allow a bidirectional flow of information without exposing competitive knowledge, as one data science professor explained: K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 373 “To build a model in order not to share the data. The model is already a risk mitigation method. With the goal, though, that you then have a flow of information in both directions.” (R4*, Professor for Data Science) The important aspect mentioned here is that exchanging models is a risk mitigation strategy, which is part of DDBMs. 4.2 Sharing Models 4.2.1 The Risk of Leaking Competitive Knowledge from Shared Model Competitive knowledge might be leaked by sharing models, as knowledge from experts (e.g., engineers) is introduced to the model in the process of creating or training (e.g., engineering knowledge about the ageing behaviour of a certain technical component). Models could also reveal information they have learned but not intended to be shared. If the model is shared in a white-box-like manner (i.e., sharing the code with parameters and configuration), competitive knowledge is likely shared, leading to a knowledge risk. For instance, models are delivered as part of a consulting or engineering project to support the customer in developing a DDBM, as one interviewee reported: “We are a service provider for model development and algorithms, and we sell those directly to our customers, then we always sell a bunch of knowledge too.” (I4*, Manager Data Analytics) The interviewee highlights that, with the model, a huge amount of knowledge is transferred to the customer. Thus, our interviewees acknowledge that competitive knowledge could spill over to other actors if models are shared. The interviewed manager is already aware of this problem and mentioned later that there are hardly any organisational guidelines to ensure that shared models are not misused regarding knowledge leakage. Reconstruct the Parameters or Configuration from a Black-Box Model Even if models are shared as black boxes, i.e., the configuration and parameters of the model are hidden, there is also the risk that knowledge can be retrieved through reengineering of the model through specific data science methods from a theoretical point. Overall this risk was perceived as low compared to white-box models. One data analytics consultant reported here on one case: “In general, you can re-engineer nearly every model if you know the input and the output. Then there are also algorithmic methods to decompose analytics models. There are methods from explainable AI to understand them. [...] We see this more, and more frequently, our customers try to better understand how our models work.” (I6*, Manager Data Analytics Consulting) This example shows that business customers are already trying to understand and re-engineer models and that providers are aware of this fact. However, similar to other security mechanisms, it is a question of effort. K 374 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 Needing to Explain how the Model Comes to Certain Decisions or Predictions The requirement of fair, accountable, and transparent AI (FAT AI) creates a demand to explain how models come to a certain decision or recommendation. One professor in Business Analytics sees this as a challenging trend from the perspective of knowledge protection and reported on one case from an industry project: “And there is a pressure here from the customer to the provider. Because you have to explain how a chatbot comes up with that conclusion. So, in that way, you are kind of exposing the algorithm behind this. [...] the openness of the algorithm means that you also expose knowledge.” (R3, Professor for Business Analytics) This example shows that providers could be forced to expose their underlying models and algorithms, and thereof knowledge could be retrieved from the exposed model. Thus, FAT AI-compliant models or explainable AI approaches could reduce the protective effect of models in DDBM. Leaking the Model to a Third Party, e.g., when Collaborating with a Startup to Build the Model A knowledge risk from sharing models could also arise when a model is jointly developed with a partner (e.g., an AI start-up) and the model is leaked there to a third party. One manager, for instance, mentioned one potential scenario: “Let’s say I have a transformer model that knows exactly how I make a chip at our company. If I lose something like that out of my hands, for example by cooperating with a startup or a partner company, whether it’s small or large. Then I lose all know-how at the push of a button.” (I17*, Manager Data Science Semiconductor Company) 4.2.2 The Risk of Inference of the Underlying Training Data Further, data science methods, such as model inversion attacks, allow someone to infer the original data used to train the model. Competitive knowledge might spill over when the model user can reconstruct the original training data from a shared model, in particular, to infer the structure of the data (e.g., particular data fields) or the structure in the data (e.g., properties of the sample and the bias in the data). Socalled model inference is technically possible in particular cases, according to data science literature (e.g., Fredrikson et al. 2015). Our experts mentioned that this can happen if a model is overfitting. This is particularly important for generative models, where not the original training data is generated, but only similar data. One of our interviewees mentioned here one hypothetical example where this model inference could happen: “[...] Then there is the risk that you are revealing information about your own data with the models. [...] Let’s assume we take two insurance companies. They want to improve fraud detection. They exchange meta-information or train models together to do that. From that, you can get the structure of the data K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 375 used for training. And that underlying structure can already give one insurance company, which of course is a competitor, a lot of information about the other.” (I9*, CEO/Co-Founder Data Science Company B) 4.2.3 Contextual Factors The risk of knowledge leakage from sharing models depends on the context, i.e., if competitive knowledge can be derived from a shared model. Especially in consulting and engineering, preserved domain knowledge from experts is leaked when whitebox models are shared, as one interviewee reported: “If I take these models and give them away, then I’ve taken the knowledge that I’ve discovered from people, from their actions, from their labelling, from their input, preserved it in the model, and sold it to the outside world. That’s a tremendous risk.” (I4*, Manager Data Analytics) This case shows that expert knowledge from employees is materialised into models. As part of an engineering business, models are shared with their customers. Moreover, through sharing the model, the materialised knowledge of their experts could spill over to their customers. One expert from the semiconductor industry (I17) also mentioned a future example in terms of generative AI and transformer models that could explain how to build a technical system (e.g., a microchip). This could be a huge risk if such a model was trained with company-specific data and leaked (e.g., through a collaboration with a start-up). The risk depends on how easily the model can be applied and transferred to other application scenarios, as one manager mentioned: “If it [the model] is very specific to a problem, I’m not afraid. [...] If the model is very generic and easily transferable to different types of problems, to a different data set, to a different context, [...], then we have to be careful.” (I4*, Manager Data Analytics) The interviewer mentions, “I am not afraid” and “we have to be careful”. Both phrases clearly indicate that this is a well-evaluated decision. Beyond abstract transferability, another organisation also needs the capability and knowledge to apply the model. Further, the availability of appropriate data sets where the model can be applied influences the risk, as one consultant mentioned: “Without the raw data, the algorithm is less useful for me. [...] has the other party also the same raw data or other data with similar formats? If yes, then that is a big risk. [...] And the highest risks are in cases in which when the algorithm is leaked, and the raw data is available or reproducible.” (I6*, Manager Data Analytics Consulting) The interviewee points to the strategy of keeping the training data in the back and just sharing the model. This is especially important, as many successful DDBMs rely on unique dynamic data sets generated through using the service (e.g., location K 376 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 data of traffic participants to predict traffic jams). Thus, the model only has value if it is used in combination with this available data. Another influencing factor is the volatility of the model: The risk increases if the model is valid for a longer period. Whereas the risk is lower if a dynamic model is constantly adjusted and updated. Models implicitly contain the knowledge represented by the data used to train the model. Building a model also comprises knowledge of how to create value-added information from raw data, as one consulting manager explained: “[You need] a combination of knowledge of the data scientist who just looks at the raw data, at the graph, very simply speaking, and the engineer who knows exactly how the machine works, who knows exactly what it means when there’s a pressure drop in the hydraulic arm of the welding robot.” (I6*, Manager Data Analytics Consulting) This statement shows that, on the one hand, domain-related knowledge, e.g., from engineering, is needed to train a model. On the other hand, knowledge from the data science discipline is also needed. Domain (expert) knowledge about a real-world phenomenon can add value to the model, such as specific casualties or relationships that cannot be discovered from the data itself but need additional contextual knowledge on the domain. Data science knowledge involves the labelling, preparing, and aggregating of the data and subsequent analytics and algorithmics and their combination. 4.2.4 Knowledge Protection Measures Our interviewees mentioned that traditional legal protection mechanisms for IPR (e.g., patents) do not work for models. As the knowledge is only implicitly contained in the model, a lawsuit to convict the guilty seems very challenging. Therefore, the owner of the know-how and IP should be defined in contracts, e.g., the IP regarding the model creation remains at the provider. Further, our interviewees mentioned defining and identifying what information should be revealed by the model and which not to build the model accordingly and ensure that the model is only used as intended. Models should be designed so that they only disclose the intended minimum amount of information (e.g., only the transfer function without revealing the influencing parameters (e.g., I17)). This, again, requires alignment and balance between sharing and protecting knowledge. The risk also depends on the balance between generated returns and the estimated risk. For instance, one expert mentioned that the monetary value of selling a model would be significantly higher than only sharing predictions, in particular, if the code of the model can be accessed. Thus, the risk can also be mitigated by adjusting the business model,ormoreprecisely,thepricing model. Thus, protecting knowledge in DDBMs is currently mainly performed via technical measures. One simple knowledge protection mechanism is to share only blackbox models: K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 377 “For example, if I share the source code, where I can see every parameter of the [decision] tree, then it’s clear that I’m selling critical knowledge. But in contrast, if I make predictions black box-like, then I would find it difficult to reconstruct the parameters.” (I2*, Data Scientist Automotive Company A) Our interviewees suggest applying data science methods to prevent model inversion attacks, such as randomisation in training, differential privacy, or other anonymization methods. For example, our interviewees mentioned using different loss functions or synthetic data for model training. Another protection mechanism is to keep the model within the organisation’s knowledge boundary and offer the model as a service via a platform or an API. However, the user of the model has to share his data now with the service provider, which could create a knowledge risk for the user. The provider shares only the results. 4.3 Sharing Predictions 4.3.1 The Risk of Reconstructing a Model from Shared Predictions Competitive knowledge might spill over when plenty of predictions are shared, and the receiver can reconstruct the model or parts of the model based on these predictions. According to computer science literature, reconstructing models based on predictions is technically possible in particular cases (e.g., Tramèr et al. 2016). However, such attacks can be mitigated easily by restricting the number of predictions or the value range. Thus, this risk was perceived as low. One way to discover knowledge is to reconstruct the underlying model by provoking lots of predictions. Moreover, the model allows inferences about the materialised knowledge. One academic expert in knowledge protection pointed to the problem: “If you sell many outcomes, yeah, then it would be even then possible to re-engineer the algorithm itself. If you are looking at what kind of results are created by what kind of data.” (R5, Senior Researcher Knowledge Management) However, the interviewee refers to “what kind of data”, and another interviewee, a data scientist, specifies this in more detail: “If you take a look at the predictions now, you’ll probably see a few features and check for which group it’s working better or worse. You’ll be able to reconstruct something there.” (I2*, Data Scientist Automotive Company A) As he says, “to reconstruct something there”, he acknowledges the big challenge of discovering competitive knowledge from a prediction-based value proposition. However, our interviewees perceived the risk of knowledge leakage through sharing predictions as low as, for instance, one interviewee said: K 378 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 “For example, the customer only gets the results back. In that case, I think the risk is very low that any knowledge will drain from the provider because the customer doesn’t have access to that knowledge.” (R8*, Researcher DataDriven Services) This statement shows that the knowledge is hidden and that the customer has no direct access to the model and the materialised knowledge in the model. Further, our experts (e.g., I17) noted that it is not so easy to derive clear conclusions—the inference is subject to probabilities. 4.3.2 Contextual Factors Reconstructing the model from predictions is possible from a theoretical point of view. However, in reality, this is not trivial and requires some prior information on the model available. How easy it is to reconstruct the model also depends on its complexity and the input data variability, as one expert in the field of DDBMs explained: “The heart of a good model is the variance of the input factors. And if I just offer an API, where I only provide a result to certain input values, but the input data that have led to that model has more variety than I’m allowing through the API, I can actually [prevent that well].” (I5*, Director Digital Business) As this quote shows, re-engineering a model based on lots of “results” that we call predictions depends on the variance of the input data if it covers the whole input space. The knowledge is hidden and is materialised in the prediction model itself. The single prediction thus offers only a small and scattered glimpse of the model. Many predictions need to be collected or even provoked in a systematic attack to re-engineer knowledge: “If you send enough different queries, you can already [reconstruct] what knowledge is materialised in the model. Depending on the complexity of the problem, this might be a task at the moment, which do not allow model reengineering due to the complexity.” (I1*, Data Analytics Consultant) This quote shows that reconstructing knowledge is possible but requires significant effort and expertise. If insights about the model are successfully collected, knowledge could be discovered. One mentioned example of knowledge that could be reconstructed is the bias that the model has learned. Further, one must balance the effort if it is worth it for the attacker. 4.3.3 Knowledge Protection Measures When predictions are shared through access to a prediction model, one simple protection measure is to control the access in terms of the number of allowed requests, the minimum time span between two requests, and the range of input values. Limiting the number of requests prevents brute force attacks for reconstruction and also denial of service attacks. Potential attacks could be recognised through atypiK Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 379 cal requests, e.g., uniformly distributed across the input space, as training and reengineering a model requires a broad range of input data. Our interviewed expert continued: “First of all, when someone penetrates me and asks me questions over the entire vector space, then I notice that this is atypical. That would be a uniform distribution in the query, which is totally atypical for such a thing, there you rather have a normal distribution in the queries.” (I5*, Director Digital Business) One protection measure is to build a prediction service that relies on dynamic data, such as real-time vehicle location data, generated through service usage and not shared with other actors. Even if the prediction model could be reconstructed based on many predictions, the knowledge cannot be applied as one malicious actor cannot access the necessary data. Another protection measure lies in the design of the business model: in prediction-as-a-service business models, pay-per-use revenue models are often used, which means that requesting lots of predictions gets expensive, and by that, even if something could be reconstructed from the model, it was compensated monetarily. 5 Discussion of Results 5.1 Discussion of Problem and Risk Relevance In our interviews, we found that knowledge risks are a relevant topic in data-driven business models. For the three types of value objects data, model and prediction, we identified five types of risks that arise when they are exchanged in a DDBM: The risk of leaking competitive knowledge from shared data, the risk of exposing competitive knowledge by using a data service; the risk of leaking competitive knowledge from a shared model; the risk of inference of the underlying training data; and the risk of reconstructing a model from shared predictions. The validation interviews confirmed the five types of risks, i.e., no additional types were suggested or emerged, and the description of the existing ones was sufficient. The risk of exposing competitive knowledge by using a data service was perceived as the most relevant risk in the validation interviews, as one expert brought it to the point: “I think that is the biggest, but also very hard to grasp, threat or fear that the management in the industry has now” (Industry Expert 14*). One problem is that the risk is very difficult to grasp. Therefore, there is sometimes a lot of fear, and as a consequence, companies are very cautious, and DDBMs may not be realised. Knowledge Risks in DDBMs Depend on Contextual Factors of the DDBM Itself The risk depends on the area of the company from which data-related value objects are shared. For instance, if data is shared to optimise an ancillary process (e.g., maintenance of production machines), the risk is perceived as less critical. Whereas, if data from their core process allows inference on their core processes, e.g., the design and configuration of products, the knowledge risk was perceived as critical. Thus, it must always be assessed if the (potential) leaked knowledge is competitive K 380 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 and business-critical. Our data also suggest that knowledge risks are particularly relevant in knowledge-intensive businesses that want to innovate towards DDBMs in addition to their existing business model. In such business models, domain expert knowledge (e.g., engineering) is materialised in models that are shared with customers and partners as part of a DDBM. Thus, competitive knowledge might be put at risk. Further, in business models with complex systems and high competition (e.g., the automotive or semiconductor industry), organisations are very restrictive about data sharing, as corporate secrets might be shared with the data. Knowledge Risks in DDBM Differ from Knowledge Risks Associated with Traditional Business Models As more areas of an organisation are digitised, there is a risk that more competitive knowledge is materialised in (AI) models. These, however, are easy to transfer compared to traditional business models, where engineers from the competition need to be headhunted or a product needs to be reverse engineered. In DDBMs, leaking a model could be sufficient for knowledge leakage. With the spread of generative AI and transformer models, we assume this aspect will become even more important in the upcoming years (cp. Tredinnick and Laybats 2023). Thus, the question of how to protect knowledge and IP in DDBMs will become more important. 5.2 Discussion of Protection Measures—How to Deal with the Risk? We found that knowledge risk in DDBMs can be mitigated by technology (which might be fast changing), by business model design options, and by ensuring transparency, building trust and contractual regulations. As a synthesis of these three areas of action, one major strategic implication of our work is that knowledge risk mitigation in DDBMs needs a differentiated and balanced assessment of whether the perceived risk has a negative economic impact or is acceptable compared to the expected return in the DDBM. 5.2.1 Technology to Mitigate Knowledge Risks A knowledge risk can often be reduced upfront by technology. Computer science literature discusses several technical attacks to retrieve something from data and models (see, e.g., Kaissis et al. 2020). Such attacks encompass training data leakage, model stealing, reverse engineering or membership inference (Hanzlik et al. 2021). Preventing such attacks or exacerbating the knowledge discovery process can be done by technical measures that relate to contemporary computer science research (see, e.g., Kaissis et al. 2020). Privacy-preserving technologies tailored to the context of big data analytics ensure the confidentiality of the data (e.g., Yakoubov et al. 2014). Examples of such privacy-preserving technologies are multiparty computation (e.g., Archer et al. 2018), data anonymization (Zeiringer et al. 2024), homomorphic encryption (e.g., Alabdulatif et al. 2020), watermarking (e.g., Regazzoni et al. 2021) or metaand transfer machine learning (e.g., Hirt and Kühl 2018), which were also mentioned by our experts as technical protection measures. Such technology, like multi-party computation, has already been found to be a proK Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 387 7 Appendix 7.1 Appendix A Table 3 List of Interviewed Experts Round ID Type of Position Industry Duration (min) Language 1 R1 Professor for Digital Business Research 36 EN 1 R2 Professor for Business Model Innovation Research 61 DE 1 R3 Professor Business Analytics Research 35 EN 1 R4 Professor for Data Science Research 56 DE 1 R5 Senior Researcher for Knowledge Management Research 67 EN 1 R6 Professor for Knowledge Management Research 60 DE 1 R7 Professor for Knowledge Management Research 59 EN 1 I1 Consultant Data Analytics Consulting 39 DE 1 I2 Data Scientist Automotive 52 DE 1 I3 CEO/Co-Founder Cyber Security 63 DE 1 I4 Manager Data Analytics Automotive 62 DE 1 I5 Director Digital Business Information Technology 76 DE 1 I6 Manager Data Analytics Consulting 67 DE 1 I7 Manager Digital Business Automotive 48 DE 1 I8 CEO/Co-Founder Data Science 70 DE 1 I9 CEO/Co-Founder Data Science 45 DE 2 R8 Researcher Data-Driven Services Research 50 DE 2 R9 Research Group Leader Data Analytics Research 53 DE 2 R10 Senior Researcher Data-Driven Services Research 38 DE 2 I10 Manager Data Analytics Consulting 55 DE 2 I11 Manager Data Science Data-Driven Service 45 DE 2 I12 Managing Director Data-Driven Service 48 DE 2 I13 Consultant Business Model Innovation Consulting 59 DE 3 I14 Consultant Data-Driven Services Information Technology 59 DE 3 I15 Consultant Data Protection, Artificial Intelligence Consulting 43 DE 3 I16 Founder and Managing Director Consulting 50 DE 3 I17 Manager Data Science Semiconductor 42 DE 3 I18 Managing Director Data-Driven Service 40 DE K 388 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 7.2 Appendix B: Interview Guideline 7.2.1 Guiding Questions Interview Round 1 Problem description Do you see this as a relevant problem? And do you know any similar examples? What other causes of risk could you imagine in this context in data-driven business models? What consequencies do you see based on these risks? Presentation of exemplary consequencies (knowledge loss, knowledge leakage, knowledge spill-over) How do you assess each of these consequences as a possible/relevant problem in data-driven business models? For each, do you know any example? What other consequences could arise from such knowledge risks? What examples from the practice of companies do you known to you where the topic of knowledge risks in data-driven business models is, was or could be relevant? 2nd part (not the scope of this paper): Presentation of a tool and evaluation questions 7.2.2 Guiding Questions Interview Round 2 Problem description How do you assess this problem of knowledge risks just described as a relevant problem in your business model/in general? Can you tell an examples from practice you aware of where the issue of knowledge risks in data-driven business models is, was or could be relevant? What potential mechanisms can you think of to reconstruct or access knowledge in the three types as customer and attacker at the same time? What would be the potential consequences of such knowledge risks for your/an organization? What factors influence the risk of knowledge leakage through the exchange of data, models or predictions? What protection measures have you implemented to avoid or prevent such knowledge risks? 7.2.3 Guiding Questions Interview Round 3 Presentation of problem knowledge risks in DDBMs Presentation of of interim results (main concepts, 5 types of risks, for each a short description and exemplary quotes from the interviews) Do you perceive these risks as relevant for your business? Are there any other types of risks missing in that context? Is the description of each risk reasonable for you? K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 389 7.3 Appendix C Table 4 Coding scheme with main categories after interview round 2 Category Description Motives for sharing This category describes motives why a type of value object is shared with other stakeholders Type of knowledge This category describes different types of knowledge that can be discovered from data-related value objects Knowledge retrieval mechanism This category describes mechanisms of how the knowledge can be discovered from the data-related value object by another party leading to a knowledge leakage Influencing factors This category describes the circumstances that make knowledge retrieval and, thus a, knowledge leakage possible. These factors influence the probability of the risk Knowledge protection measures This category describes measures of how technical or business model design measures could prevent such knowledge leakage K 390 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 7.4 Appendix D Table 5 Snapshot of the coding scheme and exemplary text segments Text segment Code Type of value object Category “Ich mach das Ganze dann als Software-as-a-Service. Das wäre so die beste Mitigation.” (I10) Offer Modelas-a-Service as a protection measure Model Knowledge protection measure “Wenn man das Modell nur als API zur Verfügung stellt, dann kann jemand zwar Anfragen stellen, da kann jemand das Modell aber noch nicht rekonstruieren.” (I9) “... dass man verschlüsselte Daten für so eine Dienstleistung verwendet.” (I1) Using encrypted data Data “Daten sollten auf jeden Fall verschlüsselt übertragen werden.” (I11) K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 391 Acknowledgements The research based on this paper has received funding from the Austrian COMET Program—Competence Centers for Excellent Technologies—under the auspices of the Austrian Federal Ministry of Transport, Innovation and Technology, the Austrian Federal Ministry for Digital and Economic Affairs and by the State of Styria. COMET is managed by the Austrian Research Promotion Agency (FFG). Conflict of interest M. Fruhwirth, V. Pammer-Schindler and S. Thalmann declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4. 0/. References Agahari, Wirawan, Hosea Ofe, and Mark de Reuver. 2022. It is not (only) about privacy: How multi-party computation redefines control, trust, and risk in data sharing. Electronic Markets 32(3):1577–1602. https://doi.org/10.1007/s12525-022-00572-w. Al-Aali, Abdulrahman Y., and David J. Teece. 2013. Towards the (strategic) management of intellectual property: retrospective and prospective. California Management Review 55(4):15–30. https://doi.org/ 10.1525/cmr.2013.55.4.15. Alabdulatif, Abdulatif, Ibrahim Khalil, and Xun Yi. 2020. Towards secure big data analytic for cloudenabled applications with fully homomorphic encryption. Journal of Parallel and Distributed Computing 137:192–204. https://doi.org/10.1016/j.jpdc.2019.10.008. Archer, David W., Dan Bogdanov, Yehuda Lindell, Liina Kamm, Kurt Nielsen, Jakob I. Pagter, Nigel P. Smart, and Rebecca N. Wright. 2018. From keys to databases—real-world applications of secure multi-party computation. The Computer Journal 61(12):1749–1771. https://doi.org/10.1093/comjnl/ bxy090. Azkan, Can, Frederik Moller, Lennart Iggena, and Boris Otto. 2022. Design principles for industrial datadriven services. IEEE Transactions on Engineering Management 71:2379-2402 https://doi.org/10. 1109/TEM.2022.3167737. Baird, Aaron, and Likoebe M. Maruping. 2021. The next generation of research on IS use: a theoretical framework of delegation to and from agentic IS artifacts. Management Information Systems Quarterly 45(1):315–341. https://doi.org/10.25300/MISQ/2021/15882. Brillinger, Anne-Sophie. 2018. Mapping business model risk factors. International Journal of Innovation Management 22(05):1840005. https://doi.org/10.1142/S1363919618400054. Brillinger, Anne-Sophie, Christian Els, Björn Schäfer, and Beate Bender. 2020. Business model risk and uncertainty factors: toward building and maintaining profitable and sustainable business models. Business Horizons 63(1):121–130. https://doi.org/10.1016/j.bushor.2019.09.009. Casadesus-Masanell, Ramon, and Joan E. Ricart. 2010. From strategy to business models and onto tactics. Long Range Planning 43(2–3):195–215. https://doi.org/10.1016/j.lrp.2010.01.004. Chen, Ying, Jeffrey Kreulen, Murray Campbell, and Carl Abrams. 2011. Analytics ecosystem transformation: a force for business model innovation. 2011 Annual SRII Global Conference., 11–20. https:// doi.org/10.1109/SRII.2011.12. Coombs, Crispin, Donald Hislop, Stanimira K. Taneva, and Sarah Barnard. 2020. The strategic impacts of intelligent automation for knowledge and service work: an interdisciplinary review. The Journal of Strategic Information Systems 29(4):101600. https://doi.org/10.1016/j.jsis.2020.101600. Dehnert, Maik, Alexander Gleiss, and Reiss Frederik. 2021. What makes a data-driven business model? A consolidated taxonomy. Proceedings of the Twenty-Ninth European Conference on Information Systems (ECIS 2021). Dorfer, Laura. 2016. Datenzentrische Geschäftsmodelle als neuer Geschäftsmodelltypus in der ElectronicBusiness-Forschung: Konzeptionelle Bezugspunkte, Klassifikation und Geschäftsmodellarchitektur. K 392 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung 68(3):307–369. https://doi.org/10. 1007/s41471-016-0014-9. Durst, Susanne, and Malgorzata Zieba. 2017. Knowledge risks—towards a taxonomy. International Journal of Business Environment 9(1):51–63. https://doi.org/10.1504/IJBE.2017.084705. Durst, Susanne, and Malgorzata Zieba. 2018. Mapping knowledge risks: towards a better understanding of knowledge management. Knowledge Management Research & Practice 17(1):1–13. https://doi.org/ 10.1080/14778238.2018.1538603. Etikan, Ilker. 2016. Comparison of convenience sampling and purposive sampling. American Journal of Theoretical and Applied Statistics 5(1):1–4. https://doi.org/10.11648/j.ajtas.20160501.11. Farayola, Oluwatoyin A., A. Abdul Adekunle, Blessing O. Irabor, and Evelyn C. Okeleke. 2023. Innovative business models driven by AI technologies: a review. Computer Science & IT Research Journal 4(2):85–110. https://doi.org/10.51594/csitrj.v4i2.608. Fassnacht, Marcel, Carina Benz, Daniel Heinz, Jasmin Leimstoll, and Gerhard Satzger. 2023. Barriers to data sharing among private sector organizations. In Proceedings of the 56th Annual Hawaii International Conference on System Sciences, January 3–6, 2023., ed. Tung X. Bui, 3695–3704. Honolulu: Department of IT Management Shidler College of Business University of Hawaii. Franke, Günter. 2020. Management nicht-finanzieller Risiken: eine Forschungsagenda. Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung 72:279-320. https://doi.org/10.1007/s41471-02000096-z. Fredrikson, Matt, Somesh Jha, and Thomas Ristenpart. 2015. Model inversion attacks that exploit confidence information and basic countermeasures. CCS ’15: Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security., 1322–1333. https://doi.org/10.1145/ 2810103.2813677. Fruhwirth, Michael, Viktoria Pammer-Schindler, and Stefan Thalmann. 2019. To Sell or Not to Sell: Knowledge Risks in Data-Driven Business Models. In Proceedings of the 2019 Pre-ICIS SIGDSA Symposium. 11. Fruhwirth, Michael, Christiana Ropposch, and Viktoria Pammer-Schindler. 2020. Supporting data-driven business model innovations: a structured literature review on tools and methods. Journal of Business Models 8(1):7–25. Fruhwirth, Michael, Viktoria Pammer-Schindler, and Stefan Thalmann. 2021. A Network-based Tool for Identifying Knowledge Risks in Data-Driven Business Models. In Proceedings of the 54th Hawaii International Conference on System Sciences, ed. Tung X. Bui, 5218-5227. Gelhaar, Joshua, and Boris Otto. 2020. Challenges in the emergence of data ecosystems. Twenty-Third Pacific Asia Conference on Information Systems. UAE, 2020, Dubai. Gieß, Anna, Frederik Möller, Thorsten Schoormann, and Boris Otto. 2023. Design options for data spaces. Thirty-first European Conference on Information Systems (ECIS 2023). Girotra, Karan, and Serguei Netessine. 2011. How to build risk into your business model. Harvard Business Review 89(5):100–105. Gordijn, Jaap, and Hans Akkermans. 2001. Designing and evaluating e-business models. IEEE Intelligent Systems 16(4):11–17. https://doi.org/10.1109/5254.941353. Gordijn, Jaap, and J.M. Akkermans. 2003. Value-based requirements engineering: exploring innovative e-commerce ideas. Requirements Engineering 8(2):114–134. https://doi.org/10.1007/s00766-0030169-x. Guggenberger, Moritz T., Frederik Möller, Karim Boualouch, and Boris Otto. 2020. Towards a unifying understanding of digital business models. Twenty-Third Pacific Asia Conference on Information Systems, UAE, 2020, Dubai. Guggenmos, Florian, Björn Häckel, Philipp Ollig, and Bastian Stahl. 2022. Security first, security by design, or security pragmatism—strategic roles of IT security in digitalization projects. Computers & Security 118:102747. https://doi.org/10.1016/j.cose.2022.102747. Günther, Wendy A., Mohammad H. Rezazade Mehrizi, Marleen Huysman, and Frans Feldberg. 2017. Debating big data: a literature review on realizing value from big data. The Journal of Strategic Information Systems 26(3):191–209. https://doi.org/10.1016/j.jsis.2017.07.003. Hallikas, Jukka, Iris Karvonen, Urho Pulkkinen, Veli-Matti Virolainen, and Markku Tuominen. 2004. Risk management processes in supplier networks. International Journal of Production Economics 90(1):47–58. https://doi.org/10.1016/j.ijpe.2004.02.007. Hanzlik, Lucjan, Yang Zhang, Kathrin Grosse, Ahmed Salem, Maxmilian Augustin, Michael Backes, and Mario Fritz. 2021. MLCapsule: guarded Offline deployment of machine learning as a service.Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 3300–3309. K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 393 Hartmann, Philipp M., Mohamed Zaki, Niels Feldmann, and Andy Neely. 2016. Capturing value from big data—a taxonomy of data-driven business models used by start-up firms. International Journal of Operations & Production Management 36(10):1382–1406. https://doi.org/10.1108/IJOPM-02-20140098. Hernandez, Exequie, G. Sanders, and Anja Tuschke. 2015. Network defense: pruning, grafting, and closing to prevent leakage of strategic knowledge to rivals. Academy of Management Journal 58(4):1233–1260. https://doi.org/10.5465/amj.2012.0773. Hirt, Robin, and Niklas Kühl. 2018. Cognition in the era of smart service systems: inter-organizational analytics through meta and transfer learning. In Proceedings of the 39th International Conference on Information Systems—Bridging the Internet of People, Data, and Things. Francisco., ed. Jan PriesHeje, Sudha Ram, and Michael Rosemann Hunke, Fabian, Christian Engel, Ronny Schüritz, and Philipp Ebel. 2019. Understanding the anatomy of analytics-based services: a taxonomy to conceptualize the use of data and Analytics in service. In Proceedings of the 27th European Conference on Information Systems—Information Systems for a Sharing Society. Stockholm Uppsala., ed. Jan Vom Brocke, Shirley Gregor, and Oliver Müller Ilvonen, Ilona, Stefan Thalmann, Markus Manhart, and Christian Sillaber. 2018. Reconciling digital transformation and knowledge protection: a research agenda. Knowledge Management Research & Practice 16(2):235–244. https://doi.org/10.1080/14778238.2018.1445427. Jennex, Murray E., and Suzanne Zyngier. 2007. Security as a contributor to knowledge management success. Information Systems Frontiers 9(5):493–504. https://doi.org/10.1007/s10796-007-9053-4. Jiang, Xu, Bao Yongchuan, Yan Xie, and Shanxing Gao. 2016. Partner trustworthiness, knowledge flow in strategic alliances, and firm competitiveness: a contingency perspective. Journal of Business Research 69(2):804–814. https://doi.org/10.1016/j.jbusres.2015.07.009. Kaiser, Rene, Stefan Thalmann, and Viktoria Pammer-Schindler. 2021. An investigation of knowledge protection practices in inter-organisational collaboration: protecting specialised engineering knowledge with a practice based on grey-box modelling. VINE Journal of Information and Knowledge Management Systems 51(5):713–731. https://doi.org/10.1108/VJIKMS-11-2019-0180. Kaissis, Georgios A., Marcus R. Makowski, Daniel Rückert, and Rickmer F. Braren. 2020. Secure, privacy-preserving and federated machine learning in medical imaging. Nature Machine Intelligence 2(6):305–311. https://doi.org/10.1038/s42256-020-0186-1. Kale, Prashant, Harbir Singh, and Howard Perlmutter. 2000. Learning and protection of proprietary assets in strategic alliances: building relational capital. Strategic Management Journal 21(3):217–237. Kanbach, Dominik K., Louisa Heiduk, Georg Blueher, Maximilian Schreiter, and Alexander Lahmann. 2023. The GenAI is out of the bottle: generative artificial intelligence from a business model innovation perspective. Review of Managerial Science https://doi.org/10.1007/s11846-023-00696-z. Khan, Freeha, Jung H. Kim, Lars Mathiassen, and Robin Moore. 2021. Data breach management: an integrated risk model. Information & Management 58(1):103392. https://doi.org/10.1016/j.im.2020. 103392. Kühne, Babett, and Tilo Böhmann. 2019. Data-driven business models: building the bridge between data and value. In Proceedings of the 27th European Conference on Information Systems—Information Systems for a Sharing Society. Stockholm Uppsala., ed. Jan Vom Brocke, Shirley Gregor, and Oliver Müller Leski, Florian, Michael Fruhwirth, and Viktoria Pammer-Schindler. 2021. Who Else do you need for a data-driven business model? Exploring roles and exchanged values. In 34th bled econference digital support from crisis to progressive change. June 27–30, 2021., ed. Andreja Pucihar, Mirjana Kljaji´ c Borštnar, Roger Bons, Helen Cripps, Anand Sheombar, and Doroteja Vidmar, 365–378. Loebbecke, Claudia, Paul C. van Fenema, and Philip Powell. 2016. Managing inter-organizational knowledge sharing. The Journal of Strategic Information Systems 25(1):4–14. https://doi.org/10.1016/j.jsis. 2015.12.002. Manhart, Markus, and Stefan Thalmann. 2015. Protecting organizational knowledge: a structured literature review. Journal of Knowledge Management 19(2):190–211. https://doi.org/10.1108/JKM-05-20140198. Mayring, Philipp. 2015. Qualitative Inhaltsanalyse: Grundlagen und Techniken, 12th edn., Weinheim: Beltz. Miorandi, Daniele, Sabrina Sicari, Francesco de Pellegrini, and Imrich Chlamtac. 2012. Internet of things: Vision, applications and research challenges. Ad Hoc Networks 10(7):1497–1516. https://doi.org/10. 1016/j.adhoc.2012.02.016. Möller, Frederik, Maleen Stachon, Christina Hoffmann, Henrik Bauhaus, and Boris Otto. 2020. Datadriven business models in logistics: a taxonomy of optimization and visibility services. In ProceedK 394 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 ings of the 53rd Annual Hawaii International Conference on System Sciences (HICSS2020), ed. Tung Bui, 5379–5388. Murray, Alex, Jen Rhymer, and David G. Sirmon. 2021. Humans and technology: forms of conjoined agency in organizations. Academy of Management Review 46(3):552–571. https://doi.org/10.5465/ amr.2019.0186. Oh, Seong J., Max Augustin, Bernt Schiele, and Mario Fritz. 2019. Towards reverse-engineering blackbox neural networks. In Explainable AI: Interpreting, Explaining and Visualizing Deep Learning,ed. Wojciech Samek, Grégoire Montavon, Andrea Vedaldi, Lars Kai Hansen, and Klaus-Robert Müller, 121–144. Cham: Springer. Opriel, Sebastian, Frederik Möller, Ute Burkhardt, and Boris Otto. 2021. Requirements for usage control based exchange of sensitive data in automotive supply chains. In Proceedings of the 54th Annual Hawaii International Conference on System Sciences, ed. Tung Bui, 431–440. Osterwalder, Alexander, and Yves Pigneur. 2010. Business model generation: a handbook for visionaries, game changers, and challengers, 1st edn., Hoboken: Wiley. Osterwalder, Alexander, Yves Pigneur, and Christopher L. Tucci. 2005. Clarifying business models: origins, present, and future of the concept. Communications of the Association for Information Systems https://doi.org/10.17705/1CAIS.01601. Perrott, Bruce E. 2007. A strategic risk approach to knowledge management. Business Horizons 50(6):523–533. https://doi.org/10.1016/j.bushor.2007.08.002. Rashed, Faisal, Paul Drews, and Mohamed Zaki. 2022. A reference model for data-driven business model innovation initiatives in incumbent firms. Proceedings of the Thirtieth European Conference on Information Systems (ECIS 2022), Timi, soara. Regazzoni, Francesco, Paolo Palmieri, Fethulah Smailbegovic, Rosario Cammarota, and Ilia Polian. 2021. Protecting artificial intelligence IPs: a survey of watermarking and fingerprinting for machine learning. CAAI Transactions on Intelligence Technology 6(2):180–191. https://doi.org/10.1049/cit2.12029. Santhosh, Gautham, Fabrizio de Vita, Dario Bruneo, Francesco Longo, and Antonio Puliafito. 2019. Towards trustless prediction-as-a-service. 2019 IEEE International Conference on Smart Computing (SMARTCOMP)., 317–322. https://doi.org/10.1109/SMARTCOMP.2019.00068. Schäfer, Fabian, Heiko Gebauer, Christoph Gröger, Oliver Gassmann, and Felix Wortmann. 2023a. Datadriven business and data privacy: Challenges and measures for product-based companies. Business Horizons 66(4):493–504. https://doi.org/10.1016/j.bushor.2022.10.002. Schäfer, Fabian, Jeremy Rosen, Christian Zimmermann, and Felix Wortmann. 2023b. Unleashing the potential of data ecosystems: establishing digital trust through trust-enhancing technologies. Thirtyfirst European Conference on Information Systems (ECIS 2023). Schüritz, Ronny, Stefan Seebacher, and Rebecca Dorner. 2017a. Capturing value from data: revenue models for data-driven services. In Proceedings of the 50th Hawaii International Conference on System Sciences. Waikoloa Village., ed. Tung Bui, 5348–5357. Schüritz, Ronny, Stefan Seebacher, Gerhard Satzger, and Lukas Schwarz. 2017b. Datatization as the next frontier of Servitization: understanding the challenges for transforming organizations. Proceedings of the Thirty-Eighth International Conference on Information Systems (ICIS), Seoul. Schüritz, Ronny, Killian Farrell, Barbara H. Wixom, and Gerhard Satzger. 2019. Value co-creation in data-driven services: towards a deeper understanding of the joint sphere. Proceedings of the Fortieth International Conference on Information Systems (ICIS), Munich. Schweihoff, Julia, Ilka Jussen, Valentin Dahms, Frederik Möller, and Boris Otto. 2023. How to share data Online (fast)—A taxonomy of data sharing business models. Proceedings of the 56th Hawaii International Conference on Systems Sciences (HICSS). Shollo, Arisa, Konstantin Hopf, Tiemo Thiess, and Oliver Müller. 2022. Shifting ML value creation mechanisms: a process model of ML value creation. The Journal of Strategic Information Systems 31(3):101734. https://doi.org/10.1016/j.jsis.2022.101734. Stachon, Maleen, Frederik Möller, Moritz T. Guggenberger, and Martin Tomczyk. 2023. Understanding data trusts. Proceedings of the Thirty-first European Conference on Information Systems (ECIS 2023), Kristiansand. Stahl, Bastian, Björn Häckel, Daniel Leuthe, and Christian Ritter. 2023. Data or business first?—manufacturers’ transformation toward data-driven business models. Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung 75:303–343. https://doi.org/10.1007/s41471-023-00154-2. Strobel, Gero, Frederik Möller, Thorsten Schoormann, and Boris Otto. 2024. Introduction to the 2nd MiniTrack on Designing Data Ecosystems: Values, Impacts, and Fundamentals. In Proceedings of the 57th Annual Hawaii Interna8onal Conference on System Sciences, ed. Tung X. Bui, 4236–4237. K Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 395 Sun, Tianxiang, Shao Yunfan, Huang Xuanjing, and Xipeng Qiu. 2022. Black-Box Tuning for Language-Model-as-a-Service. Proceedings of the 39th International Conference on Machine Learning, 20841–20855. Teece, David J. 2010. Business models, business strategy and innovation. Long Range Planning 43(2–3):172–194. https://doi.org/10.1016/j.lrp.2009.07.003. Tesch, Jan, Anne-Sophie Brillinger, and Dominik Bilgeri. 2017. Internet of things business model innovation and the stage-gate process: an exploratory analysis. International Journal of Innovation Management 21(5):1740002-1–1740002-17. https://doi.org/10.1142/S1363919617400023. Thalmann, Stefan, Ronald Maier, Ulrich Remus, and Markus Manhart. 2024. Connect with care: informal knowledge protection practices to enhance knowledge sharing in networks of organizations. VINE Journal of Information and Knowledge Management Systems, Vol. ahead-of-print No. ahead-of-print. https://doi.org/10.1108/VJIKMS-02-2022-0051 Thiel, Christian, and Christoph Thiel. 2015. Hare and tortoise: can industry 4.0 win the race against counterfeiting and piracy? Datenschutz und Datensicherheit 1(0):663–667. Thomas, Llewellyn D.W., Aija Leiponen, and Pantelis Koutroumpos. 2023. Profiting from data products. In Research handbook on digital strategy, ed. Carmelo Cennamo, Giovanni Battista Feng Zhu Dagnino, 255–272. Cheltenham: Edward Elgar Publishing. Tramèr, Florian, Fan Zhang, Ari Juels, Michael K. Reiter, and Thomas Ristenpart. 2016. Stealing machine learning models via prediction APis. Proceedings of the 25th USENIX Security Symposium, 601–618. Tredinnick, Luke, and Claire Laybats. 2023. The dangers of generative artificial intelligence. Business Information Review 40(2):46–48. https://doi.org/10.1177/0266382123118375. Vesselkov, Alexandr, Heikki Hämmäinen, and Juuso Töyli. 2019. Design and governance of mhealth data sharing. Communications of the Association for Information Systems 45(1):299–321. https://doi.org/ 10.17705/1CAIS.04518. Vetter, Oliver A., Felix S. Hoffmann, Luisa Pumplun, and Peter Buxmann. 2022. What constitutes a machine-learning-driven business model? A taxonomy of B2B start-ups with machine learning at their core. Proceedings of the Thirtieth European Conference on Information Systems (ECIS 2022), Timi , soara. Weber, Michael, Moritz Beutter, Jörg Weking, Markus Böhm, and Helmut Krcmar. 2022. AI startup business models. Business & Information Systems Engineering 64(1):91–109. https://doi.org/10.1007/ s12599-021-00732-w. Wiener, Martin, Carol Saunders, and Marco Marabelli. 2020. Big-data business models: a critical literature review and multiperspective research framework: a critical literature review and multi-perspective research framework. Journal of Information Technology 35(1):66–91. https://doi.org/10.1177/ 0268396219896811. Woerner, Stephanie L., and Barbara H. Wixom. 2015. Big data: extending the business strategy toolbox. Journal of Information Technology 30(1):60–62. https://doi.org/10.1057/jit.2014.31. Yakoubov, Sophia, Vijay Gadepally, Nabil Schear, Emily Shen, and Arkady Yerukhimovich. 2014. Asurvey of cryptographic approaches to securing big-data analytics in the cloud. 2014 IEEE High Performance Extreme Computing Conference (HPEC), 1–6. https://doi.org/10.1109/HPEC.2014.7040943. Yin, Robert K. 2009. Case study research: design and methods, 4th edn., Los Angeles: SAGE. Zeiringer, Johannes P. 2021. Tackling knowledge risks in data-centric collaborations: a tackling knowledge risks in data-centric collaborations: a literature review literature review. PACIS 2021 Proceedings. Zeiringer, Johannes P., and Stefan Thalmann. 2020. Knowledge risks in digital supply chains: a literature review. In WI2020 Zentrale Tracks: 15th International Conference on Wirtschaftsinformatik. Potsdam, March 9–11, 2020., ed. Norbert Gronau, Moreen Heine, K. Poustcchi, and H. Krasnova, 370–385. GITO Verlag. 15. Internationale Tagung Wirtschaftsinformatik. Zeiringer, Johannes P., and Stefan Thalmann. 2022. Knowledge sharing and protection in data-centric collaborations: an exploratory study. Knowledge Management Research & Practice 20(3):436–448. https://doi.org/10.1080/14778238.2021.1978886. Zeiringer, Johannes P., Stefan Thalmann, and Jürgen Fleiss. 2024. Data anonymization as instrument to manage knowledge risks in supply chains. In Proceedings of the 57th Annual Hawaii International Conference on System Sciences, ed. Tung X. Bui, 5503–5512. Zeng, Yong, Wang Lingyu, Deng Xiaoguang, Cao Xinlin, and Nafisa Khundker. 2012. Secure collaboration in global design and supply chain environment: Problem analysis and literature review. Computers in Industry 63(6):545–556. https://doi.org/10.1016/j.compind.2012.05.001. K 396 Schmalenbachs Zeitschrift für betriebswirtschaftliche Forschung (2024) 76:357–396 Zhang, Da Yong, Cao Xinlin, Wang Lingyu, and Yong Zeng. 2012. Mitigating the risk of information leakage in a two-level supply chain through optimal supplier selection. Journal of Intelligent Manufacturing 23(4):1351–1364. https://doi.org/10.1007/s10845-011-0527-3. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. K