Designing blockchain systems to prevent counterfeiting in wine supply chains: a multiple-case study

2.1 Designing anti-counterfeiting measures: theoretical underpinnings

In the literature, various definitions and nuances of product counterfeiting can be found, such as adulteration, tampering and simulation (Spink et al., 2013). Counterfeiting also includes cases where fraudsters do not follow the regulatory definitions of good manufacturing practices (GMPs), good agricultural practices (GAPs) or good hygiene practices (GHPs). According to the World Trade Organization (WTO, 2020), counterfeiting is characterized as the misrepresentation of the identity or the source of products to deceive the purchaser into believing that he/she is buying the original goods. While the literature provides a wide variety of measures to mitigate counterfeiting (see Section 2.3), there is little theoretical development of the “chemistry” of the counterfeiting phenomenon, with the exception of the situational crime prevention approach (Spink et al., 2013). According to this theory, it is by restricting the offender's opportunities for infringement (i.e. addressing the “fraud opportunity structure”) that an effective prevention of counterfeiting can be fulfilled. Thus, anti-counterfeiting measures should be designed by considering to what extent the decision-maker intends to suppress the impulse to offend. The effectiveness of measures to prevent counterfeiting can be evaluated by considering that crime is the result of a choice: more effective measures are those that deter or detect intentional infringements because offending appears less feasible or attractive (for instance, because it is easier to intercept a counterfeited product or to identify the counterfeiter). Basically, the situational crime prevention approach suggests two partially overlapping ways of restricting the offender's opportunities: (1) physically reducing opportunities (e.g. steering wheel locks are a good visual deterrent for car thieves and home security systems discourage burglaries) and (2) increasing the chances of being identified (e.g. the obligation to wear crash helmets significantly reduced thefts of motor-cycles, as thieves could be easily spotted by the police).

Situational crime prevention also establishes that the understanding of the fraud opportunity structure requires a wide-ranging awareness of the vulnerabilities of the various supply network processes, ideally including sourcing, production, distribution, retail and, in extreme situations, final disposal or reuse of packaging material (Speier et al., 2011; Spink et al., 2013). Since the fraud opportunity structure is significantly different in the steps before vs those after the producer sells the product, in this paper, we distinguish between “upstream counterfeiting”, which concerns sourcing and production activities, and “downstream counterfeiting”, which concerns distribution and retail activities. Non-compliance with compulsory international or national production specifications and with product and production process characteristics declared by the company (e.g. use of vulnerable workers, failure to employ of bio, vegan, eco-friendly agricultural or production practices) are forms of upstream counterfeiting. Some practical cases are the production of drugs with no or few active ingredients or the use of substandard aftermarket parts for automobiles or aircrafts (Hopkins et al., 2003). In downstream counterfeiting, the fraud opportunity structure is different. For instance, in the wine industry, counterfeiting can emerge after the wine company sells a bottle and takes the form of fake labels, bottle relabelling or refilling with cheaper wines. The recent large-scale wine scam uncovered by French anti-fraud bodies, in which a merchant tried to sell more than 48 million litres of wine falsely labelled as Côtes-du-Rhône, is an example of downstream counterfeiting (Vinex, 2020).

The situational crime prevention perspective was applied to our research setting where the decision-makers are brand owners who use BC as a measure that deters or detects intentional infringements and also as a traceability and communication tool that allows customers to autonomously verify that the products they buy are not affected by upstream and/or downstream counterfeiting. In this setting, the prevention of counterfeiting is pursued by deterring upstream supply network members from non-adherence to origin, production or quality control specifications (both compulsory and declared) and deterring downstream members from manipulating original goods during distribution and retail. Thus, in this paper, we assume that the desired level of upstream or downstream counterfeiting protection the brand owner intends to guarantee to customers through BC, by setting precise requirements in terms of restriction of the offender's fraud opportunity structure, is the key driver in the design of their BC systems.

2.2 Measures to prevent product counterfeiting

We identify three broad categories of approaches that companies use to mitigate product counterfeiting: product/packaging-related, customer information/education-related and process-related measures. Table 1 offers a short description of each measure together with its weaknesses and classifies each of these on the basis of its suitability to address upstream and/or downstream counterfeiting.

Like track-and-trace systems, BC can be considered a measure to prevent both upstream and downstream counterfeiting (Schmidt and Wagner, 2019). Unlike the other measures and in common with traditional track-and-trace systems, it allows customers to autonomously verify the authenticity of each product without the need to involve specific equipment/competencies or perform destructive chemical/physical/organoleptic analyses. The literature emphasizes BC's capability to overcome the limits of traditional track-and-trace systems (e.g. risk of data manipulation or the poor quality of input data) thanks to its inherent features, such as transparency and decentralized and immutable data storage (Alzahrani and Bulusu, 2019; Saberi et al., 2019). However, since companies adopt different BC feeding and reading options (Section 2.3), a precise and objective comparison between BC and existing counterfeiting approaches, in particular track-and-trace systems, would require a clearer understanding of the different BC system configurations and their advantages compared to traditional track-and-trace systems.

2.3 Blockchain systems

Although most studies recognise that BC's potential for preventing counterfeiting depends on its technical characteristics (immutability, transparency, etc.) (Wang et al., 2019), a growing number of authors are discussing different BC feeding and reading options to complement BC technology and their implications for counterfeiting. A major issue concern is ensuring the quality of the input data. According to Creydt and Fischer (2019) and Bumblauskas et al. (2020), there is currently no universal solution for BC feeding. The potential of BC technology for preventing product counterfeiting can be improved by additional security measures and associated technologies that protect against the risk of human error and potentially dishonest behaviours, thus ensuring that the data that are entered into the BC are of a higher quality (van Hoek, 2019). Drawing on the assumption that information captured by humans can be subjective and unreliable, Roeck et al. (2020) suggest pairing BC with AI to identify patterns across manually entered data that may indicate anomalies or inconsistencies. The integration of BC and AI is recommended by several SCM scholars, although the empirical evidence for a combination of these technologies is still limited to a few early use cases (Kshetri, 2018; van Hoek, 2019; Wamba and Queiroz, 2020). Other studies consider the involvement of neutral third-party certification bodies to carry out field audits and verify data quality before they are manually saved (Creydt and Fischer, 2019). Commenting on this feeding measure, Kamilaris et al. (2019) argue that the need for such intermediaries may compromise the building of decentralized trust for BC technology. An increasing number of studies suggest the use of IoT sensors to automatically enter data to ensure that the feeding process remains objective by eliminating human interaction and the risk of input errors or fraudulent behaviours (Creydt and Fischer, 2019; Kamble et al., 2019; Bumblauskas, 2020). Moreover, the IoT would eliminate the delay between data collection and data recording, which is an issue for manual entry (Zelbst et al., 2019). In this last case, periodical data transmission after data collection is a common practice to avoid network congestion and higher transaction costs (Zhang et al., 2020).

A further set of options and decisions is related to the security of smart labels that connect physical products to the related information saved on the BC (i.e. BC reading). Lo et al. (2019) point out that commonly used smart labels – such as barcodes, QR codes and serial numbers – are vulnerable to attacks, e.g. cloning. In their study, Azzi et al. (2019) argue that RFID tags can also be easily cloned, despite their more technologically advanced nature. This view is supported by Lo et al. (2019), who maintain that cloned tags contribute to the circulation of counterfeited products in supply chains. Conversely, as outlined by Alzahrani and Bulusu (2018), NFC labels provide greater guarantees of tag cloning and tampering. The security of NFC tags against tampering and replication can be improved using data encryption, although it is an expensive measure (Alzahrani and Bulusu, 2019).

A final critical point related to BC reading, although under-investigated, is that in real cases BC systems usually treat the BC as a backend database behind a centralized web server (Singhal et al., 2018). Although an ideal application that prevents potential manipulations should make information visible to final customers directly from the BC hence ensuring complete decentralization, information displayed through user-friendly web pages or mobile applications is usually stored on external servers that should ideally reflect what is written on the BC. To let customers verify the alignment between the communicated information and that saved on the BC, companies associate this information with the related BC transaction link where the information was originally saved (Montecchi et al., 2019). As clarified by Xu et al. (2019), large files (e.g. photos, videos or pdfs) are not stored on the BC due to the technology's limited capacity for handling a large quantity of data; therefore, it is usually preferred to store hash values, namely fixed-length alphanumerical strings unequivocally associated with each document by a secure encryption algorithm. The hash value allows verifying whether the information displayed on the web pages or mobile applications has been manipulated because any change in off-chain data would result in a different hash value from that saved on the BC.

Table 2 summarizes the different options that, according to the literature, could be chosen for BC feeding/reading and some open issues. Although use cases and BC studies contemplate the existence of different configuration choices for BC feeding/reading, they all lack a holistic and in-depth understanding of how a brand owner can design a BC system to assure a certain degree of security in preventing counterfeiting. In particular, there is uncertainty about which variables are relevant to the design of feeding and reading processes and how such variables can be modulated in accordance with the desired counterfeiting protection level. To address these gaps, the present research applies the concept of restriction of the fraud opportunity structure to the field of BC design to unveil the relationships between the degree of protection against counterfeiting a brand owner intends to guarantee to customers and the feeding/reading choices in the design of BC systems.

3.1 Case selection

We selected cases based on literal and theoretical replication (Yin, 2017). We deliberately searched for companies that showed differences and similarities in terms of the adopted BC systems (e.g. type of smart label, control measures over data, etc.). Furthermore, we attempted to include companies in our sample that adopted BC systems developed by different technological partners currently offering BC systems for the wine sector to present a broader picture of the phenomenon and facilitate the generalization of the results. To identify the cases, we first compiled a list of all the BC systems offered by the technology providers in the wine sector in Italy and of the winemakers involved in the implementation of these solutions. This was made possible by Internet searches and our attendance at several conferences and workshops related to BC technology. Second, we collected publicly available data and information on each BC system to decide whether it might represent an interesting case for the study.

We contacted all technology providers to present the area and purpose of the study, the research team and a brief outline of the interview protocol, and then ascertained their willingness to participate in the study. Subsequently, we identified some winemakers that were exemplar cases in the adoption of BC systems and contacted the CEOs of these winemakers via email, again presenting the team and research and ascertaining their willingness to collaborate. We also promised to give both the technology providers and winemakers the final study report and guaranteed data confidentiality. In the end, all technology providers and five Italian winemakers – pioneers in the adoption of BC technology in the wine sector – agreed to participate in our study. We will refer to the cases and related winemakers with labels from A to E.

Ideally, the unit of analysis for the study would be the supply network involved in the BC system. However, like the majority of BC-based projects at the time of this writing, those selected are in a pilot phase and therefore involve a limited number of actors, as shown in Table 3. In particular, companies C, D and E are fully integrated winemakers, as they are responsible for all production activities from grape harvesting to bottling. Companies A and B rely on external grape suppliers.

3.2 Data collection

Data collection took place between October 2018 and May 2019. Semi-structured interviews were chosen as the primary data source because they represent a highly efficient way to gather rich empirical data (Eisenhardt and Graebner, 2007).

We collected information from multiple respondents (Table 3) to capture data from different perspectives, reduce respondents' biases and reach data and theoretical saturation (Bowen, 2008). We decided to interview the winemakers' CEOs. Even if they were not BC experts, they were all directly involved in the pilot projects and were best informed about issues, such as the context, the reasons leading to BC adoption and the expected results. As often happens (even in the more structured wine companies), the winemakers' CEOs also played operational roles in the viticulture and vinification processes as agronomists or oenologists and were also involved in promotion and sales. Then, for each case, we interviewed the project manager of the technological partner to ask for more specific details and empirical evidence to clarify the technical aspects of the BC systems. With the same aim, we also interviewed the IT manager of company A; this role was not present in the other companies. It is worth noting that, in the cases investigated, the decision to adopt a BC system came from the winemaker who, with the support of the technology provider, promoted and guided the BC project. Although they were not directly involved in the decisions about the design of the BC system, in cases A and B, where the winemakers were not vertically integrated upstream, we also interviewed the CEOs of two grapes suppliers involved in the BC project. As is typical in the wine sector, they were vignerons, owners of the vineyards and agronomists expert in grape cultivation. They, therefore, represent key figures in the wine production processes, working closely with winemakers and holding crucial information on cultivation that might be of interest to customers. In case B, given the uniformity in the production activities and in the BC use, we had sufficient information to limit our interviews to one supplier.

We mainly collected data through face-to-face, on-site interviews. For logistical reasons, as suggested by Creswell (2013), we used video conferences with actors located in southern Italy (four interviewees). In total, considering all the respondents of firms and technology providers, 20 interviews were conducted for all cases.

To improve accuracy, each interview was first recorded and then transcribed. Field notes were taken by two researchers to reduce observer's bias. To enhance reliability and validity, an interview protocol was developed based on the existing literature and our research question (Yin, 2017) (see Supplementary File 1). Semi-structured interviews allowed us to create a balance between an open discussion characterized by mutual reflection and knowledge sharing and a more specific questionnaire-based conversation with the aim of finding evidence for research purposes. While all respondents were asked all the questions in the protocol, the level of detail of the discussion on the different points and the focus varied depending on the respondent (Pandey and Patnaik, 2014). For example, when interviewing representatives of the technological providers, we focused on the characteristics of the BC systems, whereas with the winemakers' CEOs, we focused on the firms' internal processes, competitive strategy, etc.

Each interview lasted on average from 90 to 120 min. To validate the information or correct any misunderstandings, we decided to provide interviewees with a structured summary of the interviews' verbatim transcripts, including all relevant information for which feedback was necessary. Providing the respondent with this structured summary, instead of the entire transcript, maximized the probability of getting feedback from him/her, and thus minimized the risk of bias.

Whenever possible, triangulation of information from different sources was used to increase the reliability of the research and the validity of the analysis (Yin, 2017). In addition to interviews, the following secondary data sources were used: (1) official documents provided by the companies and technological partners (e.g. slides of internal presentations), (2) web resources (websites, public interviews) and (3) data gathered from the implemented BC systems. Follow-up phone calls were made to clarify any doubts about the data collected.

3.3 Data analysis

The analysis was carried out in two steps: analysis of within-case data and searching for cross-case patterns (Yin, 2017).

We started the within-case analysis by creating a detailed write-up of each case. Data reduction was then used to summarize the large amount of collected information: we broke down data and characterized each case based on a series of variables that described the different BC systems and the upstream and downstream counterfeiting protections (Tables 4 and 5). We applied the data coding procedure recommended by Yin (2017) to identify variables. As is typical in theory-building case-based research, this process was cyclic and iterative (Voss et al., 2016). Starting from existing research on counterfeiting and BC (Section 2), we identified the relevant concepts in this study and, in particular, based on BC literature, we identified a number of variables potentially useful to distinguish BC systems. While some differences between BC systems clearly emerged from the use cases and BC studies (Table 2), a clear operationalization of the BC system construct was lacking in the literature. Thus, looking at evidence emerging from the data was fundamental to abductively identifying relevant variables. Identified variables were rated according to precise rules, and again, the comparison of data across cases was central to defining these rules (Section 4.1, Tables 4 and 5).

The result of this coding led to the selection of four variables to characterize BC systems: data veracity control measures, data-entry frequency, smart labels and related customer communication channels and the proportion of accessible data for which hashes are saved on the BC. Two of these variables had already been addressed by previous BC-related studies (i.e. data veracity control measures, smart labels and communication channels), whereas the other two (i.e. data-entry frequency and the proportion of accessible data for which hashes are saved on the BC) had only been briefly mentioned as potential issues in BC implementations (Table 2). Furthermore, by analysing and comparing cases, we distinguished between different desired levels of upstream and downstream counterfeiting protection (Section 4.2, Tables 4 and 5). This was strictly linked to their exposure to the risk of downstream counterfeiting by potential offenders and the extent to which they intended to insure consumers against upstream counterfeiting, guaranteeing that viticulture and vinification activities followed production and quality standards and were as declared.

As suggested by Miles and Huberman (1994), the data were organized in tabular displays (see Section 4), which was useful for both within-case analyses and cross-case comparisons. They comprised two columns: a description of the case broken down into the relevant variables and the corresponding rating of each variable according to the rules of Table 4.

Cross-case analyses were performed by structuring the data as two-variable matrices (Figure 1) to detect commonalities and differences between cases. The results derived from case analyses were then summarized in the form of propositions.

4.1 Blockchain systems

Case comparison showed the existence of commonalities as well as differences in the implemented BC systems (see Tables 4 and 5).

In all cases, the data were manually uploaded to the BC, but different measures were adopted to control data veracity: from no control measure in cases A and E, to AI in case D and third-party certification bodies in cases B and C. While these BC feeding options are well-known in the BC literature (Table 2), an important distinction emerged from the cases concerning data-entry frequency, which had received little attention in previous studies. This varied from some weeks/months (low frequency) in cases A and E, to some days/one week (medium frequency) in case D and less than 24 h (high frequency) in cases B and C. This variable appears to be relevant in the characterization of a BC system because it can influence fraud opportunity. In fact, a higher risk of manipulation occurs when data remain unrecorded and vulnerable to modification for a long period.

As to BC reading, cases differ in the use of smart labels and customer communication channels, as expected from the BC literature. In all the examined systems, product-related information displayed to customers through web pages or mobile applications is not taken directly from data saved on the BC. All data are stored on external centralized servers, but customers can verify information provided through the hashes linked to the BC transactions. This is a common practice, especially when uploading large files. However, a significant difference emerged between cases A and E and cases B, C and D, regarding the proportion of data for which hashes are saved in the BC. This has important implications for the level of protection assured by BC technology since a consumer should be able to verify at any time what information is aligned with data originally stored on the BC to avoid the risk of potential manipulation.

4.2 Upstream and downstream counterfeiting

As emerges from Table 5, cases significantly differ in the desired upstream and downstream counterfeiting protection level. This is strictly linked to their business strategy, namely how they intend to create and maintain their competitive advantage through distinctive skills and/or resources (Mintzberg et al., 1995) and to the risk of downstream counterfeiting depending on their price positioning and target market.

For companies B and C, guaranteeing grape origin, as well as growing and production processes, and product authenticity are of paramount importance. In fact, they own a special know-how related to grape growing and winemaking as well as superior quality assurance skills. They also produce in famous wine areas in compliance with stringent PDO specifications. This results in top-quality wines sold at a very high price, also in international markets. These companies are exposed to the risk of counterfeit copies and thus intend to protect customers against downstream counterfeiting (high level of downstream counterfeiting protection). Moreover, given the special product characteristics, they aim to assure them that wines are produced as declared (high level of upstream counterfeiting protection).

For companies A and E, given the wines they sell (i.e. everyday wines) and consumers they target (i.e. national large supermarkets), there is no real risk of downstream counterfeiting for their products, and thus they are not interested in reducing this risk. They also have no special skills or resources, and thus it is not important for them to provide true empirical evidence of grape origin or adherence to particular production and processing specifications. Instead, they want to reassure customers about the quality and safety of the wine (entry-level of upstream counterfeiting protection).

Finally, for case D, the level of upstream counterfeiting protection the wine producer intends to provide to customers is classified as medium, given its interest in guaranteeing the grape origin on which wine quality differentiation is based. In fact, it produces “niche” everyday wines in a narrow area with morainic hills, and its strategy is based on this distinctive resource, rather than on distinctive production skills. Given the medium-price positioning and target markets (i.e. national large supermarkets), there is no particular risk of downstream counterfeiting (null downstream counterfeiting protection).

4.3 Formulation of propositions

In this study, the data were cross-compared to derive propositions for how BC systems are designed to achieve a desired level of upstream and downstream counterfeiting protection. To this aim, we first investigated the existence of a link between the desired upstream and downstream counterfeiting protection and the BC system variables (Figure 1). Then, based on the situational crime perspective, which identifies the reduction of the fraud opportunity structure as the key driver in the design of anti-counterfeiting measures, we proposed how the desired degree of counterfeiting protection affects the design of BC systems given the varying ability of the system variables to reduce fraud opportunity.

The cross-comparison revealed different clusters of cases with similar patterns. Cases A and E, which aimed for an entry-level protection against upstream counterfeiting and no protection against downstream counterfeiting, took similar decisions in terms of BC systems, namely no control measure for data veracity, a low data-entry frequency, the use of QR codes and web pages as customer communication channel, as well as a low proportion of accessible data for which hashes were saved on the BC. Cases B and C exhibit an opposite pattern. They both desired a high level of upstream and downstream counterfeiting protection and adopted similar BC systems, with third-party certification bodies as data veracity control measure, a high data-entry frequency, NFC tags and mobile applications as customer communication channel, as well as a high proportion of accessible data for which hashes were saved on the BC.

Finally, case D is similar to cases A and E with their use of QR codes and web pages as customer communication channel, while it differs in its use of AI, a medium data-entry frequency and a high proportion of accessible data for which hashes were saved on the BC. Only this latter characteristic is common to cases B and C. Since case D has commonalities with cases A and E (no desired downstream counterfeiting protection) and differs from all the cases aiming at a medium level of protection against upstream counterfeiting, it appears evident that the precise relationships between the BC system options must be considered through a disaggregated analysis to distinguish between upstream and downstream counterfeiting.

The following proposition summarizes the results of the cross-case analysis at an aggregate level:

As for data veracity control measures, Figure 1a suggests that the desired upstream counterfeiting protection level is linked to the use of these measures: from no control when an entry-level of protection is required, to the use of AI when it is at a medium level and the use of third-party certification bodies when it is at a high level.

In cases B and C, the relevance of making customers aware of the superior quality level of the wine, grape origin and specific production processes led to the use of third-party certification bodies for data collection and data entry. Interviewed managers agreed that this choice entails higher costs, but judged this the best way to dissuade fraudulent behaviour in their context, as it can be employed to control the veracity of a large set of data ranging from process yields to cleaning, cellar temperature and so on. Both cases B and C share a significant amount of information through BC, from cultivation and winemaking processes to information related to the geographical context. The use of third-party certification bodies can insure customers against the risk of deliberate and intentional manipulation of all declared data. In contrast, company D is mainly interested in providing assurance about the grape origin on which its differentiation is based and thus aims at a medium level of protection to prevent upstream counterfeiting. It adopts AI to detect inconsistencies between some quantitative data (e.g. between hectares cultivated, tonnes of grapes harvested and bottles produced). The existence of inconsistencies means that the wine origin or even quality (e.g. a low percentage of bunch thinning) is not guaranteed; however, their absence offers a certain level of assurance of the use of grapes from declared vineyards. At the same time, AI is ineffective in verifying how some activities have been accomplished (e.g. cleaning of equipment and facilities and proper bottle storage), which is in fact not crucial for a company that aims to differentiate for grape origin rather than production processes. Thus, we can conclude that AI reduces fraud opportunity only to a certain extent and offers a counterfeiting protection level that is lower than the use of third-party certification bodies. Finally, in cases A and E, the desired upstream counterfeiting protection is at an entry-level, as they simply aim to reassure customers that the product quality is acceptable although the price is low, without any interest in providing guarantees for the grape origin or adherence to wine production specifications given that they do not possess distinctive skills and resources. Thus, they do not rely on particular data veracity control measures and can potentially save information on the BC that does not reflect reality. However, the fact that such information is shared through an immutable traceability technology represents an entry-level discouragement of fraudulent conduct.

Based on this evidence, the following proposition is advanced:

The pattern in Figure 1b is similar. It relates the desired upstream counterfeiting protection to the frequency of entering process-related information on the BC. Cross-case analysis suggests that companies B and C, aiming for high-level protection, enter data on the BC with a high frequency. In particular, after a first short period where collected information is stored on the certification body's servers or in the Italian national agricultural information system, data are synchronised with the BC. The whole process takes up to 24 h. This minimizes the time between data collection and data entry during which manipulations may occur, thus reducing the fraud opportunities. For company D, which pursues a medium level of protection, process-related data are stored and grouped on the technology provider's servers and undergo a further manual correctness check before being immutably saved on the BC. Then, the set of data is registered in a BC transaction between 24 h and a week after data collection. When the desired protection is at an entry-level – as in cases A and E – companies first use paper registers or electronic files to group information related to production processes. The timestamps of BC transactions show significant time lags for events, which range from one week to several months. In general, a higher data-entry frequency implies higher costs, as the number of transactions increases, as in cases B and C. For these latter cases, given the exclusivity of their products, it is important designing a BC system that offers higher guarantees that no manipulation can occur in the upstream network before on-chain registration, so as to provide customers with strong evidence of their distinctive skills and resources. In summary, we can conclude that the higher the data-entry frequency, the lower the fraud opportunities, as data remain vulnerable to modification for less time. Based on assumptions derived from the situational crime perspective, it follows that a higher data-entry frequency guarantees a higher protection against upstream counterfeiting.

The following proposition encapsulates this evidence:

A further relationship represented in Figure 1c links the desired downstream counterfeiting protection to the different types of smart labels and communication channels. Companies A, D and E all use QR codes printed on the wine label, which redirect to web pages. This solution is cheap, but the interviewed managers are aware that it does not significantly reduce downstream counterfeiting opportunities, as QR codes are also easily cloneable (e.g. by scanning the label, printing it and applying it to a fake bottle). However, in all these cases, downstream counterfeiting is not an issue. The target segment and the price positioning (everyday wines) in cases A and E make the risk of products being counterfeited close to zero. Similarly, managers in company D think that this risk is extremely low given the low brand recognition, the medium-price positioning and the targeted market segment. In contrast, in cases B and C, the companies decided to use NFC tags with mobile applications. This solution entailed higher costs because it required the redesign of the labelling process. However, the managers agreed that, unlike the QR codes, NFC tags significantly deter downstream counterfeiting, which was a priority for them. In fact, the NFC tags adopted by companies B and C are difficult to physically remove and provide additional guarantees of non-replicability, such as the encoded unique identifier (UID), as well as data encryption combined with the associated mobile application that decrypts the hidden data stored in the tag. These measures reduce the opportunities for fraudulent conduct, as they discourage counterfeiters from extracting the reference that redirects customers to the web content related to the genuine product from the original NFC tag and then using this link to replicate the tag. Given the exclusivity of their wines due to their distinctive skills and resources, and the consequent need for high-level protection against downstream counterfeiting, companies B and C designed a BC reading solution that is able to provide customers with a stronger guarantee of non-replicability.

The following proposition encapsulates these findings:

A final important BC system option regards the proportion of accessible data for which hashes are saved on the BC. For companies B, C and D, which aim for a medium or high level of protection against upstream counterfeiting, most of the data shared with customers is related to the hashes saved on the BC (Figure 1d). Companies B and C sell high-priced and exclusive wines with a certified grape origin, and their production is complex and meet high-quality standards. Company D's differentiation relies on grape cultivation in an exclusive terroir. They are very committed to assuring customers that the shared information is stored on the BC, which offers a higher guarantee that accessible information has not been manipulated. The aim of these companies is to reduce data manipulation and fraud opportunities by giving customers the opportunity to verify the alignment between the information related to their distinctive resources and/or skills and that saved on the BC. Conversely, companies A and E aim to guarantee an entry-level protection against upstream counterfeiting, as they sell lower priced wines in large-scale supermarket chains and do not rely on distinctive skills and resources. They only share and store information on the BC that is necessary to guarantee that their wines meet the acceptable quality standards, and then share further information without providing the related hash, assuming that customers are not concerned about adherence to declared production processes or grape origin, which is not distinctive. Thus, compared to the other cases, the restriction of fraud opportunity, and thus the protection level against counterfeiting is lower.

The following proposition summarizes this evidence:

5.1 Managerial implications

This study provides some practical suggestions to properly address the design and use of BC systems.

First, it warns managers against using the BC as a stand-alone technology and instead suggests viewing it as a part of a wider BC system where decisions on technical aspects (e.g. type of smart labels and communication channels, use of AI, etc.) must be complemented with organisational choices regarding BC feeding and reading processes (e.g. third-party involvement and data-entry frequency). A coherent set of these decisions is crucial to ensuring that data written in and read from the BC accurately reflects reality.

Second, the study suggests that BC systems are not one-size-fits-all solutions and therefore should be adapted to the required upstream and/or downstream counterfeiting protection level. Specifically, the study presents a set of variables that managers should consider when designing BC systems and shows how they can be modulated in accordance with the desired guarantees of product authenticity. Several key learnings for managers can be derived. This research indicates how to design BC systems characterized by a high level of protection against upstream and/or downstream counterfeiting. It also argues that in certain contexts (as in cases A and E), the use of BC systems without particular attention to strict data input and reading measures is a good solution, as they offer some deterrent power against intentional infringements and also operate as traceability and communication tools that allow customers to autonomously verify the safety of products they buy. For instance, in the case of upstream counterfeiting, an entry-level protection is better than no protection since the core technical features of BC technology (e.g. transparency, immutability and traceability) allow the discovery of post-entry manipulation and the detection of the counterfeiter, thus making offending less feasible or attractive. In the case of downstream counterfeiting, no protection can be a viable choice where there is no real risk of downstream counterfeiting (e.g. low price and no brand products).

The third practical contribution is directed towards policymakers who are evaluating the potential of BC vs existing measures to protect consumers, honest producers and national brands from risks due to fraudulent copies or imitations. This study corroborates the advisability of the prudential decisions of many governments to fund BC-based pilot projects to explore the real potential of this technology to defend product authenticity before extending its use on a large scale. However, it also suggests caution in considering BC a panacea that overcomes all the weaknesses of existing measures. This study demonstrates that BC systems are versatile, their real potential in reducing fraud opportunities (and defending product authenticity) in comparison with existing measures can vary and that choices on BC feeding and reading determine this variation. Thus, the configuration and management of such processes is crucial to ensure that BC technology is better able to guarantee product authenticity than existing measures.

6.1 Limitations and future research directions

A first limitation of this study is linked to the characteristics of the cases investigated (Section 3.1.1), which restricts the generalizability of this research and sets precise research boundaries. First, as in most existing BC initiatives, all the examined BC projects are in the pilot phase and thus involve a limited number of companies and products. Further studies may consider BC system implementations in more complex supply networks. For instance, a possible issue may concern the scalability of the use of third-party physical inspections, as in cases B and C. Whereas cost is not significant in simple supply chains such as those analysed in this study, third-party inspections could become economically unsustainable in more complex supply networks. Second, all the analysed cases apply public BC technologies that differ significantly from private BCs in terms of data immutability and decentralized governance, and thus conceptually in their counterfeiting protection. Further studies of private BCs are needed to understand the effectiveness of their adoption for anti-counterfeiting and whether and how BC systems may differ in terms of BC feeding and reading. Third, as argued in Section 3.1.1, the implementation of the IoT in the wine sector is still a challenge. Future research could investigate the issue of BC feeding in more automated sectors, such as the pharmaceutical sector, to understand the implications of IoT use and its efficacy in solving BC feeding issues. Although several scholars (e.g. Kamble et al., 2019; Bumblauskas, 2020) consider the IoT an effective means of guaranteeing objective data collection and recording, which reduces the risk of human manipulation; others argue that IoT sensors can also be subject to intentional manipulation (Rejeb et al., 2019; Schmidt and Wagner, 2019). In the future, more specific IoT sensors could be developed and spread to appropriately detect important production data, e.g. in the wine cellars. Further research, in the wine or luxury clothing contexts, could investigate the use of the IoT in combination with AI algorithms to support BC feeding. These solutions could eliminate the need for third-party certification bodies for data that are currently not measurable by sensors.

A further limitation of this study is that it assumes the perspective of a brand owner who configures the BC system in line with a desired counterfeiting protection level. Once BC systems become more mature and widely used, it will become essential to consider the buyers' perspective, in particular the final customers' perspective, to understand to what extent and for which clusters of customers the perceived risk of counterfeiting results is reduced through BC technology.