1.1 The Received Wisdom on the Economics of Corporate Diversification

Over the past fifty years, since the seminal work of Penrose (Reference Penrose1959) and Rumelt (Reference Rumelt1974, Reference Rumelt1982), a rich literature on corporate diversification has developed. According to this literature, the fundamental principles that explain why firms engage in diversification are the following:

• A company may possess surplus assets which are underutilized and not efficiently tradable in the open market.

• Unable to trade these assets, the company retains them and employs them internally in alternative markets.

• As the company employs these assets further from its core market, the utility derived from these assets diminishes. The assets were originally developed for a specific market, so the further the company moves away from this market, the weaker the alignment between the asset and the market, resulting in decreased productivity.

• At the same time, moving away from its core market increases coordination and transaction costs. This is due to the heightened complexity and management constraints associated with operating in distant markets.

Based on the extant economic logic, academics have argued that every firm has a limit to how much it can diversify (Markides, Reference Markides1995). The argument goes as follows: the exploitation of underutilized assets and competences in new markets creates value for the firm. If the assets are utilized in a market that is close to the core market, they can be as productive as they were in their original use in the core market and so create a lot of value in the new market. If, on the other hand, they are applied in markets that are far away from the core, then their productivity declines because they are being used for something different from their original purpose. This suggests that the further a firm diversifies from its core business, the less value it can extract from its assets, a relationship shown as a declining marginal benefits curve in Figure 3. At the same time, these benefits to diversification are not achieved without cost. In fact, the further the firm diversifies from its core, the bigger the costs associated with managing all the different businesses under one corporate umbrella. The rising marginal cost curve in Figure 3 shows this relationship between the level of diversification and the costs of diversification. The point at which the marginal benefits and marginal costs curves meet is what is known as the “optimal” level of diversification for a given firm. A firm can create value through diversification if it diversifies up to this level but will destroy value if it tried to diversify beyond this level.

Figure 3 The marginal benefits and costs of diversification

In what follows, we will argue that the digital revolution of the last twenty years has not changed this economic logic for diversification, but it has changed the context within which diversification now happens. Specifically, today’s pervasive connectivity and ever-growing (cloud) computing power allow companies to leverage a new asset – data – as well as exploit AI, and ecosystems for their diversification strategies. This has resulted in different diversification dynamics, reflected in the company behaviors we observe.

1.2 The Digital Transformation of Physical Reality and the New Context for Diversification

The economic logic of diversification was developed in the institutional and technological context of the second half of the twentieth century. The increasing diffusion of digital technologies since 2000 has changed the context in which companies operate, something which is often referred to as digital transformation. What has been the effect of this change in context?

Digital technology is an umbrella term encompassing several general-purpose technologies – such as computer vision, sensors, the Internet of Things (IoT), artificial intelligence (AI), and cloud computing. Enabled by the diffusion of these technologies, the physical reality is being radically transformed, as shown in Figure 4. At the heart of this digital transformation of the physical reality is the process of sensing, or digitizing, which involves capturing physical actions, states, or representations and converting them into a digital data format. This process often involves using various sensors and devices to collect data from physical objects, measuring parameters such as temperature, pressure, motion, and environmental conditions.

Figure 4 The digitalization of physical assets

Once sensed and digitized, these physical assets can communicate and connect with other digitized assets through the Internet of Things (IoT) or other communication technologies, such as edge computing and 5G. For instance, IoT technology enables the seamless exchange of information between interconnected devices, creating a network of “smart” assets that can interact with each other in real time.

The interconnected network of digitized assets generates vast amounts of data, often referred to as big data. This data is continuously collected, transmitted, and stored, providing a rich source of information for further analysis. The sheer volume and variety of data generated by these connected assets is overwhelming. For instance, in 2023, the world generated approximately 123 zettabytes of data. By 2028, the amount of data generated globally is expected to reach 394 zettabytes (Statista.com), marking a 220 percent increase from 2023. This upward trend does not seem to have an end, either.

To analyze this data, we can employ advanced inferencing systems such as artificial intelligence (AI) and machine learning (ML) algorithms. AI refers to the branch of computer science and engineering that focuses on the development of algorithms and systems capable of performing tasks that typically require human intelligence. These tasks encompass a broad spectrum, ranging from basic pattern recognition and problem-solving to more complex activities such as natural language understanding, decision-making, and autonomous reasoning. AI systems leverage techniques such as machine learning, neural networks, and natural language processing to simulate, virtualize, or augment the physical reality (Bailey et al., Reference Bailey, Leonardi and Barley2012; Porter and Heppelmann, Reference Porter and Heppelmann2017).

For example, the company John Deere utilizes a range of sensors placed on farming equipment and on farms to collect data on soil conditions, crop health, moisture levels, and weather conditions. The digital data is transmitted to John Deere’s cloud-based platform, MyJohnDeere, where advanced analytics and machine learning algorithms process the data to generate actionable insights such as optimal planting times, fertilizer application rates, and irrigation schedules based on real-time and historical data.

AI and ML algorithms become more powerful as they process more data. At the same time, AI and ML are also significantly enhanced by advancements in computing technologies, particularly Cloud computing. Cloud computing provides the necessary infrastructure to store, process, and analyze large datasets efficiently. It offers scalable computing power and storage resources, enabling more complex and powerful AI applications. To put the scale of digital transformation into perspective, Statista estimates that spending on digital transformation reached $2.15 trillion in 2023. This figure is projected to rise to $3.9 trillion by 2027. Additionally, worldwide spending on AI-centric systems across all industries was estimated at $154 billion in 2023.

Thus far we have briefly described the technological foundation at the core of the digital transformation of the physical reality. The environmental, social, and business impacts of these changes have been analyzed by a large and growing literature which has outlined the intricacies of such transformations. For instance, Creutzig et al. (Reference Creutzig, Acemoglu and Bai2022) note that “great claims have been made about the benefits of dematerialization in a digital economy.” Yet, they also note that the net impact of digitalization on the environment, labor markets, resource use, governance, and power relationships depend on several interaction effects and human agency, making the outcome often unpredictable. Along the same lines, Lanzolla, Pesce, and Tucci (Reference Lanzolla, Pesce and Tucci2021) show that the outcome of digitalization in enabling more efficient knowledge integration is far from being always positive.

As regards diversification, this digital transformation has resulted in four changes of interest:

• The digitalization of the physical reality creates data as a valuable new asset that diversifying firms can strategically leverage. What is unique about data in the digital age is that it is potentially so rich that it contains information that might be applicable in completely different industry contexts. In this sense, data is less (co)specialized with the legacy positioning of the company. This implies that data – in conjunction with AI which allows for the development of insights from this data – has higher potential than traditional fixed assets or traditional capabilities to create value in new markets. Data is a type of asset that can be transferred to far away markets without losing its beneficial effects as fast as other physical assets.

• Digital ecosystems mitigate the risks of executing strategies in areas beyond a company’s traditional market positioning.

• Digital technology, such as algorithmic controls, facilitates reductions in coordination and control costs.

• The trading of data in markets is subject to inherent limitations.

We explore these four points in more detail next and then discuss their collective effect on diversification.

1.3 Data as a New Asset at the Heart of Diversification

The concept of data and information as strategic assets is not a new one. Stemming from the seminal works of Akerlof, Spence, and Stiglitz (who shared the Nobel Prize in 1970), academic research has highlighted the key features of information as a strategic asset. First, unlike physical resources, information is non-depletable, meaning it does not diminish with use. In other words, once information is created, it can be used repeatedly without being exhausted, and its value can increase as more people leverage it for various purposes. However, the value of information can diminish in terms of marginal utility: as more information is acquired, the incremental value or utility of additional information might decrease. Furthermore, the value of information can decline over time if it becomes outdated or irrelevant: information is most valuable when it is current and applicable to the context at hand. Finally, in markets, the value of information can decrease if it becomes too widely available: when proprietary or exclusive information becomes common knowledge, its competitive advantage diminishes.

What is new in the digital age is that the sheer volume of data generated and collected from various sources; the velocity at which data is generated; and the variety of this data can generate information of an unprecedented scope and scale (McAfee and Brynjolfsson, Reference McAfee and Brynjolfsson2012). Furthermore, what is unique about data in the digital age is that it is potentially so rich that it contains information that might be applicable in completely different industry contexts.

Data for Increasing Efficiency in the Utilization of Physical Assets. Through their digital representation, a firm’s physical assets can be monitored, measured, and analyzed in unprecedented detail. More specifically, through their digital artefacts, the physical assets acquire characteristics which are typical of software products (e.g., Yoo, Reference Yoo2010). For instance, they become more traceable and editable which facilitates the (continuous) improvement of any elements while maintaining logical structures intact. Like software, they also become more replicable, thus enabling the effortless reproduction and dissemination of digital artifacts at minimal marginal costs – for example, consider songs which can be replicated infinitely with negligible expenditure. Similarly, modularity plays a crucial role in the decomposition and reconstitution of digital elements, fostering collaboration and adaptability across varied contexts. Finally, the granularity inherent in digital artifacts facilitates the detailed analyses and manipulations of the physical asset and of the digital artefact alike. Overall, the digital artifacts related to the physical assets create new representations that impact the ways in which physical assets are used thus changing their utilization and efficiency. It follows that more efficient asset utilization might generate the excess capacity for more diversification.

Data as a Generative Asset for Deriving Insights about Value Creation (in Related or Unrelated Markets). Through the aggregation and scrutiny of different data sources alongside the utilization of artificial intelligence, firms can procure unique insights into assets and client behavior. The academic literature has identified several mechanisms for discovery, including data fusion and integration, which combines data from multiple sources to provide a comprehensive view and uncover new insights. Pattern recognition uses statistical techniques, machine learning, and data mining to reveal hidden patterns and correlations in large datasets. Hypothesis generation and iterative learning involve the analysis of data to generate hypotheses based on observed patterns, guiding further investigation through continuous testing, experimentation, and feedback to refine models and uncover deeper insights.

Such insights can be leveraged in several ways. For instance, the anticipation of needs and the customization of offerings can manifest in the capacity to cross-promote or upsell correlated products or services to extant clientele (Bharadwaj et al., Reference Bharadwaj, El Sawy, Pavlou and Venkatraman2013; Bughin et al., Reference Bughin, Seong, Manyika, Chui and Joshi2018; Zhu et al., Reference Zhu, Kraemer and Xu2016). At a more general level, when a firm possesses an understanding of its clientele’s preferences, it can discern supplementary and more profound offerings that complement extant consumer utilization or acquisition, or it may endeavor to capture the life-cycle value of such clientele. Overall, the retention of contented clientele is comparatively less financially burdensome than the pursuit of new ones, particularly during the transition of clientele from mature products to next-generation alternatives (Bolton, Reference Bolton1998; Gupta and Zeithaml, Reference Gupta and Zeithaml2006; Kumar and Reinartz, Reference Kumar and Reinartz2016; Rust, Lemon, and Zeithaml, Reference Rust, Lemon and Zeithaml2004; Venkatesan and Kumar, Reference Venkatesan and Kumar2004), thus providing a second advantage to companies that can leverage this mechanism.

The generation of new insights about drivers of value creation is not limited to a company’s legacy positioning. For example, once Apple harvests behavioral data from its Apple watch wearers, it can analyze this data and develop insights about their nutrition, shopping, entertainment, or health. In other words, data collected via the Apple Watch is not restricted to the typical function of a watch. Similarly, once Netflix collects enough data on the types of movies its subscribers prefer, it can develop insights on viewers’ preferences and “diversify” away from just streaming movies to also producing them. That such a strategy can succeed is evidenced from the fact that eight out of the ten most successful movies produced in China in 2019 were based on insights developed from patterns of preferences extracted from Alibaba’s platform.

Data in conjunction with pervasive client accessibility has given rise to what some authors call demand-side diversification (Harrigan, Reference Harrigan2023) whereby companies seek customer-based synergies rather than production-related synergies. In this light, diversification is aimed at offering a broader value proposition through additions to existing product lines (Adner and Zemsky, Reference Adner and Zemsky2006; Schmidt, Madadok, and Keil, Reference Schmidt, Madadok and Keil2016). Amazon’s vision of “[being] Earth’s most customer-centric company, where customers can find and discover anything they might want to buy online, and endeavors to offer its customers the lowest possible prices” is a fitting example of such demand-side dynamics whereby customer data and customer base reinforce each other to create an even broader value proposition. The already cited cases of Apple (entering healthcare and entertainment) and Tesla (entering insurance) are other textbook examples of such dynamics. Manral and Harrigan (Reference Manral and Harrigan2018a and Reference Lalit and Harrigan2018b) find that this approach often increases the firms’ unrelated-diversification scores while maintaining high customer-similarity ratios.

Data as the Shaper of a New Ontology. Finally, some scholars have even hypothesized that the digitalization of physical reality fundamentally alters reality’s ontology, too. Baskerville et al. (Reference Baskerville, De Marco and Spagnoletti2020) coined the term “ontological reversal,” indicating that digital versions often precede their physical counterparts, seen in practices like 3-D printing. This shift redefines product production, with digital versions taking priority over physical ones. Traditionally, physical objects served as the basis for digital representations, but ontological reversal suggests the opposite, challenging notions of materiality. For example, digital records stored in the cloud increasingly replace physical tickets, becoming the primary proof of transaction. Ontological reversal extends beyond digitization, fundamentally restructuring reality where digital entities hold greater significance. This has profound implications across commerce, communication, and identity. In commerce, digital products may outweigh physical ones due to scalability and accessibility. Similarly, digital identities and communities may overshadow traditional social connections, reshaping interpersonal dynamics. Furthermore, it blurs temporal and spatial boundaries, making the distinction between “real” and “digital” more fluid. Traditional diversification strategies focus on physical and financial assets. With ontological reversal, data becomes perhaps the only asset, systematically altering how companies assess value and potential diversification opportunities.

1.4 Digital Technology for Reducing Strategy Execution Risks, Costs of Coordination and Costs of Control

A second significant shift lies in the structure and governance of strategy execution. Unlike diversifiers of the past, today’s digital diversifiers no longer need to handle everything on their own. In the past, companies had to build and manage a wide range of capabilities internally, leading to inefficiencies and slower response times. They invested heavily in infrastructure, research and development, and workforce training to diversify. This approach often limited the speed and scope of diversification efforts. The digital context has transformed this paradigm. By enabling companies to engage with a broader and more diverse array of partners, digital ecosystems allow companies to leverage external partnerships and collaborations, accessing specialized knowledge and innovative solutions without developing these capabilities internally (Adner, Reference Adner2017; Baldwin, Reference Baldwin2004; Jacobides, Cessano, and Gawer, Reference Jacobides, Cennamo and Gawer2018; Lanzolla and Markides, Reference Lanzolla and Markides2022). For instance, extant research has shown how different collaborative arrangements between innovators and complementors influence their ability to invest in new technology and commercialize it (e.g., Kapoor and Lee, Reference Kapoor and Mahn Lee2013; Leten et al., Reference Leten, Vanhaverbeke, Roijakkers, Clerix and Van Helleputte2013); how knowledge sharing affects inter-firm relationships and ecosystem development (e.g., Alexy, George, and Salter, Reference Oliver, George and Salter2013; Brusoni and Prencipe, Reference Brusoni and Prencipe2013; Frankort, Reference Frankort2013); and the overall health and longevity of the ecosystem (e.g., Leten et al., Reference Leten, Vanhaverbeke, Roijakkers, Clerix and Van Helleputte2013), and the strategy and management of partnerships in digital ecosystems (Lanzolla and Markides, Reference Lanzolla and Markides2022).

An increase in the number and variety of partners to execute strategy might be beneficial but might also be subject to higher coordination and control costs. In the received theory of diversification, coordination and control costs are considered as a barrier to diversification as complexity increases. However, the digital contexts provide also a second fundamental shift in governance logics (Bailey et al., Reference Bailey, Leonardi and Barley2019; Barrett et al., Reference Barrett, Oborn and Orlikowski2012; Baum and Haveman, Reference Baum and Haveman2020; Brynjolfsson and McAfee, Reference Brynjolfsson and McAfee2014; Cohen and Tripsas, Reference Cohen and Tripsas2018; Faraj et al., Reference Samer, Jarvenpaa and Majchrzak2011; Lee and Berente, Reference Lee and Berente2012) which significantly affect coordination costs. This shift is enabled by digital technologies which are often described under the label of algorithmic controls (Kellogg et al., Reference Kellogg, Valentine and Christin2020). Algorithmic controls involve the deployment of sophisticated algorithms and machine learning techniques to automate and optimize decision-making and (inter)organizational governance. Algorithmic controls represent a departure from traditional modes of organizational control and decision-making, introducing novel approaches that leverage computational power and data-driven insights to inform and guide managerial actions.

By harnessing the capabilities of AI and advanced analytics, organizations can achieve greater efficiency, accuracy, and agility in their operations, enabling them to adapt more effectively to dynamic and complex environments. The adoption of algorithmic controls facilitates real-time monitoring, analysis, and optimization of organizational processes, allowing for proactive decision-making and strategic adaptation in response to changing market conditions and competitive pressures. Algorithmic controls have far-reaching implications for organizational dynamics, power structures, and employee behaviors. As decision-making authority becomes increasingly delegated to algorithms and automated systems, questions arise concerning accountability, transparency, and ethical governance (Kellogg et al., Reference Kellogg, Valentine and Christin2020). Additionally, the reliance on algorithmic recommendations and predictions may influence managerial perceptions, attitudes, and behaviors, potentially shaping organizational cultures and norms (Faraj et al., Reference Samer, Jarvenpaa and Majchrzak2011).

Pervasive Connectivity and Network Effects. Network effects occur when a product or service becomes more valuable to a user as more people use it. The value increase relative to the number of users is often depicted as a “quadratic” function. Consequently, establishing network effects (NE) becomes a powerful mechanism for customer lock-in. While NE is not an economic force unique to the digital age, the ubiquity of connectivity in the digital context allows NE to be established across broader and more diverse user bases, encompassing varied demographics, regions, and preferences.

Received wisdom would suggest that establishing NE is easier when products are based on open standards and market mechanisms. For example, the Android operating system, which is open source, allows various manufacturers to create devices that can seamlessly interact with each other, thus leveraging open standards to establish widespread network effects. The digital context, for the relevance of data as the driver of value creation and for providing stronger internal control/coordination mechanisms, might provide strong economic advantages to companies that internalize the delivery of several complementary assets around a central value proposition.

For instance, Boudreau, Jeppesen, and Miric (Reference Boudreau, Bo Jeppesen and Miric2021) explored how network effects impacted firms using freemium strategies in the Apple App Store, finding that network effects alone do not boost revenues, but combined with freemium models, they enhance market leaders’ dominance. The study showed that while freemium strategies can expand user bases and leverage network effects, they may also reduce revenues for followers and reinforce market leaders’ positions, suggesting platform owners must carefully consider the broader market implications.

One textbook example that illustrates this mechanism is Apple’s approach with its ecosystem. Despite the dominance of open standards, Apple has successfully internalized the delivery of several complementary assets around a central value proposition – its ecosystem of devices and services. Apple’s iOS, macOS, watchOS, and other software are tightly integrated with its hardware products like the iPhone, MacBook, Apple Watch, and AirPods. Apple’s internal mechanisms counterbalance the power of open standards and market forces.

Xiaomi’s diverse product portfolio, encompassing smartphones, electronics, IoT devices, and software, highlights its strategic approach to diversification. By leveraging extensive user data, fostering engaged community participation, and building an integrated ecosystem of products and services, Xiaomi enhances customer satisfaction while effectively reducing risks associated with entering new markets and product categories.

1.5 Trading Data versus Exploiting Data through Diversification

According to the theory of diversification, excess capacity might be sold in open markets, and diversification occurs only when such commercial transactions are not possible or economically advantageous. Thus far, we have established that data is at the core of diversification in the digital age. The related question, therefore, is: “what are the considerations related to trading data versus exploiting data internally?”

Companies such as Apple, Amazon, Facebook, and Google often monetize the data they collect by providing access to third parties or directly selling it (Manyika et al., Reference Manyika2011). However, not all data can be traded. In the context of data commerce, it is essential to distinguish between data and information goods, with our focus primarily on the former. Despite the potential economic value inherent in data, several factors impede their tradability and integration into third-party business models. Notably, considerations such as data quality, compliance with regulations, and the ability to appropriate value from data assets pose significant challenges (Koutroumpis et al., Reference Koutroumpis, Leiponen and Thomas2020).

Integrating data into a company’s value proposition represents a complex endeavor, burdened with both technological and managerial hurdles (Lanzolla and Markides, Reference Lanzolla and Markides2022). The absence of a ubiquitous “eBay for data” marketplace can be attributed to formidable concerns regarding strategic behaviors, data quality assurance, and the lack of sufficient control over the utilization of data by prospective buyers (Thomas, Leiponen, and Koutroumpis, Reference Thomas, Leiponen and Koutroumpis2020). Furthermore, a significant portion of the data amassed in contemporary digital environments falls under the category of “exhaust data,” originating as by-products of various online activities such as e-commerce transactions and social interactions, rather than being intentionally generated for analytical purposes (Manyika et al., Reference Manyika2011; Mayer-Schonberger and Cukier, Reference Mayer-Schönberger and Cukier2013).

Finally, the information paradox, articulated by economist Kenneth Arrow (Reference Arrow1974, Reference Arrow1985), significantly limits the trading of data in open markets. This paradox highlights the dilemma where the buyer needs to know the value of the data before purchasing it, but once the buyer knows the information, there is no need to buy it. This creates a value assessment dilemma, as sellers are hesitant to reveal too much about their data for fear of it being used without compensation, while buyers are reluctant to purchase data without understanding its worth. Additionally, risks of data leakage and intellectual property theft further discourage detailed disclosures from sellers. This leads to trust and verification challenges, where buyers need assurance of data quality and relevance without full access, complicating the establishment of fair pricing and efficient market transactions.

1.6 Toward a New Theory of Diversification in the Digital Age?

Table 1 summarizes the differences between the theory of traditional diversification and data-driven diversification in the digital age, as discussed thus far.

Our discussion thus far – allows us to generate insights on how diversification is affected by the digital context. First, the marginal benefits curve has shifted upward primarily because firms are now increasingly using data to diversify. What is unique about data as an asset is that it contains information that might be applicable in completely different industry contexts. In this sense, data is less (co)specialized with the legacy positioning of the company. This implies that data – in conjunction with AI which allows for the development of insights from this data – has higher potential than traditional fixed assets or traditional capabilities to create value in new markets. Data is a type of asset that can be transferred to far away markets without losing its beneficial effects as fast as other physical assets do. The end result of using this type of asset to diversify is that the marginal benefits curve of diversification has shifted upward. We show this in Figure 5.

Figure 5 The marginal benefits and costs of diversification for digital diversifiers

At the same time, the costs associated with diversification have now been lowered through the use of AI and ecosystems. In contrast to diversifiers of old, today’s diversifiers can diversify increasingly through partnerships. The reason for this is that in the digital context, connectivity and automation increase the availability and type of potential partners while decreasing coordination costs. For example, Apple can move into healthcare without having to make the products itself. It can subcontract manufacturing to a medical device manufacturer. Similarly, Google can enter the banking industry without having to “manufacture” bank accounts by itself – it can do it in alliance with traditional lenders. Likewise, Vodafone can move into banking in Africa, but it does not have to manufacture and sell banking products itself. It can subcontract these activities to partners in its ecosystem. It’s not that this option was not available to diversifiers in the past, but the digital context has made it a more attractive option now. Not having to do everything under the same organizational hierarchy, in conjunction with the fact that coordination and control can be increasingly automated through AI, means that firms can now diversify without incurring the same: (a) coordination costs and (b) execution costs that previous diversifiers had to incur. This has resulted in the “cost” curve of diversification shifting downward – something shown in Figure 5.

The combined effect of the marginal benefits curve shifting upward and the marginal cost curve shifting downward is a new optimal level of diversification for firms. This optimal level is higher than what it used to be, suggesting that firms can now diversify much more than what they could twenty years ago. This explains why companies are now busily considering diversifying into seemingly unrelated businesses again.

2.1 The Capability to Develop Constant Flows of Data in Context

In the course of doing business, companies collect a wealth of data on their assets, products, and customers. Very often, this data can provide information that may be useful in a different market from the one the company is operating in. For example, knowing how customers drive their cars gives Tesla the opportunity to develop insights on how to produce customized insurance products that incorporate these customers’ driving behaviors. Similarly, having data on what movies their customers are watching can help Netflix develop insights as to what movies to make. Possession of this data may give diversifying companies an advantage over companies that do not have access to this data. This advantage may be big enough to outweigh the advantages that incumbents have in the market the new entrant is moving into.

Therefore, the first ingredient for success is not just the possession of data but rather the possession of data in context. By this we mean data that captures people or asset behaviors in contexts relevant to the market where a firm may consider entering. For example, knowing what the customer buys in a supermarket or when they buy does not give that supermarket any insights about their movie preferences. This might explain why the British supermarket TESCO failed in its attempts to offer its customers video-on-demand services through its 2011 acquisition of Blinkbox. On the other hand, knowing how their customers drive their cars gives Tesla insights on how to develop customized insurance products that incorporate their driving behaviors.

Before a company can even consider a new market entry, it must create flows of data in context. This is not easy. The company first needs to have the appropriate infrastructure to capture this kind of data. More fundamentally, it needs to have the trust of clients as well as other industry stakeholders who will provide it with access and permission to use this data. This can be a formidable challenge.

Consider, first, the challenge of developing the appropriate infrastructure to capture data in context. Apple is quietly yet relentlessly building the infrastructure to capture people’s health and fitness data, something often called “lifelogging.” For example, in September 2018, Apple launched an app that could detect when people fall down. In March 2019, it enabled the pairing of its devices with Snoww, Nike, QuardioArm (a smart blood pressure monitor) and DexCom Glucose (a glucose monitoring system). In June 2019, it improved the Apple GymKit app to include pairing with many more cardio equipment manufactured worldwide. It also launched a Cycle tracker for women and a Noise app to help users understand noise levels in high-risk environments. In June 2020, sleep trackers and automatic hand wash detectors were added to the Apple Watch, while, in September 2020, Apple Watch S6 incorporated blood oxygen sensors. These are all examples that highlight Apple’s ambition in collecting data that can be leveraged in the healthcare market. Google’s acquisition of FitBit and Amazon’s release of the Halo band (that allows for the collection of customer data through wearables) show that the healthcare industry is very much an industry of interest to these two companies as well (Guardian, Reference Guardian2022; McKenna, Reference McKenna2021; Mishra and Dastin, Reference Mishra and Dastin2022).

In addition to having the right kind of infrastructure to collect data in context, companies also need the trust of consumers to allow them access to this data. If Apple can collect data on consumers’ health through its smartwatch, it can utilize this data to move into the healthcare market. But it is not only customer trust that is needed. To offer a good enough product, the new entrant may also need data that other industry participants – including competitors – have. Gaining trust with stakeholders is not easy. The example of the UK Ministry of Justice is a case in point. Despite putting together the right infrastructure – digital bracelets on convicted felons – it proved very difficult for the UK Ministry of Justice (MoJ) to collect “data in context” from them. According to Arif Harbott, the CIO at the MoJ in 2017, the convicts simply did not “trust” the MoJ to deal with their data and did everything to interfere with the proper functioning of their bracelets.

Appropriate infrastructure and the trust of clients as well as other stakeholders to collect and use this data are both of paramount importance for building flows of data in context, and often create self-reinforcing, or self-destructing loops vis-à-vis building flows of data in context. The example of GE’s Predix is a case in point. On paper, GE had both a remarkable budget for building the Predix infrastructure and clients who seemed willing to share data to get better service and optimize their operations. According to a senior GE executive intimately involved in the development of Predix, the idea behind it – at least on paper – was brilliant. In his words:

Jeff Immelt, the CEO of GE at the time, bet the company and its legacy on this vision. Yet, Jeff Immelt’s plan to make GE a top ten software company by 2020 did not materialize. There are several reasons for this failure, but the biggest were the platform infrastructure’s failure to deliver and the subsequent breakdown in stakeholders’ trust. A senior executive commented that “it quickly became apparent that the APIs (that allow an application to access the data) and the microservices developed for the platform were unable to collect data correctly.” This created problems for the GE partners (such as application software developers) who had built applications to analyze this data and generate insights for their clients. They had difficulties accessing this data properly and even worse, could not trust its reliability. It wasn’t long before they started complaining about technical problems and delays. A tipping point was reached when GE asked its partners for a two-month “break” to fix these problems. This undermined GE’s credibility further because its partners could not allow their resources to sit idle for two months. Relations were further undermined when former GE Digital CEO Bill Ruh declared in August 2017 that from that moment on “our resources will go to our fastest-selling markets,” a clear signal of a shift in strategy. In the end, the erosion of trust between GE and its partners spelled the demise of Predix’s initial grand vision.

Big Tech platforms – the likes of Amazon, Google, Facebook, and Apple – have a clear advantage to build flows of data in context. Not only is the platform infrastructure fit-for purpose for pervasive data collection but some of these platforms have grown to become trusted brands, with both clients and broader stakeholders.

2.2 The Capability to Generate Insights from Data by Blending AI and Domain Expertise

Having flows of data in context is a good start. But this alone is not enough. What a new entrant needs to do is to convert this data into viable and scalable ideas that differentiate it relative to the established players in the market it is entering. The incumbents may not have the data the entrant has, but they have numerous other advantages simply on the basis that they have been operating in that market long before the potential entrant. To succeed with new market entry, a firm needs to leverage its data advantage in a creative way so as to give itself a fighting chance against the incumbents. This implies that entry might be a bad move even when the potential entrant has a data advantage over incumbents.

For example, consider the case of Sky UK. The company has installed millions of satellite dishes across the UK over the past twenty-five years. The signal received by these satellite dishes is affected by weather patterns, offering Sky the opportunity to collect highly granular real-time weather data across the UK. The company’s management knew that the weather data they had been collecting was valuable, and they considered the possibility to use this data to enter several new markets such as insurance, clothing, energy, and even umbrella selling. Yet, they decided against entering any of these markets primarily because they realized that having the necessary data was not enough to give them a competitive advantage or make them successful in these markets. On the other hand, Netflix decided to expand beyond content streaming to content production. They did so because they found a way to leverage their data in context to achieve a competitive advantage over other content producers. The flows of data in context collected from their large customer base allows Netflix to develop content that is informed by a deep “real time” understanding of their viewers’ tastes.

Converting data into customized insights is not easy. Simply analyzing the data or using AI algorithms to identify patterns in the data is almost never enough. For example, Universal Music possesses detailed data about people’s music preferences. They further enrich this data by combining it with additional insights on people’s habits, behaviors, and preferences that they display through their activities on social media. The combination has produced a data pool that is so huge that the company has created a dedicated division whose responsibilities include exploring this data to extract business value for artists and partners. The potential of this flows of data is high and has already materialized in new market entries spanning from the launch of a new hotel chain (UMusic Hotel) to pizza delivery (Gangstarella Pizza). Yet, according to Olivier Robert-Murphy, Executive Vice President at Universal Music Group for Brands: “… extracting/capturing insights that create value is not straightforward and requires the development of brand-new capabilities.”

Another example is the UK Meteorological Office (hereafter, Met Office). The Met Office had a lot of granular data about the weather, as well as the most powerful computing capabilities and state of the art AI experts. They could use these capabilities to develop services for a number of industries. For example, they could advise airliners such as British Airways the best route to take on any given flight to avoid weather turbulence or head winds. They could also offer consulting services to transportation companies such as Maersk as well as farmers and other agricultural workers. According to the Met Office’s CIO, Charles Ewen, they did experiment with a few of these services – such as advising airliners on route selection that optimized fuel consumption. Yet, after a few initial attempts, the Met Office decided to abandon its forays in these areas.

According to Ewen: “What the Met Office realized was that mastering weather forecasts is only one ingredient for success in offering services such as crop selection or fuel optimization, to farmers or airlines.” The other ingredient is having domain expertise in the various markets they were entering – which the Met Office did not have. As a result, Ewen said that “[the Met Office] decided to focus even more on the core business of making meteorological predictions alongside developing our competencies in collaboration and partnerships with domain experts.”

The mistake companies often make is to assume that building an AI infrastructure and hiring lots of AI experts will allow them to deliver on their data advantage. AI is invaluable for automating and scaling but using some machine learning algorithm to identify patterns in the data will rarely deliver the creative insights that a company needs to outsmart the incumbents in the markets it is entering. AI experts may master automation but often lack the knowledge and experience to contextualize the insights. The lack of contextualization may prove costly. It is the fusion of AI expertise with domain expertise that underpins the capability that allow for the delivery of customized insights at scale. Many companies do not have the domain expertise in the markets they wish to enter. For example, Apple has made no secret of its desire to move into the healthcare industry. Its CEO, Tim Cook, recently described the healthcare industry as an area where the company could make a “meaningful impact” for humanity. But Apple lacks the domain expertise – it knows very little about the healthcare business. If it is to succeed in this new market, it will need the help of partners that have the necessary expertise. For example, it has recently partnered with Stanford Medicine to conduct a study with over 400,000 participants from all 50 states in the span of 8 months.

Alphabet’s Waymo faced a similar challenge. There is no doubt that Alphabet’s Waymo has a pool of the best AI scientists in its autonomous vehicles’ projects. But to exploit its technology in the car industry, it needs domain expertise. Its recently departed leader, John Krafcik, is a former auto industry executive, who brought his industry experience into the equation. In addition, Mr. Krafcik has put the company on a path of increasing partnerships with auto companies such as Daimler and FCA to help them merge their huge data pool and cloud/AI capabilities with industry expertise.

Netflix is another example of a company that uses industry experts to help it. Insights from viewing behaviors are not simply translated into new production scripts. Such insights are used as input factors to identify trending patterns but then the probing and the production of the scripts is delegated to script writers: AI cannot yet write scripts. Traditional television players such as the BBC found it increasingly more difficult to compete because the likes of Apple, Amazon, and Netflix were hiring script writers and content producers at sky-high salaries, at a level unseen in the industry.

Blending AI and domain expertise is crucial, yet not easy. After he left GE, Jeff Immelt candidly admitted that what he got wrong was thinking that the success of Predix was only based on technology and AI. The integration of GE Digital into GE Energy and the embedding of AI capabilities in the new companies – GE aviation and GE Healthcare – show once again the importance of fusing AI and domain expertise to generate useful insights from data. This hard-learned lesson was recently summarized in GE’s new mantra: “to understand the insights hidden in data: process engineers don’t have to be a data scientist.” This was a point also emphasized by Giordano Albertazzi, President of Europe, Middle East, and Africa at Vertiv, a critical infrastructure provider. He said:

2.3 The Capability to Manage Ecosystem “Partners” in the Digital Context

The blending of AI and domain expertise helps to develop bespoke insights, but to succeed in the new market, the company needs to deliver on its value proposition. For this to happen, many other capabilities are needed, and the company needs to decide on how to acquire or develop these additional capabilities. Fortunately, the digital context has made this task a little easier for potential new entrants: there are now many more potential partners with whom they can partner to execute their strategies. There is no question that digital connectivity has provided companies with many more partners to work with in order to enter and compete in new markets (Jacobides, Cennamo, and Gawer, Reference Jacobides, Cennamo and Gawer2018).

For example, in an effort to become a software industry champion and boost adoption of Predix, GE launched a Global Digital Alliance Program in 2016 which was “ … dedicated to building the digital industrial ecosystem across global systems integrators, independent software vendors, telecommunications service providers and technology providers.” According to a GE top executive: “We were able to sign 464 partners in a single year. The ecosystem (strategy advisors, software integrators, software houses, hardware companies, etc) literally jumped at us. Predix had the potential to build a multibillion-dollar market for all and they all wanted a piece of the pie.”

Axel Springer, the German media company, now operates a very successful portfolio of online digital classified businesses, including the job portal StepStone and property sites such as Immonet. A senior executive at Axel Springer acknowledged that they have been able to enter into these new markets without having to build all these different businesses. As he put it: “We do not have to be the providers of these products. We simply have to find the best providers and connect them to customers through our platform.”

It is therefore noncontroversial to propose that companies today have a wide choice of potential partners to work with in the new markets they enter. This, however, has given rise to a new challenge for firms: not all of these partners contribute equally to the relationship. This is important because different types of partners require different approaches from the digital diversifier. How can a firm determine what type of partner is needed and what kind of challenges it will face in managing each type?Footnote ¹

The answer to this question will be greatly influenced by two key factors:

• the partner’s contribution in offering data in context to the new entrant; and

• the partner’s contribution in helping the new market entrant compete in an entirely new market.

Leveraging of data in context is critical for digital diversifiers. As argued already, data in context is not just any data – it is, instead, data that captures people or asset behaviors in contexts relevant to the market where a firm may consider entering. It is this data that gives the new entrant unique insights on what to offer the customers in the new market that even the established players in that market do not possess. A potential partner can add enormous value to the entrant by providing it with data in context that complements what the entrant already possesses. However, the data in context provided by a partner can vary in importance from being marginal to being of essence to the value proposition of the new entrant. For example, the data that Apple’s app developers offer to Apple are of peripheral value to Apple. By contrast, the data that Predix’s clients provide GE are absolutely essential to what GE Predix is trying to do – without this data, Predix has little valuable to offer.

The second area where a partner can add value is in helping the digital diversifier operate in the new market at scale, quickly and efficiently. For example, the partner can develop and manufacture the products that the entrant intends to offer, sparing the entrant the need to develop its own manufacturing operations and capabilities. The partner may also contribute in other ways by, for example, offering distribution and marketing services to the entrant or providing quick access to a large customer base. As was the case with data in context, the value that a partner brings to the operations of the new entrant can vary in importance from limited to significant. For example, farmers may provide John Deere with valuable data for it to develop new services but apart from that, they are not involved in how John Deere operates. By contrast, the car companies that partner with Waymo in the driverless car market will help Waymo not only with valuable data but also with the manufacture and distribution of these cars.

Putting these two factors together provides us with a classification scheme for what kind of partners each new entrant may need. This is shown in Table 2. The two dimensions of the table are the two areas that the partner can contribute to the entrant – data in context (on the vertical axis) and contributions in creating value in the new market (on the horizontal axis). The contributions in each of these areas can go from Low to High. This produces the 2X2 matrix shown in Table 2.

In the upper-left quadrant, we have partners that can be labeled as satellites – the data they offer to the new entrant is marginal and so is the support they can offer to the operations of the entrant. Apple’s developers can be classified as such. In the upper-right quadrant, we have partners that can be called complementors – they offer limited or marginal data in context, but they provide significant support to the new entrant in running its operations in the new market. The script writers for Netflix or the insurance companies that will help Tesla develop and sell customized insurance policies to Tesla drivers are examples of complementors. Salesforce’s ecosystem provides another example. If we make 100 the total value delivered through Salesforce’s platform, complementors such as system integrators and specialized app developers contribute a significant part of this value, some say up to 80 percent (with the remaining 20 percent contributed by Salesforce itself).

In the bottom-left quadrant, we have partners that can be called suppliers – they provide valuable data in context but little else. The UK Meteorological Office (Met office) is a good example of such a supplier, offering its meteorological data to consultancies to develop services for their customers. Finally, in the bottom-right quadrant, we have partners that can be called strategic partners – they provide both data in context that is of critical importance to the new entrant but also services and support that the diversifier needs to operate in the new market. The car companies partnering with Waymo can be classified as such.

This classification shows that there are different types of “partners” with whom a digital diversifier can cooperate with and that treating them all as one and the same will be a mistake. Some of the recent work on digital ecosystems has failed to make the simple, yet important, point, that not all partners are the same. Although leveraging digital ecosystems is a common thread for all new market entrants, the argument here is that successful market entrants need to be clear about the role that their ecosystem partners have in their value proposition and differentiate their partner strategy accordingly. As Jose Salas, Global Head Strategic Alliances & Partnerships at Wolters Kluwer, commented: “ … everybody uses the word partner, but they mean very different things. Be clear about what you need a partner for before devising your partnership management approach.” Being able to identify the right partners is the key to a constructive working relationship and can be the difference between success and failure in new market entry.

2.4 New Management Capabilities Needed for Managing Ecosystem Partnerships

Digital ecosystems help digital diversifiers enter and compete in entirely new markets but create problems of their own. Given the plethora and diversity of partners available, firms can collaborate with an array of companies, not all of which contribute equally to the relationship. This implies that the diversifier must first understand what types of partners it is building relationships with and then develop customized strategies for each. To make matters worse, the digital context has made the management of these partnerships even more complex and challenging than those in the pre-digital era. The additional challenges created by the digital context highlight the need to broaden and elevate partnership management capabilities in all firms. The skills that worked well in a pre-digital world may not be sufficient in the new context.

3.1 Unrelated Diversifiers Need to Avoid Head-on Competition

One of the many insights that have emerged from the academic literature on diversification over the past forty years has to do with the strategy that a firm should utilize to improve the odds of success of entry in a new market. It is now generally accepted that the probability of entering a new market successfully is higher if the new entrant adopts a strategy that allows it to avoid attacking the established players in that (new) market head-on (Christensen, Reference Christensen1997; Geroski, Reference Geroski1991; Porter, Reference Porter2008). For example, Michael Porter (Reference Porter2008: 514) proposed that “… the cardinal rule in offensive strategy is not to attack head-on with an imitative strategy, regardless of the challenger’s resources or staying power.” Similarly, Kim and Mauborgne (Reference Kim and Renee2014: 4) argued that “The only way to beat the competition is to stop trying to beat the competition.” And Christensen’s theory of disruptive innovation (Reference Christensen1997) is based on the premise that to establish a foothold in a new market, entrants need to play the odds and avoid head-on competition with better-resourced incumbents. Examples of companies that support this proposition include IKEA, Canon, Amazon, EasyJet, Enterprise, Netflix, and Skype, among many others.

The rationale for this becomes apparent when we consider the academic findings on the track record of companies that attempted entry into new markets (e.g., Audretsch, Reference Audretsch1995; Geroski, Reference Geroski1991, Reference Geroski1995). One of these findings is that most new entrants fail within a short period of entry. For example, Geroski (Reference Geroski1991) reported that about 5–10 percent of new entrants fail within a year of entry, 20–30 percent fail within two years, and about 50 percent fail within five years of entry. Similarly, Dunne, Roberts, and Samuelson (Reference Dunne, Roberts and Samuelson1989) found that 64 percent of their sample of new US firms had exited within five years of entry while a full 79 percent had stopped trading ten years after entry. The failure rate of new entrants is so high that the new entry process has been compared to using a revolving door – new entrants are like people going through a revolving door, exiting the room as soon as they enter it (Audretsch, Reference Audretsch1995).

A second finding from this literature is that most entrants imitate the strategies of the incumbents in the new markets they enter. For example, Geroski (Reference Geroski1991) estimated that imitative entry accounts for 90 percent of all entry, with only 10 percent of new entrants adopting innovative strategies. It is not surprising, therefore, that most new entrants fail or that it takes a long time for the profitability of new entrants to be equal to that of the established players (Biggadike, 1979).

The high failure rate of new entrants is a reflection of the fact that when they enter a new market, they have to fight it out with incumbents in that market who enjoy first-mover advantages. Unless the new entrants utilize an innovative strategy to attack, their chances of success are limited. Since we know that 90 percent of all entrants use imitative entry to attack, the high failure rate is not a surprise (Geroski, Reference Geroski1991: 230). This implies that the probability of success in entering a new market is increased if the entrant adopts an innovative strategy.

Even though these insights were developed in the pre-digital era, there is no reason to believe that they are not equally applicable in the digital era. Several high-profile digital examples support this argument. For example, in 1999 Amazon launched an auction site to compete with eBay. It failed because its strategy tried to duplicate eBay and its successful strategy (Kim and Mauborgne, Reference Kim and Mauborgne2011). Similarly, Google has withdrawn from many markets where they could not offer products or services different enough to avoid head-on competition. Google +, Google Wallet Card, and Google News and Weather are just some examples on a very long list of failed ventures. Microsoft had an equally miserable experience when it tried to diversify into the search business, attacking Google. The following quote from a May 2009 Forbes.com article captures the essence of the argument well (Barret, Reference Barrett2009):

All these examples show that the need to enter a new market on the back of a differentiated business model is as relevant in the digital era as it was forty years ago. What is new in the digital context is the increased number and wider variety of strategies or business models that companies can utilize to enter new markets. This is where the digital context has made a difference.

3.2 The Digital Context Has Led to the Emergence of New Business Models

A business model has been defined as a system of interrelated activities that a firm puts together to execute its high-level strategy (Afuah, Reference Afuah2003; Amit and Zott, Reference Amit and Zott2020; Zott and Amit, Reference Zott and Amit2010; Zott, Amit, and Massa, Reference Zott, Amit and Massa2011). These activities can be grouped into three main categories: who to target as customers, what products and services to offer these customers, and how to serve these customers, that is, what value chain activities to put in place to allow the firm to deliver value to customers (Abell, Reference Abell1980; Markides, Reference Markides2008).

Business model innovation has been defined as the discovery of a new Who-What-How combination in an industry (Markides, Reference Markides2023). To qualify as a new business model, it has to be more than just new to a firm – it has to be new to the world or new to the industry. Similarly, it has to be more than the discovery of new activities that make up the system. As Teece (Reference Teece2010) suggested, to qualify as a new business model, it will have to be a new architecture that links the various activities of the system together. In this sense, we can have a new business model even if none of the existing activities in the system change – it can be a system that combines either the existing activities or a set of different activities in a new way (Markides, Reference Markides2023).

The business model concept became very popular in the last twenty years, yet the emergence of new business models is not a new phenomenon. IKEA introduced a new business model in the retailing business in the 1950s and Nucor did the same in the steel business in the 1960s as did Dell in the personal computer industry in the 1970s. However, the emergence of digital technologies has had a profound effect on how business model choices are made and how they are implemented. At its most basic, the digital context has led to a significant increase in the number of viable options available for each activity in the system and that, in turn, has led to an increase in the number of combinations these activities can create when they are put together. Put another way, digital technologies have led to a proliferation of viable business models or strategies available to any given firm (Lanzolla and Markides, Reference Markides2021; Menz et al., Reference Menz, Kunisch and Birkinshaw2021). Whereas in the pre-digital era, a firm could have the choice of a limited number of viable strategies to choose from, the digital era has expanded the number (and variety) of viable choices enormously.

This simple fact alone has enormous implications for Strategy – for example, potential entrants now have a wealth of strategic options to attack incumbents, increasing their probability of success. Similarly, new sources of competitive advantage have been created – a firm can now enjoy a competitive advantage that is not based on its position in an industry or the strength of its internal resources but it’s based instead on the business model it has developed and “how it plays the game” in its industry (Lanzolla and Markides, Reference Markides2021). The proliferation of business models has also affected how value is created and distributed in a given industry and has led to the erosion of traditional sources of competitive advantage. There are other implications that flow from the simple fact that digital technologies have led to a proliferation of viable strategies and business models in any given industry, but for the purposes of this manuscript, we will focus on what this means for digital diversifiers.

For a firm considering diversifying into a new market, the availability of many more business models to choose from increases the probability that the firm will avoid head-on competition with established incumbents. This, in turn, improves the probability of successfully diversifying into new markets, something that encourages companies to undertake investments that even twenty years ago would have been considered unattractive.

Perhaps the most obvious new business model of the digital era is the platform business model (Cusumano et al., Reference Cusumano2019; Marshall et al., Reference Marshall, Mueck and Shockley2015). Several companies, both de novo entrants, such as Uber and Airbnb, and diversifying established companies, such as Apple and Axel Springer, have utilized it to enter new markets successfully. A classic example is Apple’s successful entry into the mobile phone market. In 2007, Apple was a computer manufacturer when it decided to enter the mobile phone market, in the process attacking huge competitors such as Nokia, Samsung, and Motorola. It looked like an uneven battle but the fight was short-lived – by 2015, the iPhone accounted for almost all of the profits in the industry (Marshall et al., Reference Marshall, Mueck and Shockley2015). A big reason for its success was its application store – a platform that connected app developers with iPhone users. By exploiting the power of the platform strategy and in the process leveraging network effects, Apple was able to neutralize the incumbents’ huge first-mover advantages which included massive scale, huge R&D budgets, and trusted brands.

A similar example is the German media company Axel Springer. As part of its digital transformation strategy, the company has moved into the online classified job and real estate markets through portals such as StepStone Group, AVIV Group, and Immonet. These platforms have helped Axel Springer become a market leader in the job and real estate markets in a number of European markets. Other companies that have used the platform strategy to attack bigger competitors in established industries include Amazon in book selling, Airbnb in short- and long-term accommodation, Ebay in retailing, Tinder in dating, Etsy in retailing of handmade or vintage items, and Uber in taxis.

3.3 The Strategy of Personalization at Scale

While it should be obvious that aspiring diversifiers now have a number of potentially attractive business models to exploit in entering new markets, there is one business model that seems perfectly suited to leverage and exploit the new realities of the digital age – such as the availability of data, the drastically reduced costs of searching, analyzing, storing, and sharing information and the ease and speed with which huge number of people could now be reached and served, wherever in the world they may be located (Foss, Reference Foss2005). This is the business model of personalization at scale, a model that allows for the provision of personalized products or services at volume and at mass market prices (Arora et al., Reference Arora, Dreze and Ghose2008). As its name suggests, this is a business model that can offer differentiation and low cost at the same time, something that, as Porter (Reference Porter1980) suggested long time ago, is particularly difficult to pull off.

A good example of this is provided by GE’s Predix unit (Lal, Rajiv, and Johnson, Reference Lal and Johnson2017; Weber, Reference Weber2017). This was created by GE to provide tailored and personalized services to its clients in an effort to achieve the goal set by the then CEO Jeff Immelt of making GE a top ten software company within ten years. According to one of the senior executives who was part of this effort: “Predix as a strategy was brilliant: to make the GE assets installed around the world (worth trillions of dollars) more productive through bespoke predictive maintenance and optimization services … what was not to like in it?” The effort eventually failed, but the strategy underpinning it was clear – offer personalized services at affordable prices. In a similar vein, Vodafone has spent years to develop a pet tracker service which, through cellular tracking of pets, could develop personalized services for pet owners, including daily activity-based food plans. And Alix Partners, the consulting firm, has developed a database of cases that allow for personalized diagnostics of their clients.

Another example is provided by Netflix when it moved from content streaming into content production (Allen et al., Reference Allen, Feils and Disbrow2014; Amatriain, Reference Amatriain2013; Gibbons, Reference Gibbons2019; Hadida et al., Reference Hadida, Lampel, Walls and Joshi2021; Richin, Reference Richin2022; Taylor, Reference Taylor2013). There are many legacy companies that are in content production, yet most of them are struggling to respond to Netflix. What gives Netflix a competitive advantage over legacy companies such as the BBC or Disney is the fact that Netflix has the customer data and the geographic footprint that allows it to produce and deliver content in ways that existing content providers cannot match. Collecting and clustering viewing data has allowed Netflix to infer viewers’ tastes and preferences. As their co-founder Reed Hastings told Forbes (Chmielewski, Reference Chmielewski2020): “We fundamentally want to be better at creating stories people want to talk about and watch than any of our competitors.” More importantly, once it develops movies based on these insights, it has the footprint to distribute these movies on a global scale. As the CEO of Netflix Ted Sarandos told investors in April 2020 (Berman, Reference Berman2020): “One thing that’s not widely understood is that we work really far out relative to the industry, because we launch all our shows, all episodes, at once. And we’re working far out all over the world.”

Ping An in China offers another compelling example (Yu et al., Reference Yu, Feng and Lavoie Orlick2020). By pioneering the “Internet + AI + 1,000 in-house physicians” service model, Ping An Good Doctor collaborates with offline pharmacies and hospitals to provide users with 24/7, one-stop medical healthcare services. This approach aims to address diverse and personalized healthcare needs, deliver authentic medical care, and build user trust. Building on this foundation, Ping An Good Doctor introduced an innovative strategic product, Private Doctor of Ping An Good Doctor, in August 2019, forming partnerships with twenty-nine global industry leaders. This service offers users personalized, one-on-one online consultations, round-the-clock medical support, and additional features such as an e-health profile and a tailored healthcare management plan. Doctors can monitor users’ health over the long term, fostering trust and closer relationships. Through these strengthened connections, users benefit from more precise, consistent, and personalized healthcare services.

Another example is Tesla’s entry in the auto insurance industry (Mullaney, Reference Mullaney2022; Orsoni, Reference Orsoni2022; Zarifis, Reference Zarifis2020; Zawacki, Reference Zawacki2021). Legacy insurance companies do not understand how to price risk for autonomous cars. This results in premiums that would be unsustainable for customers. In 2019, Tesla launched its own insurance program, starting with availability in California, where the automaker claims up to 30 percent cheaper premiums than the competition because of its ability to nail down precisely the risk profile of every and each driver. Finally, Apple is utilizing this strategy in its efforts to enter the healthcare industry (Balasubramanian, Reference Balasubramanian2022; Farr, Reference Farr2020; Gurman, Reference Gurman2022; Nellis, Reference Nellis2022; Samanta, Reference Samanta2021). Apple is building a business model that revolves around (hyper) personalized and always-available advice to people about their health. This is a business model significantly different from the asynchronous and based-on-averages approach more common in legacy national health systems.

3.4 The Underlying Capabilities Needed

Personalization at scale allows new entrants to differentiate themselves relative to the legacy incumbents, something that enhances their chances for success. It is a strategy that – unsurprisingly – became popular at the beginning of the twenty-first century, with the advent of digital technologies (Piller, Reference Piller2004). This is not surprising because the successful implementation of this strategy is dependent on the three organizational capabilities identified in the previous section, all three being derivatives of the digital context: the capability to collect flows of data in context; the capability to derive value-creating insights out of this data through the fusion of AI and domain expertise; and the capability to develop customized partner strategies for the many and diverse partners in the firm’s ecosystem.

The three capabilities identified here represent a significant departure from earlier frameworks about personalization at scale. While in the past the hurdles in implementing personalization at scale were mostly linked to operational flexibility and logistics (Arora et al., Reference Arora, Dreze and Ghose2008; Zipkin, Reference Zipkin2001), our discussion here suggests that the key to successful implementation of personalization at scale in a digital context is the capability to integrate knowledge and technology quickly and efficiently. Data, AI and ecosystems allow firms to overcome most of the legacy operational trade-offs, but these potential benefits remain untapped without the appropriate capabilities. It is perhaps not surprising that most of the successful cases of personalization at scale have been implemented by technology companies. Another way of saying this is that only a subset of companies will possess the capabilities needed for adopting the strategy of personalization at scale – and these are likely to be the firms “born digital,” such as Amazon, Google, Alibaba, and Meta. These are the firms that will diversify the most by exploiting this strategy. By contrast, most other firms that do not possess the requisite capabilities will refrain from diversification altogether. The empirical evidence on diversification in the last twenty years shown in Section 1 seems to support this assertion.

The ability to execute strategies of personalization at scale does not imply that customers will be willing to adopt any personalized products/services. Other researchers (Kumar et al., Reference Kumar, Rajan, Venkatesan and Lecinski2019) found that artificial intelligence is aiding personalized engagement marketing and they proposed that consumers are ready for (Kumar et al., Reference Kumar, Rajan, Venkatesan and Lecinski2019): “a new journey in which AI is a tool for endless options and information that are narrowed and curated in a personalized way.” While this seems to be the trend, the pace at which – and the extent to which – personalization at scale strategies might be appealing for potential customers is perhaps an empirical question. What is obvious is that customers make several trade-offs when considering adoption of such products and services. Trade-offs include the extent to which they are willing to trade on their privacy to get more personal service/product and the extent to which personalized products add to their self-worth vs. impacting their external/social projects. Nevertheless, when designed and executed right, personalization at scale strategies offer a formidable way to enter new markets without facing head-to-head competition.

Finally, a note of caution. The fact that digital technology has allowed for the emergence of the strategy of personalization at scale does not mean that all personalized services will be valued by clients to such an extent that a sizeable market will be created. On the one hand, consumers and industrial companies make trade-offs between privacy and personalized services and what they finally decide will be contingent on several factors, both cultural and economic. On the other hand, the market for a given personalized service might simply not be big enough to pay for the cost of delivering on the strategy. The case of Sky UK refraining from entering the market for personalized weather forecasts is just one of several examples. Overall, any considerations on adopting the strategy of personalization at scale to diversify into a new market should build on realistic assumptions on the market attractiveness and market size of the personalized service being considered.

4.1 New Challenges for the Corporate Center

The digital context has brought to the fore new problems that the diversified firm needs to worry about as it attempts to manage its units. Four such problems are particularly strategic in nature: (i) customer data governance, especially now that customer data is increasingly being leveraged for diversification and is also increasingly being shared outside the boundaries of the firm with ecosystem partners; (ii) the appropriate use of AI in making decisions, ensuring that final critical decisions are always made by human beings, despite the technological superiority of AI; (iii) managing the risks that emerge by having multiple and different ecosystem partners to work with; and (iv) identifying mechanisms to ignite and scale network effects by leveraging business units that are operationally independent.

4.2 The Role of the Corporate Center in Digital Diversifiers: Making the Case for the CM-form

So far, we have discussed the new challenges that the digital environment has created for the corporate center of a diversified firm. Given these new challenges, we will now propose that compared to pre-digital diversifiers, the corporate center of digital diversifiers must assume a more active and more interventionist role in managing business units. There are three reasons why we propose this change.

First, digital diversifiers may appear like unrelated diversifiers of old, but they are not. Their separate units are connected together by a common asset – data – and, often, by the enactment of network effects, and this makes them more like related diversifiers than unrelated ones. This implies that the center must become active in managing the synergies among these “related” units, something that requires a shift in the organizational structure of the diversified firm from the M-form to the CM-form (Williamson, Reference Williamson1975).

Second, the types of synergies that digital diversifiers need to exploit – namely data-enabled information synergies and client-side synergies – require both a unified strategy for technology development, data management, and digital infrastructure and a more active management of such synergies themselves. This basically means that the need for exercising central control and identifying strategic directions is now of a magnitude that is much higher than before – even when compared to the need for exercising central control in related diversifiers in the pre-digital era.

Finally, the institutional vacuum around data and AI creates significant legitimacy risks for digital diversifiers. This implies that digital diversifiers must centralize controls on the use of data and AI. The need for these controls – and the associated advantages for digital diversifiers – will decrease as regulation becomes more prescriptive thus reducing the legitimacy risks.

The need for adopting the CM-form organizational structure follows directly from the literature on related diversification. Pre-digital unrelated diversifiers had adopted the multidivisional organizational structure (Chandler, Reference Chandler1962) and operated primarily as passive holding companies (Chandler, Reference Chandler1962; Rumelt, Reference Rumelt1974; Williamson, Reference Williamson1975). This essentially meant that the corporate center exercised strong strategic and financial control over the divisions but did not interfere in their operating decisions, nor did it try to actively coordinate the activities of the various units – it simply operated a competitive internal capital market which, for various reasons, was assumed to be more efficient than the external capital market. In Williamson’s (Reference Williamson1975) terminology this was the M-form organizational structure which was perfectly appropriate for unrelated diversifiers. Chandler (Reference Chandler1962) documented the rise of this organizational structure in the twentieth century, in conjunction with the rise of the diversification strategy.

Today’s digital diversifiers are increasingly moving into markets that seem totally unrelated to their core businesses, so one would assume that the M-form structure might also be the most appropriate organizational arrangement for them. However, despite appearances, digital diversifiers cannot be considered unrelated diversifiers. Underpinning many of their diversification moves is at least one common asset (i.e., data) and very often a second one as well – technology and web infrastructure. For example, Amazon’s online shopping operations appear to have no relationship with its operations in the streaming of movies (Amazon Prime), but contrary to appearances, this is not the case. Thanks to its dominance in online shopping, Amazon has a goldmine of data on customer behaviors that can be utilized by its advertising business. And increasingly this is the case. For example, Amazon has now started showing commercials to viewers in several countries such as the United States, Britain, Canada, and Germany. Viewers who want to opt out must pay a fee, something that has been estimated to add around $3 billion to Amazon’s ad sales in 2024. Its access to customer data from its online shopping operations has helped Amazon’s ads business become the world’s third largest after Alphabet and Meta. Similarly, Airbnb transformed itself from a hospitality company to a travel company by adding real-world experiences to its offering. This might seem like an unrelated diversification move, but it’s not. Airbnb is leveraging the same underlying technology, web infrastructure, and data flows to access users. Amazon and Airbnb are just two examples of a more general point: seemingly unrelated diversifiers are actually more related than they look.

This implies that today’s seemingly unrelated diversifiers are very often related diversifiers whose units share assets that are just different from those shared by related diversifiers of the past. Granted, the relatedness stems from different assets than before, but it would be a mistake to ignore it or underestimate it. This, in turn, implies that a more appropriate organizational structure for today’s digital diversifiers will be the centralized M-form structure (i.e., the CM-form structure in Williamson’s (Reference Williamson1975) terminology). This is because the CM-form structure allows the corporate center to not only exercise strong strategic and financial control over the units but, more importantly, to get involved in the operating decisions of the SBUs. This is vital because the center needs to be active in setting governance rules and in transferring skills and competences across units and in helping the units exploit interrelationships and synergies. It would be a mistake if the corporate center assumed a passive role (as conglomerates of old did), or if it gave the units total operational autonomy. Doing so will lead to dysfunctional behaviors, and the under-exploitation of the benefits of digital diversification.

Recent literature supports and complements our key tenet that the CM-form structure is the most effective way of organizing for digital diversifiers. First, extant literature argues that centralization enables more efficient exploitation of data-enabled information synergies. For example, Anjos and Fracassi (Reference Anjos and Fracassi2015) showed that conglomerates that are more centralized relative to specialized firms create more value than specialized firms. Second, as discussed earlier, for digital diversifiers, client-side synergies are often the key driver of cross-divisional extra value creation (see also Mawdsley and Somaya, Reference Mawdsley and Somaya2018). In this vein, Tarziján and Snihur (Reference Tarzijan and Snihur2024), synthesizing existing research on managing cross-selling and indirect network effects in multisided platforms (MSPs) such as Uber and Cornershop, theorize that the resulting increasing complexity requires higher levels of centralization.

4.3 A CM-form Organizational Structure That Is Different from Before

The CM-form structure that digital diversifiers need to adopt will be different from the CM-form structure that pre-digital diversifiers had adopted in two ways: (i) the degree of centralization and (ii) the types of synergies being exploited and the integration mechanisms being used to exploit these synergies.

5.1 Managing the New Challenges

Challenge #1: Balancing more centralization for efficient exploitation of synergies between the different businesses in the portfolio with more decentralized decision-making at the business level.

As we already argued in the previous section, digital diversifiers need a more active corporate center to fully and efficiently exploit synergies between the various businesses in their portfolio. However, the need for more central control comes at the same time when more decentralized decision-making (necessary in today’s volatile external context) is much more possible as a result of the introduction of digital technology in the workplace which has allowed for enhanced access to information, more real-time collaboration between people in different locations and more decentralized decision-making by individuals and teams lower down in the organizational hierarchy (Garicano, Reference Garicano2000; Hanelt et al., Reference Hanelt, Bohnsack, Marz and Antunes Marante2021; Menz et al., Reference Menz, Kunisch and Birkinshaw2021; Schafheitle et al., Reference Schafheitle, Weibel and Ebert2020). How can this tension be managed, and how can the diversified firm achieve a balance between centralization and decentralization?

One possible way is to put in place boundary-setting parameters to guide decision-making (Ancona and Isaacs, Reference Ancona and Isaacs2019). This is not a new idea and the Strategy literature has already identified several parameters or guardrails that can be employed (Ancona and Isaacs, Reference Ancona and Isaacs2019; Eisenhardt and Sull, Reference Eisenhardt and Sull2001; Hamel, Reference Hamel2011; Mankins and Garton, Reference Mankins and Garton2017; Stone, Deci, and Ryan, Reference Stone, Deci and Ryan2009). The question for us is: “are there any guardrails that are particularly suited to the digital diversifiers of today?”

Given today’s fast-changing environment, a good guardrail to use will be the company’s clear and clearly communicated Strategy choices in three areas – the customers it will target (and those it will not), the products and services it will offer (and the ones it will not), and the value chain activities it will utilize to compete in its markets (Ancona and Isaacs, Reference Ancona and Isaacs2019; Hamel, Reference Hamel2011; Markides, Reference Markides2000). The idea is that employees will have autonomy to make any decisions that do not move the organization away from these Strategy choices but will not have autonomy to make decisions that deviate the company from its stated Strategy choices. In other words, employees will have autonomy to make decisions on things that improve the firm’s operations but will not have autonomy to make decisions on things that will change the firm’s strategic direction.

This sounds simple enough but survey after survey reports that the majority of employees do not know their organization’s strategy choices (e.g., Devinney, Reference Devinney2013). This may be because many companies fail to make the difficult choices that strategy demands (Vermeulen, Reference Vermeulen2017) or because they fail to communicate them clearly to their employees (Northcraft and Neale, Reference Northcraft and Neale1986). Whatever the reason, the result is that employees do not seem to know their organization’s strategy or its strategic choices. This is unfortunate because without knowing the choices, it will be dangerous to give them autonomy.

A second way to manage the tension between centralization and decentralization is to change the role that middle managers play in guiding people in the organization. For example, Girod and Kralik (Reference Girod and Králik2021) found that self-steering teams would not deliver on their objectives if middle managers were not actively involved in guiding and supporting them. Similarly, Annosi and Lanzolla (Reference Annosi and Lanzolla2023) showed that higher levels of autonomy and engagement are accompanied by more governance and a leadership style from line managers which leverages normative and cognitive actions and a new social contract. In seeking to align autonomous teams to firm strategy, a key challenge for managers is acquiring information from their teams. Not surprisingly, autonomous teams are protective of their autonomy and can see attempts to collect information as an intrusion. Annosi and Lanzolla (Reference Annosi and Lanzolla2023) showed that line managers overcame this problem by creating a new social contract with their teams. Granted, teams value their autonomy, but they also understand that they need to build a better understanding of “what’s going on” in the organization. Leveraging this need – that is, that even autonomous teams need to feel part of a bigger picture – middle managers can position themselves as brokers of both organizational connection and strategic information for teams, in exchange for the teams’ benevolence in sharing information about themselves.

Challenge #2: Faster AI-enabled decision-making versus the need to keep humans in the loop: the role of augmented strategic controls and higher stakeholder engagement.

As already pointed out, digital technologies have revolutionized organizational control, enabling unprecedented levels of digitization, and datafication in managing workplace behaviors and processes. Scholars have explored how these advancements are reshaping organizations, influencing both the structure of work and the mechanisms of control (Bailey et al., Reference Bailey, Leonardi and Barley2019; Benkler, Reference Benkler2007; Cameron, Reference Cameron2022; Cameron and Rahman, Reference Cameron and Rahman2022; Cardinal et al., Reference Cardinal, Kreutzer and Miller2017; Castells, Reference Castells2000; Edwards, Reference Edwards1978; Huber, Reference Huber2003; Kellogg et al., Reference Kellogg, Valentine and Christin2020; Parent-Rocheleau and Parker, Reference Parent-Rocheleau and Parker2022; Rhymer, Reference Rhymer2023; Thompson, Reference Thompson1989; Van Maanen and Barley, Reference Van Maanen and Barley1984; Zuboff, Reference Zuboff1988, Reference Zuboff2019).

In this context, the rise of AI within organizations has taken central stage and scholars have even coined concepts such as “algo controls” to describe the growing role of algorithms in directing and monitoring employee behavior. While algo controls can standardize processes and enhance efficiency, they come with significant risks related to employee autonomy, privacy, bias, transparency, and overall workplace culture (Kellogg, Valentine, and Christin, Reference Kellogg, Valentine and Christin2020; Lindebaum, Vesa, and den Hond, Reference Lindebaum, Vesa and Den Hond2020; Minbaeva, Reference Minbaeva2021; Murray, Rhymer, and Sirmon, Reference Murray, Rhymer and David2021; Parent-Rocheleau and Parker, Reference Parent-Rocheleau and Parker2022; Raisch and Krakowski, Reference Raisch and Krakowski2021; Tambe, Cappelli, and Yakubovich, Reference Tambe, Cappelli and Yakubovich2019). One challenge for managers is how to balance the use of these technologies with the human aspects of work to avoid unintended negative consequences. For instance, Amazon’s use of algorithmic management in its warehouses, where AI is used to monitor worker productivity in real time, exemplifies the power and pitfalls of these systems. While this approach has driven efficiency gains, it has also drawn criticism for fostering a high-pressure work environment, where workers feel under constant surveillance and where stress levels are elevated. This highlights the need for careful consideration of how such technologies impact employee well-being and organizational culture.

A second critical feature of managing in the digital context is the high visibility of employee actions and the immediate feedback loops that technologies enable. The effects of extreme transparency, where every action is visible and instantly recorded, are not always straightforward (Bernstein, Reference Bernstein2017). For example, Roberts (Reference Roberts2009) argued that transparency, rather than merely shedding light on behaviors, can actively reshape them – sometimes in unexpected ways. Employees aware that they are being monitored might alter their behavior to align with what they believe is expected, potentially stifling creativity or leading to superficial compliance rather than genuine engagement. Similarly, Bernstein (Reference Bernstein2012), through an in-depth study at one of the world’s largest mobile phone factories in China, identified what he termed the “transparency paradox.” His research revealed that while increased observability of workers might seem to enhance performance, it often led to efforts by employees to obscure their actions, thereby reducing overall productivity. This paradox underscores a key challenge in digital management: the need to balance visibility with autonomy. Bernstein’s study further found that when zones of privacy were introduced, allowing workers to operate without constant oversight, their performance actually improved.

In summary, the introduction of AI and algorithmic controls is a double-edged sword. On one hand, these technologies offer powerful tools for enhancing efficiency, ensuring compliance, and optimizing processes. On the other hand, they bring new challenges that require managers to be vigilant about the broader impact on employee behavior and organizational culture. Navigating such challenges requires a significant shift in traditional corporate governance structures. Filatotchev and Lanzolla (Reference Filatotchev and Lanzolla2023) suggested a move toward an “open source” governance model, which emphasizes adaptability, inclusivity, and strategic foresight to better manage the complexities of the digital age.

Central to their approach is the idea of fostering greater stakeholder engagement and decentralizing decision-making. Unlike traditional hierarchical governance, this model involves collaboration with a wide range of stakeholders, such as employees, customers, and regulators, to create governance practices that are more responsive to the rapidly changing digital environment. This inclusive approach enables organizations to better identify and address emerging challenges and opportunities. Balancing short-term compliance with long-term legitimacy is another key recommendation. In a world where regulations are constantly evolving, organizations must not only adhere to current laws but also proactively build trust and credibility. This forward-looking approach to corporate governance allows organizations to adapt more smoothly to future regulatory changes and societal expectations.

The focus on strategic controls is also crucial in the digital context. Traditional governance often prioritized financial controls and short-term metrics, but Filatotchev and Lanzolla (Reference Filatotchev and Lanzolla2023) argued for a shift toward controls that emphasize innovation, adaptability, and the effective management of digital resources. Aligning governance with long-term strategic goals supports sustained growth and competitiveness in the digital era. Decentralizing decision-making is essential to responding to the speed and complexity of digital markets. By empowering local teams to make decisions, organizations can respond more quickly and effectively to new opportunities and challenges, while still maintaining alignment with overall strategic objectives. Building digital capabilities at the board level is another important strategy. Organizations should ensure that their leadership includes individuals with a deep understanding of digital technologies and their implications. This digital literacy at the top ensures that governance decisions are informed by the latest technological insights and are aligned with the needs of a rapidly changing environment.

Effective engagement with polycentric regulatory environments is also critical. As organizations operate across multiple jurisdictions with varying regulations, flexible governance structures are needed to navigate this complexity. This proactive regulatory engagement helps ensure compliance and supports innovation. Finally, addressing the ethical risks associated with digital transformation is paramount. Issues such as data privacy, algorithmic bias, and surveillance can pose significant risks to an organization’s reputation and trust. Incorporating strong ethical guidelines into governance practices helps organizations maintain compliance with legal standards and uphold societal values.

Challenge #3: More behavioral visibility leading to more accountable management versus more visibility leading to gaming behaviors.

Behavioral visibility has become increasingly important as organizations undergo digitization, digitalization, and datafication (Albu and Flyverbom, Reference Albu and Flyverbom2016; Bernstein, Reference Bernstein2012). This concept refers to the degree to which individuals’ actions and behaviors within an organization can be observed, thanks to the widespread use of digital technologies. These technologies, through enhanced connectivity and data collection, enable the monitoring, analysis, and influence of behaviors on an unprecedented scale. The concept of behavioral visibility is rooted in earlier scholarly work on visibility and transparency in organizations. For instance, Leonardi (Reference Leonardi2014) introduced “communication visibility,” which describes how social media platforms make individual communications within organizations more visible, thereby affecting knowledge sharing and innovation. Similarly, Flyverbom et al. (Reference Flyverbom, Hemmingsen and Hansen2016) explored the “management of visibilities” in the digital era, emphasizing how visibility is crucial for organizational control and coordination.

Behavioral visibility goes beyond merely observing actions; it also encompasses the power dynamics that arise from this increased visibility. Bernstein (Reference Bernstein2012) highlighted the “transparency paradox,” where greater visibility can lead to unintended consequences such as reduced privacy and increased control over employees. Additionally, foundational theories by Goffman (Reference Goffman1959) on self-presentation and Brighenti (Reference Brighenti2007) on visibility as a social category offer insights into how individuals manage their visibility within organizational contexts. These theories suggest that visibility is not just a passive condition but something actively managed by both individuals and organizations. Complementing these insights, more recent literature rooted in organizational behavior and information systems points to the emergence of active management of visibility by employees. In this vein, instead of fostering transparency and desired behaviors, the increased visibility can prompt some employees to engage in deceptive practices to manipulate how they are perceived by others. Consequently, top executives receive misleading feedback, which undermines the effectiveness of management in high behavioral visibility environments.

Given the tensions created by behavioral visibility – such as the potential for reduced privacy, increased control, and shifting power dynamics – management must adapt in several key ways. Scholars like Albu and Flyverbom (Reference Albu and Flyverbom2016) discuss the challenges of managing visibility, emphasizing the need to balance the benefits of transparency with the protection of individual autonomy and privacy. As such, it is crucial for management to foster a culture of trust where visibility is used to support employees rather than to control them. This involves clearly communicating how data and visibility will be used and ensuring that privacy is respected. Second, insights based on the cases analyzed in this Element – in Sections 2 and 3 – suggest that management should balance transparency with autonomy by setting clear guidelines that protect employees’ rights while still enabling the organization to benefit from increased visibility. This can involve establishing ethical frameworks for data use and ensuring that visibility does not undermine individual agency and lead to work-related stress.

Third, and perhaps even more crucially, behavioral visibility implies that top management too must increase consistency in the way they lead given that any inconsistencies are immediately visible to everyone in the organization, thus potentially triggering diverging behaviors. This is even more important for managers of diversified firms where the variance in contexts, businesses, and institutional environments amplifies the risk of triggering diverging behaviors. Perhaps in this vein, it might help if leaders could embrace a more inclusive, responsive style that facilitates open communication and collective problem-solving.

5.2 Do We Need New Managers?

In addition to the challenge of managing ecosystem partners discussed in Section 2, in this section, we have outlined three other challenges facing managers of diversified companies and proposed potential strategies for addressing them. But are managers truly prepared to tackle these challenges? How must leadership styles and skills adapt to effectively manage the complexities of a digitally diversified organization, especially when balancing central control with decentralized decision-making and maintaining consistency in a highly transparent environment? What new knowledge and competencies should managers prioritize to lead successfully in the digital era, and how can we equip future leaders to navigate the intricacies of AI, big data, and heightened behavioral visibility? These and related questions should be central to a renewed academic focus on (re)defining management for digital diversifiers and beyond.

6.1 Will History Repeat Itself?

Whether the current diversification boom by certain (mostly technology) firms like Amazon, Google, and Apple will continue and grow in the future (or burst like it did in the 1960s) will depend primarily on how successful these early pioneers of digital diversification are with their diversification. If they are successful, then it’s reasonable to expect the boom to continue, with other firms rushing in (Abrahamson and Rosenkopf, Reference Abrahamson and Rosenkopf1993; DiMaggio and Powell, Reference DiMaggio and Powell1983; Lanzolla and Suarez. Reference Lanzolla and Suarez2012; Meyer and Rowan, Reference Meyer and Rowan1977). If, on the other hand, these pioneers of digital diversification run into problems, then pressures to refocus will intensify and we are likely to see a repeat of history.

The question, therefore, that we must address is: “what will influence the success of these digital diversifying pioneers?” In this manuscript, we have emphasized the importance of one set of factors, all of them firm-specific. First, digital diversifiers must excel in collecting constant flows of data in context, then blending AI with domain expertise to generate value-creating insights out of this data, and finally managing a plethora of new and diverse partners within their ecosystems to succeed in the new markets they are entering. Second, they must strategically avoid direct competition with incumbents when entering new markets, often by deploying the strategy of personalization at scale. Third, they must put in place an active corporate center to exploit synergies among their units, and they must employ appropriate integrating mechanisms to leverage these synergies, especially client-side synergies. Fourth, they must apply new management priorities, including balancing centralization with autonomy, integrating AI-driven efficiencies with human oversight, and addressing the tensions that arise from increased organizational transparency – tensions that can lead to suboptimal behaviors and undermine organizational resilience.

Doing all these things and doing them in a self-reinforcing way will increase a firm’s chances of success. But this is not enough. The strategic management literature has further argued that a company’s ability to sustain its competitive advantage also relies on creating barriers to imitation (Porter, Reference Porter1980) by internalizing resources and capabilities into structured organizations (Barney, Reference Barney1991), and by establishing a web of self-reinforcing, opaque activities (Lanzolla and Markides, Reference Lanzolla and Markides2022). This means that each factor identified in this manuscript as important for success should not be viewed in isolation but as part of a broader, integrated strategy where each factor strengthens and amplifies the others. For instance, the ability to collect and analyze vast amounts of data effectively will enhance personalization at scale strategies, allowing the firm to better meet customer needs and avoid direct competition with incumbents. At the same time, an active corporate center that efficiently exploits synergies among units can enhance the firm’s ability to manage diverse partnerships within its ecosystem, creating a feedback loop where data-driven insights lead to more effective partnerships, which in turn generate even more valuable data. Furthermore, balancing centralization with autonomy will enable units to operate with agility while still aligning with overarching corporate goals. These self-reinforcing dynamics would allow digital diversifiers to continuously adapt and evolve, making them more resilient and better positioned to capitalize on opportunities in digital markets. Thus, the sustainability of digital diversifiers’ advantages depends on how well and how quickly they integrate these firm-specific factors to make imitation more difficult.

Although we have emphasized the importance of firm-specific factors in determining the success of digital diversifiers, there is no question that there is another set of factors, external to the firm, that also play a big role. Two external factors in particular stand out: the weakness or, in some cases, the total absence of regulatory frameworks around digital markets (e.g., Birkinshaw, Reference Birkinshaw2024) leading to what academics have called an institutional void (e.g., Khanna and Palepu, Reference Khanna and Palepu1997); and investor attitudes and expectations that have so far been positive toward diversification investments, something that has resulted in abundant capital being provided to diversifying firms (e.g., Filatotchev, Lanzolla, and Syrigos, Reference Filatotchev, Lanzolla and Syrigos2025). As Sorkin (Reference Sorkin2017) commented: “… investors are seemingly willing to give these behemoths a free pass in the name of growth and innovation.”

Lack of proper regulation has created an environment where “everything goes,” enabling firms to experiment with new business models and markets, and often take positional advantages. Birkinshaw (Reference Birkinshaw2024: 1) argued that “… some [firms] have finessed regulations by identifying gaps and inconsistencies they could take advantage of, some have sidestepped regulations by arguing that those regulations do not apply to their situation, and others have nullified those rules by denying their existence altogether.” For instance, Uber leveraged gaps in traditional taxi regulations to rapidly expand its ride-hailing service; Amazon capitalized on the lack of e-commerce rules to dominate online retail before regulations caught up; Jumia addressed Africa’s logistical and financial service voids by building its own infrastructure and payment systems; Tencent filled China’s payment infrastructure gap with WeChat Pay, embedding itself deeply in everyday transactions; and M-Pesa revolutionized financial services in Kenya by creating mobile money solutions in the absence of widespread banking access. Needless to say that these institutional-based positional advantages are clearly dependent on changes in the regulatory environment (e.g., Lanzolla and Frankort, Reference Lanzolla and Frankort2016).

Digital diversifiers have also benefited from positive investor expectations. When investors are optimistic about a firm’s potential to succeed in new markets, they are more willing to provide the capital needed to finance such expansions. In this sense, positive investor attitudes act as an enabler to diversification. Filatotchev, Lanzolla and Syrigos (Reference Filatotchev, Lanzolla and Syrigos2025), building on a systematic study of S&P 500 companies in the period 2006–2020, showed that investors paid a premium over their peers to companies whose CEOs were signaling an orientation toward exploiting digital technology in their strategy. Higher premiums or, more broadly, broader access to financial resources might have created an environment favorable for experimenting with new business models and expanding into new markets.

This discussion suggests that whether the digital diversifiers of today will succeed with their diversification investments will depend on these internal and external sets of factors and how they evolve over time. First, will the diversifiers be successful in doing all the firm-specific things we identified earlier and will they be successful in doing them well and in protecting them from imitation? Second, will the lax regulatory environment and positive investor attitudes continue to prevail in the market or would the outside environment turn sour on digital diversifiers?

Putting these two factors together provides us with a classification scheme that helps us answer the question: “will the digital diversifiers of today be successful with their diversification investments?” This is shown in Table 7. The two dimensions of the table are the two sets of factors identified earlier: (i) will the diversifiers be successful in doing the firm-specific things discussed in this manuscript well (horizontal axis), and (ii) will the regulatory environment and investor attitudes change for the worse or will they remain supportive of diversification as they are today (vertical axis)? This produces the 2X2 matrix shown in Table 7.

In the upper-left quadrant (A), the external environment remains lax and benign but the firms fail to make good use of the firm-specific factors necessary for success. The likelihood of sustained success is uncertain and in the balance. By contrast, in the upper-right quadrant (B), the environment remains benign and the firms succeed in doing all the firm-specific things we identified earlier, as well as doing them well, and protecting them from imitation. The firms will likely be very successful with their diversification.

In the bottom-right quadrant C, the external environment changes for the worse but the firms succeed in making good use of the firm-specific factors necessary for success. In all likelihood, firms will continue to enjoy success with their diversification but not to the same extent as in quadrant B. Finally, in the bottom-left quadrant (D), the external environment changes for the worse and the firms fail to make good use of the firm-specific factors necessary for success. Digital diversifiers will most probably struggle with their diversifications and their financial results will be bad.

What does this tell us about the question of interest to us, which is: “will history repeat itself?” The future of digital diversification hinges on which of these four scenarios plays out. If the conditions described in Quadrant B prevail, the current wave of digital diversification will continue and may even accelerate, attracting new aspiring digital diversifiers. Conversely, if the conditions in Quadrant D prevail, we may see a repeat of the 1960s, where diversification strategies faced significant challenges and firms ended up de-diversifying. In either of the remaining two scenarios, the outcome remains uncertain and in the balance.

It remains to be seen which of these outcomes materializes. But rather than attempt to make a prediction or argue about the likely outcome, we believe that future research should focus on developing theory further as well as empirically testing some of the ideas and propositions made in this manuscript. For example, we have argued that digital diversification does not alter the traditional economic logic of diversification. Further research, both theoretical and empirical, might be required to test this thesis. As the context becomes more digitized and digital technologies expand their scope and depth of application, researchers may want to explore if new internal factors – beyond the ones described in this manuscript – might impact the economics of diversification, thus, for instance, changing the relationship between diversification and performance from an inverted U-shaped to an S-shaped curve, as proposed by Aversa and Hueller (Reference Aversa and Hueller2023).

Similarly, additional research is necessary to explore and test whether – as claimed in this manuscript – new types of synergies (such as client-based synergies) have emerged to supplement the traditional synergies associated with diversification. For example, Jacobides et al. (Reference Jacobides, Ma and Romanenkov2023a) argued that ecosystem synergies will be critical for digital diversifiers. How do these synergies interact with one another? Are there optimal configurations, perhaps even self-reinforcing? Or will new tensions emerge? More broadly, will the new technologies help with managing the inevitable tensions among synergies, or will complexity undermine any advantage?

Along with economic, strategic, and organizational considerations, we also believe that research should (re)consider the role of management in the diversified firm. In this manuscript, we have brought forward some new critical managerial tensions. Additional work is needed to understand the many implications for leadership and management that modern digital diversifiers present. In summary, a lot of work remains to be done for us to develop a more comprehensive understanding of the digital diversification phenomenon. Yet, we believe that the research and ideas presented in this manuscript are a robust foundation to build upon and develop our understanding of the phenomenon further.