DIALOGUE

对话

Artificial Intelligence, Data-Driven Learning, and the Decentralized Structure of Platform Ecosystems

人工智能、数据驱动学习与平台生态系统的去中心化结构

Gregory, Henfridsson, Kaganer, and Kyriakou (2021) highlight the important role of data and AI as strategic resources that platforms may use to enhance user value. However, their article overlooks a significant conceptualdistinction:theinstalledbaseofdecentralizedusers who connect with a platform lie outside the boundaries of the platform-owning firm, whereas the accumulated data derived from that installed base exists internal to the boundaries of the firm and under firm control. Accounting for this distinction brings forth two key departures from their theory. First, the decentralized structure of a platform ecosystem makes value capture bytheplatform an essential consideration when analyzing the implications of data-driven learning for users. Because AI and data allow a platform to increase the share of value the platform owner captures from the users, the value perceived by users can often decline as the user base grows. Second, as an internal asset of the platform firm, data from users and complementors exhibits different dynamics compared with the dynamics that govern the installed base itself. As a result, the quantity and quality of the platform’s stock of data are only loosely coupled with the size of the platform’s installed base. We highlight the strategic implications ofthis distinction for a manager launching a new multisided platform. 格雷戈里、亨弗里德松、卡加内尔和基里亚库（2021）强调了数据和人工智能作为战略资源的重要作用，平台可能会利用这些资源来提升用户价值。然而，他们的文章忽略了一个重要的概念区别：与平台相连的去中心化用户的既有基础（installed base）位于平台所属公司的边界之外，而从该既有基础中积累的数据则存在于公司边界之内并受公司控制。考虑到这一区别，其理论会出现两个关键偏差。首先，平台生态系统的去中心化结构使得在分析数据驱动型学习对用户的影响时，平台对价值的捕获成为一个关键考量因素。由于人工智能和数据使平台能够增加平台所有者从用户那里捕获的价值份额，随着用户基数的增长，用户感知到的价值往往会下降。其次，作为平台公司的内部资产，来自用户和互补者的数据表现出与管理既有基础本身的动态不同的动态。因此，平台数据存量的数量和质量仅与平台既有基础的规模松散关联。我们强调这一区别对启动新多边平台的管理者的战略意义。

We read with great interest the article “The Role of Artificial Intelligence and Data Network Effects for Creating User Value” by Gregory, Henfridsson, Kaganer, and Kyriakou (2021), and agree with the authors that data is a strategic resource that enhances platform value and, as such, data-driven learning deserves greater attention within platform strategy research. Their article touches on many of the important strategic considerations that affect how platform owners can use AI capabilities and data to create value. However, the article overlooks a key conceptual difference that distinguishes the installed base of users and complementors on a platform from the data derived from that installed base. The users and complementors who connect with a platform lie outside the boundaries of the platform-owning firm, and the platform ecosystem has a decentralized governance structure. In contrast, the data accumulated by the firm is internal to the boundaries of the firm and centrally controlled by the firm. 我们饶有兴致地阅读了Gregory、Henfridsson、Kaganer和Kyriakou（2021）发表的文章《人工智能与数据网络效应在创造用户价值中的作用》，并认同作者的观点，即数据是一种提升平台价值的战略资源，因此，数据驱动的学习在平台战略研究中值得更多关注。他们的文章涉及了许多影响平台所有者如何利用人工智能能力和数据创造价值的重要战略考量。然而，该文章忽视了一个关键的概念差异，即平台上的用户和互补者的安装基础（installed base）与从该安装基础中衍生的数据之间的区别。与平台建立连接的用户和互补者位于平台拥有企业的边界之外，且平台生态系统采用分散式治理结构。相比之下，企业积累的数据则位于企业边界之内，并由企业集中控制。

We highlight two consequences of this essential but overlooked distinction. First, the decentralized structure of a platform ecosystem makes value capture— not just value creation—an essential consideration when analyzing an innovation ecosystem (Adner & Kapoor, 2010; Brandenburger & Nalebuff, 1996). We discuss how AI and data allow a platform to increase the share of value the platform owner captures from the ecosystem, such that the value perceived by users declines. Second, as an internal asset of the firm, data from users and complementors exhibits different dynamics compared with the dynamics that govern the installed base itself. As a result, the quantity and quality of the platform’s stock of data are only loosely coupled with the size of the platform’s installed base. Collectively, our arguments constitute an important refinement of the theory put forward by Gregory et al. (2021). 我们强调了这一本质但被忽视的区别所带来的两个后果。首先，平台生态系统的去中心化结构使得价值捕获——而不仅仅是价值创造——成为分析创新生态系统时的一个关键考量因素（Adner & Kapoor, 2010；Brandenburger & Nalebuff, 1996）。我们将讨论人工智能和数据如何使平台能够增加其从生态系统中捕获的价值份额，从而导致用户感知的价值下降。其次，作为企业的内部资产，来自用户和互补者的数据表现出与控制安装基础本身的动态不同的动态特性。因此，平台数据存量的数量和质量仅与平台安装基础的规模松散关联。总的来说，我们的论点构成了对Gregory等人（2021）提出的理论的重要改进。

DATA-DRIVEN VALUE CAPTURE

数据驱动的价值捕获

Gregory et al. (2021) focus their theory on how datadriven learning increases value creation for users. While we agree that value creation deserves attention, we believe this focus misses a key strategic implication of digitization: specifically, that the impact of data-driven learning on value capture is as large as, or larger than, its impact on value creation. The outcome variable in the Gregory et al. (2021) framework is “perceived user value,” and the framework predicts that AI capabilities and data increase perceived user value. But, any analysis of the value perceived by a user, we suggest, must inherently take value capture into account. Decentralized users and complementors lie outside the boundaries of the platform-owning firm, creating tensions between the value they capture anhevaluheapture.First, plaor cptures a larger share of value—say, by increasing price—the user will perceive less ofthe total value created. Second, value capture by the platform can also have a negative effect on total value created if the mechanism of value capture reduces the intrinsic quality orinstalledbase ofthe platform. In this section, we show that AI capabilities and data allow a platform owner to increase the share of value they capture from interactions with platform users. 格雷戈里等人（2021）的理论聚焦于数据驱动型学习如何提升用户的价值创造。虽然我们认同价值创造值得关注，但我们认为这种聚焦忽略了数字化的一个关键战略含义：具体而言，数据驱动型学习对价值捕获的影响与对价值创造的影响相当，甚至更大。格雷戈里等人（2021）框架中的结果变量是“感知用户价值”，该框架预测人工智能能力和数据会提升感知用户价值。但我们认为，任何对用户感知价值的分析都必须固有地考虑价值捕获。去中心化的用户和互补者处于平台拥有方的边界之外，这在他们捕获的价值与平台捕获的价值之间造成了张力。首先，如果平台捕获更大份额的价值（例如通过提高价格），用户将感知到更少的总创造价值。其次，如果价值捕获机制降低了平台的内在质量或安装基数，平台的价值捕获也可能对总创造价值产生负面影响。在本节中，我们将展示人工智能能力和数据如何使平台所有者能够增加从与平台用户互动中捕获的价值份额。

Capturing Value with User-Level Price Discrimination

以用户级价格歧视捕获价值

A basicresult in economics is that a seller can capture more value from its customers when the seller is able to price discriminate. If a seller can only charge the same price to all customers, the customers with higher willingness to pay gain a greater consumer surplus from the transaction. Two factors limit a firm’s ability to price discriminate: (1) how well it can categorize specific consumers as members of a certain segment and (2) how well it ascertains the willingness to pay of each segment (e.g., Varian, 1989). For example, cinemas price discriminate by offering lower prices to students only when a customer’s status as a student can be verified with a piece of ID. 经济学中的一个基本结论是，当卖家能够实施价格歧视时，卖家可以从客户那里获取更多价值。如果卖家只能向所有客户收取相同的价格，那么支付意愿更高的客户将从交易中获得更大的消费者剩余。限制企业实施价格歧视能力的两个因素是：（1）企业对特定消费者进行特定群体分类的能力，以及（2）企业确定每个群体支付意愿的能力（例如，Varian，1989）。例如，电影院通过向学生提供更低的价格来实施价格歧视，但前提是客户的学生身份可以通过身份证件进行验证。

By using accumulated data, a platform can overcome both factors that limit price discrimination. Platforms identify and track specific users through several methods, such as user account logins, device identifiers, web browser cookies, and IP addresses (Athey, Calvano, & Gans, 2016; D’Annunzio & Russo, 2020). Platforms create detailed databases that allow for the granular categorization of a specific customer along numerous dimensions. Platforms can then estimate the customer’s willingness to pay for a good or service based on the platform’s accumulated data on their general activity (e.g., web-browsing history), past transactions, and A/B tests conducted on similar customers (Shiller, 2020; Thomke, 2020). This allows the platform to customize the prices displayed to a given user in ways that extract value from the user—namely, individual-level price discrimination, a practice that is well established in the pricing of flights (Sengupta & Wiggins, 2014) and e-commerce retail (Garbarino & Lee, 2003). 通过使用积累的数据，平台可以克服限制价格歧视的两个因素。平台通过多种方法识别和追踪特定用户，例如用户账户登录、设备标识符、网页浏览器Cookie和IP地址（Athey, Calvano, & Gans, 2016；D’Annunzio & Russo, 2020）。平台创建详细的数据库，允许从多个维度对特定客户进行精细分类。然后，平台可以基于其积累的关于用户一般活动（例如网页浏览历史）、过往交易以及针对类似客户进行的A/B测试的数据，估算客户对商品或服务的支付意愿（Shiller, 2020；Thomke, 2020）。这使得平台能够以从用户身上提取价值的方式，定制显示给特定用户的价格——即个人层面的价格歧视，这种做法在机票定价（Sengupta & Wiggins, 2014）和电子商务零售（Garbarino & Lee, 2003）中已得到广泛应用。

Data also allows the platform to optimize its value capture based on the user’s present psychological state. Facebook told its advertisers that its data and algorithms can identify “moments when young people need a confidence boost” and target ads to teenagers when they feel “worthless,” “insecure,” “defeated, “anxious,” useless,” stupid," “over whelmed, “stressed,” and “a failure” (Reilly, 2017). Perhaps the optimal (in the purely economic sense) time to capture value from users is when they are at their lowest, which some might interpret as opportunistic or predatory. 数据还使平台能够根据用户当前的心理状态优化其价值捕获。Facebook 告诉其广告商，其数据和算法可以识别“年轻人需要信心提升的时刻”，并在青少年感到“无价值”、“不安全”、“挫败”、“焦虑”、“无用”、“愚蠢”、“不知所措”、“压力大”和“失败”时向他们定向投放广告（Reilly，2017）。也许从用户那里获取价值的最佳（从纯粹经济意义上讲）时机是他们处于最低谷时，这可能会被一些人解读为投机或掠夺性的行为。

Value Capture at the Expense of Value Creation

以牺牲价值创造为代价的价值捕获

It is common practice to design platforms in ways that capture value for the platform owner at the expense of the total value being created (Cusumano, Gawer, & Yoffie, 2019; Parker, Van Alstyne, & Choudary, 2016; Zhu & Liu, 2018); this design choice impacts how platforms use the data they collect. Consider that many platforms deliver services to users at a price of zero and generate revenue by showing users advertisements (Ghose & Yang, 2009). Here, the platform owner has an incentive to create value for users in order to attract them to the platform and retain them as recurring users of the platform. But the platform’s goal should not be misconstrued: the user is valuable to the platform because they pay attention to the advertisements posted on it. A decision that may reduce the value a user captures, while providing more value to advertisers, is something the platform would rationally enact, even if some users leave because of the advertisements, thereby reducing the total value created by the platform ecosystem. 设计平台时，以牺牲总创造价值为代价来获取平台所有者价值是常见做法（Cusumano, Gawer, & Yoffie, 2019; Parker, Van Alstyne, & Choudary, 2016; Zhu & Liu, 2018）；这种设计选择会影响平台如何使用所收集的数据。许多平台以零价格向用户提供服务，并通过向用户展示广告来创收（Ghose & Yang, 2009）。在此，平台所有者有动力为用户创造价值，以吸引用户并将其保留为平台的重复用户。但不应误解平台的目标：用户对平台有价值，是因为他们关注平台上发布的广告。一项可能减少用户获取价值、同时为广告商提供更多价值的决策，平台会理性地实施，即使一些用户因广告而离开，从而降低平台生态系统创造的总价值。

A platform has an incentive to manipulate the psychological state ofits own users to keep them addicted to the platform. App developers write and share playbooks of tips and techniques for retaining users by taking advantage of psychological reward mechanisms (Zichermann & Cunningham, 2011). For example, some video games with freemium business models surreptitiously let players win more often immediately after the player makes a monetary in-game purchase, known as a “microtransaction”; this conditions the user to make repeated purchases in order to maintain the flow of dopamine-inducing victories (Drummond & Sauer, 2018; King & Delfabbro, 2018). Recently, there has been controversy around in-game purchases of “lootboxes” that go further and introduce a gambling-like element of chance to the experience, facilitating gambling addiction (Zendle & Cairns, 2018) and drawing the attention ofregulators (Griffiths, 2018). 一个平台有动机操纵自身用户的心理状态，以让他们对平台上瘾。应用程序开发者编写并分享了利用心理奖励机制来留住用户的技巧和方法手册（Zichermann & Cunningham，2011）。例如，一些采用免费增值商业模式的电子游戏，会在玩家进行游戏内小额货币购买（即“微交易”）后，偷偷让玩家更容易获胜，这种做法会让用户为了维持多巴胺诱导的胜利而重复购买（Drummond & Sauer，2018；King & Delfabbro，2018）。最近，关于游戏内“开箱”（lootboxes）购买的争议不断，这类购买进一步引入了类似赌博的随机元素，助长了赌博成瘾（Zendle & Cairns，2018），并引起了监管机构的关注（Griffiths，2018）。

As the platform collects more data on how users respond to attempts at manipulation, it learns to manipulate its users more effectively. This data-driven learning may induce the user to spend more time and/or money on the platform, even if the user derives less overall enjoyment from their experience using it. Sometimes, users sense when they are being exploited. In 2017, Electronic Arts (EA) released its game Star Wars Battlefront II. To “unlock” popular characters such as Darth Vader, users either had to play the game for an overwhelming number of hours or pay to purchase the character in-game. EA utilized data on user behavior to dynamically make the hours-of-play requirement unachievable for most users. Many expressed disdain for EA’s monetization strategy: 随着平台收集到更多关于用户如何应对操纵尝试的数据，它会学会更有效地操纵用户。这种数据驱动的学习可能会诱使用户在平台上花费更多时间和/或金钱，即使用户从使用平台的体验中获得的整体乐趣减少。有时，用户会察觉到自己正在被利用。2017年，艺电（EA）发布了游戏《星球大战前线II》。为了“解锁”达斯·维达等热门角色，用户要么必须玩极大量的游戏时长，要么花钱在游戏内购买角色。EA利用用户行为数据，动态地让大多数用户无法达到所需的游戏时长。许多人对EA的盈利策略表示不满：

The truth is [EA] know[s] very few people are going to sink a full work week into this game and you’re hoping that somebody is desperate enough to buy credits to unlock the character. It has nothing to do with providing a “sense of pride and accomplishment.” This is a flat-out lie and [EA] know[s] it. (bookem_danno, 2017) 事实是，[EA]知道很少有人会投入整整一周的工作时间在这个游戏上，你只是希望有人会迫切到愿意购买信用点来解锁角色。这与提供“自豪感和成就感”毫无关系。这是彻头彻尾的谎言，[EA]自己也清楚。(bookem_danno, 2017)

In this case, the situation turned out poorly for EA: user complaints snowballed, earning EA the Guinness World Record for the most “downvoted” comment in the history of Reddit, a popular internet forum (Leskin, 2019). While one lesson is that you should not get between Star Wars fans and Darth Vader, another is that user or complementor backlash against platforms using data against them increasingly occurs in highstakes settings (Zhu & Liu, 2018), such as drivers suing Uber for access to their personal data because they worry Uber uses it to manipulate them into driving longer hours or to cut their pay to a bare sustenance level (Holder, 2019). 在这种情况下，EA（艺电）的处境变得很糟糕：用户投诉愈演愈烈，使EA获得了吉尼斯世界纪录——在知名网络论坛Reddit的历史上，其评论获得了最多的“踩”（Leskin, 2019）。一个教训是，你不应该在《星球大战》粉丝和达斯·维达（Darth Vader）之间制造冲突；另一个教训是，在高风险场景中，用户或利益相关者对利用数据损害自身利益的平台的抵制越来越普遍（Zhu & Liu, 2018）。例如，司机起诉优步（Uber），要求获取其个人数据，因为他们担心优步会利用这些数据操纵他们延长工作时间，或削减他们的工资至仅够维持基本生活的水平（Holder, 2019）。

Ultimately, firms make a strategic choice of whether to use data to improve value creation or to use it for value capture. Data-driven learning implies a (monotonically) increasing relationship between the volume of data and the capacity for both creation and capture. Firms strategically decide whether more creation or capture is realized. 最终，企业会做出一个战略性选择：是利用数据来提升价值创造，还是利用数据来获取价值。数据驱动的学习意味着数据量与创造和获取的能力之间存在（单调）递增的关系。企业会战略性地决定实现更多的创造还是更多的获取。

DECOUPLING OF DATA AND INSTALLED BASE

数据与已安装基数的解耦

Gregory et al. (2021) coin the term “data network effects,” which they describe as “a new category ofnetwork effects.” We agree with the authors that the phenomenon of data-driven learning by platforms deserves careful scholarly attention. However, we question whether referring to data-driven learning as a category ofnetwork effects is accurate. The “network effects” label brings with it an existing set of mental models and strategic frameworks that may not apply to data-driven learning. It is well established that network effects stem from the size and structure of a platform’s installed base (Afuah, 2013; Farrell & Saloner, 1986; Katz& Shapiro, 1985); hence, the label “datanetwork effects” implies there is a tight coupling between data-driven learning and the installed base of users and complementors on a platform. We believe the coupling is weaker than this label suggests. The decentralized structure of a platform ecosystem means that the dynamics—that is, changes in stock over time—of the installed base of users differ from the dynamics of the quantity and quality of platformowned data. As a result, the extent of a platform’s data-driven learning is only weakly coupled with its number of users. Hence, prescriptions derived from treating data-driven learning as a network effect may be misleading. 格雷戈里等人（2021）创造了“数据网络效应”这一术语，他们将其描述为“网络效应的一个新类别”。我们同意作者的观点，即平台的数据驱动学习现象值得学术界的仔细关注。然而，我们质疑将数据驱动学习称为网络效应的一个类别是否准确。“网络效应”这一标签带来了一套现有的心智模型和战略框架，这些模型和框架可能不适用于数据驱动学习。众所周知，网络效应源于平台安装基础的规模和结构（Afuah，2013；Farrell & Saloner，1986；Katz & Shapiro，1985）；因此，“数据网络效应”这一标签暗示数据驱动学习与平台上用户和互补者的安装基础之间存在紧密联系。我们认为这种联系比该标签所暗示的要弱。平台生态系统的去中心化结构意味着用户安装基础的动态变化（即存量随时间的变化）与平台拥有的数据的数量和质量的动态变化不同。因此，平台的数据驱动学习程度与其用户数量之间仅存在弱耦合关系。因此，将数据驱动学习视为网络效应而得出的建议可能具有误导性。

There are two fundamental differences between the dynamics of data-driven learning and the dynamics of direct and indirect network effects. First, data-driven learning is much more path dependent than direct and indirect network effects. The data and behavioral insights that platforms derive from users and complementors accumulate over time, whereas direct and indirect network effects depend on the potential interactions that could take place between users and complementors at any given moment (Gawer, 2014; McIntyre & Srinivasan, 2017). A well-known mental model for direct network effects is the “critical mass” threshold for an installed base: a platform with a user base above that threshold comprises a self-sustaining ecosystem (Afuah & Tucci, 2003), and prospective users might join a platform on the expectation that its installed base will cross the threshold in the future (Fang, Wu, & Clough, 2021). Similar thresholds exist under indirect network effects. Research has yet to establish whether data-driven learning exhibits a similar critical threshold; if such a threshold exists, it would likely be defined by the platform’s total accumulated experience with a set of users—akin to a learning curve—rather than the current or anticipated future size of the user base. 数据驱动学习的动态性与直接和间接网络效应的动态性存在两个根本区别。首先，数据驱动学习比直接和间接网络效应更具路径依赖性。平台从用户和互补者那里获得的数据和行为洞察会随着时间累积，而直接和间接网络效应则取决于在任何特定时刻用户和互补者之间可能发生的潜在互动（Gawer，2014；McIntyre & Srinivasan，2017）。直接网络效应的一个著名心智模型是安装基数的“临界质量”阈值：用户基数超过该阈值的平台构成一个自我维持的生态系统（Afuah & Tucci，2003），潜在用户可能会基于平台的安装基数未来会跨越该阈值的预期而加入平台（Fang，Wu，& Clough，2021）。间接网络效应下也存在类似的阈值。研究尚未确定数据驱动学习是否表现出类似的临界阈值；如果存在这样的阈值，它可能由平台与一组用户的总累积经验定义——类似于学习曲线——而不是当前或预期的未来用户基数规模。

Second—ongoing legal disputes notwithstanding—data is presently treated as a proprietary and tradeable asset of a firm, whereas the installed base of users and complementors on a platform make autonomous decisions to join or leave platforms at will. In the current legal regime, platform users agree to lengthy terms and conditions that most do not read and that, in effect, allow platforms a great deal of leeway in utilizing and owning users’ data. In many cases, those terms and conditions allow the platform to sell that data: one of Twitter’s highest-margin businesses is licensing user data in the form of subscriptions to historical and real-time data on the platform (Bary, 2018). As an asset, data can be purchased from third parties prior to establishing a user base;1 they can also be retained even if users leave the platform. 尽管存在持续的法律纠纷，数据目前仍被视为企业的专有且可交易资产，而平台上的用户和互补者的既有用户群会自主决定是否加入或离开平台。在当前的法律制度下，平台用户同意冗长的条款和条件，而大多数用户并未阅读这些条款，实际上这些条款允许平台在利用和拥有用户数据方面拥有很大的自由度。在许多情况下，这些条款允许平台出售数据：Twitter的高利润业务之一就是以订阅形式许可平台上的历史和实时数据（Bary，2018）。作为一种资产，数据可以在建立用户群之前从第三方购买；1它们即使在用户离开平台后也可以被保留。

To illustrate the strategic implications of these differences, consider the decision facing a manager launching a new multisided platform. If the manager believes that direct and indirect network effects are the key source of competitive advantage, they are likely to adopt a strategy to “get big fast” (e.g., Afuah, 2003; Eisenmann, 2006; Wu, Clough, & Kaletsky, 2019). On the other hand, if the manager believes that data-driven learning will be their key source of advantage, they are likely to prioritize learning over growth (Ries, 2011). Rather than rapidly achieving scale, they may instead spend effort trialing multiple—possibly incomplete—product alternatives on small subsets of users (Ghosh, 2021). They may seek to purchase an existing data set to mine for insights on user behavior, or, if they work for an incumbent firm, capitalize on the firm’s pre-existing data accumulated from its users and complementors. 为了说明这些差异的战略含义，我们来考虑一位管理者在推出新多边平台时面临的决策。如果管理者认为直接和间接网络效应是竞争优势的关键来源，他们可能会采取“快速做大”的策略（例如，Afuah, 2003；Eisenmann, 2006；Wu, Clough, & Kaletsky, 2019）。另一方面，如果管理者认为数据驱动的学习将是其关键优势来源，他们可能会优先考虑学习而非增长（Ries, 2011）。与快速实现规模扩张不同，他们可能会转而在小部分用户中试用多种可能不完整的产品替代方案（Ghosh, 2021）。他们可能会寻求购买现有数据集以挖掘用户行为的见解，或者，如果他们效力于一家现有企业，就利用该企业从用户和互补者那里积累的既有数据。

As Gregory et al. (2021) suggest through their choice of label, data-driven learning and direct and indirect network effects are often positively correlated. Their article provides valuable insights that apply to those situations. We draw attention to instances when the dynamics of data-driven learning differ from the dynamics of direct and indirect network effects. 正如Gregory等人（2021）通过其标签选择所指出的，数据驱动学习与直接和间接网络效应通常呈正相关。他们的文章提供了适用于这些情况的宝贵见解。我们关注数据驱动学习的动态与直接和间接网络效应的动态不同的情况。

CONCLUSION

结论

We are grateful to Gregory et al. (2021) for initiating an important conversation about the role that data and artificial intelligence capabilities play in the creation of value within platform ecosystems. In this dialogue essay, we highlight a defining attribute of platform ecosystems: the decentralized nature of users and complementors. Recognizing this attribute changes some of the theory’s predictions with respect to the value perceived by users and qualifies its assumptions on the extent to which data-driven learning is coupled with network effects. 我们感谢 Gregory 等人（2021）开启了一场关于数据和人工智能能力在平台生态系统价值创造中所扮演角色的重要讨论。在这篇对话性文章中，我们强调了平台生态系统的一个核心特征：用户和互补者的去中心化本质。认识到这一特征会改变一些理论对用户感知价值的预测，并修正其关于数据驱动学习与网络效应耦合程度的假设。

REFERENCES

参考文献

Adner, R., & Kapoor, R. 2010. Value creation in innovation ecosystems: How the structure of technological interdependence affects firm performance in new technology generations. Strategic Management Journal, 31: 306333.
Adner, R. 和 Kapoor, R. 2010. 创新生态系统中的价值创造：技术相互依赖的结构如何影响新技术代际中的企业绩效。《战略管理杂志》，31：306-333。

Afuah, A. 2003. Redefining firm boundaries in the face of the internet: Are firms really shrinking? Academy o Management Review, 28: 3453.
Afuah, A. 2003. Redefining firm boundaries in the face of the internet: Are firms really shrinking? Academy of Management Review, 28: 3453.

Afuah, A. 2013. Are network effects really all about size? The role of structure and conduct. Strategic Management Journal, 34: 257273.
Afuah, A. 2013. 网络效应真的只关乎规模吗？结构与行为的作用。《战略管理杂志》，34：257-273。

Afuah, A., & Tucci, C. L. 2003. Internet business models and strategies: Text and cases (2nd ed.). New York, NY: McGraw-Hill.
阿福阿, A., & 图奇, C. L. 2003. 互联网商业模式与战略：文本与案例（第2版）. 纽约, NY: 麦格劳-希尔.

Athey, S., Calvano, E., & Gans, J. S. 2016. The impact of consumer multi-homing on advertising markets and media competition. Management Science, 64: 15741590.
Athey, S., Calvano, E., & Gans, J. S. 2016. 消费者多归属对广告市场和媒体竞争的影响。《管理科学》, 64: 15741590.

Bary, E. 2018, April 25. Twitter earnings growth relied on selling user data. MarketWatch. Retrieved from https://www.marketwatch.com/story/twitterearnings-growth-relied-on-selling-user-data-2018-04- 25
Bary, E. 2018年4月25日。Twitter的收益增长依赖于出售用户数据。MarketWatch。检索自https://www.marketwatch.com/story/twitterearnings-growth-relied-on-selling-user-data-2018-04- 25

bookem_danno. 2017. Re: Seriously? I paid 80$ to have Vader locked? [Online comment]. Retrieved from https:// www.reddit.com/r/StarWarsBattlefront/comments/ 7cffob/seriously_i_paid_80_to_have_vader_locked
bookem_danno. 2017. Re: 真的吗？我花了80美元把维达锁起来？[在线评论]. 取自 https:// www.reddit.com/r/StarWarsBattlefront/comments/ 7cffob/seriously_i_paid_80_to_have_vader_locked

Brandenburger, A. M., & Nalebuff, B. J. 1996. Co-opetition. New York, NY: Currency Doubleday.
Brandenburger, A. M., & Nalebuff, B. J. 1996. 竞合. 纽约, NY: Currency Doubleday.

Cusumano, M. A., Gawer, A., & Yoffie, D. B. 2019. The business of platforms: Strategy in the age of digital competition, innovation, and power. New York, NY: HarperCollins.
Cusumano, M. A., Gawer, A., & Yoffie, D. B. 2019. The business of platforms: Strategy in the age of digital competition, innovation, and power. New York, NY: HarperCollins.

D’Annunzio, A., & Russo, A. 2020. Ad networks and consumer tracking. Management Science, 66: 4921 5484, iii-iv.
D’Annunzio, A., & Russo, A. 2020. 广告网络与消费者追踪。《管理科学》，66：4921-5484，iii-iv。

Drummond, A., & Sauer, J. D. 2018. Video game loot boxes are psychologically akin to gambling. Nature Human Behaviour, 2: 530532.
德拉蒙德（Drummond, A.）和索尔（Sauer, J. D.），2018年。电子游戏中的战利品箱在心理上类似于赌博。《自然·人类行为》，2（期）：530532。

Eisenmann, T. R. 2006. Internet companies’ growth strategies: Determinants of investment intensity and longterm performance. Strategic Management Journal, 27: 11831204.
Eisenmann, T. R. 2006. 互联网公司的增长战略：投资强度和长期绩效的决定因素。《战略管理杂志》，27：11831204。

Fang, T. P., Wu, A., & Clough, D. R. 2021. Platform diffusion at temporary gatherings: Social coordination and ecosystem emergence. Strategic Management Journal, 42: 233272.
方, T. P., 吴, A., & 克拉夫, D. R. 2021. 临时集会中的平台扩散：社会协调与生态系统涌现。《战略管理杂志》, 42: 233272.

Farrell, J., & Saloner, G. 1986. Installed base and compatibility: Innovation, product preannouncements, and predation. American Economic Review, 76: 940955.
法雷尔，J.，& 萨洛纳，G. 1986. 已安装基数与兼容性：创新、产品预告与掠夺。《美国经济评论》，76: 940-955.

Garbarino, E., & Lee, O. F. 2003. Dynamic pricing in internet retail: Effects on consumer trust. Psychology and Marketing, 20: 495513.
加巴里诺（Garbarino, E.）和李（Lee, O. F.），2003年。互联网零售中的动态定价：对消费者信任的影响。《心理学与市场营销》，20：495-513。

Gawer, A. 2014. Bridging differing perspectives on technological platforms: Toward an integrative framework. Research Policy, 43: 12391249.
Gawer, A. 2014. Bridging differing perspectives on technological platforms: Toward an integrative framework. Research Policy, 43: 12391249.

Ghose, A., & Yang, S. 2009. An empirical analysis of search engine advertising: Sponsored search in electronic markets. Management Science, 55: 16051622.
Ghose, A., & Yang, S. 2009. 搜索引擎广告的实证分析：电子市场中的赞助搜索。《管理科学》，55：16051622。

Ghosh, S. 2021. Think before you act: The unintended consequences of inexpensive business experimentation. Academy of Management Proceedings, 2021. doi: 10.5465/AMBPP.2021.97
Ghosh, S. 2021. 三思而后行：廉价商业实验的意外后果。《管理学会会议论文集》，2021年。doi: 10.5465/AMBPP.2021.97

Gregory, R. W., Henfridsson, O., Kaganer, E., & Kyriakou, H. 2021. The role of artificial intelligence and data network effects for creating user value. Academy of Management Review, 46: 534551.
格雷戈里（Gregory, R. W.）、亨弗里德森（Henfridsson, O.）、卡加内尔（Kaganer, E.）和基里亚库（Kyriakou, H.）. 2021. 人工智能和数据网络效应在创造用户价值中的作用. 《管理学会评论》（Academy of Management Review）, 46: 534551.

Griffiths, M. D. 2018. Is the buying of loot boxes in video games a form of gambling or gaming? Gaming Law Review, 22: 5254.
格里菲斯，M. D. 2018. 电子游戏中购买战利品箱是否属于赌博或游戏？《游戏法律评论》，22：5254。

hiQ Labs, Inc. v. LinkedIn Corp. 2019. U.S. Court of Appeals for the Ninth Circuit, vol. 938 F.3d: 985. (9th Cir. 2019).
hiQ Labs, Inc.诉LinkedIn Corp.案，2019年，美国第九巡回上诉法院，938 F.3d卷：985页（第九巡回法院，2019年）。

Holder, S. 2019, August 26. For ride-hailing drivers, data is power. Bloomberg. Retrieved from https://www. bloomberg.com/news/articles/2019-08-26/citylabdaily-for-ride-hailing-drivers-data-is-power
Holder, S. 2019年8月26日。对于网约车司机而言，数据就是力量。彭博新闻社。取自https://www.bloomberg.com/news/articles/2019-08-26/citylabdaily-for-ride-hailing-drivers-data-is-power

Isaac, M., & Lohr, S. 2017, April 24. Unroll.me service faces backlash over a widespread practice: Selling user data. New York Times. Retrieved from https://www. nytimes.com/2017/04/24/technology/personal-datafirm-slice-unroll-me-backlash-uber.html
艾萨克，M.，& 洛尔，S. 2017年4月24日。Unroll.me服务因一项广泛存在的做法——出售用户数据——遭到反对。《纽约时报》。Retrieved from https://www. nytimes.com/2017/04/24/technology/personal-datafirm-slice-unroll-me-backlash-uber.html

Katz, M. L., & Shapiro, C. 1985. Network externalities, competition, and compatibility. American Economic Review, 75: 424440.
Katz, M. L., & Shapiro, C. 1985. 网络外部性、竞争与兼容性。《美国经济评论》，75：424440。

King, D. L., & Delfabbro, P. H. 2018. Predatory monetization schemes in video games (e.g. “loot boxes”) and internet gaming disorder. Addiction, 113: 19671969.
King, D. L., & Delfabbro, P. H. 2018. 电子游戏中的掠夺性货币化方案（例如“开箱”）与网络游戏障碍。《成瘾》，113：1967-1969。

Leskin, P. 2019, September 9. EA’s comment on a Reddit thread about “Star Wars: Battlefront 2” set a Guinness World Record for the most downvoted comment of all time. Business Insider. Retrieved from https://www. businessinsider.com/reddit-world-record-downvotesea-star-wars-battlefront-2-2019-9
Leskin, P. 2019年9月9日。EA在Reddit上关于《星球大战：前线2》的帖子下的评论创下了有史以来被点踩最多的评论的吉尼斯世界纪录。Business Insider。取自https://www.businessinsider.com/reddit-world-record-downvotesea-star-wars-battlefront-2-2019-9

McIntyre, D. P., & Srinivasan, A. 2017. Networks, platforms, and strategy: Emerging views and next steps. Strategic Management Journal, 38: 141160.
麦金泰尔，D. P.，& 斯里尼瓦桑，A. 2017. 网络、平台与战略：新兴观点与下一步行动。《战略管理杂志》，38：141160。

Neuburger, J. D. 2020, March 12. LinkedIn files petition to the Supreme Court in HiQ web scraping case. National Law Review. Retrieved from https://www. natlawreview.com/article/linkedin-files-petition-tosupreme-court-hiq-web-scraping-case
Neuburger, J. D. 2020年3月12日。LinkedIn在HiQ网页抓取案中向最高法院提交请愿书。《国家法律评论》。取自https://www.natlawreview.com/article/linkedin-files-petition-tosupreme-court-hiq-web-scraping-case

Parker, G. G., Van Alstyne, M. W., & Choudary, S. P. 2016. Platform revolution. New York, NY: W. W. Norton & Company.
Parker, G. G., Van Alstyne, M. W., & Choudary, S. P. 2016. 平台革命. 纽约, NY: W. W. 诺顿公司.

Reilly, M. 2017. Is Facebook targeting ads at sad teens? MIT Technology Review, May 1. Retrieved from https:// www.technologyreview.com/2017/05/01/105987/isfacebook-targeting-ads-at-sad-teens
Reilly, M. 2017. Is Facebook targeting ads at sad teens? MIT Technology Review, May 1. Retrieved from https:// www.technologyreview.com/2017/05/01/105987/isfacebook-targeting-ads-at-sad-teens

Ries, E. 2011. The lean startup: How today’s entrepreneurs use continuous innovation to create radically successful businesses. New York, NY: Crown Business.
里斯，E. 2011. 精益创业：当今企业家如何通过持续创新打造极具颠覆性的成功企业。纽约，纽约州：皇冠商业出版社。

Sengupta, A., & Wiggins, S. N. 2014. Airline pricing, price dispersion, and ticket characteristics on and off the internet. American Economic Journal. Economic Policy, 6: 272307.
森古普塔（A.）和威金斯（S. N.），2014年。航空公司定价、价格分散以及互联网上和线下的机票特征。《美国经济期刊：经济政策》，6：272307。

Shiller, B. R. 2020. Approximating purchase propensities and reservation prices from broad consumer tracking. International Economic Review, 61: 847870.
希勒，B. R. 2020. 从广泛的消费者跟踪中近似购买倾向和保留价格。《国际经济评论》，61: 847870。

Thomke, S. H. 2020. Experimentation works: The surprising power of business experiments. Boston, MA: Harvard Business Press.
Thomke, S. H. 2020. 实验有效：商业实验的惊人力量。马萨诸塞州波士顿：哈佛商业出版社。

Varian, H. R. 1989. Price discrimination. In R. Schmalensee & R. Willig (Eds.), Handbook of industrial organization, vol. 1: 597654. New York, NY: North-Holland.
Varian, H. R. 1989. 价格歧视。载于 R. Schmalensee 与 R. Willig（编），《产业组织手册》，第 1 卷：597-654。纽约，纽约州：North-Holland。

Wu, A., Clough, D. R., & Kaletsky, S. 2019, May. Nascent platform strategy: Overcoming the chicken-or-egg dilemma (Technical note 719-507). Boston, MA: Harvard Business School Press.
吴，A.，克劳夫，D. R.，& 卡莱茨基，S. 2019年5月。新兴平台战略：克服先有鸡还是先有蛋的困境（技术说明719-507）。马萨诸塞州波士顿：哈佛商学院出版社。

Zendle, D., & Cairns, P. 2018. Video game loot boxes are linked to problem gambling: Results of a large-scale survey. PLoS One, 13: e0206767.
Zendle, D. 和 Cairns, P. 2018. 电子游戏开箱与问题赌博相关：大规模调查结果。PLoS One, 13: e0206767.

Zhu, F., & Liu, Q. 2018. Competing with complementors: An empirical look at Amazon.com. Strategic Management Journal, 39: 26182642.
朱, F., & 刘, Q. 2018. 与互补者竞争：亚马逊公司的实证研究。《战略管理杂志》, 39: 2618-2642.

Zichermann, G., & Cunningham, C. 2011. Gamification by design: Implementing game mechanics in web and mobile apps. Sebastopol, CA: O’Reilly Media. 齐赫曼（Zichermann, G.）和坎宁安（Cunningham, C.）. 2011. 游戏化设计：在网页和移动应用中实施游戏机制. 塞瓦斯托波尔，加利福尼亚州：奥莱利媒体公司（O’Reilly Media）.

David R. Clough University of British Columbia David R. Clough 英属哥伦比亚大学

Andy Wu Harvard University https://doi.org/AMR_20200222 Andy Wu 哈佛大学 https://doi.org/AMR_20200222

David R. Clough (david.clough@sauder.ubc.ca) is an assistant professor in the OBHR Division and the Entrepreneurship and Innovation Group at the University of British Columbia’s Sauder School of Business. He received his PhD from INSEAD. His research interests include new venture emergence, organizational learning, and technological change in innovation ecosystems. David R. Clough（david.clough@sauder.ubc.ca）是英属哥伦比亚大学尚德商学院组织行为与人力资源部（OBHR Division）及创业与创新研究组的助理教授。他在欧洲工商管理学院（INSEAD）获得博士学位。其研究兴趣包括新企业的诞生、组织学习以及创新生态系统中的技术变革。

Andy Wu (awu@hbs.edu) is an assistant professor in the Strategy Unit at Harvard Business School. He received a PhD and MS in applied economics from the Wharton School of the University of Pennsylvania, where he is a senior fellow at the Mack Institute for Innovation Management. 吴安迪（awu@hbs.edu）是哈佛商学院战略系的助理教授。他拥有宾夕法尼亚大学沃顿商学院应用经济学博士和硕士学位，同时在沃顿商学院的创新管理麦卡恩研究所担任高级研究员。

Data Network Effects: Key Conditions, Shared Data, and the Data Value Duality

数据网络效应：关键条件、共享数据与数据价值二元性

Clough and Wu (2022) provide an interesting and thought-provoking response to our article (Gregory, Henfridsson, Kaganer, & Kyriakou, 2021) on the role of artificial intelligence (AI) and data network effects for the creation of user value. We welcome the debate around data network effects as a new category of network effects. In this response note, we build upon the points raised by Clough and Wu (2022) to outline three clarifications to our theory of data network effects, concerning (1) conditions under which data network effects accrue, (2) the importance of theorizing shared data, and (3) the model’s ability to explain the cumulative effect of data-driven learning on value creation and value capture. Clough and Wu (2022) 对我们关于人工智能 (AI) 和数据网络效应在创造用户价值中作用的文章 (Gregory, Henfridsson, Kaganer, & Kyriakou, 2021) 给出了一个有趣且发人深省的回应。我们欢迎围绕数据网络效应作为一种新型网络效应展开的讨论。在本回应中，我们基于 Clough 和 Wu (2022) 提出的观点，对我们的数据网络效应理论进行三点澄清，分别涉及（1）数据网络效应产生的条件，（2）理论化共享数据的重要性，以及（3）该模型解释数据驱动学习对价值创造和价值捕获的累积效应的能力。

KEY CONDITIONS FOR DATA NETWORK EFFECTS

数据网络效应的关键条件

Clough and Wu’s (2022) response calls for further clarification about the conditions under which we can refer to “data network effects.” As recently explained by Cennamo (2020), building on the ideas presented by Hagiu and Wright (2020), there are two key conditions for data network effects. The first condition is that learning from one user should translate into a better product or experience for other users, not just that single user. In other words, as more users use the product, the product must improve the experience for all users. The second condition is that the product experience enhancement from learning should happen fast enough to affect the current value of the product. It should benefit its current users, not the next product generation’s users. In other words, the product improves over the consumption lifetime with more users adopting it. Clough和Wu（2022）的回应要求进一步明确我们何时可以提及“数据网络效应”。正如Cennamo（2020）最近基于Hagiu和Wright（2020）提出的观点所解释的，数据网络效应存在两个关键条件。第一个条件是，从一个用户那里获得的学习应该转化为对其他用户而非仅仅单个用户的更好产品或体验。换句话说，随着更多用户使用该产品，产品必须改善所有用户的体验。第二个条件是，从学习中获得的产品体验提升必须足够快，以影响产品的当前价值。它应该惠及当前用户，而不是下一代产品的用户。换句话说，随着更多用户采用该产品，产品在使用生命周期内会不断改进。

Clough and Wu (2022) suggest that the term “data network effects” is misleading, assuming that the two conditions stated above are typically not met. This view, however, overlooks the role of AI that is central to the activation of data network effects. As stated in our original article (Gregory et al., 2021), the widespread use of AI on modern platforms makes a significant contribution to fulfilling the two conditions for data network effects stated above. First, a core characteristic of AI is learning from individual cases to identify and translate patterns into predictive models that feed into the iterative improvement of products and experiences for other users, not just one single user from which data are collected and analyzed. Second, another core characteristic of AI is the ability to efficiently scale data-driven learning and instantly release the resulting improvements to the product experience to affect the current value of the product for each user. While we explained these characteristics of AI in our original article (Gregory et al., 2021), we had not explicitly linked them to the key conditions for network effects, and we thank Clough and Wu (2022), Hagiu and Wright (2020), and Cennamo (2020) for triggering us to add this clarification to the debate. Clough和Wu（2022）认为“数据网络效应”这一术语具有误导性，前提是上述两个条件通常不满足。然而，这一观点忽视了人工智能在激活数据网络效应中核心的作用。正如我们在原文（Gregory等人，2021）中所述，现代平台上人工智能的广泛应用对满足上述数据网络效应的两个条件做出了重大贡献。首先，人工智能的一个核心特征是从个体案例中学习，识别模式并将其转化为预测模型，这些模型不仅用于从单个收集和分析数据的用户，还用于迭代改进其他用户的产品和体验。其次，人工智能的另一个核心特征是能够高效地扩展数据驱动的学习，并立即将由此产生的改进应用到产品体验中，从而影响每个用户当前使用产品的价值。虽然我们在原文（Gregory等人，2021）中解释了人工智能的这些特征，但我们没有明确将它们与网络效应的关键条件联系起来，我们感谢Clough和Wu（2022）、Hagiu和Wright（2020）以及Cennamo（2020）促使我们在这场辩论中补充这一澄清。

DATA: FROM FIRM RESOURCE TO SHARED DATA

数据：从企业资源到共享数据

Another key argument presented by Clough and Wu (2022) as a foundation for their ideas about value creation and capture is that the accumulated data derived from the installed base of decentralized users exists internal to the boundaries of the firm within centralized data structures. This argument resonates well within the realm of the resource-based view of the firm in which resources under control of the firm, including intangible resources such as databases, serve as potential sources of competitive advantage (Hall, 1993). However, data should not be treated as any other type of resource or production factor such as capital, labor, or oil (Parra-Moyano, Schmedders, & Pentland, 2020). We surmise that Clough and Wu’s (2022) argument potentially breaks down when taking into consideration at least three emerging characteristics of data Clough和Wu（2022）提出的另一个作为其关于价值创造和获取观点基础的关键论点是，来自去中心化用户安装基础的累积数据存在于企业内部边界内的集中式数据结构中。这一论点在企业资源基础观（resource - based view of the firm）领域中引起了强烈共鸣，在该领域中，企业控制的资源（包括数据库等无形资源）是竞争优势的潜在来源（Hall，1993）。然而，数据不应被视为任何其他类型的资源或生产要素，如资本、劳动力或石油（Parra - Moyano、Schmedders和Pentland，2020）。我们推测，当考虑到数据至少三个新兴特征时，Clough和Wu（2022）的论点可能会不成立。

Copyright of Academy of Management Review is the property of Academy of Management and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder’s express written permission. However, users may print, download, or email articles for individual use. 《管理学会评论》的版权归管理学会所有，未经版权所有者明确书面许可，其内容不得复制、通过电子邮件发送至多个网站或发布到讨论组。但是，用户可以为个人使用打印、下载或通过电子邮件发送文章。