Artificial Intelligence Augmented Decision-making: A Case Study on Processes and Challenges in Nascent Firms
Metadata only
Author
Show all
Date
2022-06-16Type
- Other Conference Item
ETH Bibliography
yes
Altmetrics
Abstract
The last two decades have witnessed the rapid adoption of AI-augmented decisions in various spheres of business processes (Shrestha et al., 2021). This rapid adoption is facilitated by the ease of curating digital data, advancement of algorithmic methods, and related infrastructures such as GPUs and TPUs (von Krogh, 2018). On the one hand, there is increasing optimism that algorithmic decisions bring diverse benefits to the organization by increasing efficiency in decision-making, improving the accuracy of decisions, and freeing managerial attention for other activities (Brynjolfsson et al., 2011; Provost & Fawcett, 2013; Brynjolfsson & McElheran, 2016a; Shrestha et al., 2019; Stobierski, 2021). On the other hand, we have also witnessed pushbacks from various sources, including popular media and academic commentaries indicating that the traction on algorithmic decision-making is a mere hype and does not bring measurable performance gains (Wilson, 2004; Shieh, 2015; Kiron, 2017; Wamba et al., 2017; Popovič et al., 2018; Mikalef et al., 2019).
Echoing the business interest and public opinion, we have witnessed an escalation of various academic commentaries about the implementation of AI augmentation (Brynjolfsson et al., 2011; Brynjolfsson & McElheran, 2016b). Given the interdisciplinary nature of the problem, these commentaries feature viewpoints and perspectives from an amalgamation of scholars across diverse disciplines (von Krogh, 2018; Rahwan et al., 2019; Tarafdar et al., 2019; Townsend & Hunt,2019; Davenport, 2020; Ma & Sun, 2020; Tinguely et al., 2020; Kundu, 2021). Despite there being numerous conceptual and theoretical research identifying the benefits or challenges facing AI-augmented decision-making (AIADM), there is a lacuna of literature on how these are exactly designed and implemented in practice, the challenges managers face in their adoption, and whether they bring about proposed gains and performance guarantees. A small set of papers that examine this phenomenon empirically relies on data from large incumbent firms or smaller technological firms (Brynjolfsson et al., 2011; Brynjolfsson & McElheran, 2016a; Tarafdar et al., 2019; Senoner et al., 2021; Shrestha et al., 2021). Even though AIADM is expected to generate essential value for small business firms, we do not know how processes and challenges of AI adoption emerge in those small firms and how the phenomenon is the same or distinct as compared to incumbents. First, nascent firms operate under comparatively higher uncertainty as compared to incumbents, thus adding further complication in the adoption of AIADM. Second, as they feature resource constraints and limited manpower, the trade-off between the resources required to adopt AIADM and the benefit they produce by compensating for the lack of human resources becomes crucial (Townsend & Hunt, 2019; Weber et al., 2021).
To examine this research gap, we study a nascent online fashion retailing company that decided to adopt data-driven decision-making to increase sales and the participation of its customers in product co-creation. In our extensive field study, we follow the company's entire deployment process. Our examination of the case is informed by a heterogeneous set of primary data including, interviews with the firm's executive members, data scientists, analysis of their archival data, field notes, experiments in their customer portals, and customer surveys.
We generate novel insights into the entire process that unfold in practice and the challenges managers face. In contrast to the extant literature, which provides a linear and unidirectional view of AI deployment in business decisions, we obtain an iterative model of the process. We identify four phases, Genesis, Onboarding, Auditing, and Sustaining of AIADM. We observe that the outcomes of one phase are influencing the feasibility and/or the outcomes of another phase(s). Hence, the challenges arising in these phases affect other phases creating feedback loops and back-and-forth workflows between the phases we identified. Moreover, when the case company observes success in AIADM initiatives it extends AIADM applications to other different use cases using the same 4-phase process. Our findings suggest that the AI implementation process in nascent firms is a connected, iterative, and complex cycle involving genesis, onboarding, auditing, and sustaining phases. Show more
Publication status
unpublishedEvent
Organisational unit
03719 - von Krogh, Georg / von Krogh, Georg
Funding
197763 - Sustaining Knowledge Creation in Online Communities: Enabling, Creating, and Maintaining (SNF)
Notes
Conference lecture held on June 16, 2022.More
Show all metadata
ETH Bibliography
yes
Altmetrics