Agent-Based Data Curation Practices: Customer Responses to Human versus Algorithmic Data Requesters in Established Business-to-Business Relationships
Martin Adam, Abhay Nath Mishra, Alexander Benlian
Information Systems Research · Tue
Abstract
With the increasing value generated by data curation and the rise of artificial intelligence (AI) agents that converse and act like human agents, vendor companies in established business-to-business (B2B) relationships increasingly delegate data curation tasks to algorithmic data requesters (ADRs) rather than human data requesters (HDRs). Firms employ these data requesters to email existing customers either to collect new information for improved services (data enrichment) or to update outdated information to maintain existing ones (data reconciliation). Despite the growing use of agents in data curation, little is known about how customers respond to these practices—particularly how their responses vary by requester type (HDR versus ADR) and by the nature of the data work practice (enrichment versus reconciliation). Drawing on the effort-accuracy framework and gain-loss message framing, we investigate customer agreement with email-based data requests from HDRs (versus ADRs) in data enrichment (versus data reconciliation) in a multimethod approach. Evidence from a randomized field experiment with a leading European pharmaceutical company followed by an online experiment reveals that customers have a reduced inclination to agree (versus disagree) with a data request from an HDR (versus ADR) in data enrichment because of their preference for minimizing effort in interactions with the data requester. However, for data reconciliation, customers prefer an HDR (versus ADR) because they have lower concerns about errors that could arise in these interactions. A post hoc analysis reveals that although the findings on customer agreement are supported for data enrichment (i.e., the customer completion rate is higher for ADRs versus HDRs), we find only marginal support for data reconciliation (i.e., the customer completion rate is only marginally higher for HDRs versus ADRs). Qualitative responses from a follow-up online survey and interviews with customers of the pharmaceutical company corroborate and complement the main quantitative findings. Overall, this research expands our understanding of continued customer engagement in data curation practices and has implications for vendor companies seeking to deploy ADRs instead of HDRs in data work and established B2B relationships. History: Jeffrey Parsons, Senior Editor; David Xu, Associate Editor. Funding: This work was supported by the Deutsche Forschungsgemeinschaft [Grant 522190413]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/isre.2023.0478 .
Keywords
AI agents, human-AI interaction, algorithm aversion, data enrichment, data error, data reconciliation, data curation practice, field experiment, gain-loss framing, multimethod
Enable JavaScript for citation tools and additional paper interactions.