What does it take to get Copilot working well with finance data? New work led by Shane Lee sheds light on the “invisible labour” often required for AI productivity gains…
We all recognise the marketing discourse around GenAI promising staff empowerment and rapid productivity gains. However, Shane Lee’s work injects a dose of reality, documenting how sociotechnical realities on the ground can introduce ‘friction’ before such gains are realised, requiring various “workarounds” (an established concept in the information systems literature).
Shane is a member of the professional staff in the Management Discipline Group, UTS Business School. Supported by his manager Stella Ng, he is reflecting on the introduction of generative AI to professional staff in the university, documenting this critically in his interdisciplinary Master of Research, supervised by Baki Kocaballi (Engineering & IT), Bhuva Narayan (Design & Society) and Simon Buckingham Shum (CIC).
Last month he presented to the Australasian Conference on Information Systems (ACIS 2025). His paper, Making AI Functional with Workarounds: An Insider’s Account of Invisible Labour in Organisational Politics (details below), documents an initiative to build a chatbot for querying and reporting financial data.
In Shane’s case, he wanted to push the envelope of what was possible within a novel AI infrastructure, which was in the process of being laid within the university (Microsoft Azure AI, and Copilot). This required navigating organisational structures and politics, as well as technical capabilities and permissions. Once he finally secured access to the right technological capabilities, the working solution he implemented focused on engineering the data pipeline, using a programmatic approach to pre-process and structure the data before it reached the AI, ensuring the system received high-quality, relevant context. This is now adding value to his work, and to the hundreds of academics now receiving timely, accurate financial statements.
Shane’s talk struck a chord with the conference audience, sparking discussion on how staff must often navigate not only personal learning and technical obstacles, but organisational territoriality and assertions of positional power, to advance innovation. The study suggests that the workarounds users create to bypass these blockers represent vital “articulation work” necessary to complete the system and make it functional.
Dive deeper into the full paper…
Lee, S. C., Narayan, B., Buckingham Shum, S., Ng, S., & Kocaballi, A. B. (2025). Making AI Functional with Workarounds: An Insider’s Account of Invisible Labour in Organisational Politics. Australasian Conference on Information Systems (ACIS 2025), Brisbane. Open Access: http://hdl.handle.net/10453/190882
Abstract: Research on the implementation of Generative Artificial Intelligence (GenAI) in higher education often focuses on strategic goals, overlooking the hidden, and often politically charged, labour required to make it functional. This paper provides an insider’s account of the sociotechnical friction that arises when an institutional goal of empowering non-technical staff conflicts with the technical limitations of enterprise Large Language Models (LLMs). Through analytic autoethnography, this study examines a GenAI project pushed to an impasse, focusing on a workaround developed to navigate not only technical constraints but also the combined challenge of organisational territoriality and assertions of positional power. Drawing upon Alter’s (2014) theory of workarounds, the analysis interprets “articulation work” as a form of “invisible labour.” By engaging with the Information Systems (IS) domains of user innovation and technology-in-practice, this study argues that such user-driven workarounds should be understood not as deviations, but as integral acts of sociotechnical integration. This integration, however, highlights the central paradoxes of modern GenAI where such workarounds for “unfinished” systems can simultaneously create unofficial “shadow” systems and obscure the crucial, yet invisible, sociotechnical labour involved. The findings suggest that the invisible labour required to integrate GenAI within complex organisational politics is an important, rather than peripheral, component of how it becomes functional in practice.