Method · Chapter 3
How the study was done
The study is qualitative. It combines a detailed account of one workflow I took part in with interviews about other people's work, and compares them through the same three questions while keeping their evidence separate.
Two parts, three settings
The two parts are treated differently because my relationship to the material differs. I took part in the SDF workflow. I did not take part in the work interviewees described; I asked about it and analysed their accounts.
In depth
Autoethnographic analysis
One recurring SDF reporting workflow (Chapter 4). It can trace how one arrangement developed and became part of regular work.
Breadth
Thematic analysis of 11 interviews
Participants' accounts of data, governance, review, task fit, adaptation and checking in their own work.
What each source can show
| Source | Can support | Does not, by itself, establish |
|---|---|---|
| Reflective journal main source for the SDF account | How I saw my experience, problems, decisions and responses, looking back | My understanding at the time. Entries were written in February 2026 about 2025 events and are not independently confirmed |
| Technical materials processor, anonymisation utility, processing logic, diagrams, selected non-identifying traces | Specific claims about what I built and how parts of it operated | What it meant for the organisation, or that every figure was right in context |
| Dated messages | Dates and the order of events, such as the IT requests | Not analysed as a separate source; not, for example, whether a suitable environment was later obtained |
| Interview transcripts IP01–IP03, SP01–SP08 | How participants described their work, constraints, changes and checking | Implementation, duration or outcomes; events in another setting |
| Literature | Frames the analysis | Anything about these settings |
No setting's evidence is used to establish what happened in another. Thesis Sections 1.5, 3.5 and 3.7.2
Studying a workflow I built
Strength
Insider access to decisions, uncertainty and judgement that finished outputs do not show.
Risk
Selective attention, and events looking more coherent in hindsight. The journal was written after the events.
Safeguards
Keeping apart what I recorded, the dated messages, what the code shows and my later interpretation. Writing in the first person so my role stays visible.
The analytic approach follows Anderson, L. (2006). Analytic autoethnography. Journal of Contemporary Ethnography, 35(4), 373–395. https://doi.org/10.1177/0891241605280449 on using full membership of a setting to develop theory, and O’Raghallaigh, P., McCarthy, S., & Adam, F. (2016). The complementary role of story and theory in IS autoethnography. In Pre-ICIS Workshop: Advances in Qualitative IS Research Methodologies. https://research.ucc.ie/en/publications/the-complementary-role-of-story-and-theory-in-is-autoethnography/ on combining personal account and theory. In the interviews my position was different: I avoided filling gaps with my own knowledge of the university, and where my question introduced an idea, I checked whether the participant developed, confirmed or qualified it. Thesis Sections 3.2.1 and 3.4
The interviews
11
consented UTS staff volunteers, recruited through the university's Microsoft 365 Copilot Teams channel
22 Apr – 14 May 2026
semi-structured interviews on Microsoft Teams, recorded and transcribed with consent
IP01–03 · SP01–08
transcripts reviewed and de-identified; the code register stored separately
Everyday staff were asked about their role, how they used GenAI tools, the difficulties they met, what they did when a tool did not fit, what checking remained, what less visible work arose, and what support would help. Institutional support staff were asked about where AI-related workflows meet friction, the workarounds people create, hidden work, governance and scaling.
Because volunteers came through a Copilot channel, the group may lean towards people interested in GenAI, and it does not represent all staff. Some questions used research terms such as “workarounds” and “hidden work”, so I checked whether participants supported or qualified those ideas. Exact licensing entitlements, permissions, organisational reach and participant proficiency are as participants described them and were not independently audited. Thesis Sections 3.3.1–3.3.2 and 3.7.2
How I analysed the interviews
- Start from the frameworkAn initial codebook drawn from the research questions and literature
- Pilot on four transcriptsIP02, IP03, SP01 and SP03, to refine unclear or overlapping codes
- Code all elevenIn NVivo, using eight parent codes
- Review in contextEach passage read back in its transcript before use
- Decide what it supportsQuote, paraphrase, keep as context, or leave out
Eight parent codes
Friction; workaround formation or local adaptation; invisible labour or less visible work; checking and judgement; governance, assurance and risk; access, capability and responsibility; support, explanation and workload shift; and other or unexpected patterns. These are coding categories, not findings, and were used only for the interviews.
Records kept
Transcript memos, a codebook change log, evidence-selection records and claim matrices linking passages to claims. Coding counts helped retrieval and were not used as measures of how common something was. Grouping episodes by where friction came from and what it did is my analytical classification, not a participant severity rating. Proposed changes were recorded separately from changes reported in use.
The approach is thematic analysis (Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77–101. https://doi.org/10.1191/1478088706qp063oa) with hybrid deductive and inductive coding (Fereday, J., & Muir-Cochrane, E. (2006). Demonstrating rigor using thematic analysis: A hybrid approach of inductive and deductive coding and theme development. International Journal of Qualitative Methods, 5(1), 80–92. https://doi.org/10.1177/160940690600500107), and an audit trail (Nowell, L. S., Norris, J. M., White, D. E., & Moules, N. J. (2017). Thematic analysis: Striving to meet the trustworthiness criteria. International Journal of Qualitative Methods, 16(1), 1–13. https://doi.org/10.1177/1609406917733847). I did the coding myself; no second researcher reviewed it. Thesis Sections 3.3.3 and 3.6.2
Ethics and confidentiality
The study had University of Technology Sydney Human Research Ethics Committee approval ETH25-10884. Participants gave informed consent and are identified only by code. Exact titles, units, projects and systems are generalised where they could make someone recognisable, which matters especially for institutional support roles. In my own account, colleagues appear only as far as my experience requires, and their intentions are not inferred. Workplace emails and tickets are not used as a separate evidence source. Thesis Sections 3.7–3.7.1
How GenAI was used in the research
| Where | How Microsoft 365 Copilot was used |
|---|---|
| Throughout the thesis | To review structure, clarity, consistency and grammar and to suggest wording. Suggestions were reviewed and revised or rejected; I remain responsible for the argument, sources, evidence boundaries and final wording. |
| Interview analysis | Bounded help checking coded-passage boundaries and links between evidence and claims. Suggestions were checked against transcripts and records. I applied all coding and made every interpretation. Automatic meeting summaries were not kept or used. |
| The SDF workflow | Querying selected prepared files, April to December 2025, with responses checked and reconciled before staff use. Copilot was part of the workflow being studied, not a source of research evidence. |
Thesis front matter, GenAI use disclosure; Sections 3.6.2 and 3.7.2
Limits of the design
- One workflow in one university, examined in depth.
- A retrospective journal; my involvement shapes interpretation.
- A self-selected interview group; accounts rather than observation.
- Interviews varied in depth; some described proposals or short experiences.
- One researcher coded the interviews.
- Confidentiality limits the detail that can be reported.
These limits set the scope of the findings: a bounded qualitative account, not evidence of prevalence, productivity effects or transferability. Thesis Sections 3.7.2 and 7.8