A teamwork framework for preventing breaches of academic integrity and improving students' collaborative skills in the AI era

Synthesis: Issa and Hall (2024) designed and ran a teamwork framework in a postgraduate Business Project Management unit at an Australian university for five semesters (408 students enrolled, Semester 1 2021 to Semester 1 2023), built to make Academic Integrity observable rather than to detect breaches. The mechanism is deliberate supervision of the process: a 12-week team project plan worth 40% of the unit, a 30-minute virtual meeting with every team to review their work breakdown structure, each member asked directly whether their contribution is original, one peer-reflective form per teammate plus a self-reflective form, and week-nine conversations in which the lecturer establishes each student's own contribution before the final mark is set. Reported outcomes: the project-plan teamwork average rose from 27/40 (67.5%) in Semester 1 2020 to 31/40 (77.5%) in Semester 2 2021 and held at 30–31 through Semester 1 2023, and the unit's end-of-semester evaluation items scored between 89% and 100% against university averages of 80% to 86%. The abstract's claim that misconduct and extension requests "decreased dramatically" is asserted without a single count anywhere in the paper.

Key Findings

1. The integrity mechanism is a conversation, not a detector. In week nine the lecturers met each team and, from presentations given by role, established whether the work was the students' own or produced by AI tools — and those meetings are described as the basis for confirming each student's original work before the final mark.
2. The self/peer structure is per-person, not per-team. Every student writes one peer-reflective form for each team member and one self-reflective form about their own contribution, and submits notes on the items they produced or acquired while completing the assessment.
3. Marks moved and then plateaued. The project-plan teamwork average went 27/40 (Semester 1 2020), 28/40 (Semester 2 2020), 30/40 (Semester 1 2021), 31/40 (Semester 2 2021), 30/40 (Semester 1 2022), 31/40 (Semester 2 2022), 30/40 (Semester 1 2023).
4. End-of-semester evaluation ran well above university averages. Unit versus university on the eVALUate items: learning experience 100 versus 85 and later 89, 95, 94, 98; assessments 95 versus 86 and later 89, 90, 97, 92; feedback 90 versus 81 and later 95, 90, 100, 98.
5. Participation was modest and voluntary. Of 408 enrolled students, 146 completed the informal feedback survey and 129 responded to the voluntary eVALUate instrument, with per-semester response rates between 29% and 36%.
6. The cohort was mostly international. 25% of the BPM cohort were Australian students and 75% were international students from China, Indonesia, India, Iran, Japan, Malaysia, Mongolia, Pakistan, Singapore, Sri Lanka, Taiwan, Thailand and Vietnam, which the authors treat as material to the group work itself.
7. Mixed methods with light instrumentation. Anonymous formal and informal student comments manually coded into themes, quantitative analysis in Excel and Qualtrics, single unit, single institution, ethics approval HRE2019-0389.

What the framework actually does

The design turns the assessment into a supervised process with several observable checkpoints. The team project plan carries 40% of the unit and runs 12 weeks from orientation to submission, with the rubric, self/peer forms and presentation criteria released on day one. The 30-minute team meeting on the work breakdown and organizational breakdown structures is where formative feedback is given, and it is also where the integrity question is asked member by member. Requiring the presentation content — including the Group Work artifacts such as the RACI matrix, critical path and risk register — means each student has to speak to their own role.

That is the paper's real argument: the assessment asks for something a model cannot hand over intact, because the student has to explain and defend a share of the work in front of the instructor who assigned it. Whether or not AI was used, the meeting is where the instructor finds out what the student knows.

The two headline claims are not the same kind of claim

The mark increase is measured, semester by semester, in one table. The misconduct decrease is not measured at all: it appears in the abstract, in the discussion and in the conclusion as something the authors observed, with no counts of integrity cases, no before-and-after figures and no comparison against another unit. An assessment-integrity reader should treat it as the authors' professional judgment from a supervisory position rather than as an effect size, and it is the claim most likely to be quoted.

The mark series has its own confound, which the authors state plainly. A new lecturer joined the unit in Semester 1 2021, and the paper attributes the rise to "the BPM lecturer's ongoing formative feedback, the assessment's clear guidelines, and the ACD lecturer's presentation" — three changes at once, with no control group and no attempt to separate the collaboration framework from the teaching that accompanied it. The plateau from Semester 2 2021 onward is as informative as the initial jump: if the framework were producing continuous gains, the later semesters should not sit flat at 30–31 out of 40.

What the evaluation scores can and cannot show

The unit-level eVALUate items reached 100 on learning experience against a university average of 85, which is a striking gap and also a self-selected one: response rates were 29% to 36%, giving 19 to 50 responses per semester out of cohorts of 55 to 166, and the instrument is voluntary, anonymous and about the student experience rather than about learning. The university comparison is a same-institution average for whatever mix of units the students took, not a matched control, and the paper loses the university average entirely from Semester 2 2022 because the platform changed.

The qualitative material is the strongest part of the evaluation and the weakest evidence of integrity: students describe learning project management software, gaining presentation skill and gaining confidence as team members, which supports the Collaborative Learning claim and says nothing directly about whether AI use fell.

What this means for practice

• Make the process the evidence. If a student must present and defend their share of a team plan in a short meeting with the instructor, an AI-generated submission becomes visible without any detector being involved.
• Ask per person, not per team. One self-reflective form plus one peer-reflective form per teammate gives the instructor an individual record; a single team submission gives none.
• Give formative feedback before the integrity question is settled. The paper's sequence is feedback first, verification at week nine, final mark last, which keeps the meeting hospitalable rather than adversarial.
• Do not repeat the misconduct claim as a finding. The paper offers no numbers for it, and a policy audience will ask.
• Expect instructor time to be the real cost. A 30-minute meeting for every team, repeated across five semesters in a unit of up to 166 students, is the load-bearing resource in this design.

Limitations

• One unit, one institution, one assessment task. A postgraduate Business Project Management unit at a single Australian university, with the teamwork assessment as the only object of study; the authors name this and defer generalization to future work.
• No integrity counts. Misconduct and extension-request reductions are asserted rather than measured, so the framework's central promise is unevidenced in its own data.
• Confounded mark series. A new lecturer and a redesigned assessment arrived together in Semester 1 2021, and the authors credit three changes jointly.
• Voluntary, low-response, experience-focused evaluation. 129 eVALUate responses out of 408 enrolled, response rates of 29% to 36%, and items about learning experience and feedback rather than about learning gains or AI use.
• Manual coding without reported agreement. Comments were coded into themes manually, with no inter-coder reliability figure, and the quantitative analysis was run in Excel on reporting-platform output.
• 2024 data predating current models. The study spans Semester 1 2021 to Semester 1 2023, so the generative tools its framework was designed against are earlier ones than those now in use.

Citation

Issa, T., & Hall, M. (2024). A teamwork framework for preventing breaches of academic integrity and improving students' collaborative skills in the AI era. Heliyon, 10, e38759.