AI in Marking

Every year, over 550 first‑year business students at Henley Business School take part in an intensive “Vibes Week” entrepreneurship bootcamp, where they pitch a business idea in a 7‑minute group presentation that counts towards their final mark. With more than 120 teams and a large marking panel, it was very hard for staff to take notes, mark fairly and consistently, and still give useful written feedback to every group.​ We built a system where markers use a detailed digital rubric during each presentation to record scores and short comments. After the presentation, AI transforms that rubric information into a clear, tailored feedback report. Markers review the AI‑drafted feedback first; the module convenor then samples and moderates before release.​ This means students now receive faster, more consistent and easier‑to‑understand feedback, while staff spend less time on repetitive writing and more time on academic judgment and moderation. The AI helps summarise and phrase feedback, but humans remain fully in control of marking decisions.

This project is an example of an organization identifying the opportunity to implement a new and cutting-edge technology for automation that could improve many facets of an existing process. Great leaps forward however, are not without risk. This project applied best-in-class process and principles from the assessment industry to ensure the move forward put their program on the leading edge and not the bleeding edge. Implementing automated marking with IntelliMetric allowed the program to meet the triadic, holy-grail of assessment (validity, reliability, equity) and improve upon second-tier considerations. IntelliMetric marking resulted in maintained validity of the writing assessments, increased reliability of marking, equitable rating of papers free from bias while also reducing the cost of marking and decreasing the time to deliver results to candidates. By following the scientific practices ingrained in the assessment community each step forward included stakeholder input, collaboration with the vendor and most importantly an evaluation of the evidence, in the form of data prior to making a decision. This project focused on getting the decision right over getting the decisions fast.

Safety Training Awards (STA) has transformed its assessment process by replacing manual portfolio reviews with sAInaptic's AI-assisted marking model, directly integrated into STA Online, their own e-assessment platform. Previously, tutors spent up to an hour manually evaluating a single lesson plan - a process prone to fatigue and inconsistent feedback. The new AI system reduces this time to an average of just 8 seconds per plan. In practice, tasks that once took an entire afternoon can now be completed in 30 minutes, including human review of AI marking. Key Advantages - Efficiency: Accelerated marking allows students to qualify and start their careers sooner. - Consistency & Precision: AI evaluates work against strict technical rubrics, eliminating human bias and catching minute errors (e.g., missing equipment listings) that a tired eye might overlook. - Constructive Feedback: The system generates detailed, supportive guidance to help trainees improve their teaching skills and reduce re-submissions. - Human-Centric Focus: By automating the administrative burden, tutors can shift their focus from paperwork to poolside mentoring, enhancing the practical quality of teacher training. This digital transformation ensures faster results and higher standards, ultimately producing more well-prepared swimming instructors.

Mock exams are essential preparation for students, but marking them creates an impossible workload. A single round of mocks can take teachers 50+ hours, with marking often delayed for weeks. By the time students receive feedback, learning momentum is lost. The core advantage of Serenity is that it doesn't require digital transformation, something schools lack capacity for. Instead, Serenity solves the problem of turning chaotic, messy handwritten responses into text that can be marked. This technical breakthrough allows schools to keep paper-based exams. Students write on paper as they will in their actual exams. Teachers scan scripts using phones or existing scanners. Serenity transcribes the handwriting, marks each answer, and produces detailed feedback, all without changing exam format, training staff on AI systems, or disrupting established practice. Teachers remain in control. They review all marks before students see them and can override anything. Subject experts calibrate the system beforehand using human-marked papers to ensure accuracy. The result: students receive meaningful feedback within days whilst it's still fresh and actionable. Teachers reclaim time. Mock exams fulfil their intended purpose: timely, meaningful preparation for the real thing. In addition, Serenity has the potential to reverse the recruitment challenge for schools.

When teachers mark student work, we often worry about consistency—would two different teachers give the same grade to the same essay? In the world of Artificial Intelligence, we face a similar problem: different AI models (like ChatGPT or Google Gemini) can sometimes disagree on a score, or even give a different mark if asked twice. Our project solves this by creating a 'Digital Moderation Panel.' Instead of relying on just one AI, we have five of the world’s leading AI models mark the same piece of student work simultaneously. We then use a 'majority voting' system to decide the final grade. Our research proved that when these different AI 'experts' agree, the score is highly accurate and matches human expert standards. If they disagree, it flags the work as needing human review. This approach makes AI marking much fairer and more reliable, giving schools and students the confidence that their work is being graded with the same rigor as a panel of human experts, but with the speed and efficiency of modern technology.

When teachers mark student work, we often worry about consistency—would two different teachers give the same grade to the same essay? In the world of Artificial Intelligence, we face a similar problem: different AI models (like ChatGPT or Google Gemini) can sometimes disagree on a score, or even give a different mark if asked twice. Our project solves this by creating a 'Digital Moderation Panel.' Instead of relying on just one AI, we have five of the world’s leading AI models mark the same piece of student work simultaneously. We then use a 'majority voting' system to decide the final grade. Our research proved that when these different AI 'experts' agree, the score is highly accurate and matches human expert standards. If they disagree, it flags the work as needing human review. This approach makes AI marking much fairer and more reliable, giving schools and students the confidence that their work is being graded with the same rigor as a panel of human experts, but with the speed and efficiency of modern technology.