At the end of every reporting cycle, teachers faced the same task: write individual, personalised comments for every student in every class. For a class of 20, that meant 3.5 hours of work — minimum. Multiply across five classes and three reporting periods and you’re looking at over 50 hours a year spent on report writing alone.

The comments had to be specific, rubric-referenced, and differentiated. Copy-paste wasn’t an option. Generic AI output wasn’t good enough. And the clock was always ticking.

We built a report writing workflow in n8n that runs entirely from a spreadsheet teachers already maintain. No database. No technical setup. No training beyond a 20-minute walkthrough.

• Marks in the spreadsheet — Teachers record student marks throughout the term as normal. One tab for marks, one tab for rubrics. Nothing new to learn.

• Rubric pulled automatically — The workflow reads the rubric for each mark level directly from the spreadsheet. Update the rubric tab and the workflow updates with it — no one touches the automation.

• Strengths summary generated — The AI analyses performance across all marks and rubrics to write a personalised strengths summary grounded in the actual assessment criteria.

• Targets from the next level up — The workflow identifies rubric descriptors at the next level above the student’s current performance and turns them into specific, achievable improvement goals.

• Full class in 2.5 minutes — Trigger the workflow. All 20 report comments are generated in under three minutes. Teachers then spend 15–20 minutes proofreading and personalising.

The workflow Google Sheets → Loop Over Items → Intro comment → Strengths → Improve → back to Sheets. Three AI stages, each grounded in the rubric. Runs on Ollama locally — no data leaves the building.

Before the workflow, a set of reports for one class took between three and four hours. After, the same task took roughly 27 minutes from clicking run to completing the proofread.

That’s an 87% reduction in time per reporting cycle. Across five classes and three reporting periods a year, that returns 45 hours to every teacher who uses it.

Because every comment is grounded in the actual rubric descriptors for that student’s mark, it reads as specific and considered — not AI-generated boilerplate.

Student marks and names are handled with the same care you’d expect from any school system.

• Local AI — The language model runs on school hardware via Ollama. No student data leaves the building.
• Spreadsheet-based — Data stays in the same Google Sheet or Excel file teachers already use. No new systems, no new data stores.
• No third-party processing — Unlike ChatGPT or Gemini, student performance data is never submitted to an external service.
• GDPR and PDPA compliant by design — Local processing removes the compliance risk entirely rather than managing it through policy.

There is no database to configure, no IT project to scope, and no technical knowledge required from teachers.

Rubrics are updated directly in the spreadsheet by the teacher. Change the rubric, and the workflow automatically uses the updated version on the next run. No one needs to touch the automation.