Administrative work — schedules, notices, forms

The template-once-reuse-mechanically pattern

Open your sent folder or your school-comms folder for last term. Read the subject lines. You will see a pattern: a handful of message types repeat over and over, with the dates and details changed. The Friday notice home. The monthly parent update circular. The sports-day or annual-day schedule. The exam timetable. The library-overdue reminder. The scholarship-application instruction. The transfer-certificate cover letter. For most teachers, seven to ten templates cover ninety percent of the term’s outgoing paperwork.

The pattern that pays off is simple. Pick one template at a time. Sit with AI for fifteen minutes. Build a clean version — bilingual where it should be, with placeholders for the bits that change (date, time, room, names of officials, deadline). Save the result somewhere you can find it again — a Google Doc, a Drive folder, a notebook section. From that point on, when the same comms goes out next month, you fill the placeholders and send. You do not redraft. You do not “ask AI to make it sound nicer.” You do not start over. The template carries the format; only the facts change.

The first hour of this work — building five or six templates — is the most valuable hour you will spend on admin all year. Every subsequent reuse takes ninety seconds instead of fifteen minutes. Over a term, the saving is in the order of ten to fifteen hours.

Bilingual by default for school notices

Most school notices in Nepal benefit from a bilingual format. Nepali primary, English secondary, both visible on the same sheet. The reason is the same as in parent updates — different members of the household read in different languages, and a monolingual notice excludes someone in almost every family. The diaspora older sister who actually fills out the form is on English; the grandmother who decides whether the child attends the meeting is on Nepali. Both versions, same page, no guessing.

Before AI, bilingual notices were aspirational. Producing them by hand meant doing the work twice, every time. With a fact-first prompt that says “produce this notice in Nepali first, then English second, on the same page, in matching tone and identical content,” the model gives you both versions in one go. You skim both — your Nepali catches what the model’s Nepali misses — and the notice is done. The time cost of bilingual is now roughly the same as the time cost of monolingual used to be.

A useful refinement: ask the model for a third version when it matters — a short SMS-length summary of the notice, suitable for sending on Viber or WhatsApp to families who will never read the full paper version. “Then produce a two-line Nepali summary for SMS, with the date, time, and place only.” That short version reaches the parents who do not read the school bag.

The data-discipline rule

This rule is non-negotiable, and it gets violated the most often precisely because admin work feels routine. Student lists with full names. Mark sheets. Attendance registers with roll numbers. Family contact details. Citizenship numbers on scholarship forms. Photographs of any school document. None of this goes into a public chatbot. Ever.

The reason is structural. The free tiers of ChatGPT, Gemini, and similar consumer tools may use your inputs for training, and even the paid tiers route data through servers outside Nepal that you cannot audit. A school’s administrative records are a small but very real surveillance trove — the names, ages, addresses, family circumstances, and academic histories of children. If a stranger could rebuild those from your prompt history, you have failed in a basic professional duty.

The practical workarounds are simple and they almost always work. Aggregate before you prompt. Instead of pasting the full register, write “thirty-two students in Class 6, average attendance eighty-nine percent, three with attendance below seventy percent.” Redact identifiers. Write “Student A is eligible for a scholarship; Student B has an overdue library book” rather than the names. Use generic placeholders. Write “[parent name]” and “[child name]” in the template; substitute the real values only after the template comes back to your own document, on your own computer, where the data never leaves your control.

For schools that handle sensitive material at scale — boarding schools, large private schools, schools with national-level scholarship pipelines — the right answer is a school enterprise account with a no-training data agreement, or an on-premise model. For everyone else, redaction at the prompt is the floor.

Where AI saves the most admin time

Not all admin work benefits equally from AI. Some tasks — entering marks into the school portal, ticking attendance, filing physical forms — are not language tasks at all, and AI saves you nothing. The tasks where AI saves the most time are the ones that turn rough material into formal written output. Four kinds, in particular, repay the effort.

Meeting minutes from rough notes. You scribble during the parent-teacher meeting or the staff meeting — half-sentences, names, decisions, action items. After the meeting, you paste those rough notes (with names redacted to roles) into the chatbot and ask for “a clean, formal minute in Nepali, with sections for Attendees, Agenda, Discussion, Decisions, and Action Items.” The output is a usable first draft. You spend five minutes correcting names and tightening; you save forty minutes you used to spend writing minutes from scratch.

Summary of a parent meeting into a written record. After a one-on-one with a parent — a behaviour issue, a fee question, a scholarship request — you owe the file a written summary. You speak into your phone for ninety seconds, get a transcript (or just type rough notes), and ask the model to produce a paragraph-length record for the file: date, attendees, topic discussed, agreement reached, follow-up date. The model produces it in the register a file expects. You read it, fix one or two things, save it. The whole record takes three minutes.

Multi-language versions of a notice. You have a Nepali notice already. You ask for an English version that matches in content and tone, then a short Viber-suitable summary. Two outputs in ninety seconds. Before AI, the English version alone would have taken twenty minutes of looking up phrases.

A six-paragraph explainer of a policy change. The school is changing its phone-in-classroom policy, or its homework policy, or its attendance threshold for scholarships. Parents need to understand it. You give the model a fact block — what the old policy was, what the new policy is, why, and what families need to do — and ask for “a six-paragraph explainer in plain Nepali, suitable for parents with mixed education levels, with each paragraph having a clear heading and a bullet of the action required.” The output is a notice that actually communicates instead of formally announcing. Parents read it. Compliance goes up. The headmaster looks good. You saved an hour.

In each of these four cases the input is rough material you already have, and the output is the formal version someone else expects. That is precisely the kind of work the model is best at, and precisely the kind of work that has been quietly stealing your evenings.

The time-back rule

There is one principle without which all of this is pointless. Every minute AI saves on admin should go back into teaching, planning, or the relational work that no AI can do.

The trap, otherwise, is real. AI saves you forty minutes on a Friday evening. You fill those forty minutes with more admin — answering more emails, drafting another form, polishing another circular — because admin is the kind of work that expands to fill any space you give it. The saving disappears. You are no less tired on Monday. The class is no better prepared.

The discipline is to take the saving as a saving. Some of it goes back into your teaching — the lesson you were going to wing on Monday becomes one you have actually thought about. Some of it goes back into a child you have been meaning to call home about — a phone call to a parent that an email could not have done. Some of it goes back into rest, which is a teaching input you have probably been underspending on for years. None of it gets reinvested into more paperwork. That is the entire point.

If the saving goes into more admin, the AI has not served you; it has expanded the volume of your job. If the saving goes into the work that only you can do, the AI has done something quietly important — it has bought you back a part of the profession you trained for.

Check your understanding

Quick check

—

True or false. You need to write a clean summary of last week's parent-teacher meeting for the school file. To save time, you paste the meeting attendance register — with full names, roll numbers, fees-paid status, and parents' phone numbers — into a public chatbot like ChatGPT and ask it to produce a formal minute.

1. ATrue — this is exactly the kind of routine admin task AI is meant to speed up.
2. BFalse — student names, roll numbers, fee status, and parent contact details are sensitive data, and pasting them into a public chatbot risks leaking children’s and families’ private information. The correct approach is to redact identifiers (use roles or placeholders), or use a school enterprise account with a no-training agreement, before asking the model for the formal minute.
3. CTrue — the chatbot would forget the information after the session ends.
4. DFalse — but only because the chatbot’s Nepali is not good enough; the privacy issue is fine.

Check answer→

What comes next

You now have a workflow for parents (messages), for the term (reports), and for the school’s paperwork (notices and forms). The remaining question is the one this whole course has been quietly circling — what the rise of AI does to the meaning of learning itself. The next chapter is about academic integrity, the ethics of AI use in school, and the harder question underneath both: in a world where the machine can write the essay, what exactly are we teaching for?