Where AI belongs in a school office, and where it does not

The pitch every school now hears

Many software pitches to schools now include AI: AI that writes the notice, AI that predicts which family will pay late, AI that answers parents at midnight. Some of it is genuinely useful. Some of it puts a machine where a school needs a fact.

Plausible is not the same as correct

A school office deals in facts that people act on: a fee balance, a mark, an attendance percentage, whether a parent received a notice. Language models are very good at producing text that sounds right. In a school office, sounding right is not enough. A balance that is almost right, or a question that is nearly on the syllabus, does more harm than no help at all, because it looks finished and nobody checks it.

The rule Campus One is built on

In the Campus One modules described on this site, the work is split in two:

• Facts of record come from arithmetic over the school's own records, with tests: balances, receipt numbers, attendance percentages, mark totals.
• Where AI is used, it may help a person choose among things the school already has, and the person decides. The work must still get done without it: in the exam paper generator, for example, the standard method runs when no AI is available, and the screen says so.

The sections below show where that rule is enforced in the product.

Where the rule is enforced

Money

In the fee ledger, balances, ageing, receipt numbers and how a payment is split across instalments are all worked out by arithmetic over the ledger. No AI and no estimate ever writes a balance. Receipts are numbered without gaps for each campus and financial year. Our guide to fee management explains why that matters at the counter.

Question papers

The exam paper generator does not need AI to make a paper: its standard method, which uses no AI, builds a paper from questions already approved in the school's own bank, to a blueprint whose marks must add up. Where an AI model is set up, it may help choose questions from that bank, and its choice is checked against the same blueprint. It can only point to questions that exist; it never writes question text. If no model is set up, or the AI budget is used up, the standard method runs and the screen says so. Our guide to question paper blueprints shows the checks.

Parent messages

Notices are written from the school's own templates; the messaging module has no AI drafting. No message is shown as delivered unless a messaging provider confirms it. In the demo no provider is connected, so every notice is marked simulated, and the screen says so.

Records and decisions

When a new enquiry's phone number matches a family already in the school, Campus One shows the match as evidence and stops there. The family record is reused only when a person admits the child. In the store, the run-out forecast is plain arithmetic rather than AI, and it still only produces a draft recommendation with the evidence behind it; it never orders anything.

The product's engineering rules require every number on a dashboard to open the records behind it, and say plainly: no fabricated numbers. The record history is designed to include actions taken by an automation or an AI, written like any other entry.

What is not built

The Complete plan on the price list names an AI office assistant. It is not built, and the pricing page marks it as not built. We would rather show that gap than describe a feature that does not exist.

Questions to ask any vendor about AI

1. Which features use AI, by name?
2. Can AI change a balance, a mark or an attendance record? It should not be able to.
3. Does each AI feature still work, in a standard way, with AI switched off?
4. Does AI write anything that goes to parents or students, and who approves it first?
5. Is every AI action recorded on the record it touched?
6. Which of our data is sent to an AI model, and where does that model run?

Ask for the answers in writing. Our article on where school data should be stored has the questions to ask about the data itself.

Who builds by these rules

Campus One is built by AskShiva Private Limited, whose leadership team is Divya Prakash Jha (Chief Executive Officer), Vaibhav Jain (Chief Operating Officer) and Zeeshan Khan (Chief Marketing Officer). The rules in this article are written into the product's engineering notes and its code. They are the company's position, set out here in the company's voice; nothing in this article is a quotation from any one person.

This describes how the product is built. It is not a certification or a compliance claim.

The outcome: every number has an explanation

When facts come from records and AI only helps people choose, a school can explain every number it gives a parent: where it came from, who entered it and when. That is a better foundation for using AI than any feature list.

The takeaway

Use AI to help people choose and to save typing, never to produce a fact of record, and ask every vendor the six questions above. The exam paper generator page shows how AI help with question choice works in Campus One, and our checklist for choosing school software covers the rest of the decision. To see the standard method and its checks on the demo school, ask us for a walkthrough.