Every school ERP added a chatbot. We did the opposite. We did the opposite.

It is 9:40 in the morning and a principal in Kanpur has a management-committee meeting at 10. She needs three numbers: how much fee is still pending for the senior classes, how many new admissions closed this month, and how many students are short of documents before the board inspection. All three answers exist inside her school software, correct and current. Getting them out means opening three screens, running two reports and pasting it all into a sheet. She will walk into that meeting with two of the three, and a promise to "send the rest by evening."

Here is the thing I kept coming back to: her school never had a data problem. It had a distance problem — ten clicks between the question in her head and the answer on her screen. So when we thought about what "AI" should mean for a school, we did not build a chatbot. We closed that distance.

The chatbot era was a dead end

Walk through any Indian school-software expo and you will see the same thing on every stall: a chat bubble in the corner of a website, or a camera that recognises faces at the gate. That is what "AI-powered" has come to mean. It photographs well in a brochure and helps almost no one in the front office.

The problem with a bolted-on chatbot is that it only knows what its makers scripted. Ask it something slightly off-script — "who in class 9 still owes second-term fees?" — and it hands you a link to a help article. It cannot see your school's actual data, so it cannot answer your school's actual questions. It is a mascot, not a member of staff.

Meanwhile the assistants that can reason — the ChatGPTs and Claudes that your own children and teachers already use daily — had no safe way to reach the one place a school's answers actually live: its ERP. That gap is the whole opportunity, and almost nobody in Indian school software was looking at it.

The real unlock: let the AI you already use ask your school

The shift that made this possible is quiet but enormous. AI assistants recently gained a safe, standard way to connect to trusted software — to read specific things, with permission, and answer questions about them. The moment that arrived, the right move for a school ERP was obvious: stop building a worse chatbot, and instead let the good assistants ask your school directly.

So that is what Inkwelly does. A school connects its account to ChatGPT or Claude once. After that, anyone the school allows just asks — in English, in Hindi, in the Hinglish that real offices speak — and the assistant looks it up in the school's own live data and answers. "Who hasn't paid this term." "This month's collection." "How many students, board-wise." No export, no report, no training. The answer is never a day old, because it is reading the same live records the office works from.

“A school that can't get a straight answer out of its own software doesn't have a data problem. It has a distance problem — ten clicks between the question and the answer. We deleted the ten clicks.”

Why "read-only" is a feature, not a limitation

The first question every serious principal asks is the right one: can the AI change or delete anything? The answer is no, by design. The assistant can read and answer. It cannot collect a fee, edit a record, remove a student or message a parent. Every change still happens the old-fashioned, accountable way — by a named person, inside Inkwelly, with a confirmation.

This is not us being timid. It is us being correct. An assistant that answers is enormously useful and carries almost no risk — the worst it can do is tell you a number. An assistant that can act on live school records without a human watching is a liability no school should take on, however clever the demo looks. We would rather ship the safe, useful 90% today than a reckless 100% that gets a school burned once and abandoned forever.

What this actually changes for a Tier-2 school office

Strip away the technology and look at the day. The front-desk assistant stops being a human search engine for forty "has my fee gone through?" calls. The accountant produces the defaulter list before the reminder calls, not by staying late to build it after. The principal walks into the committee meeting with all three numbers, current to the minute, because she asked for them on the way in.

None of this requires anyone to learn new software. That is the part I care about most. Indian schools do not switch ERPs because of a shiny feature; they switch because the last one made their staff feel stupid. An assistant you talk to in your own words is the opposite of that — it meets your office where it already is.

Where I'd be wrong

The honest counter-argument is that this only matters if schools trust an AI with their questions at all, and plenty of principals are rightly cautious. If a school's data is messy, the answers will be too — the assistant is a mirror, not a magician. And there are questions it cannot yet reach, because we have deliberately started narrow, with fees and students, rather than promising everything and delivering a fragile everything. If you were betting against us, that is where you would push: adoption caution and early breadth.

I think caution is exactly why the read-only, permission-bound, start-narrow design is right. Trust is earned one correct answer at a time, and we would rather earn it slowly than lose it fast.

So no, we did not add a chatbot. We made the school itself answerable — to the assistant the principal already trusts, in the language she already speaks. That is what an AI-native school ERP should mean, and as far as we can tell, we are the first in India to build it that way. The rest of the industry is welcome to catch up.

See it answer your own school's questions

Book a 20-minute demo, bring the questions your office asks every day, and watch them get answered in seconds.