Fee Reconciliation Agents for Finance Offices

On the third day of odd-semester registration, the finance officer of a state university had a bank scroll, an ERP that thought 400 students were unpaid, a scholarship portal that had sanctioned 180, and a queue of parents holding screenshots. A vendor offered a chatbot that would 'answer fee queries in natural language'. She asked whether it could tell her why the same UTR appeared twice. It could not. She asked us for a matcher, not a mouth.

This case-shaped guide is for finance officers, deputy registrars (accounts), scholarship cells and internal auditors. The use case is reconciliation: bank versus demand versus waiver versus scholarship versus hostel versus exam fee. The agent proposes matches and lists orphans. It does not issue a no-dues, does not refund, and does not threaten a student with debarment. Those are speaking orders.

Not audit advice. Your accounts code, GFR where it applies, and the CAG's interest in fee income still govern. An agent that 'clears' a balance is a cashier you did not appoint.

Three ledgers, one student, no poetry

The bank knows a UTR, a date and an amount. The ERP knows a demand: tuition, development, hostel, exam, late fee. The scholarship or DBT advice knows a sanctioned head and a beneficiary. Reconciliation is the join. Failures are ordinary: a parent paid the wrong academic year, a UTR was pasted twice, a waiver was approved in a dean's email but not in the fee head, a scholarship arrived net of a deduction nobody mapped.

The agent should speak in those objects. 'UTR 3221 probably matches demand 2026-27/CSE/4418, amount equal, name fuzzy' is a proposal. 'Student is cleared' is a decision. Keep the verbs apart.

Late-fee logic is campus law. Retrieve the live notification. Do not let the model invent a condonation because the parent was polite in the chat.

Finance verbs. If the accounts code does not give the agent the verb, it does not have it.

Verb	Agent	Officer
Match UTR to demand	Propose, with confidence and the fields used	Accept or reject; never auto-accept fuzzy name matches
Scholarship / waiver apply	Show sanction letter and the head it should hit	Post the head
Refund	List duplicates and credit balances	Sanction under delegation
Write-off / condone late fee	Retrieve the notification and the student's timeline	Speaking order
No-dues for exam / leaving	Show open heads	Sign the slip

The parent queue is not the ledger

A status bot is allowed after the matcher works. It may say: paid, pending, this UTR is unseen, go to counter 3 with this slip. It may not negotiate. It may not take a new payment card number into a chat. Payments stay on the official gateway.

Screenshots are not UTRs. The agent can try to read one into a tray for a clerk. It cannot treat an edited PNG as money.

Debarment threats — 'you will not sit the exam' — are legal acts. The agent links the live notification. It does not freelance a consequence.

Data, the cashier, and the demo tenant

Fee files are personal data plus money. Bank account numbers, Aadhaar in scholarship advices, and parent mobiles are not training material. On-prem. No public chatbot. No SI copy 'for the demo'.

Split the service account: the matcher reads bank and demand; the status bot reads only the student's own heads after login. A helpdesk identity that can see every unpaid list is a leak of who is poor.

Retention follows accounts and audit, not the vendor default. Internal audit should sample matches the way they sample a cashier.

What the case looked like after six weeks

The state university ran the matcher on two programmes. Fuzzy name matches stayed in a human pile. Exact UTR-amount-date matches posted after a daily officer click on a batch list. Orphan UTRs fell by half because parents were told, by a clerk using the list, which reference to put next time. The chatbot the vendor wanted was never the pilot. A one-line status after login arrived in week five.

The finance committee understood the orphan graph. They did not ask about tokens. That is how you know you picked the right use case.

Hostel, exam fees and late-fee clocks

A student can be clear on tuition and blocked on hostel. The status line must name the head, not say 'pending'. 'Pending' is how a parent pays the wrong head twice. Name the head, the amount, and the desk. If two heads are open, list both. Do not let the model pick the one that is easier to say.

Late-fee clocks start from a notification, not from a model's sense of fairness. If the syndicate condones a window after a flood or a portal outage, pin that speaking order as the only condonation object. The agent retrieves it. It does not extend it to a friend of a dean who wrote a polite paragraph.

Refunds after a withdrawal of admission sit under a different notification. The matcher can list a credit. The refund sanction is still a human under delegation. Mixing a refund into an auto-clear is how you pay someone who still owes the hostel.

Objections you will hear — and what to do with them

These are the lines that stall the file. Answer them in the room, then put the answer in the note. A spoken answer without paper will be forgotten by the next officer.

Auto-post everything above 95 percent confidence.

Then define confidence without names, and still exclude refunds, write-offs and any match that uses a screenshot. High confidence on a wrong year is still wrong. Start with exact keys and a daily batch click.

Students want to argue in the chat.

Argument is a ticket to a clerk. The agent can open that ticket. It cannot settle it. Settlement is how unofficial condonation is born.

Our bank already has an AI product.

The bank is not your fiduciary for student demands. You may take a structured scroll. You should not give the bank your ERP in return. Join the scroll on your side.

CAG will not understand an agent.

CAG will understand a match table, a batch approval, and a speaking order. They will not understand a chat that wrote off a hundred rupees. Build the table.

A six-week matcher on two programmes

Do not start on the whole university. Pick two programmes whose demands are clean.

1. Week 1: name heads, bank accounts, scholarship sources, and the no-dues rule. Pin the late-fee notification.
2. Week 2: read-only joins. Produce orphan and double-UTR lists. No posting.
3. Week 3: officer batch-accept of exact matches only. Fuzzy pile stays human.
4. Week 4–5: add scholarship heads with academic year as a hard key.
5. Week 6: optional login status line. Still no negotiation, no write-off, no card numbers.

How this shows up in the file

Finance standing order: matcher proposes, officer posts, no goodwill head, no chat condonation, no training on scrolls, service accounts split, internal audit samples weekly in fee season.

Attach one worked match and one rejected fuzzy match so the next officer can see the standard.

This article is informational field guidance for Indian public universities and colleges, not legal, regulatory, academic, labour or procurement advice. Confirm against the live AICTE / UGC / NAAC notice, the DPDP Act and Rules as commenced, your Act and statutes, reservation orders, labour counsel and the syndicate before you file it.

How this works on an Indian campus

A P3 University should be able to run “Fee Reconciliation Agents for Finance Offices” without importing a US playbook. “fee reconciliation automation” hits UGC/AICTE/NAAC clocks, exam secrecy, reservation rules, and students who may be minors.

Fee season is three ledgers that do not agree. Automate the match. Humans keep the write-off, the refund and the demand. DPDP applies to student personal data. Chatbots are not a strategy. Exam and admissions writes stay with officers. Affiliated colleges need isolation, not one shared index.

• No production student data in a vendor SaaS sandbox.
• Write the academic integrity policy before the tool.
• Consent and purpose tags on student-facing agents.
• Budget for staff training, not only licences.

Close this loop before the next CAB

Put “Fee Reconciliation Agents for Finance Offices” on the next change-advisory or bid-opening agenda as a single line item with an owner. If it cannot earn a line item, it will not earn a control. The owner should be a P3 University, not “the vendor.”

Revisit the item when the model, the GeM term, the region, or the SI changes. “fee reconciliation automation” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.

What the next file must contain

“Fee Reconciliation Agents for Finance Offices” earns a line in the noting only if a P3 University can attach proof of “fee reconciliation automation.” A heading is not proof. A vendor slide is not proof. A workshop photograph is not proof.

Write three dated sentences: what was decided, who owns it after the next posting order, and when it will be re-checked. If you cannot write the three sentences, you are not ready to buy, to sell, or to go live.

Leave unsourced percentages out of the note. DPDP is not a blanket localisation statute. The November 2025 AI governance text is guidance, not an Act. CERT-In’s 28 April 2022 directions still set specified incident and log clocks. A PAC, when lawful, lives in GFR Rule 166.

• Name the designation that owns “fee reconciliation automation.”
• Attach one artefact a stranger can open next year.
• Record the instrument you are actually using.
• Revisit when the model, the SI, the notice or the posting changes.