Student support agents for admissions and fees

A student support chatbot is worth building for one reason: the questions arrive in floods, and they are the same questions. Admissions season brings tens of thousands of "what documents do I need" and "when is the last date"; fee deadlines bring "why does my portal show a penalty"; the first week of term brings "where is my class". The admissions office cannot staff for the peak, so students wait, call repeatedly and post on social media. An agent that answers from the institution's own rules, on the channels students use, in the languages they speak, removes most of that volume and hands the rest to a person with the context attached.

This article sets out what a university AI assistant for admissions and fees should do, what it must never do, how it connects to the student information system, how to run it on WhatsApp and the web, and what it costs. It is written for registrars, admissions heads and IT leads at colleges, universities and school networks.

What a student support agent does

The agent answers questions and performs a small number of safe actions. On the question side: admission eligibility and process, document lists, deadlines, fee structures, scholarship rules, hostel and transport, timetables, exam schedules and results release dates. On the action side, within policy: sending a fee receipt, raising a ticket, booking a counselling slot, sending a document checklist, updating a contact number after verification. Anything that changes an academic or financial record, such as a fee waiver, a course change or a deadline extension, is proposed to a staff member rather than executed.

FAQ bot vs grounded support agent

Grounding in the institution's rules

Institutions run on documents: the prospectus, the fee circular, the academic calendar, hostel rules, examination regulations, and a stream of notices that supersede each other. The agent must answer from the current versions of these, cite the document, and know when a notice has changed a date. That means a retrieval layer with versioning, so the fee deadline the agent quotes is from this year's circular and not last year's, and a content owner in the registrar's office who approves what the agent may use. The engineering is the same as our retrieval and knowledge engineering practice, and the reason it matters is in why basic RAG fails in production.

Connecting to the student record

"Why does my portal show a penalty" cannot be answered from a document; it needs the student's ledger. The agent connects to the student information system or ERP through a small, read-mostly API: identity verification (roll number plus OTP to the registered phone), programme and year, fees due and paid, timetable and exam schedule. Write actions are few and each has a policy: a receipt can be sent to the registered contact only; a contact-number change needs OTP verification on both numbers; a ticket can be raised for anything. Every action is logged with the policy that allowed it. The pattern is described in policy-gated actions.

Applicants who are not yet students

Admissions traffic comes largely from people who have no record yet. The agent handles them as prospects: answers eligibility and process questions from the prospectus, captures name, programme of interest and contact with consent, and books counselling slots. Once an application exists, it verifies the applicant against the application ID and can report status. The lead capture feeds the admissions CRM; institutions on TheEazy CXM get this as a native integration.

Languages and channels

An education WhatsApp bot is often the first channel to add, because it is where students and parents already are, and WhatsApp handles document images and voice notes naturally. The web widget and the student app share the same agent and conversation history. Language handling is practical rather than clever: detect the language, answer in it, keep official terms (programme names, fee heads) in the form used in the institution's documents, and let the student switch at any time. Regional-language answers must come from reviewed content, not on-the-fly translation of the prospectus, because a mistranslated eligibility rule is a complaint. The WhatsApp Business Platform documentation sets out template and session-message rules that shape what the agent can send proactively, such as deadline reminders.

Escalation with context

Complaints, fee disputes, grievances, anything involving a medical or personal circumstance, and any request the agent cannot ground go to a person. The hand-off carries the student's identity, the question, the transcript and the documents retrieved, and routes to the right desk: admissions, accounts, examinations, hostel. Staff answer from their existing helpdesk or from a simple queue we provide. The weekly review of escalation reasons is where the agent improves: a cluster of "scholarship eligibility for lateral entry" questions means a document is missing or unclear, and fixing the document fixes the agent.

Measuring it

Measure resolution by intent, not deflection: a student who gave up is not a success. Track reopens (the same student asking the same thing within a week), escalation reasons, actions executed within policy (with zero outside), and satisfaction asked once at the end. Report separately for the admissions peak and the fee-deadline peak, because they are different populations with different questions. The argument for resolution over deflection is in AI ticket deflection is the wrong metric.

A worked example

A private university group with several campuses handled admissions and fee questions through a call centre that was overwhelmed for six weeks a year and idle for the rest. We built a grounded agent on WhatsApp and the admissions website, fed by the prospectus, fee circulars and the academic calendar with version control, connected read-only to the ERP for verified students and to the admissions CRM for prospects. Actions were limited to receipts, ticket creation and counselling bookings. The agent ran in shadow mode for two weeks, drafting answers that staff reviewed, before going live in English, Hindi and Kannada. During the following admissions season the call centre handled disputes and complex cases with transcripts attached, and the registrar's office used the escalation report to rewrite two confusing circulars. The multilingual voice agent for a hospital network shows the same language and escalation approach on the phone.

Team and timeline

A student support agent is typically an AI engineer for retrieval, prompts and evaluation, an integration engineer for the ERP and WhatsApp connections, and a content owner from the registrar's office, over six to ten weeks. The first two weeks assemble and version the source documents and build the evaluation set from real questions; the middle weeks build retrieval, actions and channels; the last weeks run shadow mode and go live before the peak. Plan to launch at least six weeks before admissions open. Customer service agent builds start from $12,500 / ₹8L, and a Standard Care Plan covers the peaks with 24×5 support. Current figures are on the pricing page, and the education sector page describes the rest of what we build for institutions.

Before you start: a checklist

• List the top fifty questions from last year's admissions and fee peaks, with the correct answers
• Name a content owner who approves the documents the agent may use
• Confirm how the ERP or SIS exposes student data, and how identity is verified
• Decide which actions the agent may take and write the policy for each
• Choose the languages and have the reviewed content ready in each
• Set up the WhatsApp Business account and templates early; approval takes time
• Agree the escalation desks and their response targets
• Fix the launch date at least six weeks before the peak

Questions clients ask

• Can it answer results queries? It can report release dates and, after verification, direct the student to their result; it should not read out marks on WhatsApp unless the institution's policy allows it.
• What about parents? Parents can be verified as registered guardians and given the same read access; anything beyond that goes to staff.
• Will it replace the call centre? No; it takes the repetitive volume so the call centre can handle disputes and complex cases properly.
• How do we keep it current? The content owner uploads each new circular; the retrieval layer versions it and the agent quotes the latest.

Related reading

See multilingual customer support with AI for Indian businesses for language handling, from FAQ bot to support agent: a migration plan if you already have a bot, and learner data protection for consent and access rules when students are minors.

Ground the agent in your own rules, connect it to the record, gate every action, and the admissions peak becomes manageable.