AI Student Enquiry-to-Counselor Playbook

When a prospective learner asks about fees, intake dates, and eligibility in one message, some facts may be publishable while the individual decision is not. This playbook answers the verified part, collects only what the next step needs, and moves the unresolved question to a counselor without pretending to decide admission.

Copy the pilot card below into a document or spreadsheet and begin with one real enquiry type. If the official source or counselor route is missing, stop and repair it before automating an answer.

Copy the pilot card and define completion

Enquiry transition	Source to check	Allowed result	Stop or human handoff
When [learner question] occurs, move from [start] to [end]	[program, intake, and calendar source]	[verified answer, session, or no action]	[decision/data conflict and receiver]

• Start: one enquiry break has an owner, baseline, official program corpus, approved data fields, and audience rule.
• End: the learner receives a verified answer, accepted counselor handoff, confirmed information session, documented application-support state, or stop.
• Evidence: content version, question, approved fields, permission, answer/action log, handoff receipt, correction, and outcome.
• Time horizon: one intake cycle with changed deadlines, missing evidence, international questions, minors/sensitive data, duplicate records, and opt-out.

Roles and prerequisites

Role	Owns	Approves	Receives handoff
Program owner	Curriculum, intake, prerequisites, published fees	Program answer	Content conflict
Admissions owner	Application process and decision boundary	Admissions action	Applicant case
Counselor/student services	Human guidance and accessibility support	Conversation next step	Complex enquiry
Privacy/safeguarding owner	Approved data and audience controls	Sensitive exception	Risk signal
Workflow owner	Pilot, QA, and expansion	Keep/pause/expand	Cross-team issue

Inventory official program pages, intake/deadline, fee and refund wording, prerequisite definitions, application status, counselor roster, event calendar, consent, accessibility, language, age/audience rule, and retention. Define what the assistant cannot decide.

Phase 1: map the enquiry and enrollment boundary

• Owner: program owner with admissions.
• Inputs: official corpus, application states, calendars, service queues.
• Actions: trace discovery, program question, counselor route, application support, decision boundary, enrollment, and stop.
• Output: source/state map with owner, freshness, and prohibited actions.
• Exit gate: every automated answer has an official source; every decision has a human/system owner.
• Escalation/rollback: stop where program versions or applicant states cannot be reconciled.

Phase 2: choose one low-risk transition

• Owner: workflow owner with privacy/safeguarding review.
• Inputs: enquiry evidence, baseline, approved fields, representative exceptions.
• Actions: select one job such as published-course question to counselor; minimize data; define no-answer and no-collection rules.
• Output: bounded transition and acceptance criteria.
• Exit gate: no admission, scholarship, credential, accommodation, or safeguarding decision is delegated.
• Escalation/rollback: route rather than infer when age, identity, accessibility, immigration, financial aid, or eligibility is material.

UNESCO's guidance frames generative AI in education around a human-centered approach. It supports governance review, not a claim that one workflow is compliant or effective. Review the source.

Phase 3: validate program facts and minimize data

• Owner: program owner for facts; privacy owner for collection.
• Inputs: current intake, prerequisites, published fees, deadlines, approved form schema.
• Actions: retrieve the versioned answer; ask only the minimum necessary field; label uncertainty; avoid requesting documents in chat unless the approved secure process requires it.
• Output: eligible answer/context or secure human route.
• Exit gate: answer scope, source date, applicant state, and data channel agree.
• Escalation/rollback: retract stale information, delete/contain wrongly collected data under policy, and route the case.

A Vietnamese public warning about personal-information exposure during admissions supports verified-channel and data-minimization controls. It is not a complete legal analysis. Review the source.

Phase 4: execute and reconcile the handoff

• Owner: counselor or admissions owner accepts the next step.
• Inputs: eligible context, approved response, live appointment/application state.
• Actions: send one answer/action; preserve the original question and source; confirm a session only after calendar write; record any correction.
• Output: answered, booked, accepted, stopped, or human-owned state.
• Exit gate: learner-facing, calendar/application, and owner states match.
• Escalation/rollback: suppress confirmation after partial write and return to the last verified state.

Phase 5: review and expand by program cell

• Owner: workflow owner with privacy and admissions.
• Inputs: answer validity, reconciliation, handoffs, corrections, opt-outs, sensitive events.
• Actions: review by program, intake, language, audience, channel, and enquiry type; inspect all high-risk cases.
• Output: signed keep/pause/expand decision.
• Exit gate: pre-set quality and containment thresholds hold.
• Escalation/rollback: disable only the affected program/intake/question cell.

Fictional completed run

A fictional continuing-education provider pilots enquiries for an adult data-analysis certificate. A prospective learner asks about prerequisites, tuition, and a September intake. The official source confirms the fee and intake but defines prerequisite evidence ambiguously. The assistant answers the two verified facts, does not judge eligibility, and sends the original prerequisite question to a counselor. When the learner attempts to paste a passport number, the collection rule stops the field and points to the approved application portal. The counselor accepts the handoff and records the missing-evidence instruction. No admission, completion, or enrollment-lift claim is made.

Enquiry transition	Source to check	Allowed result	Stop or human handoff
Certificate enquiry to verified answer and counselor route	Current fee/intake page and approved application channel	Answer fee and intake; offer information-session step	Ambiguous prerequisite goes to counselor; passport data is blocked and moved to the secure portal

Handoff, QA, and measurement

The Trigger column is the join key between the two compact tables.

Trigger	Severity	Owner
Eligibility/decision ambiguity	High	Admissions/counselor
Sensitive or excessive data	Critical	Privacy/safeguarding
Deadline/calendar conflict	High	Program owner

Trigger	Required context	Response target	Fallback
Eligibility/decision ambiguity	Program/version, original question, evidence state	Before advice	Human decision
Sensitive or excessive data	Field, channel, audience, action withheld	Immediate	Stop collection, secure route
Deadline/calendar conflict	Intake/date/version, requested slot	Before confirmation	Withdraw and verify

QA checks: official content version; intake/deadline timezone; fee scope; minimum fields; minor/audience rule; secure document channel; accessibility; opt-out; decision boundary; accepted handoff; rollback.

The Metric column is the join key between the two compact tables.

Metric	Definition	Baseline
Supported-answer rate	Answers traceable to a current approved program source / attempted answers	Pre-pilot test set
Handoff acceptance rate	Required handoffs accepted with complete context / required handoffs	Manual baseline
Sensitive-data containment	Detected unapproved fields not retained or repeated by the workflow / detected fields	Safety test set

Metric	Decision threshold	Owner	Cadence
Supported-answer rate	Set before launch	Program owner	Weekly
Handoff acceptance rate	Set before launch	Counselor owner	Weekly
Sensitive-data containment	100% required	Privacy owner	Weekly

Failure	Early signal	Corrective action
Invented eligibility or stale program fact	Answer lacks a current source or conflicts with the program version	Retract the answer and route the question to the program or admissions owner
Excessive/insecure collection or audience risk	Unapproved field, document, minor, or sensitive context appears in chat	Stop collection, contain it under policy, and move to the approved privacy/safeguarding route
Decision, access, or handoff failure	Chat produces an admissions outcome, next step is inaccessible, or no counselor accepts	Suppress confirmation and require an accepted counselor/admissions alternative

Pause the affected program, intake, or question cell until the corrective action is verified.

Frequently asked questions

Can the assistant tell someone whether they will be admitted?

No. It may explain published criteria and process, then route any individual decision to the authorized admissions owner.

What information should it collect first?

Only approved minimum fields needed for the selected next step, such as program interest, preferred contact channel, and session time. Do not collect documents by default.

Can it answer tuition or deadline questions?

Only from a current official source with scope and date. Route scholarships, exceptions, refunds, and individual eligibility to their owners.

How should Vietnamese localization differ?

Use the institution's official Vietnamese program terms, local intake/channel practice, and approved privacy wording; do not merely translate foreign admissions concepts.

What to do after completion

Limits. This playbook is not educational, admissions, immigration, privacy, or legal advice. It claims no enrollment benchmark, decision accuracy, integration, or Easy AI capability.

Use the customer-data guide to govern the minimum applicant context, the lead assignment template for counselor ownership, and the AI-to-human handoff guide for context completeness.