Where AI fits in automation
Rule-based automation handles predictable steps: if a fee is overdue by seven days, send a reminder. Many administrative tasks do not fit clean rules. A student email might be a transcript request, a complaint or a fee question; a scanned file might be a mark sheet or an identity proof. AI steps classify and extract from this unstructured input so that rule-based workflows can take over, with people reviewing anything uncertain.
Task types and the right level of automation
| Task | AI contribution | Human checkpoint |
|---|---|---|
| Sorting incoming requests | Classifies emails and portal requests by type and urgency | Staff can re-route; low-confidence items go to a person |
| Extracting document data | Reads names, dates and marks from uploaded documents | Staff verify extracted values before they update records |
| Drafting routine replies | Prepares answers from approved knowledge | Staff approve before sending |
| Summarising long threads | Condenses case history for hand-over | Receiving staff member reviews |
| Tagging feedback | Groups survey comments by theme | Analysts check themes before reporting |
Planned building blocks
Trigger
An event or schedule starts the workflow, such as a new request arriving in the student portal.
AI step
Classify the request, extract details or draft a response.
Confidence gate
Above your threshold the workflow continues; below it the item goes to a person.
Approval
Named roles approve drafts or decisions, with deadlines and escalation.
Action
Update the record, send the approved message or create a task.
Log
Inputs, AI outputs, approvals and final actions are stored for audit.
Worked example: a planned transcript request flow
A student writes, “I need my transcript sent to a university abroad by Friday.” The AI step classifies it as a transcript request and extracts the destination and deadline. The workflow checks finance and library clearance automatically. If cleared, it generates the transcript and prepares a secure sharing link, and a registrar approves the release with one click. If not cleared, it drafts a message explaining what is outstanding for staff to send.
Controls planned for administrators
- Enable AI steps per workflow, not globally
- Set confidence thresholds per step
- Require approval for any outbound message or record change
- Review a sample of AI outputs each week
- Switch AI steps off instantly without breaking the workflow
Frequently asked questions
What is an AI workflow?
A workflow that includes AI steps, such as classifying, extracting, drafting or summarising, alongside rules and human approvals.
What happens when the AI is unsure?
Items below your confidence threshold are routed to a person instead of continuing automatically.
Can AI steps be switched off?
Yes, that is the plan. Administrators can disable AI steps per workflow at any time without breaking the rest of the process.