AI chat bots and agents can help student success teams respond to routine questions, draft replies, and route conversations without making every interaction depend on manual staff follow-up. Use them where the question, data source, and escalation path are clear.
Before you configure a bot, define the support process it belongs to. Staff should know what the bot can answer, when an agent can draft or send a reply, and how students reach a person when the conversation needs review.
Guiding principles
Use these principles to keep AI support useful and reviewable:
Start narrow: Give each bot a focused purpose, audience, and set of approved sources.
Set visible boundaries: Tell students when they are interacting with AI and what the bot can help with.
Plan escalation: Route sensitive, unclear, or unresolved conversations to staff.
Protect data: Limit knowledge sources and permissions to the information the bot needs.
Review behavior: Test conversations before launch and review live interactions after launch.
Step 1: Define the support scope
Start by deciding where AI support belongs in the student experience. A good first bot handles a predictable support need, such as answering questions about deadlines, office hours, appointments, or required forms.
Define the scope before configuration:
The student audience the bot should support.
The questions the bot should answer.
The topics the bot should avoid.
The point when staff should take over.
A narrow scope makes the bot easier to test and easier for students to understand.
Step 2: Choose between a chat bot and an agent
AI chat bots and agents support different types of work. Choose the tool based on whether the interaction should answer a question or move an Inbox conversation forward.
Need | AI chat bot | Agent |
|---|---|---|
Primary role | Answer questions from supplied knowledge sources. | Draft or send replies in an Inbox group, depending on the selected agent mode. |
Common channel | Embedded web chat or SMS. | SMS or web conversations associated with an Inbox group. |
Best first use | High-volume questions with stable answers. | Conversations where staff want AI-drafted responses or controlled auto replies. |
▶️ Action item: Review AI Chat Bots and Agents before deciding which tool fits the support process.
Step 3: Prepare knowledge sources and instructions
AI chat bots answer from the knowledge sources and system instructions you provide. Keep those sources focused on the student-facing task. A bot that answers registration questions, for example, should not need broad access to unrelated institutional pages.
Strong instructions should tell the bot how to handle uncertainty. Include rules for:
When to answer directly.
When to provide a link or office referral.
When to stop and escalate to staff.
Which topics are outside the bot's scope.
Example opening:
"I can help with registration deadlines, required forms, or booking an advising appointment. What would you like to do?"
Step 4: Configure escalation and agent mode
Escalation should be part of the design, not an afterthought. For AI chat bots, select an Inbox group when conversations should move to staff. For agents, associate the agent with an Inbox group and choose the mode that matches your review model.
Agent modes support different levels of staff control:
On-Demand: Staff select when the agent drafts a reply.
Auto Draft: The agent drafts replies automatically, but staff still send them.
Auto Reply: The agent drafts and sends replies automatically when the conversation fits the boundaries in the system prompt.
Start with staff review when the process is new. Move toward automatic replies only after the team has tested common and edge-case conversations.
▶️ Action item: Review Bots Overview and Inbox when AI support needs to hand conversations to staff.
Step 5: Connect AI support to student actions
The most useful AI interactions point students toward a clear next step. That next step might be a portal page, form, appointment link, office referral, or staff conversation.
Plan the handoff before launch. For each supported topic, document the student action and the Slate object that supports it. This helps staff test the bot and keep responses aligned with the current process.
▶️ Action item: Review Portals and Deliver when bot guidance connects students to self-service pages or follow-up communication.
Step 6: Test before launch
Use preview conversations to test whether the bot follows the intended boundaries. Include straightforward questions and questions that should trigger escalation.
Test the bot against scenarios such as:
A student asks a question that the bot should answer from an approved source.
A student asks for help outside the bot's scope.
A student provides incomplete or contradictory information.
A conversation includes sensitive language that should stop automated replies.
Update the system instructions when the bot answers too broadly, cites the wrong source, or continues a conversation that should move to staff.
Step 7: Measure and refine
Review bot activity after launch so the process improves over time. Reporting should help staff understand whether students are getting useful answers and whether escalations are happening at the right point.
Useful measures include:
Questions answered without staff intervention.
Conversations escalated to an Inbox group.
Topics that produce repeated unanswered or unclear responses.
Student actions completed after the conversation, such as form submissions or booked appointments.
Staff changes to drafted replies before sending.
▶️ Action item: Use Queries to plan reporting for bot activity, escalation, and follow-up outcomes.
Recommendations
A strong AI support process helps students get timely answers while preserving staff control over sensitive or uncertain interactions.
Start with one support process: Build the first bot around a focused student need.
Use trusted sources: Keep knowledge sources current and limited to the bot's purpose.
Make escalation obvious: Give students a clear path to staff help.
Review before automating: Use On-Demand or Auto Draft modes before enabling Auto Reply.
Keep improving: Use conversation review and staff feedback to refine instructions.
Further reading
Use these articles when you move from planning to configuration: