Students Have Questions at 11 PM
A prospective student is researching your university at 11pm. She has questions about the application deadline for international students, whether her qualifications are accepted, and what scholarship options exist. Your admissions office opens at 9am tomorrow.
She has four other universities open in browser tabs. The one that answers her tonight gets her attention. The rest are waiting for a morning she may not come back to.
That's not a hypothetical. UCAS data consistently shows that applicants who receive fast, specific responses to their initial enquiries are significantly more likely to firm their choice at that institution. The conversion window for many applicants — particularly international students choosing between multiple countries — is hours, not days.
Education is a relationship-driven sector — students choose institutions based on trust, quality, and feeling supported. The admin layer around that relationship — the queries, the paperwork, the logistics — doesn't need human expertise. It needs speed, accuracy, and availability.
AI agents handle that layer around the clock, so your staff can focus on the human work that actually builds the relationships.
Where AI Agents Deliver Value in Education
Admissions and Enquiry Handling
Prospective students ask hundreds of questions during the admissions process: entry requirements, application deadlines, course content, fees, scholarship availability, campus facilities, student life. Most have clear, factual answers.
An agent handles the initial enquiry layer immediately — answering specific questions, guiding students through the application process step by step, and routing complex or personal queries to an admissions officer who can give them proper attention.
A mid-sized UK university we've worked with was handling around 2,400 inbound enquiries per month during the September UCAS cycle. Their admissions team of six people was visibly underwater — average response time had drifted to 18 hours, and the agents' feedback consistently flagged that they were spending most of their day answering the same 40 questions repeatedly rather than doing actual admissions work. The AI agent brought response time to under two minutes on standard queries and freed the team to focus on borderline applications and personal statement guidance. First-choice firm rate improved by 14% in the following cycle — not solely because of the agent, but the team attributed a meaningful portion of it to faster, more reliable initial communication.
For institutions with large international applicant pools, this is particularly valuable. Queries come in from multiple time zones, in multiple languages, at all hours. The agent handles all of them at once without needing someone manning an inbox at 3am Manchester time.
Enrolment and Registration Support
Once a student is accepted, the enrolment sequence begins. Module selection, accommodation applications, finance forms, IT account setup, library registration. Each step has questions, and often confusion.
An agent guides new students through enrolment systematically — answering questions at each stage, reminding them of deadlines, chasing outstanding steps, and escalating to the relevant department when a student is stuck on something that genuinely needs human resolution.
Consider what this looks like for a student starting a computer science degree: they need to select 120 credits of modules, some of which have prerequisites or timetable clashes. They need to apply for student finance, set up their institutional email, register for accommodation, and confirm their enrolment by a specific date. Each step involves a different system and a different team. A well-built agent holds the whole sequence in view — it knows which steps the student has completed, which are outstanding, and which deadlines are approaching — and communicates that clearly without the student needing to figure out which department to contact about each thing.
Dropout at the enrolment stage — students who accept an offer but never complete registration — is a measurable problem for many institutions, and a quiet one because it's easy to miss until you look at the numbers. Proactive support during this window tends to reduce it.
24/7 Academic and Administrative Support
Current students generate a constant stream of admin queries across the year: timetable questions, assessment submission, grade queries, library access issues, IT problems, financial aid questions.
Most have straightforward answers that don't need an adviser. An agent handles them at any hour — the student studying at midnight who can't figure out how to submit their assignment gets an answer immediately, not at 9am the next morning, by which time the deadline may have passed.
For a university with 15,000 students, that's not a trivial number of queries. Student services teams at institutions that size typically handle 300–500 tickets per week during term, a significant fraction of which are questions with factual answers already documented somewhere the student couldn't easily find. An agent trained on the institution's policies, procedures, and systems can handle the bulk of those without human involvement.
Course and Content Guidance
For EdTech platforms and online learning providers, the agent can go further: helping students navigate course content, recommending next modules based on their progress, answering questions about specific topics covered in the course, and providing guidance on their learning path.
Take a professional certification platform with 50,000 active learners across a data science curriculum. The most common drop-off points are predictable — learners stall when they hit concepts they don't understand and have no obvious next step. An agent embedded in the platform can identify when a learner has been inactive for three days, check where they stopped, and send a targeted message: not a generic "We miss you!" email, but "It looks like you stopped at the probability module — that one trips a lot of people up. Here's a shorter explainer video that might help, and here are two practice exercises that approach it differently." That's a qualitatively different intervention than anything a team of five support staff could deliver across 50,000 users.
This isn't replacing instructors. It's the scaffolding that helps students keep moving between instructor interactions — which is where most online learners get stuck and quietly drop off.
Student Wellbeing Check-Ins
Agents can be configured to proactively check in with students at high-stress points in the academic year — exam season being the obvious one. A simple message asking how someone is doing, providing links to support resources, and offering to connect them with student services.
This is not a replacement for counselling. It's an early-touch system that helps students who might not proactively seek help know that support is available and easy to access.
Any check-in agent must have an immediate escalation to human support for students showing signs of serious distress. That's a design requirement, not a nice-to-have. We've watched this fail when teams underspecified the escalation logic, and the only acceptable failure mode here is to escalate too readily, not too rarely.
Alumni Engagement
For universities, alumni relationships drive donations, mentorship programmes, and employer partnerships. An agent handles the communication layer — event invitations, updates, reunion coordination, fundraising campaigns — personalised to each alumnus's history and interests.
A 1996 alumna who studied engineering and now works in renewable energy is more likely to engage with a mentorship request from a current engineering student than with a generic newsletter. An agent can identify that match, draft the personalised outreach, track whether she responded, and follow up appropriately — at a scale a small alumni relations team simply cannot replicate manually.
What AI Agents Should Not Do in Education
Clear boundaries matter as much here as in healthcare or financial services.
Agents should not make academic judgements — grading, assessment feedback that counts toward qualification, decisions about academic misconduct. These require human expertise and carry professional responsibility.
They should not replace pastoral care or counselling. A student in genuine distress needs a human. The agent's only job in that moment is to make sure they reach one as quickly as possible.
And they should not make admissions decisions. Eligibility information, process guidance — fine. The decision itself, no.
Off-the-Shelf Chatbot vs Custom-Built AI Agent
Many institutions start by trying a generic chatbot platform before realising it doesn't fit how they actually work — the gap is the same one we cover in AI agents vs chatbots. Here's where the distinction matters:
| Capability | Off-the-shelf chatbot | Custom AI agent |
|---|---|---|
| Answers institution-specific policy questions | Partial — needs significant manual FAQ loading | Yes — trained on your actual documents |
| Integrates with your SIS / LMS / CRM | Rarely without custom dev work | Built to integrate with your specific stack |
| Handles multi-step enrolment flows | No | Yes |
| Escalates to the right department based on query type | No — routes to one inbox | Yes — routes to admissions, finance, IT, etc. |
| Supports multilingual students | Varies by platform | Yes, configurable by language |
| Proactive outreach (deadline reminders, check-ins) | No | Yes |
| Built-in escalation logic for distress signals | No | Yes, required |
| Timeline to deploy | 1–4 weeks for basic setup | 6–8 weeks for full integration |
| Ongoing customisation | Limited | Full control |
The off-the-shelf route makes sense for very small institutions with modest query volume and no integration requirements. For anyone with a complex student information system, multiple departments, and meaningful international student populations, custom builds consistently outperform.
What to Expect in Practice
The first two weeks of a deployment are almost always revealing. You discover that 30% of the query volume you thought was about admissions is actually about fees and finance — which means your finance team needs to be part of the content build, not just your admissions team. You find edge cases in your own policies you didn't know existed until the agent had to give a definitive answer about them.
That process is useful, not a problem. Building the agent forces institutions to document and clarify policies they've been handling case-by-case through institutional memory. The discipline of training the agent often improves consistency across the institution — staff start getting clearer answers from the agent than they were getting from each other.
A typical timeline for a well-scoped deployment:
- Week 1–2: Map the student journey, identify highest-volume query types, gather existing FAQ and policy content
- Week 3–4: Build and integrate with your student information system, CRM, or learning management platform
- Week 5: Testing with staff playing the role of students — including adversarial testing for boundary cases
- Week 6: Soft launch to a subset of students with close monitoring
- Week 7–8: Full deployment
Eight weeks to a production-ready student support agent. Wellbeing check-in features should go through additional review and probably a longer pilot.
Common Mistakes Institutions Make
Loading the agent with policy documents and calling it done. An agent trained only on PDFs from your website will answer questions about the documents, not about how things actually work. You need to build in the procedural knowledge that lives in your team's heads — the exception cases, the escalation paths, the things your admissions officers know but have never written down.
Skipping the adversarial testing phase. Students are creative, sometimes frustrated, and occasionally trying to find loopholes. A student asking "what happens if I miss the fee deadline" needs a clear, accurate answer. A student asking the agent to "tell me a way I can get an extension without penalty" needs a firm, honest redirect. You won't catch those scenarios unless someone on your team actively tries to break the agent before it goes live.
Underspecifying what happens when the agent doesn't know something. Every agent has knowledge gaps. The question is whether it says "I don't know, here's how to find out" or confidently produces a plausible-but-wrong answer. The latter is worse than no agent at all. Confidence calibration is a design choice, not a default.
Treating wellbeing check-ins as a simple add-on. This feature carries more responsibility than any other part of a student support deployment. It needs dedicated review, input from your counselling and student services teams, clearly documented escalation criteria, and probably legal review. Institutions that rush it because it seems simple have found themselves in difficult positions when a student in crisis received an automated response that wasn't equipped to handle the situation.
Where This Doesn't Fit
If your institution genuinely runs on small-cohort, high-contact teaching and your admin volume is modest, an agent is probably over-engineering. The right fit is institutions with meaningful enquiry and student support volume — where the routine queries are visibly drowning the people you'd rather have doing other work.
Related guides
- AI agents for nonprofits: do more with less
- AI agents for customer support
- How AI agents handle multilingual support
- AI agents for membership organisations
- Our AI agent development services
Ready to Support Every Student, Any Hour?
The questions your students are asking at 11pm deserve answers. The admin load your admissions and student services teams carry doesn't all have to fall on people.
Talk to us about your business — we'll help you identify the highest-impact use case for your institution and tell you honestly if a deployment makes sense yet.
Frequently Asked Questions
How much does it cost to build an AI agent for a university or EdTech platform?
A custom AI agent for a mid-sized university — covering admissions enquiries, enrolment support, and student services — typically costs between £25,000 and £65,000 to build, depending on the number of systems it needs to integrate with and the complexity of the escalation logic (our AI agent development cost guide breaks down the drivers behind that range). Ongoing costs (hosting, maintenance, retraining as policies change) usually run £500–£2,000 per month. Off-the-shelf chatbot platforms start cheaper but often require significant customisation work that closes the gap, and they don't integrate as cleanly with complex student information systems.
Will students actually use an AI agent, or will they just ask to speak to a human?
Adoption depends almost entirely on quality. An agent that gives fast, accurate, specific answers gets used — and students stop asking for a human on routine queries because the agent gives them what they need. Institutions that have deployed well-trained agents typically see 70–80% of standard enquiries fully resolved without human escalation. The requests for human escalation that remain tend to be genuinely complex cases, which is exactly what you want.
How do you make sure the agent gives accurate answers about our specific policies?
The agent is trained on your institution's actual documents — policy handbooks, course pages, fee schedules, accommodation terms — and tested extensively against real query scenarios before launch. The build process includes a dedicated phase where your staff review answers to hundreds of test questions and flag anything inaccurate. Accuracy on policy-specific questions is higher than most institutions expect, because the agent isn't guessing — it's referencing the source material directly.
Can the agent handle international students querying in languages other than English?
Yes. A properly built agent can respond in the language the student writes in, with no configuration required from the student's side. For institutions with significant French, Mandarin, Arabic, or Hindi-speaking applicant pools, this is one of the more straightforward features to implement and one of the most impactful — it removes a barrier that previously meant international students either wrote in imperfect English or didn't enquire at all.
What happens when the agent gets a question it can't answer?
The agent should acknowledge that it can't answer, explain what it can do, and provide a clear path to a human who can help — whether that's an email address, a phone number, or a direct handoff to a live chat if your team is available. The wrong behaviour is a confident wrong answer, and preventing that is a core part of how we build escalation and confidence logic into the system. If the agent isn't sure, it says so.
How long does it take to get an AI agent live for student support?
A focused deployment covering admissions enquiries and basic student support typically goes live in six to eight weeks. That includes integrating with your student information system, training the agent on your policies and FAQs, internal testing, and a monitored soft launch before full deployment. More complex scopes — multiple department integrations, wellbeing features, multilingual requirements — extend that timeline, usually to twelve weeks.
What are the data privacy considerations for storing student conversations?
Student conversation data must be handled in compliance with GDPR (for UK and EU institutions) or FERPA (for US institutions), which means clear data retention policies, appropriate storage location, and no use of student interaction data to train third-party models without explicit consent. Any reputable implementation partner should have this documented before deployment starts — if they don't raise it, that's a red flag. We build with data residency and retention policies agreed and documented before any system goes live.