When the Answer Exists But Nobody Finds It: The Case for Smarter Virtual Agents in Education

The numbers behind the frustration

According to Zoom’s research, 43% of users say chatbots fail to resolve their issues. Industry data reinforces this: over 7 in 10 adults find customer service chatbots frustrating, according to Backlinko, and 85% believe their problems usually require a human to solve.

These findings reflect the general state of first-generation chatbot deployment: rule-based, rigid, and poorly matched to the complexity of real questions.

For educational institutions, that complexity is considerable. A prospective student asking about psychology programmes has a fundamentally different need from a current student who cannot find their timetable, an international applicant navigating visa requirements, or a parent trying to understand tuition costs. One digital front door must serve all of them, simultaneously, accurately, without friction.

The stakes are higher than they might appear. In the Netherlands, student satisfaction is measured and published by the Nationale Studenten Enquête. A digital experience that consistently fails students does not stay invisible; it shows up in scores that shape enrolment decisions.

Why most chatbots fall short

Research suggests that 29% of chatbots fail due to poor intent recognition or a lack of contextual understanding (source: Genuitysystems). They answer the question that was typed, not the question that was meant. And they tend to break down precisely when a user most needs help.

In a retail context, a frustrated user moves on. In an educational context, a student who cannot obtain an answer to a question about a deadline or a financial aid requirement may miss something important. In healthcare, the consequences can be even more serious.

What a well-designed virtual agent actually does

The distinction between a first-generation chatbot and a modern virtual agent is significant.

A rule-based chatbot follows decision trees. A modern virtual agent uses natural language understanding to interpret what a user is actually asking, draws on a broader knowledge base, and routes complex queries to the right human, while preserving full context.

Zoom Virtual Agent 3.0, which Duppal deploys and manages for its clients, is built on this architecture. It maintains context across every touchpoint and draws on existing content, websites, knowledge bases, and internal documents. Not everything lives on a webpage, and every institution is structured differently. That is a design reality, not an obstacle. A well-configured virtual agent accounts for it from the start.

Security and privacy: a prerequisite, not an afterthought

For educational institutions and healthcare organisations, data privacy is a fundamental requirement. Zoom has worked with SURF, the collaborative organisation for IT in Dutch higher education and research, to adapt its policies specifically for education customers in the Netherlands and the EEA. Zoom Virtual Agent has its own dedicated Data Transfer Impact Assessment under GDPR. For healthcare, Zoom’s compliance framework includes HIPAA-aligned controls and Business Associate Agreements for the handling of protected health information.

At Duppal, security configuration is part of every deployment, and not a module added later.

Education: a multi-audience challenge

Universities, universities of applied sciences, and large secondary schools simultaneously serve an unusually wide range of visitor types. Each has different questions, different familiarity with the institution, and different tolerances for friction.

Research highlights the significant role well-designed AI tools play in supporting international students, a population that often faces both language barriers and unfamiliar administrative processes. When a virtual agent handles routine queries accurately, it also frees advisors and support staff for interactions that genuinely require a human.

What good implementation looks like

The technology is only part of the answer. Research consistently shows that to maximise the potential of AI tools, implementation must be supported by a solid framework (source MDP), one that accounts for the specific context, user population, and content landscape of the institution.

At Duppal, that means starting with a thorough mapping of an institution’s information architecture. Designing conversation flows that reflect the real questions real users ask. And ongoing management: monitoring performance, updating content, and refining the agent as the institution evolves.

A virtual agent is not a project with a go-live date and a handover. It is a living system. That is precisely what managed deployment provides.

The answer has always been there

Educational institutions and healthcare organisations have tons of information available. However, they lack systems that make that information genuinely accessible to every visitor, at any hour, without requiring them to know in advance where to look.

The technology to close that gap exists. What it requires is implementation that takes the complexity of these environments seriously.

That is the work Duppal does. And it starts with a conversation.