Page 269 - CyberSecurityEssentialsEnglish

TABLE 6-15: QUALITY CHECK FOR A LEARNER SUPPORT ASSISTANT

Dimension The review question

Does it rest on an approved guide, a service, and a clear update
Accuracy
date?

Does it distinguish general information from an individual
Limits
decision?

Does it state that it is an automated system and offer access to a
Transparency
person?

Privacy Does it request the least possible data?

Accessibility Does it work in clear language and across several means of access?

Safety Are there escalation rules, and have sensitive cases been tested?

Fairness Do we review differences in service between groups?

Is the outcome measured, rather than the number of
Impact
conversations?

6.6.1 Case Study: A Busy Support Center
◇ A composite teaching case

This is a composite teaching case provided for explanation, not an account of any
particular institution.
Background: a university center receives hundreds of questions about registration and
appointments. Problem: the human reply to complex cases is delayed by the volume of
repeated questions. Decision: build an assistant that draws on an official guide, answers
general questions, and assembles a structured summary before referral. Result: the
team’s time moved to exceptions and human cases. Lesson: the success was not in
replacing advisers, but in improving triage and context before a case reaches them.

The team began by gathering the actual questions rather than assuming them. Every answer
was tied to an official page with a review date. They then designed escalation categories:

conflicting data, a case not covered by the guide, a complaint, an accommodation tied to a
disability, and a personal crisis. The system was not allowed to issue a decision, only to
assemble the request number, the type of problem, and what had been tried, with the user’s

consent.

Testing showed that some short Arabic questions were misunderstood because of missing
context. Clarifying questions and visible options were added instead of asking the learner to

write a paragraph. The team also found that long replies increased repetition, so they

268