Page 49 - CyberSecurityEssentialsEnglish

Step Three: Prepare a Local Test Set
Collect five to ten cases that represent the real work; do not choose only easy examples.
Include a long case, a short case, one with missing information, and one containing local

terminology or a complex table. Write the acceptance criteria before running the tool so
that you do not change them after seeing the result.

Step Four: Run the Experiment and Measure
TABLE 1-10: TEMPLATE FOR MAPPING AI USE IN A REAL TASK

Impact of Acceptance Time Time
Task Role Notes
error criterion before after

Failed on one
Example:
Every figure is table and needs
Summarize Augment
Medium linked to the 45 min 20 min clearer guidance
a weekly ation
correct page or a better
report
source.

Step Five: Make a Limited Decision
After the experiment, make a clear, bounded decision: use the tool in a low-risk task;
improve the experiment by adding examples or sources; use it as an assistant without

delegating the decision; or do not use it because its risks exceed its benefits. Choosing not
to use the tool may be the right decision. The goal is to improve reliable work, not to increase

the number of tools.

Calculate value across the whole workflow. Drafting time may fall while review time rises.
Measure total time, corrections, errors, and the proportion of outputs accepted on the first

attempt. Add the cost of the tool and training where relevant. The adoption decision will
then rest on clear results, not an impression from a single experiment.

◇ Case study: Accelerating a weekly report
This composite teaching case is provided for explanation and does not describe a specific
event.
A team used a model to summarize notes scattered across messages and tables. In the

first trial, the system cut the time in half but confused two dates. The team redesigned
the workflow: it standardized the data source, required a reference for every number,
and kept final approval with the report owner. After ten test cases, preparation time fell
and less rework was needed to correct errors. The lesson is that improvement came from
a clear task, reliable source, and effective review, not from prompt wording alone.

? Question to consider: What must remain human?

In your chosen task, which part requires contextual knowledge, responsibility, or

48