Page 200 - CyberSecurityEssentialsEnglish

Chapter Five
From Tasks to Workflows

Email, documents, data, presentations, and decisions
How to turn the use of AI from an individual experiment into a process that can be repeated,
measured, and improved
Chapter question: how do we stop a success with AI from remaining an individual

experiment, and turn it into a way of working that can be repeated, measured, and
improved?
An early summary: a tool produces an output, while a workflow makes that output

deliverable, reviewable, and improvable.
Story: When the Experiment Succeeded but the Process Did Not Hold
In 2024 McDonald’s announced that it would end a test of automated voice ordering in its

drive-through lanes, a trial it had developed with IBM since 2021 across a number of
locations. The problem was not that AI cannot help. It was that a wide operational trial
exposed how hard it is to rely on one clever step without a clear route for human

intervention, error handling, the logging of exceptions, and the improvement of the system
when orders are misunderstood [11].

In a fast-food restaurant, the task is not merely understanding a customer’s sentence. There
is noise, there are accents, a compound order, an unavailable item, a last-second change,

payment, and handover. If the system gets it wrong, the employee must know when to
intervene, how to regain control, how the case is corrected, and where recurring errors are
recorded. Without those links, AI looks impressive in a first demonstration and fragile in daily

operation.

This story serves as an entry point to every institutional use of AI. A model may succeed at
writing a draft, summarizing minutes, or extracting a figure, but an organization gains no
lasting capability unless it knows the approved inputs, the steps, the roles, the decision

points, the quality gates, the records, and who owns the final decision.

“A good output saves time once. A good workflow saves time and protects the quality every
time.”

◆ Key idea: value is measured by repetition, not by a single output.
The professional value of AI is not measured by the elegance of one answer. It is
measured by its ability to support a clear process that can be repeated, measured,
handed to another person, checked for errors, and improved over time.

199