Purpose
Use this lesson to produce helpful replies that respect policy and escalation limits. The aim is to produce work that a colleague can review and act on, not merely text that sounds convincing.
Lesson Explanation
Scenario: A customer requests a refund outside the normal window.
Better instruction: Draft a warm reply that explains the policy, offers the approved alternative, and escalates to a supervisor if the customer disputes the charge.
Support prompts must include the approved policy, permitted remedies, tone, and escalation trigger. Do not let an empathetic draft turn into an unauthorized financial commitment. The reply explains policy, offers an approved option, and identifies escalation.
Work in a short cycle: provide the relevant input, state the result and limits, inspect the draft against the acceptance criteria, then revise the instruction when a requirement is missing. Keep the final human decision with the person responsible for the work.
Practice Questions
-
A prompt asks an AI to “respond kindly to this angry customer” with no mention of company policy. What does this lesson say support prompts must include?
View Answer
The approved policy, permitted remedies, tone, and escalation trigger.
-
An AI-drafted support reply is warm and empathetic but accidentally promises a refund outside company policy. What does this lesson warn about this outcome?
View Answer
An empathetic draft can turn into an unauthorized financial commitment if the approved policy and permitted remedies aren’t explicitly stated in the prompt.
-
Write a prompt for a support reply to a customer requesting an exception to a return policy, following this lesson’s guidance.
View Answer
Something like: “Draft a warm reply explaining that returns after 30 days aren’t normally accepted, offering a store credit as the approved alternative, and escalating to a supervisor if the customer disputes this.”
-
A support reply prompt states the tone and the policy, but never specifies when the issue should be escalated. What is missing, per this lesson?
View Answer
An escalation trigger.
-
What does this lesson say a support prompt must include, alongside the approved policy and tone?
View Answer
Permitted remedies and an escalation trigger.
-
A drafted reply to a customer dispute offers a remedy that isn’t actually part of the approved policy options. What does this lesson say went wrong?
View Answer
The prompt likely didn’t clearly state the permitted remedies, allowing an unauthorized option to be offered.
-
A completed support reply explains the relevant policy, offers an approved alternative, and identifies when to escalate to a supervisor. What does this reflect?
View Answer
A well-guardrailed support reply consistent with this lesson’s full guidance.
-
Why does this lesson specifically caution that “an empathetic draft” can become risky?
View Answer
Empathetic language alone doesn’t prevent overstepping into unauthorized commitments – guardrails (policy, remedies, escalation triggers) need to be explicit regardless of tone.
-
A support prompt names the tone (warm, apologetic) but doesn’t specify what should happen if the customer becomes hostile or disputes the resolution. What is missing?
View Answer
An escalation trigger.
-
A customer support template explicitly states which specific remedies are approved for a given situation, rather than leaving remedy options open-ended. What principle does this reflect?
View Answer
Stating permitted remedies explicitly, one of the required elements this lesson identifies for support prompts.
-
Why might a support reply that’s well-written and empathetic still create a real business problem, according to this lesson?
View Answer
If it inadvertently makes an unauthorized commitment (like an unapproved refund), the tone quality doesn’t prevent the practical, financial consequence of that overstep.
-
A support reply prompt states the policy and remedies but not the tone. What risk might this create, based on this lesson’s broader principles about tone (from Part 2)?
View Answer
The reply might come across as cold or inappropriate for the situation, even if the factual content (policy, remedy) is accurate.