PUT IT INTO PRACTICE

Final project From customer inquiry to reviewed reply

By Andrew Heldenmuth, founder and operator of Ascension AI.

Suggested time: 15 minutes.

Build a repeatable workflow for turning a routine customer inquiry into a reviewed reply. Use the fictional bicycle shop and approved facts from Lesson 6, or an entirely fictional business like your own. Finish with a saved draft and a completed review checklist; sending the reply is outside this exercise.

Use these two fictional inquiries. They are similar but not identical, so the second attempt does not simply copy the first:

Inquiry A: “I use my bicycle to get to work. Could you look at the gears on Thursday? Can you tell me the price before I book?”
Inquiry B: “My rear gear keeps slipping on hills. Could I come in Friday afternoon, and roughly what will it cost?”

Availability, cost, and completion time are unconfirmed for both. Each reply should acknowledge the request, explain the inspection requirement, and avoid promising the requested day or giving any price, even a rough one. Flip a coin to decide which inquiry gets the manual attempt and which gets the AI-assisted attempt.

Build your workflow card

Copy and complete these fields:

Copyable prompt

Task:
Who will use the output:
Allowed input information:
Information that must stay out:
Tool and account approved for this work:
Reusable prompt:
Checks before use:
Person responsible for approval:
Situations that require escalation:
Total time for a manual attempt:
Total time with AI, including review and corrections:
Corrections needed:
Decision: Keep testing, revise, or stop:

Spend three minutes choosing the task and writing the prompt. Spend five minutes writing a reply to one inquiry manually. Spend five minutes writing a reply to the other inquiry with AI assistance, including review. Use the final two minutes to compare quality and effort. If the task needs longer, finish the comparison rather than relying on an incomplete result.

For repeated trials, record actual time saved as manual minutes minus AI-assisted minutes. A negative result means the AI-assisted attempt took longer. One attempt is a starting observation, not proof of future savings. Because the two inquiries differ slightly, repeat the comparison later with the methods swapped before you trust the result.

Ready to complete the course

I can explain what AI helps with and why its answers need checking.
My prompt states the task, context, boundaries, and output format.
My practice data is fictional and contains no real confidential information.
I have checked every important claim and commitment in the output.
A named person reviews and approves the work before it is used.
I have compared the full effort with a manual attempt and recorded the result.

Your next step: Repeat the same low-risk task with a few different fictional examples. Keep the workflow only if it consistently produces work you can trust after review. Before using real business data, confirm the tool, data permissions, and review process with the person responsible for your business policies.

← Build a workflow Next: Sources →