AI QA SCORING WORKFLOW

Score support transcripts and draft coaching notes.

A support QA workflow that scores transcripts against a fixed rubric, links findings to evidence, and drafts coaching notes for supervisor review.

HOW IT RUNSWeb input → OpenAI → validation → review packet

The live demo runs on the website. A matching, importable n8n version is available to download.

01 · INPUT

Paste a support transcript

Customer: Hi. My ClearAir 500 started making a loud rattling sound last night. I use it in my son's room and I am worried it could be unsafe.

Agent: I can help with that. Before we go further, please turn the unit off and unplug it.

Customer: Okay, it is unplugged now. I bought it about eight months ago.

Agent: Thank you. I am sorry this happened, especially with the unit in your son's room. I need the serial number and your order email so our product safety team can review it.

Customer: The serial number is CA5-88421. The email is customer@example.com.

Agent: I have noted both. I am flagging this for our safety team today. They will review the case before anyone asks you to use the unit again. Is there anything else you noticed, such as heat, smoke, or a smell?

Customer: No smoke or smell, only the rattling.

Agent: Understood. Keep it unplugged while we review the case. You will get an email with the case details. Is there anything else I can help with today?

Customer: No, that is all.

Agent: Thank you for reporting it. I am sorry again for the concern.

02 · PROCESS

A clear path from transcript to review

Download the n8n version ↓

1Receive transcriptWeb input→

2Score fixed rubricOpenAI→

3Check every fieldValidation→

4Flag uncertaintyReview gate→

5Prepare coachingSupervisor review

03 · REVIEW PACKET

Score with the reason attached

high confidence

91out of 100

Overall finding

The agent handled the safety concern well, showed empathy, and set a clear next step. The main gap is a missing case number and exact follow-up window.

Suggested score. A supervisor should confirm the rubric and evidence.

met

Safety and policy

100/100

“please turn the unit off and unplug it”

The agent gave the safe first step before troubleshooting.

30% of total

met

Empathy and ownership

100/100

“I am sorry this happened, especially with the unit in your son's room”

The apology is specific to the customer's concern.

20% of total

met

Discovery

95/100

“Is there anything else you noticed, such as heat, smoke, or a smell?”

The agent checked for the signs that would change the risk level.

20% of total

partial

Resolution and next step

75/100

“I am flagging this for our safety team today”

The owner is clear, but the agent did not give a case number or review window.

20% of total

met

Close and confirm

90/100

“Keep it unplugged while we review the case”

The agent restated the safe next step and checked for more needs.

10% of total

04 · COACHING DRAFT

Turn one gap into a useful next step

Editable · not sent

KEEP

What worked

You led with the safe action and made the apology fit the customer's concern.

CHANGE

Focus area

Make the handoff easier to track by giving the case number and a clear follow-up window.

PRACTICE

Next action

Role-play the last 30 seconds of a safety call. Close with owner, case number, time window, and the action the customer should take while waiting.

Supervisor noteReview policy, context, and the source transcript before using this note.