INDUSTRY FIELD GUIDE
Contact center call escalation and quality rubric data for AI
Quality teams can test whether an evaluation follows the correct branch for a caller’s intent.
NAICS 561422 · Telemarketing Bureaus and Other Contact Centers
Quality teams can test whether an evaluation follows the correct branch for a caller’s intent. AWS Contact Lens publicly supports conditional questions on evaluation forms. The FTC provides separate guidance for telemarketing. Together these sources explain why a scripted sample should record call type and which evaluation applies. AWS Contact Lens evaluation forms · FTC Telemarketing Sales Rule guidance
Write paired fictional calls. One caller asks about a policy and gets an answer. The other disputes a prior action, triggering a rubric branch for escalation. Give each transcript a version and ask the model to quote what triggered the branch and whether the agent actually completed the scripted transfer. An independent human evaluation anchors the expected labels.
Obtain written rights for actors and original scripts. Exclude real recordings, customer identifiers, account details and operational playbooks. Confirm the proposed AI use before delivering audio.
Describe a synthetic contact center evaluation
Original Fiund exercise
Caller intent selects the rubric branch, while a promised transfer needs completion evidence
Two original fictional calls are text scripts with adult roles. No real audio, account, customer, operational playbook or outbound contact is included.
| Record | Source | What it records |
|---|---|---|
| CCQ1 | Practice rubric v2 | Information-only question uses INFO branch. Explicit dispute of a prior action uses ESCALATE; completed transfer requires a receiving-agent acknowledgment in the script. |
| CCQ2 | Call A script | Caller: 'What is the policy?' Agent explains original mock policy. Caller asks no dispute. No transfer event. |
| CCQ3 | Call B script | Caller: 'I dispute the earlier change.' Agent: 'I will transfer you.' Script ends before any receiving-agent acknowledgment. |
Question for the model
Which branch applies to each call, which exact words trigger B's branch, and was its scripted transfer completed?
Show the reference answer and checks
CCQ1 selects INFO for CCQ2's policy question. CCQ3's explicit 'I dispute the earlier change' triggers ESCALATE. The agent promises a transfer, but the script lacks CCQ1's required receiving-agent acknowledgment, so completion is unverified. Score branch selection and action completion separately rather than penalize A for no transfer or treat B's future-tense promise as an executed handoff. This evaluates the original rubric, not actual regulatory compliance.
- Applying escalation checks to an information-only branch.
- Treating an agent's transfer promise as a completed receiving-agent handoff.
Fields to preserve
- Script and rubric version
- Defines the evaluation basis.
- Intent-trigger span and selected branch
- Makes classification inspectable.
- Promised action and completion evidence
- Separates intention from execution.
Test another case
Add a dispute paraphrase without the word dispute and a completed receiving-agent reply. Require semantic intent matching and independent completion scoring.
Use original scripts and written actor rights for any audio. Exclude real calls and customer details; confirm the intended AI use before delivering recordings.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.