> ## Documentation Index
> Fetch the complete documentation index at: https://koreai.mintlify.app/llms.txt
> Use this file to discover all available pages before exploring further.
# By Hold Etiquette Metric
The By Hold Etiquette metric automatically evaluates how agents manage customer holds during voice interactions, ensuring consistent adherence to hold standards across all interactions.
The metric assesses three aspects:
| Aspect | Required | Description |
| ---------------------------- | --------- | ----------------------------------------------------------------- |
| **Pre-hold Notification** | Mandatory | Verifies that agents inform customers before placing them on hold |
| **Hold Duration Compliance** | Optional | Checks whether hold times stay within acceptable limits |
| **Call Resumption Behavior** | Optional | Evaluates how agents reconnect with customers after a hold |
The metric integrates with telephony systems to detect hold events and can evaluate multiple hold instances in a single conversation.
## Key Benefits
* Evaluate 100% of hold interactions automatically.
* Identify coaching opportunities and compliance gaps.
* Improve customer experience during hold scenarios.
* Maintain regulatory compliance across all agent conversations.
This metric is available for voice channel interactions only. One Hold Etiquette metric is allowed per evaluation form. Telephony integration is required to detect hold events accurately.
***
## Create By Hold Etiquette Metric
1. Navigate to **Quality AI** > **Configure** > **Evaluation Forms** > **Evaluation Metrics**.
2. Select **+ New Evaluation Metric**.
3. From the **Evaluation Metrics Measurement Type** dropdown, select **By Hold Etiquette**.
4. Enter a descriptive metric **Name** (for example, "Holding Customer for a Long Time").
5. Select the applicable **Languages** for this metric. You must select at least one. Languages can be added or removed later.
6. Enter an evaluation **Question** describing what the metric measures (for example, "Did the agent inform the customer before placing them on hold?").
7. Configure **Hold Notification**-verifies whether agents properly informed customers before placing them on hold.
Select an **Evaluation Method**:
### Gen AI-Based Method
Uses generative AI (LLM) to evaluate whether the agent's speech meets defined etiquette criteria.
* In **Description**, enter the expected behavior (for example, "The agent must inform the customer before placing them on hold using courteous language").
* Set the **Resumption Behavior Evaluation Window**-time period after a hold ends within which resumption behavior is assessed.
* Configure **Advanced Settings**-define the detection window for hold notification:
| Option | Description |
| -------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------- |
| **Last utterance before hold initiation** | Evaluates only the final statement before the hold. Silences and dead air are excluded. |
| **Configured window before hold initiation** | Set a custom detection window (1-100 seconds) before the hold. Utterances within this window are valid for compliance. |
### Deterministic ML Method
Uses BGE-m3 embeddings to compare agent utterances against sample phrases.
This method uses BGE-m3 embeddings to compare agent utterances to sample phrases with configurable similarity thresholds.
* In **Answer**, add sample utterances (for example, "I need a moment to get the details. May I put you on hold?").
* Add multiple utterances per configured language. Each language has its own set.
* Check/uncheck utterances to include or exclude them.
* Set the **Similarity** threshold (0-100%, default 60%):
* **Orange** = below threshold (0-60%)
* **Green** = above threshold (60-100%)
8. Configure **Sub-Criteria** (optional settings to assess hold-related behaviors):
**Hold Duration Compliance:**
* Toggle on to enable duration-based evaluation.
* Enter the **Maximum Acceptable Hold Duration** (1-300 seconds, default: 30 seconds).
**Call Resumption Assessment:**
* Toggle on to evaluate how effectively agents resume the conversation after a hold.
* Evaluates whether the agent acknowledges the delay, reconnects context, and proceeds smoothly.
9. When Call Resumption Assessment is enabled, select an **Evaluation Method**:
**Gen AI:**
* In **Description**, define expected resumption behavior (for example, "Agent smoothly resumes the conversation, acknowledges the wait time, and proceeds with relevant information").
* Set the **Resumption Behavior Evaluation Window** (1-120 seconds, default: 10 seconds).
* Assign **Sub-Criteria Weightage**-weights must total 100%:
| Sub-Criterion | Description |
| --------------------- | -------------------------------------------------------- |
| **Hold Notification** | Weight for courtesy and pre-hold notification compliance |
| **Hold Duration** | Weight for compliance with hold time limits |
| **Call Resumption** | Weight for smooth and contextual post-hold resumption |
**Deterministic ML:**
* Add sample utterances agents should use after resuming from hold.
* Set the **Similarity** threshold (0-100%).
* Set the **Resumption Behavior Evaluation Window** (1-120 seconds, default: 10 seconds).
10. Select **Create**.
The Create button stays disabled until all sub-criterion weights total exactly 100%. Once saved, this metric becomes active and appears in the Evaluation Metrics dashboard.
***
## Scoring Logic
### Sub-Criterion Weightage Example
| Sub-Criterion | Example Weight | Description |
| ----------------- | -------------- | ----------------------------------------------- |
| Hold Notification | 40% | Courtesy and pre-hold notification compliance |
| Hold Duration | 30% | Compliance with the maximum allowed hold time |
| Call Resumption | 30% | Smoothness and clarity of post-hold interaction |
**Instance score formula:**
(Hold Notification Result × 40%) + (Hold Duration Result × 30%) + (Call Resumption Result × 30%)
### Deterministic ML (Binary)
* **Pass (1)** — duration ≤ configured threshold or utterance matches sample
* **Fail (0)** — duration > threshold or no utterance match found
**Example:** Hold Notification (40%): Pass, Call Resumption (30%): Pass, Hold Duration (30%): Fail → Total Score = 70%
### GenAI (Continuous)
Each sub-criterion produces a semantic outcome (Yes/No or graded response). The outcome is multiplied by its assigned weight to generate a continuous, non-binary score.
***