Work with a custom insight
If you use a summary configuration, then custom insights provide a way to better understand and extract essential information that participants discuss in an interaction. These insights appear at the bottom of the interaction’s summary in the order as listed in the Custom Summary configuration.
Insights can include entities such as the price charged to a customer for a product or service, an order, or part number. Your insights can also contain pieces of information to derive from the interaction, such as the greeting used by the agent or customer, or questions that relate to the content of the transcription.
You can also use the custom insights description to add mini-prompts with conditions that force the display of the labels.
| Label | Description |
|---|---|
| Part number | Display the part number if mentioned, otherwise display “Not Provided.” |
| ID verification | Did the agent verify the customer’s name and a call back number? State “yes” or “no.” |
| Policy type | The type of insurance policy mentioned in the conversation, or “not mentioned,” if not mentioned. |
| Policy number | The number that identifies the customer’s policy, if mentioned on the call. |
| Activation Date | The date that the customer’s policy first became active. If not mentioned in the call, put “unknown.” |
| Upsell | Did the agent attempt to offer the customer more coverage? Answer “yes” or “no.” |
| Customer name | The first and last name of the customer. |
| Calling on behalf | If the caller is calling on behalf of someone else, provide the name of the person or organization. |
| Amount | Monthly cost of the customer’s current subscription, otherwise put “unknown.” |
| Next steps | The next steps mentioned by the agent, if discussed. |
| Thank you | Did the agent say “thank you” to the customer at the end of the call? Yes or no. |
| Order type | Was the order a new order or a restocking order? Put unknown if not mentioned. |
Custom insight tips and recommendations
When you create a custom insight description, make sure that you use wording that is understandable by a human. A person who looks over the transcript may not be familiar with industry jargon unless it is explicitly called out in a transcript. A clear description is critical to ensuring that the custom insight appears in the summary.
- Use as few words as necessary for the custom insight description.
- Add a binary option after a question. For example, “Display ‘yes’ or ‘no’.”
- Provide instructions for situations when the custom insight is not detected or answerable from the transcript. For example, “Leave blank if not found in transcript,” or “Add ‘n/a’ if not mentioned.”
Ask someone who is not familiar with the topics discussed on the queue to identify the insights of a sample transcript based on the description of a custom insight. If they find it difficult to understand what they should identify, consider rewriting the description.
Add a custom insight
You can add up to 10 custom insights per summary configuration.
- Click Menu > Orchestration > AI Studio > Summaries.
- Open an existing summary or create a new one.
- In the Insight label field, add the name of the custom insight.
- In the Insight description field, add the details about the insight that you want to extract from the transcription.
Test your custom insights
Large Language Models (LLMs) are non-deterministic, and the output of custom insights can vary. Make sure that you test the custom insights on various transcripts that contain different words and phrases. If a custom insight does not perform as expected, update the description and retest the insight. An iterative process is normal when working with generative AI.
If the evidence from the transcript produces a low confidence in the result, then the custom summary does not appear. For example, if you create a custom insight for Part Number, but that insight is not explicitly mentioned in the transcript, then the insight does not appear in the summary.
