Why it matters
Data parsing allows your AI agent to extract structured fields from call transcripts, such as summaries, outcomes, yes or no answers, or any key details your workflows rely on. Correct configuration ensures consistent tracking and enables automated downstream actions. Although agents can also handle messaging, data parsing is designed specifically for call analysis.
Key Concepts
Data field: A structured piece of information extracted at the end of a call.
Information type: The format of the extracted value such as text, number, boolean, or selector.
Field description: Guidance that tells the parsing model what the field represents and how to extract it.
Format example: An optional example that clarifies the expected output.
Parsing model: The model used to analyze call transcripts and generate structured fields.
Step-by-Step: Set Up Data Parsing and Analysis
Open your AI agent and go to the Editor tab.
Scroll to the Data parsing and analysis section.
Click Add field to create a new extraction field.
Select the information type: text, selector, boolean, or number.
Enter the field label.
Add a clear description explaining what the model should extract.
(Optional) Add a format example to show the model the expected output structure.
Repeat for each field you want to extract.
Drag fields to reorder them as needed.
Use the pencil icon to edit a field or the trash icon to delete it.
Choose the parsing model from the dropdown menu.
Click Save to apply your parsing configuration.
Tips and Best Practices
Designing Effective Fields
Use short, readable labels.
Add descriptions that specify where the information appears in the call and what phrasing to look for.
Include format examples for structured or styled outputs (summaries, categories, or numerical formats).
Choosing Information Types
Use text for summaries or multi-sentence explanations.
Use selector when you need predefined categories and consistent reporting.
Use boolean for clear yes or no values.
Use number for totals, durations, or counts.
Improving Extraction Accuracy
Keep descriptions specific and unambiguous.
Avoid overlapping fields that describe similar concepts.
If extraction is inaccurate, refine the description or add a clearer example.
Review extracted fields in View data during call review to monitor accuracy.
Selecting a Parsing Model
Choose the model that best fits your accuracy requirements.
Select models optimized for extraction rather than creativity if precision is important.
Keep parsing models consistent across agents when outputs need to match.
Troubleshooting
Issue | Possible Cause | Fix |
Extracted fields are empty | Descriptions unclear or too general | Rewrite descriptions with specific cues about what to extract |
Wrong values appear in View data | Model misinterprets language | Add format examples or refine descriptions |
Selector outputs are inconsistent | Options too broad or unclear | Narrow the selector choices and update field descriptions |
Data extraction varies across calls | Parsing model not ideal for your domain | Switch to a more accurate or domain-specific model |