Skip to main content
The analysis configuration page at /platform/{chatbotId}/analysis-config controls how OpenBat processes and annotates your conversations. It has five tabs, each managing a different aspect of the analysis pipeline.
Analysis configuration page showing the Metadata tab with discovered fields

​
Metadata

Manage custom metadata fields sent via your SDK. When OpenBat receives metadata keys it hasn’t seen before, they appear in the Discovered fields table. For each discovered field, you can:
  • Track — add it to managed fields, making it available as a filter in the conversations table
  • Accept — acknowledge the field without tracking
  • Deny — reject the field
Common auto-discovered fields include: browser, device, IP, language, OS, referer, and userAgent.
Track the fields you want to use for filtering and segmenting conversations. Deny fields you don’t need to keep the interface clean.

​
Translation

Translation settings with auto-translate toggle and primary language selector
Enable Auto-translate messages to translate non-primary-language messages before analysis. This ensures consistent sentiment and intent scoring regardless of the original language. Select your Primary language (e.g., English) from the dropdown. Messages in other languages are translated to the primary language before analysis. The original message text is preserved alongside the translation.
Auto-translation adds one additional AI call per non-primary-language message, which may increase costs for chatbots with high volumes of multilingual conversations.

​
User analysis

User Analysis tab showing sentiment toggle, flag toggles, and custom flag creation
Control what signals OpenBat detects in user messages:

​
Sentiment scoring

Toggle sentiment analysis on or off. When enabled, every message receives a sentiment score.

​
User flags

Enable or disable detection for each built-in flag:
FlagDescription
Churn RiskUser language suggesting they may leave
Frustrated UserExpressions of frustration or anger
Buying SignalInterest in upgrading or purchasing
Competitor MentionReferences to competing products
Urgent RequestTime-sensitive or high-priority needs
Confused UserSigns of confusion or misunderstanding
Click + Add custom to create your own flags. Define a name and description — OpenBat uses the description to guide AI detection.

​
Assistant analysis

Assistant Analysis tab with behavior classification toggles for 9 behavior types
Classify assistant response behaviors. Toggle detection for each behavior:
BehaviorDescription
HelpfulProvides useful, accurate responses
Yes-ManAgrees with everything without critical thinking
DodgingAvoids answering the question directly
Over-ApologizingExcessive unnecessary apologies
HallucinatingProviding incorrect or fabricated information
RedirectingSends user elsewhere instead of helping
VerboseUnnecessarily long responses
RoboticUnnatural, template-like responses
OtherCatch-all for unclassified behaviors
Click + Add custom to define additional behavior classifications.

​
Prompts

Prompts tab showing editable AI prompts for Sentiment, Intent, Topics, and Flags
Edit the AI prompts that power the analysis pipeline. Four prompts are available:
  • Sentiment — controls how sentiment scores are calculated
  • Intent — defines how user intents are classified
  • Topics — guides topic extraction from conversations
  • Flags — determines how user flags are detected
Unmodified prompts show a Default label. Click on a prompt card to edit it. You can always revert to the system default from the editor.
There is no built-in prompt versioning. Before editing a prompt, copy the current text so you can revert manually if needed. Unmodified prompts show a Default label.
Poorly written prompts can produce unreliable analysis results. Test changes with a sample of conversations before relying on modified prompts in production.

​
Next steps

Customize prompts

Step-by-step guide to editing analysis prompts.

Manage metadata

Track custom metadata from your SDK.

Analytics dashboard

See the results of your analysis configuration.