Intelligence Engine 1.4 Release Notes
Enhancements
Historical Interaction Analytics for Cases (Including Emails)
This release introduces unified historical processing for Emails and Cases. The system now processes cases spanning mixed channels and multiple agents as a coherent whole. Instead of analyzing isolated interactions, analytics unifies long cases into a single timeline to categorize, summarize, and extract signals for comprehensive reporting. Key capabilities include:
Whole Case AI Summaries
Analysts can now access a historical summary of the entire case that encompasses all associated interactions.
Holistic Categorization and Discovery
Categorization and Discovery tools now analyze the entire case lifecycle. This allows you to identify contact reasons, issue types, and risk signals (such as potential churn) that develop over the course of a long conversation. Options remain to analyze only the first or last interactions when specific focus on entry or resolution is required.
Sentiment and Quality Prediction
Extract emotional signals and predict quality markers, such as First Contact Resolution (FCR) or CSAT scores, directly from the conversation text. This provides quality insights even when customers do not complete post-interaction surveys.
Per-Message Autoscoring
Configure Autoscoring to evaluate individual interactions as they are associated with a case. Granular, rolling analysis allows for early intervention in risky cases to identify compliance issues or negative sentiment immediately without waiting for the entire case to be resolved.
Transparent AI Evidence
To ensure analytics are explainable, the system highlights the exact phrases in the transcript or email (including body, subject, or attachment names) where an intent or category was detected. Transparent attribution removes the "black box" effect, allowing human reviewers to verify why the AI made a specific classification.
New Grammar Intent to Allow Context Matching
By design, Grammar Intents only find matches in the specific message being evaluated. The new PARTY[ENABLE_CONTEXT] syntax allows you to extend the search scope to the preceding history of the case. For example, using CUSTOMER[ENABLE_CONTEXT]: "keyword" instructs the system to trigger the intent if the customer used "keyword" at any previous point in the thread, not just in the current reply. This supports rules that account for information established previously, such as a product type mentioned in the initial email or transcript but not repeated in subsequent replies.
Language Groups
To manage intents across different dialects (e.g., US, UK, and Australian English), you can now group dialects under a shared "Base Language." Previously, intents had to be defined separately for every specific dialect, requiring redundant configuration.
Language Groups allow for default definitions independent of region, while retaining the ability to create specific overrides. The system now prioritizes intent definitions in the following order:
- Region-Specific: Matches the specific dialect (e.g.,
English - Australia).
- Base Language: If no region-specific definition exists, falls back to the default definition for the language (e.g.,
English - Any).
- First Available: If no base language definition exists, uses the first available definition for that language.
|
Existing configurations with identical definitions across dialects will be automatically migrated to the base language to reduce duplication. | |
Set Advanced Options in AI Accounts
The Intelligence Engine now supports Advanced Options defined for AI Provider Integration accounts. This allows you to integrate the latest models that require specific parameters and expand the utility of current models.
For more details, see the Advanced Configuration Options for AI Integration Accounts in the Bright Pattern v5.34.2 release notes.