Smart Topic Builder
Functionality, configuration, and workflows for genAI topic discovery, definition, and auditing with Smart Topic Builder
Smart Topic Builder uses generative AI to discover, define, and evaluate the precision of standard Text Analytics topic hierarchies and rules. Smart Topic Builder is designed to make admin-driven topic development more efficient, not to replace or automate it entirely. Admins can edit AI-suggested topics and rules. Admins retain full control of topic publication and activation, and they have final acceptance of topic and rule proposals.
Access Smart Topic Builder from the Topics screen in Admin Suite.
Role permissions and workflows
To use Smart Topic Builder, the user's role must have the Manage Topics and the Manage Topics with Smart Topic Builder administrative permissions.
There are 3 distinct Smart Topic Builder workflows:
- Topic discovery
- Enables admins to specify an untagged dataset, and to provide a description and example comments, for genAI topic discovery. Topic names and hierarchies are generated.
- Topic rule generation
- Enables admins to describe, specify example comments and phrases for, and use genAI to generate rules for a topic. Topic rules and word groups are generated.
- Topic precision audit
- Enables admins to assess topic performance, identify comment mis-classifications, and prioritize topic rule refinement with genAI.
Data input, output, processing, and retention
- Processes only untagged English comment data for topic generation
- Processes both tagged and untagged conversational data for topic generation
- Uses existing word groups and creates new ones as needed
- Uses existing user features, but does not create new user features
- Outputs topic names, rules, and word groups in English only
- Adheres to existing MEC data retention policies and region-specific processing requirements
- Supports any signal type currently used with tag pools and topics, including survey, social, digital, and conversational contact center and chat interactions
Suggested topics are queued for admin approval before they are added to the Topic Hierarchy as a draft. Generated topic and topic definition drafts, rule drafts, and word groups can be edited prior to activation and publication.
Topic discovery results are retained for 15 days or until a new discovery job is processed.
Service level objectives (SLOs)
- Topic discovery
-
Processing time: Around 15-30 minutes; sends an email notification when complete
Contributing factors that may increase processing time:- Number of existing topics in the tag pool
- LLM errors and error correction
- Rate limits
- Topic rules generation
-
Processing time: Usually 1-2 minutes per rule
Contributing factors that may increase processing time:- Number of existing word groups and user features
- LLM errors and error correction
- Number of suggestions considered before being discarded for irrelevance
- Rate limits
- Topic precision audit
-
Processing time: Usually less than 2 minutes
Preparing your system and program for Smart Topic Builder
To optimize Smart Topic Builder output for all workflows, review your existing topic sets, topics, specifications, and overall Text Analytics setup:
- Review your existing topic sets and tag pools
-
- For each topic set, ensure all relevant comment fields are included in the tag pool of comment data to be analyzed.
- Create and update specifications for your topic sets
- A topic set specification is distinct from Smart Topic Builder's Topic Definition, and should include these elements for each topic:
- Level 1 topic
- Required. Parent topic that does not contain rules
- Level 2 topic
- Required. Child topic that contains rules
- Description
- Required. A clear, simple statement of what the topic is meant to capture
- Example sentences
- Optional. 1-2 sentences for examples of statements that the topic should capture, and 1-2 example sentences of what the topic should not capture
- Example keywords and phrases
- Optional. Comma-delimited lists of terms and phrases found in sentences that should and should not match the topic
- Remove outdated program components
-
- Delete unused topics from active topic sets so they are not included with discovery processing
- Delete unused topic sets and word groups for the system
- Request a Text Analytics health check on your Text Analytics system settings
- A Text Analytics health check is a process performed outside of Experience Cloud for identifying opportunities in a Text Analytics implementation to optimize reports and topics to get the best insights from Text Analytics.
Text Analytics health checks are performed by the Medallia Text Analytics team. For more information, contact Medallia support.
Sensitive data rules and personally identifiable information
Smart Topic Builder processes all untagged comment data, and all tagged and untagged conversational data, to look for new topics, and it respects configured discovery filters, PII field properties, and sensitive data rules. Fields with the PII property enabled are excluded from tool access, and sensitive data rules are applied to comment data before that data is processed by the large language model.
Proposed topics are generated from aggregated data and typically are about broad subjects. Example comments appear as part of the topic definition, and although these examples respect sensitive data rules, they may contain PII. Admins should review comments for PII when approving topic drafts.
Activate and publish AI-generated topics
The process of activating and publishing topics and topic changes is the same as for topics developed without AI, but with one exception: users must confirm that they want to activate or publish AI-generated topics using the dialog that displays.
For more information, see and Activate topics and Publish updated topics.
AI-generated data transparency
When AI-generated content or data is displayed in Experience Cloud, the generative AI icon displays with the module title or data, and a text disclaimer displays in the module's footer or after the relevant content.
Smart Topic Builder displays the generative AI icon with generated topics, rules and components, word groups, audit annotations, and buttons that initiate AI-powered workflows. Editing topic components and publishing a draft AI-assisted topic removes the AI-generated designation and icon.
