Group Action Framework

  • Release version: Australia
  • Updated July 21, 2026
  • 3 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Group Action Framework

    The Group Action Framework (GAF) is an intelligence feature within the ServiceNow AI Platform that leverages large language models (LLMs) to group related records—such as incidents, cases, and knowledge base articles—and apply actions to these groups efficiently. It operates through two key processes: grouping, which clusters similar records, and actioning, which maps new records to these clusters and selects representative records for further processing. Topic-labeling during setup assigns human-readable labels to clusters for easier identification.

    Show full answer Show less

    GAF enhances AI agents and generative AI capabilities by improving consistency and quality through using representative examples, reducing the cost of LLM calls by focusing only on representative records, and scaling effectively to handle large datasets.

    Key Features

    • Modular Skills: GAF comprises several specialized skills that work together but are exclusive to GAF workflows:
      • Grouping Skill: Uses machine learning to cluster related records.
      • Topic-Labeling Skill: Applies LLMs to assign readable names to clusters.
      • Action Strategy Skill: Chooses representative records within clusters.
      • Action Mapper Skill: Runs LLM inference on representative records to produce summaries.
      • Action Reducer Skill: Consolidates summaries from the mapper into a single cluster summary.
    • Integration with AI Search: GAF leverages AI Search to enhance clustering capacity (up to 500,000 records) and recall speed, using AI Search as a fallback when GAF does not return results. Configuring AI Search is recommended for improved performance.
    • Configurable Grouping Inputs: Customers can customize:
      • Source tables and specific fields used for clustering (default settings work well for most cases).
      • Filter conditions to select records (a minimum of 2,000 filtered records is recommended for effective clustering).
      • Options to apply taxonomy for better cluster categorization and HTML preprocessing to remove markup.
      • Parameters controlling clustering behavior and dimensionality reduction (UMAP and HDBScan), which affect cluster formation and similarity thresholds.

    Diagnosing and Troubleshooting

    ServiceNow customers can monitor GAF activity and troubleshoot issues by reviewing logs in the Generative AI Log [sysgenerativeailog] table filtered by GAF-related skills, or in the Log [syslog] table filtered by the sn-gaf scope. Common errors and resolutions are documented separately to assist with problem-solving.

    Additional Considerations

    GAF currently does not support domain separation, which is important to note for customers managing multi-domain instances.

    Group Action Framework (GAF) is an intelligence feature on the ServiceNow AI Platform that groups related records and applies actions to them using LLMs.

    GAF overview

    GAF is composed of two processes. The grouping process identifies clusters of similar records (incident, cases, KB articles, and the actioning process maps) new records to clusters and selects the representative record. Topic-labeling occurs during the setup process and labels the different clusters, which form the Description field of records on the GAF record groups [sn_gaf_record_group] table. GAF processes together benefit your AI agents and generative AI features in multiple ways.

    • Improves consistency and quality of agentic and generative AI features by using the best examples from groups of records.
    • Reduces the cost of LLM calls by only executing on the representative records.
    • Scales to accommodate large amounts of data because selected records can represent any size cluster.

    Skills used in GAF

    Multiple skills are involved in GAF setup and execution. They are modular, so not all executions will use all skills, but they can be used together in tandem. These skills are used exclusively for GAF and currently cannot be included on their own in custom agentic workflows.

    Grouping skill
    Clusters related records using machine learning techniques.
    Topic-labeling skill
    Adds human-readable names to the clusters using an LLM to make the clusters easier to identify.
    Action strategy skill
    Selects representative records from each cluster for the mapper and reducer skills to use.
    Action mapper skill
    Runs LLM inference calls for the selected representative records, producing a record summary for the selected records.
    Action reducer skill
    Uses the generated summaries created by the mapper skill to produce a single summary for the entire cluster.

    GAF and AI Search

    GAF uses AI Search to improve its effectiveness and uses it as a fallback option in case GAF does not return any results. When AI Search is enabled, then GAF can utilize its optimized prediction. The optimized prediction feature increases clustering capacity up to 500,000 records and improves recall speed.

    See Install AI Search and Set up AI Search for Group Action Framework for more information on configuring AI Search for GAF.

    Grouping inputs

    When you configure GAF, the Grouping inputs record in the Now Assist Skill Config Var Set defines what records are clustered and how clustering behaves. The default values work for most configurations, but the following parameters are available if you need to adjust them.

    Table
    The source table whose records are clustered. For example, the Incident [incident] table for ServiceNow Otto for ITSM.
    Fields
    The field or fields used as the basis for clustering. GAF compares the content of these fields across records to identify similar ones.
    Filter
    Conditions that determine which records from the source table are included in the grouping. You should have at least 2,000 records after filters are applied for clustering to complete successfully.
    Use taxonomy
    When enabled, applies a taxonomy to improve the categorization of clusters.
    HTML Preprocessing
    When enabled, strips HTML markup from field content before clustering. Enable this if your source records contain HTML-formatted text.
    Dimensionality reduction parameters (UMAP)
    Controls how GAF compares records to each other before clustering. The n_neighbors parameter [2–100] sets how many similar records GAF considers when deciding which cluster a record belongs to. The n_components parameter [5–100] sets how many characteristics of each record GAF tracks during that comparison.
    Clustering parameters (HDBScan)
    Controls how clusters are formed from the compared records. The min_samples parameter [2–10] sets how many similar records must exist before GAF treats a pattern as a real cluster rather than noise. The min_cluster_size parameter [2–100] sets the minimum number of records required to form a cluster. The cluster_selection_epsilon parameter [0.01–0.99] controls how similar two clusters need to be before GAF merges them into one.
    Note:
    The default values for UMAP and HDBScan parameters work for most configurations. Larger, denser clusters represent more common issue patterns and are surfaced more frequently. GAF's retrieval logic also handles smaller clusters, so less common patterns are still accessible.

    Diagnosing and troubleshooting GAF

    To find logs where GAF is used on your instance, go to the Generative AI Log [sys_generative_ai_log] table and filter based on the skill that used GAF. You can also filter the Log [syslog] table for the scope containing sn-gaf. Actions performed by skills, AI agents, and agentic workflows that use GAF all generate logs.

    For more information about common errors and possible resolution steps, see Troubleshooting GAF.

    Additional information

    GAF does not support domain separation at this time.