To identify natural groupings and anomalies in your data quickly, cluster records dynamically without training or maintaining a reusable model. Unsupervised clustering uses pre-trained algorithms to group similar records. Use the capability to explore data, combine information from multiple columns into a single variable, or summarize millions of rows into a few meaningful clusters. For example, group related product or category records to preprocess data, and use the resulting cluster labels as variables in downstream forecasting or multi-class models.
Where: This change applies to Lightning Experience in Developer, Enterprise, Performance, and Unlimited editions.
When: This functionality is available starting in August 2026.
How: In Data Cloud, go to the AI Models tab and select Cluster. Select a clustering model that best fits your data and requirements.
-
K-Means: Divide records into a specified number of distinct clusters.
-
HDBSCAN: Discover natural groupings in data with irregular patterns and identify outliers without specifying the number of clusters in advance.
Run a predict job to generate dynamic cluster assignments, such as cluster IDs and labels. Because clustering doesn’t require a separate training step, you can generate results without first creating a reusable model.
