Model Reference
This section is the per-model configuration reference. For each model type it documents how the model works, the full JSON configuration (every field, type, whether it's required, and defaults), a complete example, the output it writes, and feature-engineering guidance.
- For the step-by-step workflow (create → configure → activate), see Configure a Model.
- For the input data your tables must contain, see AI Model Data Requirements.
- For how to use the outputs in audiences and signals, see Predictions.
Pages
| Model | Type | Reference |
|---|---|---|
| Action Prediction | Predictive | Action Prediction |
| pcLTV (Transactional) | Predictive | pcLTV Transactional |
| pcLTV (Subscription) | Predictive | pcLTV Subscription |
| RFM Clustering | Descriptive | RFM |
| Interest | Descriptive | Interest |
:::info Models read the reconciled tables
Model configurations reference the _bpp enriched tables — the reconciled copies BPP writes alongside your source tables, which carry the bpp_user_id column (see Event Reconciliation). That is why user_id_column is typically bpp_user_id and source_table_id points at a *_bpp table.
:::
:::note Infrastructure fields are auto-populated
Fields such as data_src.region, project_id, dataset_id, bucket_name, and the output table names are filled in automatically by the platform from your connected data source. They are documented here for completeness, but you normally only provide the modelling parameters (filters, features, thresholds, columns).
:::