Asset Service Prediction Considerations

Prerequisites

To set up your AI model, you must have Data 360 and Agentforce Service (formerly Service Cloud) enabled within the same Salesforce org. The standalone Data Cloud One edition isn't supported.

Data Requirements and Formatting

• Repair Data Source: The model is designed to use work orders as the primary source of repair information.
• Repair Types: For the model to work correctly, use standardized repair types.
  • Use a defined list of categories, such as Screen Replacement or Battery Malfunction, instead of allowing free-text entries.
  • Each AI model supports a maximum of 200 unique repair types. If you require more repair types, create and train additional models.
• External Data: If you store repair data in external systems, ensure it references the original work order or repair record in your CRM.
• Sufficient Data Columns: The model requires a minimum of three columns with data for training. Although it's configured to select eight columns by default, empty columns are automatically dropped.

Model Training & Maintenance

• Duplicate Entries: It's normal to see duplicate primary key values in the data after it has been transformed. This doesn't affect the quality or accuracy of your model.
• Troubleshooting Failures: If the model fails because of bad data, you can fix it by reviewing the Asset Service Action Data Lake Object. Use it to filter out problematic entries or adjust the data fields as needed.
• Optional—For Fresher Predictions: To get predictions based on the most recent information, schedule your data preparation tasks to run more often than your model's scoring schedule.

Note To customize access to Data 360 objects used by Asset Service Prediction, grant object access according to your chosen Data Governance policy. If you have more than one data space, you must grant access to objects in each data space. Object names in non-default data spaces follow the convention of <DataSpace Prefix>_<Object Name>. For help, see Assign Data Access.