You are here:
Install and Enable Asset Service Prediction
During the Asset Service Prediction install, you can configure the tool and optimize its AI model for performance.
Installation and Configuration
After the prerequisites are met, you can proceed with the installation.
- Navigate to the location for installing Data 360 apps and templates and select Asset Service Prediction. Click Install.
- An installation wizard will prompt you to configure two critical settings. These values
define how the model behaves and can be fine-tuned later.
- Days between Consecutive Repairs: This number determines if
two repairs on the same asset are considered related. For example, if you set this to
90 days, a repair on January 1st and another on March 15th are treated as a
consecutive pair.
Your data pull timeframe must be long enough to support this setting. If you set this value to 180 days, your Asset Service Action data transform should be configured to pull at least a year's worth of historical data to ensure the model can identify these patterns.
- Minimum Prediction Confidence Level: This sets a threshold to filter out low-confidence predictions. For instance, a value of 75% means you will only see predictions that the model is 75% confident or more in. After installation, review your model's performance metrics in Einstein Studio to determine the best threshold for your business needs and adjust it later in the CreateAssetServiceActionScoringData data transform.
- Days between Consecutive Repairs: This number determines if
two repairs on the same asset are considered related. For example, if you set this to
90 days, a repair on January 1st and another on March 15th are treated as a
consecutive pair.
- After configuring the settings, complete the installation. You can monitor its progress from the App Install History page.
Review and Refine Your Model
A successful installation is just the beginning. The key to valuable predictions is a well-trained and high-quality model.
- Go to Einstein Studio: Access your newly created prediction model.
- Assess Model Performance: Review the model metrics to understand its accuracy and effectiveness. If the performance isn't meeting your expectations, the next steps are to check your data.
- Ensure Sufficient Data: AI models require enough data to learn effectively.
Einstein Studio needs a minimum of 400 records (historical asset repairs) to generate a
model.
If you have fewer than 400 records, go to the AssetServiceActionCreation data transform and adjust the filters to broaden the date range, pulling in more historical repair data.
- Improve Model Quality:
- Data Quality is Key: Investigate your source data for any inconsistencies or errors that could be impacting model performance.
- Optimize Variables: For better results, you can enhance your model by editing it in Einstein Studio to exclude variables that are not necessary for making predictions.
- Retrain the Model: After making any changes to your data or model configuration, be sure to re-run the relevant Batch Data Transforms (BDTs) to retrain the model with the updated information.
Troubleshooting and Advanced Considerations
- Installation Failure: If the app setup fails, any Batch Data Transforms (BDTs) created during the process are not automatically removed, and Data 360 credits may still be consumed. You must manually delete these components before attempting to reinstall the app.
- Reinstalling: To perform a clean reinstallation, you must first manually find and delete the original installed app and all its associated BDTs.
- Multiple Installations: If you need to install the Asset Service Prediction app more than once in the same org, you must rename the BDTs for each installation to avoid conflicts. A best practice is to append the installation date to the BDT names.
- App Label: Keep the application label under 30 characters for best results.
