Your NLP bot is more effective when you update it regularly to react to the latest customer interactions. You can rebuild your active bot’s intent model and automatically go live with the updated model as soon as the build is complete. Or you can build and test your new model in a sandbox with change sets.
Required Editions
| View supported editions. |
| User Permissions Needed | |
|---|---|
| To build and manage Einstein Bots: |
Customize Application OR Modify Metadata OR Manage Bots |
| To manage intent sets and bot training: |
Customize Application OR Modify Metadata OR Manage Bots OR Manage Bots Training Data |
To update your active bot, add utterances via Einstein Intent Sets. When you’re ready to build, access the Model Management page from the Bot Builder, and then click Build Model. Your updated model is automatically active when the new build is complete. After a new build is complete, you can’t revert to the previous model within the bot. To keep your current model as a backup, clone your bot before you start a new build.
To test while customers are actively using the bot, we recommend using change sets to move bots, bot versions, and intent sets between sandbox and production.
- In the sandbox, to create a version of the bot, click the dropdown next to the latest active version of the bot, and then click Clone.
- Add new utterances to a bot via Einstein Intent Sets. Utterances can be added manually to an intent set or loaded via CSV file.
- Train your bot in the sandbox. Differences in the training scores of the same bot in sandbox and in production is expected behavior. To evaluate the model performance, the bot samples a portion of the utterances for training, so the training sets aren’t identical.
- Bots, bot versions, and intent sets can be included in an outbound change set. To add Bot Versions to a change set, add the bot first, and then click View/Add Dependencies.
- Deploy to production.