Enhanced Bot setup uses the same Bot Builder UI shell as Classic but with the Enhanced template selected at creation. The configuration steps mirror Classic with the added Topics layer and multi-language support.
- Create a new Enhanced Bot
Setup, Einstein Bots, New Bot. Pick Enhanced Bot Builder. Configure name, language(s), and channels. The bot starts with a default Topic structure (Welcome, End Conversation, Transfer to Agent).
- Build Topics and group Dialogs
Create a Topic per business area: Order Management, Account Servicing, Billing. Add Dialogs under each Topic. Topics are visible to the customer as conversation starters; a Welcome Dialog can list "I want to ask about ..." options that map to Topics.
- Configure multi-language
Bot Settings, Languages. Add each supported language. For each Dialog, provide translated response text inline or via translation file. Test language detection by greeting the bot in different languages.
- Enable Conversation Repair
Bot Settings, Conversation Repair. Toggle on. The feature scans live conversations daily and produces a suggestion queue. Review the queue weekly and accept suggestions that genuinely improve the bot.
- Wire Einstein GPT (optional)
Bot Settings, Generative Responses. Enable per-Dialog or per-bot. Configure grounding sources (Knowledge articles, CRM records). Test GPT responses on internal-only Dialogs before exposing customer-facing GPT.
Logical groups of related Dialogs (Order Management, Account Servicing). Customers can pick a Topic at conversation start; admins manage Dialogs per Topic.
Multi-language within one bot. Add languages in settings; provide translated Dialog text. NLP detects customer language automatically.
Daily auto-detection of mishandled utterances. Suggestions land in a queue for admin review and one-click accept.
Optional Einstein GPT integration for dynamic responses grounded on org data. Configure carefully on sensitive use cases.
- Enhanced Bot and Classic Bot are separate products. Migration is manual; there is no one-click conversion.
- Multi-language requires translation effort per Dialog. NLP detects language automatically but does not translate Dialog response text for you.
- Conversation Repair suggestions need human review. Accepting all suggestions blindly can degrade the bot by overfitting on edge cases.
- Generative Responses can hallucinate on sensitive use cases. Limit to low-stakes conversations and ground on curated Knowledge articles.
- New channel features ship to Enhanced first. Orgs on Classic miss out on Apple Business Chat rich messages, WhatsApp List Messages, and other recent additions.