New Updates include: Conversation Assist Enhancements
➡️ Exact delivery dates may vary, and brands may therefore not have immediate access to all features on the date of publication. Please contact your LivePerson account team for the exact dates on which you will have access to the features.
🛑 The timing and scope of these features or functionalities remain at the sole discretion of LivePerson and are subject to change.
Conversation Assist
Features
Setup is now rule-based and offers more flexibility
Previously, many Conversation Assist settings were account-level, and this limited your options regarding which recommendations were offered and to whom.
We’ve changed this by redesigning the Conversation Assist “setup experience.” It’s now rule-based, which is much more intuitive and flexible:
Example of a knowledge base rule
Recommendations are still skill-based, so every rule must specify at least one skill. But every rule can have one or more “add-ons.” Essentially, the add-on completes the rule’s definition. And here’s where things get very flexible: Within an add-on, you can specify:
- The agent groups and/or profiles to which to offer recommendations
- The recommendation sources (knowledge bases or bots) to use
- The minimum confidence threshold to use
- Knowledge bases: The number of articles to retrieve from each knowledge base
- Answers enriched via Generative AI: Whether this is turned on and the prompt(s) to use
Again, these options are all per add-on.
Important notes
- Rules affect 1) recommendations offered inline in conversations and 2) recommendations offered via the On-Demand Recommendations widget.
- The order of a rule’s add-ons matters: At runtime, add-ons are evaluated in order, and the first one that’s matched is executed.
Call to action
LivePerson is in the process of migrating all brands to this new architecture on a brand-by-brand basis. We're focusing on brands with a low number of recommendation sources first, followed by those with a high number.
To avoid conflicts due to the new architecture, only your recommendation sources (knowledge bases and bots) that are turned on (enabled) will be migrated to the new architecture. You'll need to manually create new configuration rules for any knowledge bases or bots that were turned off (disabled) in Conversation Assist.
Once your account is migrated, you'll be notified by us. Please do the following:
1. Verify that the new rules are configured as you expect.
2. Give each new rule a name and description that's meaningful to you. (By default, the names are "Migrated rule 1," "Migrated rule 2," etc. And the default descriptions list the relevant knowledge bases|bots and skills.)
Our recommendation engine just got smarter (APAC region only)
The behavior discussed below is in Beta release. We've only rolled this out to brands with accounts in the APAC region so far. We're working to gain learnings from this limited release before rolling it out to all regions.
When a consumer is messaging with an agent, sometimes those messages aren’t intentful. Sometimes they’re just messages that offer information, for example:
- My name is Jane Doe.
- My account number is abc1234567.
- I bought the item last month, and the order number is GA2342345.
- My address is 123 4th Avenue, Apt. 2A, New York, New York 10010.
The system can now identify when a consumer message only offers information. Now, when a message is classified as such, recommended answers are no longer offered.
Important notes
- This change only affects recommended answers. It doesn’t affect recommended bots. A bot is still recommended if the agent picks up a conversation that’s assigned to the relevant skill (i.e., that same skill is used in a rule for the knowledge base in Conversation Assist).
- This change affects both 1) recommended answers offered in-line in conversations with consumers and 2) those offered in the On-Demand Recommendations widget.
- Recommended answers are still offered in response to small talk from the consumer (hi there, how are you, etc.)
- A consumer message might offer information and contain an intent, such as, “My account number is abc1234567. Can you tell me my balance?” In these cases, recommended answers are still offered.
