Just getting started? Complete the Using Conversation Assist tutorial.

It takes up to 3 hours for changes in Conversation Assist configuration to take effect.

Limitations

You can use any type of knowledge base in KnowledgeAI™ as a recommendation source in Conversation Assist. However, keep in mind the following limitations:

  • "Rich content" answer recommendations aren't supported if the knowledge base is external (without LivePerson AI).
  • LLM-enriched answers: A single rule can't contain a mix of knowledge bases that enrich answers via Generative AI and knowledge bases that don't do this. Don't set up a rule this way, not even if you put one type in one add-on and another type in another add-on within the same rule. Instead, set up different rules assigned to different skills to support your use case.

Step 1: Create the KB and articles

Access KnowledgeAI and create at least one knowledge base (KB) from your content source or from scratch. You can create and use any type of knowledge base, and the knowledge base can be public or private.

Also create at least one article therein, so you can verify that your setup is complete and working. You can continue to add more articles at any time after setup.

High-quality knowledge content is critical for ensuring the successful adoption of answer recommendations by your agents. During setup of your knowledge bases, take care to follow our best practices on structuring knowledge bases and adding articles.

At this point, in KnowledgeAI verify that the desired articles are active.

New to KnowledgeAI and knowledge bases? Check out the Meta Intents & Knowledge Bases tutorial.

Step 2: Create KB-level recommendation rules

In this step, you create the rules that determine when answers from the knowledge base are offered as recommendations to agents.

  1. Access Conversation Assist, and click Recommendation Sources.

    The Knowledge Bases tab is displayed by default.

  2. Click Add rule.
  3. Define the rule that determines when answers from the knowledge base(s) are recommended to agents. Each rule element is described farther below.
  4. Click Save and activate.
  5. Add additional rules as you require.

Rule elements - general

General attributes of a rule

  • Name: Enter a short, meaningful, and unique name that highlights the rule’s basic function and purpose. It's important to name the rule well, so you can leverage reporting effectively.
  • Description: If desired, provide a more in-depth description of the rule: rationale, approach, i.e., anything that’s useful.

  • Skills: Select the Conversational Cloud skills that you want this rule to apply to. You must specify at least one skill. A skill can be used in only one knowledge base rule.

    In conversations routed to these skills, articles in the knowledge bases specified in this rule are returned by KnowledgeAI™ to Conversation Assist. (A conversation is routed to the skills assigned to the campaign's engagement.)

    When specifying the skills, carefully consider which sets of knowledge you want to expose through each skill serviced by human agents.

Rule elements - add-ons

A rule add-on completes the rule’s definition. You must define at least one rule add-on because, at a minimum, that’s where you specify the knowledge bases to use in the rule.

If you define multiple add-ons, the order of the add-ons matters: At runtime, the add-ons are evaluated in order, and the first one that’s matched is executed. So, order the add-ons as you require.

Add-on attributes of a rule, with a callout to the move icon that can be used for reordering add-ons

  • For agents belonging to: You can further limit recommendations from the knowledge base to specific Conversational Cloud agent groups and/or profiles. Carefully consider which sets of knowledge you want to expose to specific groups and/or profiles. Or, if the rule will be only skill-based, leave these blank.

  • Offer recommendations: Select where the recommendations should be offered. You can select “in conversations only,” “in On-demand widget only,” or “in conversations and in On-demand widget.”

    Keep in mind that agents only receive recommendations in conversations if the Inline recommendations setting on Conversation Assist's Settings page is turned on. That setting takes precedence over all knowledge base rules.

    The next setting (Enhance user's query using conversation context) is in Early Access release.

  • Enhance user's query using conversation context (in conversations only): This setting is only available if, for Offer recommendations, you select one of the options to offer recommendations in conversations. That's because this setting only applies in that case.

    What does this setting do? Often the user’s query doesn’t include enough context to retrieve a high-quality answer from a knowledge base. To solve this, you can turn on this setting if you want the system to gather additional conversation context (conversation turns) and use it to rephrase the user’s query before searching the knowledge base. Rephrasing is done using KnowledgeAI’s Query Contextualization feature.

    Queries that are only small talk (chitchat) are not enhanced. But remember that this setting applies only to recommendations offered in conversations. Your agents can continue to use the On-Demand Recommendations widget to search for responses to queries that are only small talk.

    Additionally, queries that are not in English are not enhanced. So, if your solution is not in English, it doesn’t make sense to turn on this setting because queries that aren’t in English are never enhanced. However, some brands support multi-lingual queries, such as English and Spanish queries against English knowledge bases. In such a case, go ahead and turn on this setting, so you can take advantage of query enhancement in cases where the query is in English.

  • Retrieve best articles from: Specify here the knowledge bases to search for matched articles. Ensure your knowledge bases follow our best practices in terms of both structure and content. Also specify the minimum confidence score that articles must have to be retrieved. Keep in mind that the higher the score, the more relevant the match. To increase the likelihood of a matched article, try a lower score.

    If you don’t intend to use the articles to generate an answer that’s enriched via an LLM and Generative AI, then only 1 article is retrieved from each knowledge base.

  • Enrich answers via Generative AI: Not ready to use Generative AI and LLMs? No problem. You can leave this setting off if you choose. This setting is discussed in the next section.
Enrich answers via Generative AI

If you want to offer your agents recommended answers that are enriched via Generative AI, turn on the enrich answers via Generative AI setting.

A callout to the toggle for turning on enriched answers via Generative AI

  • Send to LLM up to N articles for enrichment: Select the number of articles to retrieve from each knowledge base and send to the LLM for an enriched response. The system always returns 1 enriched answer per knowledge base.

    Providing more knowledge coverage (not just a single article) to the LLM for an enriched answer often results in a response that's superior.

  • Enrichment prompt: Select an enrichment prompt from the Prompt Library.
  • No Article Match prompt: Optionally, select a No Article Match prompt from the Prompt Library.

Example rule

Our example rule below is for a fictitious, national automotive brand named Acme Auto. The rule is for a single skill named Support, which the brand assigns to all of its customer support agents.

Name, description, and assigned skill for an example rule

Acme Auto agents are highly specialized, so the brand divides its agents into two Conversational Cloud agent groups:

  • ICE Support for handling FAQs about cars with an internal combustion engine (ICE)
  • EV Support for handling FAQs about electric vehicles (EV)

Thus, the rule shown above includes two add-ons: The first add-on is for offering answer recommendations from the ICE FAQs knowledge base to agents in the ICE Support agent group. The second add-on is for offering answer recommendations from the EV FAQs knowledge base to agents in the EV Support agent group.

So, for example, for an agent to receive answer recommendations from EV FAQs, the following must happen:

  • The agent must pick up a conversation that is routed to the Support skill. (A conversation is routed to the skills assigned to the campaign's engagement.)
  • The agent must be in the EV Support agent group.

As mentioned earlier, the order of the add-ons matters: At runtime, the add-ons are evaluated in order, and the first one that’s matched is executed. So, in our example here, if the agent were a member of both groups, the agent would never receive recommendations from the EV FAQs knowledge base because the first add-on (for the ICE Support agent group) always evaluates to true for the agent.

Step 3: Configure settings

  1. Access Conversation Assist, and click Settings.
  2. Configure relevant settings.