What is Conversational AI 2.0?

Conversational AI 1.0 is any bot you create without enabling the Conversational AI 2.0 option in the Advanced tab of your bot settings.

What is the difference between Aisera Conversational AI 1.0 and 2.0?

Some comparisons between 2.0 and 1.0 are highlighted below.

Event	Process
Request Analyzer	The Request Analyzer only considers Small talk and Toxicity
Intent Extraction/Query Generation	As in version 1.0, the logic to generate the query is the same in both systems. This process is handled outside the Fulfillment Engine.
Search	Using the generated search queries, the system searches through: • Private KBAs/Service Catalogs (Neural Search). This is the same as in version 1.0. • Public KBAs (Public Search Service): Different logic - 2.0 the bot will search public sites in parallel. As in 1.0, Public KBAs are only used when the bot did not find a response in the Private KBA. For Public KBAs, if there are 20 allowlisted sites, the Aisera Gen AI platform uses Google to determine which sites are best to search for the given query. The platform uses the top 3 sites that Google recommends and provides the website content as context to the LLM and pulls the relevant chunks. • Flow Search (ICM+): Searches in all systems are conducted in parallel, and the results are then sent to the validation step.
ACL filtering	The validated chunks of information gathered are filtered by ACL to ensure that only the allowed content is used. In Conversational AI 2.0, only the Private KBA and Service Catalog items are considered for ACL.
Validation step	The final search results are passed through the LLM for evaluation. If the content is not relevant, these results are filtered out. If no results are left at this point, then bot will go to Fallback.
Decision Step	The validated chunks are passed to the decision engine to determine if the Fulfillment was provided or if a Context Disambiguation Service (CDS) question is needed. • Private/Public validated chunks: These either provide the RAG answer or trigger a CDS question. • Service Catalogs/Workflows: These are simply shown without triggering CDS questions.
Response Generation	The final response provided by the chatbot.

Intentless Workflow Search

With Intentless deployment, since there are no Intents, the traditional ICM search (1.0) is not used to match workflows with requests.

Instead, the intentless workflow search engine uses Workflow Descriptions to match a workflow to a user request. These are descriptions in natural language, specifying when a workflow will be triggered. The Intentless workflow calculates which user inputs may be required to execute targeted workflows. This is achieved using LLMs to intelligently review workflow descriptions and determine the mandatory and optional input data fields needed to execute the workflow. This process of conversationally collecting required input variables is called slot filling.

See Creating Intentless Workflows for more information.

User Awareness

Aisera applications/bots can extract input parameter values directly from the application/bot user's Request and prior conversation history. This reduces the need for application/bot users to repeat information that they have already provided.

1. If a user mentions a value in their request (such as, "What is the leave policy in Greece?"), the country value Greece is automatically bound to the relevant input variable in any workflow that requires it, without asking the user again.

2. If the same or a related value was provided in an earlier sequence of the conversation, the application/bot retrieves it from the conversation history and uses it for slot filling in subsequent workflow interactions.

Contextual Awareness

When your application/bot cannot determine a direct answer to a request, it will respond by providing relevant contextual information in addition to the existing fallback options.

When an application/bot user lacks access or no exact information is available for their request, the application/bot will display the reason the request is not fulfilled.

For example, it might return: "You do not have access to this information; however, I found some related topics that might help." instead of displaying the configured static fallback response.

The application/bot responds with any related information that might still be useful, providing more context before presenting the fallback options.

By delivering related information with the configured fallback, the application/bot describes why the primary information is unavailable while still offering helpful alternatives.

Limitation for multi-intent queries: if any single intent can be fulfilled, the bot will display that successful response and ignore the other intent if it triggers a fallback. However, if both intents trigger a fallback, the bot will surface the contextual fallback for both intents simultaneously.

Citation Feature

The Conversational AI 2.0 Citation Feature allows you to view reference documents for each segment of the knowledge response you receive.

See Analyze Output of Knowledge Generation for more details.