When a guest sends a message, the Aplysia 3 chatbot uses a multi-step decision-making process to determine the best response. This process combines topic logic with document retrieval to ensure flexibility, accuracy, and control.
Step 1: Topic Matching
The first stage in Aplysia3’s response process is Topic Matching. The system evaluates the guest's message to determine if it corresponds to any pre-configured Topics. These Topics are created and trained by property or brand administrators to address frequent or critical guest inquiries.
-
If a matching Topic is found and set to "Replace AI-generated Answer", Aplysia3 immediately responds using the predefined content. This can include text, images, and interactive buttons. No further document search is required in this case.
-
If a matching Topic is set to "Support AI-generated Answer", Aplysia3 proceeds to retrieve supporting text from property documents. The final response blends AI-generated content with the visuals and actions defined in the Topic Response.
-
If a matching Topic exists but lacks configured content, it is ignored, and the system continues to the next step.
-
If no matching Topic is found, Aplysia3 automatically advances to document retrieval.
This stage allows for full control over certain interactions while still enabling AI flexibility when needed.
Step 2: Document Retrieval
When a message is not fully answered by a Topic, Aplysia3 activates its core AI capability: Retrieval-Augmented Generation (RAG).
-
The chatbot searches across the Property Knowledge Document and the Company Knowledge Document for up-to-date, structured information.
-
Only documents explicitly marked as “Generating live answers” are used in this process.
-
RAG enables Aplysia3 to extract relevant content in real time and synthesize it into context-aware, accurate responses that are tailored to the user’s question.
This step ensures Aplysia3 can address a wide variety of inquiries, even those not anticipated in pre-configured Topics, while staying grounded in verified, structured data.
Step 3: Fallback Behaviour
If neither a matching Topic nor relevant document content is available, Aplysia3 moves into fallback mode.
Depending on configuration, this may involve:
-
Displaying a generic fallback message
-
Transferring the conversation to a human agent
-
Offering general support options
This ensures the guest always receives a response or a path to further help, even if the system cannot resolve the issue independently.
Additional Decision-Making Features
While following the main three-step process, Aplysia3 also runs parallel checks to enhance interaction quality:
-
Triggering conversation flows such as “Book a Room” or other custom journeys.
-
Suggesting relevant Topics based on the guest’s current intent, using its Proximity Feature.
-
Escalating to a human agent when the complexity or tone of the conversation indicates that further support is required.
Why This Matters
This layered architecture allows Aplysia3 to offer:
-
High accuracy for property-specific and branded topics.
-
Dynamic, up-to-date responses via real-time document integration.
-
Flexibility and control through configurable Topic overrides.