HALO can leverage conversation history as an additional knowledge source, provided certain conditions are met. This enables HALO to utilize historical data to enhance its ability to answer incoming questions.
Preconditions
Conversation history can be used as a knowledge source when all of the following conditions are met:
-
Compatibility: You are using both HALO and Mobile Service Cloud (MSC).
-
Conversation volume: The profile has at least 500 relevant* conversations in its source language from the last 60 days.
*) HALO uses a probabilistic prompt to classify conversations as relevant. Its goal is to identify conversations that include informative solutions that could be useful to other customers.
Please note that this prompt is proprietary and may change over time. If you have questions, contact an AI specialist.
Process Overview
If the above conditions are met, historic conversations are automatically incorporated into HALO’s knowledge base. Here is how it works:
-
Processing Historic Conversations:
-
Historic conversations with non-AI agents are read and processed.
-
-
Clustering:
-
Conversations are clustered into groups of similar inquiries.
-
Examples of clusters include questions about delivery times or payments.
-
-
Summarizing Question-Answer Pair Formation:
-
Each cluster is transformed into a single summarizing question-answer pair.
-
Example: "Q: My package has not arrived yet; when will it arrive? A: That's unfortunate. Our delivery times range from 2 to 5 days."
-
-
Upload to Knowledge Database:
-
These generic question-answer pairs are uploaded to the knowledge database.
-
Impact on AI Responses
When a new question closely resembles an existing question-answer pair, the AI will likely use this historical information to formulate its response. This integration maximizes the utility of conversation history, enabling HALO to deliver more accurate and contextually relevant answers.