What Data Sources Power Now Assist's "Chat Recommendation"?

[kanakoA]

Which data sources does the Now Assist skill "Chat Recommendation" refer to?

Does it pull from past incident resolution history, or from knowledge articles?

Based on some testing, it seems to reference past incident history, but not knowledge articles.

If anyone has insights or experience with this, I’d appreciate your input.

[VedS]

Hi @kanakoA ,

ServiceNow's Now Assist uses a combination of data sources to power its "Chat Recommendation" feature, leveraging the vast amount of information available within the ServiceNow platform and external Large Language Models (LLMs). Here's a breakdown of the key data sources:

• Your ServiceNow Instance Data: This is the primary and most crucial data source. Now Assist draws heavily from your organization's specific data within the ServiceNow platform, including:
• Past chat conversations: The history of interactions between agents and users provides valuable context, common issues, and successful resolutions.
• Knowledge articles: These are a rich source of documented solutions, FAQs, troubleshooting steps, and policy information.
• Cases and incidents: Data from resolved and ongoing cases/incidents helps Now Assist understand problem patterns and effective solutions.
• Other related records: Depending on the context (e.g., IT Service Management, Customer Service Management, HR Service Delivery), Now Assist can access data from various linked records to provide more relevant recommendations. This could include user profiles, asset information, service catalogs, etc.
• Customer-specific configurations and workflows: Now Assist adapts to your organization's unique processes and terminology.
• Large Language Models (LLMs): Now Assist integrates with leading LLMs, which are trained on massive datasets of text and code. These LLMs provide the foundational natural language processing (NLP) capabilities that allow Now Assist to:
• Understand the nuances of human language in chat conversations.
• Generate coherent and contextually relevant responses.
• Summarize complex information.
• Refine existing text.
• ServiceNow natively supports integration with various LLMs, including OpenAI (GPT-3, GPT-3.5 Turbo, GPT-4), Azure OpenAI, Google Cloud AI (Vertex AI & MakerSuite), Aleph Alpha, and IBM WatsonX.
• Now Assist Data Kit (NADK): This is a feature within ServiceNow that allows organizations to create, curate, and maintain datasets specifically for training and evaluating custom AI models and skills, including those used for chat recommendations. This enables:
• Custom datasets: Organizations can create datasets from their ServiceNow tables (e.g., knowledge articles, incidents) to tailor the AI's understanding to their specific needs.
• Ground Truth: NADK allows for defining "ground truth" or the desired output for a given record, which is used to evaluate the accuracy of AI-generated responses.
• Feedback and Continuous Improvement: Now Assist is designed to learn and improve over time. Agent feedback on recommendations (e.g., marking a suggestion as helpful or unhelpful) contributes to the ongoing training of the underlying AI models, leading to more accurate and relevant recommendations in the future.

Please mark this as "correct" and "helpful" if you feel this answer helped you in anyway.

Thanks and Regards,

Ved

[anubhavkapoor76]

@kanakoA - If you are referring to "chat reply recommendation" skill then here is a breakdown:

Step 1: Transcript of the current chat is passed to the chat summarization skill which returns a brief summary of it.

Step 2: From Step 1 summary its inputted to AI search call which retrieves KA's, Incidents, cases and other chats. This is RAG call.

Step 3: The chat reply recommendation skill then takes the outputs from Step 2 and transcript of current chat and recommends a chat reply.