In the last post, I explained how semantic search improves discoverability by helping users find the most relevant content based on meaning, not exact wording. But what if we could take this even further? What if, instead of giving users a list of articles to sift through, we could provide a direct, conversational answer to their specific question?

That’s the promise of generative AI in knowledge management, and it’s reshaping our entire perspective on self-service.

The Conversation Expectation

Today’s customers don’t just want faster search results; they want experiences that feel as natural as asking a colleague for help. They want to describe their problem in their own words and receive a clear, actionable response without having to sift through multiple articles or navigate complex documentation structures.

This shift in expectations isn’t just about convenience – it’s about reducing cognitive load. When customers face a technical issue or urgent problem, the last thing they want is more homework. They want solutions.

RAG: The Technical Foundation

Enter retrieval-augmented generation (RAG), our method for delivering AI-powered answers that are both intelligent and trustworthy. Unlike standalone AI models that might hallucinate or provide outdated information, RAG systems ground their responses on your organization’s actual knowledge base.

Here’s how it works: when a customer asks a question, the system first uses semantic search to find the most relevant content from your knowledge base. Then, generative AI synthesizes this information into a natural language response that directly addresses the user’s specific question. The result feels conversational and immediate, but it’s actually sourced from vetted, authoritative content.

For example, instead of returning three articles about “database connection errors,” the system might reply: “It sounds like you’re experiencing a timeout issue with your database connection. This usually happens when the connection pool is exhausted. Here’s how to fix it: First, check your connection pool settings in the admin console…” The answer is specific, actionable, and feels like it was written just for that customer.

Channel-Specific AI Strategies

But this is where strategy becomes essential: different channels demand different approaches to AI-generated content, and misjudging this can erode trust.

For public self-service channels, I recommend restricting RAG to verified, published content such as official knowledge articles and documentation. Customers expect authoritative information, and any uncertainty in AI responses can undermine confidence in your entire support ecosystem. The guardrails should be strict, and the system must clearly indicate when it cannot provide a definitive answer.

For agent-assisted channels, the guardrails can be more flexible. In this case, RAG can include useful but non-curated content like previous service requests, chat transcripts, or internal troubleshooting notes. Why? Because the agent acts as a human filter, reviewing and refining the AI-generated response before it reaches the customer. This creates a powerful hybrid model where AI offers speed and agents add judgment.

The Human-AI Workflow

This results in what I call “role-aware AI delivery,” a model where human oversight is intentionally integrated into the process. Agents act as curators of AI output, reviewing, refining, and personalizing responses before sending them to customers.

In practice, this might look like: AI suggests a response based on similar past cases, the agent reviews it for accuracy and tone, makes necessary adjustments (possibly adding customer-specific context), and then sends the refined answer. The customer receives a response that’s both quick and thoughtful, while the agent focuses on high-value review and customization rather than starting from scratch.

Measuring Success Beyond Speed

As organizations adopt AI-generated answers, traditional support metrics need to evolve. First-contact resolution rates might drop initially, not because the system is failing, but because RAG is resolving easy-to-solve questions without the customer submitting a chat or SR. Customers are now more comfortable asking complex and nuanced questions they previously wouldn’t have bothered to submit.

Instead, focus on metrics like:

  • Answer satisfaction scores: Are customers finding the AI responses helpful?
  • Escalation rates: Are fewer cases requiring human intervention?
  • Self-service completion rates: Are customers able to resolve issues end-to-end without needing to create a ticket?
  • Agent efficiency: Are agents able to handle more complex cases when AI handles routine inquiries?

Looking Ahead

The next frontier isn’t just about providing better answers – it’s about anticipating questions before they’re asked. In the next post, I’ll show how Oracle uses AI not only to retrieve or generate answers but also to proactively identify gaps in your knowledge and predict what content customers will need before they even realize they need it.

Want to see How AI is providing value in the real world? Don’t miss Powering Sustainable Energy with Fusion Service: ESS Inc.’s GenAI Journey where you’ll hear firsthand how ESS Inc. is using AI-generated SR summaries and Knowledge Article authoring to transform their service operations.