The AI recommendations feature

Kevin has recently joined U+ Communications and is still finding his way around the tools available to help customers. He receives a call from John, who went on a cruise trip and is worried about roaming charges on his bill, even though he had a travel pass. John wants to understand how the charges occurred and what can be done.

In the past, this kind of situation would require Kevin to search through documentation or reach out to peers and seniors to find the right answer. Now, Kevin has access to Pega Knowledge Buddy directly within the customer service application. He types his question about roaming charges and travel pass billing directly into the Knowledge widget and receives a quick, summarized response he can share with John. The response also includes a link to the source document, so Kevin can verify the information and share it with confidence.

John is still not satisfied. He is sure he had airplane mode enabled throughout his entire trip. Kevin immediately asks a follow-up question to Knowledge Buddy without repeating the full context of the conversation. Because of the conversational capability Knowledge Buddy understands the conversation context from within the same session. Kevin receives a follow-up answer that addresses the airplane mode scenario specifically.

Kevin explains the possible reasons to John clearly and confidently, drawing on the relevant policy document referenced in the Pega Knowledge Buddy response. John is now satisfied with the explanation.

After the interaction, Kevin reflects on the conversation. He feels that some additional resolution options could have been offered to the customer. He provides this feedback directly on the response within the Knowledge widget. Kevin can give feedback on any message in the conversation, and this submitted feedback is what triggers the recommendations engine.

How the recommendations engine works

After Kevin submits his feedback, the recommendations engine, a background agent that runs continuously, analyzes the feedback together with the full context of the interaction: the question, the generated answer, and the specific document chunks used to produce that answer. The engine does not generate recommendations for interactions where no feedback has been submitted.

Approximately ten minutes after Kevin provides feedback (based on the scheduler frequency), the engine generates a recommendation case. This case identifies what was missing or unclear in the source content, for example adding additional resolution options for travel pass billing disputes. Before creating a new case, the engine checks whether a similar recommendation already exists. If it does, the new feedback is tagged to the existing recommendation ID rather than generating a duplicate case.

A knowledge author reviews the recommendation on the AI Recommendations landing page, or directly at the content level, where all suggestions gathered from various users for a specific document are visible. The home page dashboard also displays the volume of content recommendations. After reviewing the suggestion, the author manually updates the content or source document to include the missing information and re-ingests the new version to Pega Knowledge Buddy. To validate the improvement, the author can navigate to Preview Buddy in the Knowledge Buddy portal to test if the Buddy response now includes the additional resolution options.

The next time a CSR asks a similar question, Pega Knowledge Buddy returns an improved, more complete answer -- without anyone needing to manually flag the issue again.

Summary

This scenario demonstrates how the recommendations feature in Pega Knowledge Buddy closes the loop between CSR interactions and knowledge base quality. By allowing Kevin to provide feedback directly within the customer service application, the recommendations engine analyzes that feedback -- along with the full context of the interaction -- and generates targeted improvement suggestions. Knowledge authors can review and act on those suggestions, making every future interaction faster, more accurate, and more helpful.