Human service agents resolving product and warranty cases had to search for a fragmented knowledge base and rely on institutional knowledge locked inside past brand expert-resolved cases. Quality and speed depended heavily on individual tenure, and the work spanned four languages. The goal was faster, more consistent resolutions — and a way to capture and reuse expert knowledge at scale. 

An Agentforce Employee Agent is embedded directly into Salesforce and is available to selected customer service representatives through a permission set. 

To provide accurate recommendations, we first identified a group of experienced support agents whose case resolutions represented best practice. We then transformed historical cases handled by these experts, together with knowledge articles and technical documentation, into a trusted AI knowledge base. Knowledge articles were migrated as RAG-optimised text (structured problem-solution format, consistent tagging and search-friendly phrasing), while AI-generated summaries of expert-approved historical cases turned years of support experience into a searchable institutional knowledge base. 

When a new case is opened, Agentforce uses RAG and similarity search to find the most relevant content. It first matches on product or article number, then searches generic knowledge articles, and finally similar expert-approved historical cases. Based on these sources, it generates a structured recommended resolution and a ready-to-send customer email draft in both the customer’s language and the representative’s language. 

Every recommendation includes links to the underlying knowledge article or historical expert case, allowing the representative to quickly verify why the answer was generated. Only validated knowledge and expert-approved cases are used as sources. 

In addition, we built a guided decision-tree tool that walks representatives through warranty qualification questions step by step to determine whether the request is covered under warranty, out of warranty, a crash replacement, or another outcome. 

The representative always stays in control – they can refine the AI response through chat, review the suggested email before sending, and provide thumbs-up/down feedback to continuously improve the solution.