A manufacturing and distribution client's support team was spending substantial time on repetitive return, replacement, and refund inquiries. Policies were scattered across departments, and finding answers took several minutes. This case study describes a four-week forward deployed engineering project.
The Challenge
- RMA policies and exceptions were spread across more than six documents and wikis.
- Onboarding was slow and answer quality varied by representative.
- Repeated inquiry types accounted for more than half of support requests.
The Approach: Forward Deployed Engineering
We began with support logs and real workflows. We identified the most frequent inquiries, built a RAG knowledge base, and embedded evidence-backed answer drafts directly in the representatives' existing interface.
- Week 1: Review logs and documents and identify the most common inquiry types.
- Weeks 2–3: Build and demonstrate a hybrid RAG assistant MVP.
- Week 4: Incorporate feedback, refine citations and approval flows, and deploy.
Results
- 70% reduction in average response time.
- MVP demonstrated within two to three weeks.
- Answer sources fully traceable.
Representatives moved from searching documents to reviewing and refining evidence-backed drafts. Human confirmation remained in place for sensitive actions such as refund approval.
Key lesson: Remove repetitive work so people can focus on judgment. Visible sources helped representatives trust the assistant and encouraged adoption.