Background and Challenges
The client, an industrial-equipment manufacturer with roughly 200 employees, kept policies, product manuals and after-sales tickets scattered across shared drives, mailboxes and paper binders. New hires shadowed senior colleagues for close to a month before answering customer questions on their own. When support or pre-sales staff hit a hard question, finding an answer took ten-plus minutes — and was not always the right one.
An off-the-shelf knowledge-base tool had been considered, but it could not align with internal permissions and sensitive data could not go to a public cloud, so the initiative stalled.
Our Approach
We built a privately deployed knowledge base with RAG question answering. Policies, product and after-sales documents are parsed, chunked and embedded, then served by an on-premises large language model that answers strictly from authorized material.
Three design points mattered. Permissions mirror the internal org structure, isolating sensitive content by department. Every answer cites its source for one-click verification. And two entry points — an API and a web console — plug into the ticketing system and the intranet portal respectively.
Results
- Time to independent handling for new hires fell from about 4 weeks to 2.5, with onboarding cost down accordingly (sample data, to be verified);
- Average time to find an answer dropped from 10+ minutes to under 1, lifting first-contact resolution (sample data, to be verified);
- All data and models run inside the client’s intranet and passed their information-security review (sample data, to be verified);
- The knowledge base now works as a living asset: answers follow document updates automatically.