The organisation wanted an AI Virtual Agent for branch managers and staff, delivered through its internal branch service portal, so they could self-serve and stop calling the central support desk. The catch was that the solution had to be tested and tuned without putting the live production environment at risk.

The project ran in two phases.

Phase 1: Safe pilot, grounded answers. Konversational cloned the branch portal to create a dedicated, access-controlled pilot environment, so the beta stayed separate from the live service. Now Assist in Virtual Agent was set up to understand plain-English questions and answer them from the client’s own knowledge base and catalogue content. Retrieval-augmented generation kept answers grounded in approved content rather than generic internet data. Guardrails were in place from day one: sensitive-topic filtering, synonym matching for branch terminology, and a feedback loop that sends unresolved queries to the support desk.

Phase 2: Skills and handover. Alongside the technical build, Konversational worked with the CSM team to deliver Impact Accelerator training, so the client’s team built its own skills as the pilot progressed. At close, Konversational delivered full documentation and walked the client’s teams through it, setting them up to own the service day-to-day ahead of a wider rollout.

The AI Virtual Agent went live in production for pilot users within five weeks of kick-off. Internal engagement was strong, and leadership began preparing to demo the pilot to the CEO. Because the pilot ran in production rather than a separate test environment, the Knowledge Management team could make content changes and see the impact in real time, which helped them strengthen content ahead of a wider go-live. The organisation came away with a safe, production-tested foundation ready to extend beyond the first user group, and a team fully equipped to run it themselves.