When rolling out an enterprise AI assistant, technology is rarely the bottleneck. In our projects, the same three questions consistently determine whether the system is still in use six months later.
1. Who maintains the knowledge base? The assistant only knows what's contained in the documents behind it. If the policy, price list, or product description gets updated but the knowledge base doesn't, the assistant will confidently state outdated information — which is worse than not answering at all. It's worth naming a responsible person and an update schedule before rollout, and choosing a solution where content updates aren't a development task.
2. Does the user know they're talking to a machine? Our experience is that open disclosure doesn't reduce trust — it increases it: if the assistant explains upfront what it can and can't do, users ask realistic questions and experience less disappointment. This also applies to source citation — if the answer shows which document it drew from, it stays verifiable.
3. What happens when the assistant can't answer? Most dissatisfaction comes from conversations hitting a dead end. There needs to be a clear handoff point to a human: contact information, a callback request, or ticket logging — ideally in a way where the colleague can see the conversation so far and doesn't need everything repeated.
If these three questions can be answered at the start of the project, the post-rollout period typically becomes about fine-tuning rather than firefighting.