Resolution rate: the only metric that matters at first
In the first few weeks a single number tells you whether the rollout is working: the share of conversations closed without an agent. If it is low, the cause is almost always the content, not the model — the knowledge base does not cover the real questions. The gap report shows you exactly which those are, and filling them raises the rate quickly. Once the metric stabilises, attention shifts to quality: of the conversations resolved automatically, how many led to a complete request and how many ended with no action at all.
Frequent topics are a map of your website
When the same topic appears in hundreds of conversations, it means the information is either missing from the site or hard to find. The frequent-topics list therefore becomes a content priority list: the pages that are missing, the sections that should move higher, the questions that deserve a page of their own. Not infrequently, the most frequent chat question points to a page that would bring considerable organic traffic if it existed — conversations tell you what people are looking for before a keyword tool does.
Keeping credit usage under control
Usage is metered in AI credits, per conversation processed, not per agent. The daily report shows usage per channel and a forecast for the end of the month, so you do not discover an overrun only at invoicing. If the forecast climbs unexpectedly, the usual causes are visible in the same place: a traffic increase, an answer length set too high by default, or conversations dragging on because the bot is missing information and the customer keeps rephrasing. Each of those has a direct fix in the panel.
