The “Analytics” section shows how the bot really works: how many people write, where they abandon the conversation, how much AI costs and how the operators cope. Looking at the numbers is useful not for the report’s sake, but to understand what to improve.
What the numbers are computed from
Messages, scenario steps and AI calls are recorded once as facts, and all the on-screen metrics are computed from them. That is why the numbers on different screens always agree: the same activity figure cannot differ in two places.
The overview and the funnel
The overview screen shows activity for a period and the state of the bots. The funnel answers the question “where are clients lost”: by scenario steps it shows how many people reached each step and how many dropped off. It is the fastest way to find a bottleneck — a question that is too long, or a confusing button.
By node
For every scenario step you can see how many times it ran, how it ended and how many errors there were. That is how steps that do not work as intended get found.
AI cost and knowledge base quality
AI spending is visible by bot and model for the chosen period — it is clear where the money goes. Next to it is knowledge base quality: how many answers were grounded in your documents, and how many times no answer was found in the base. If the second number grows, it is time to extend the documents — draft answers are conveniently taken from the gaps report in the Knowledge Base section.
Operators and broadcasts
For operators you can see how many conversations were handed to humans, how many waited for an answer and how many were closed — a hint on whether another person is needed. For broadcasts — the progress and per-recipient statistics.
The activity heat map
The heat map shows on which days and hours people write most often. It helps pick the time for broadcasts and understand when the operator should be around.
Logs
Logs are a detailed record of everything the bot did. There is search with filtering by bot, conversation, node, severity, errors and time, paged browsing, and the ability to open a conversation in full. The live feed shows entries at the moment the scenario executes — convenient when you are checking something.
Records have a retention period: steps are kept for thirty days, for example, then deleted. The summary metrics remain — they have already been computed and are stored separately.
Health
The “Health” screen shows for every bot whether it is running or stopped, its last error and when, how much message intake is lagging and how many messages are queued. The data updates in real time; the state of the platform itself is shown alongside. It is the “is everything alive” screen.
Live conversations
The “Live” section is a list of conversations happening right now: which step the person is on, what they last wrote and whether they are waiting for an answer. You can watch the system work as it is, without reports and without delay.
Next: model spending is configured in the Models & Prompts section, working with people — in Inbox, and checking a bot before publication — in Testing & Debugging.