Pulse
Sessions, Micro-insights, conversations, content gaps, and flow numbers for one app.
Pulse is the day-to-day surface for what's happening in your product, in this order:
Sessions
A single 30-day session count at the top.
Insights
The Micro-insight surface: short, cited statements, each carrying a place, a number, and
an action you can click. A number always links through to the sessions behind it. Some carry
a clearly marked Hypothesis note, a model-proposed cause, never presented as a fact, and
always tied to the sessions it was read from. Nothing renders here if there's nothing notable
to say; if your attributes are missing plan_value/plan_cycle, this section prompts you to add
them instead, since revenue-weighted insights are impossible without them.
Conversations
A searchable transcript browser over every conversation the agent has had.
Visitors can rate any answer good or bad from the widget, and those ratings show up here. Each conversation in the list carries a count of how many answers were rated each way, so you can spot a bad answer without opening every thread, and the rating itself sits under the answer it is about when you open the transcript. Only answers a visitor actually rated show anything: an unrated answer is silence, not a neutral score.
A run of bad ratings on the same question is usually a Sources problem rather than a persona one. Check the content gap below first.
Content gap
Questions the agent couldn't answer, grouped by exact text and ranked by frequency, each with a count and when it was last asked. This is your content backlog: what to add to Sources next.
Flows
This section measures your own authored flows, distinct from Lift, which measures the underlying task regardless of whether anyone used the widget.
The date range picker at the top of the section covers the last 7, 30 or 90 days, or any two dates you pick. Whatever you choose drives everything below it at once: the three numbers, both charts and the table.
The three numbers
- Runs started: how many flow runs began in the window.
- Completion rate: one rate across every flow.
- Live flows: how many flows are published in this environment right now.
How every run ended
A stacked column per day. The height of a day is the runs that started that day, so volume and quality read from one shape. Every run counts toward exactly one outcome, so the five colours always add up to that day's total:
- Completed: the person finished. A run that broke halfway and was then finished counts here too, because the flow did its job. The error is not lost, it still shows in the Errors column of the table.
- Failed: the run hit an error and never finished.
- Dismissed: the person closed the flow themselves.
- Left unfinished: no dismissal, no error, just a long silence. A run with nothing happening for thirty minutes and no ending lands here.
- In progress: the run is still open. It is kept apart so a run that started two minutes ago does not read as an abandonment.
Runs by flow
One bar per flow. The length is that flow's run count and the colours inside it are the same outcome split as the chart above, so you can see which flow carries the traffic and which one is stuck in a single glance. A flow with three runs looks small rather than looking like a perfect 100%. The ten busiest flows get their own bar and the rest are rolled into one Other flows bar, so the bars still add up to the total above them.
Clicking a bar opens that flow's own Pulse page.
The table
Per flow: started, completed, completion rate, weakest step, errors and a small trend chart. The name in each row leads to that flow's step-by-step funnel.
This section counts flows that are published in this environment and flows that ran inside your window. Drafts you never published stay out of the numbers. A flow you unpublished last week still keeps last week's runs, because a report about the past should not change afterwards.
A very wide range can hit ARCY's scan limit. When it does, a notice appears above the section saying the numbers cover only the most recent events. Picking a shorter range gives a complete answer.