Run a Signal on demand
Once a Signal exists, click Run from its detail page to open the Run Signal dialog and score it over a fresh date range.
Date range — Today, Yesterday, Last 7 days, Last 30 days, or Custom.
The dialog shows a live count of sessions in range as you adjust the date range.
Channels and Keyword filter are read-only here — they reflect what the Signal was already configured for in Scope, not something you change at run time.
Click Run to start the evaluation. Results appear on the Signal's Overview tab when the run completes.
Run a test while building a new Signal
You can also kick off a first test run directly from Step 3 (Schedule & Run) while building a new Signal.
Understand pre-filters and post-filters
Signals uses two kinds of filters, and it's worth knowing the difference before you run.
Pre-filter — set in the Scope step when building the Signal, determines which sessions get evaluated in the first place (by Agent, Guide, Channel, keyword, or date). Use a pre-filter when you have a specific business need, like evaluating only chat sessions or only sessions handled by a particular guide.
Post-filter — the Agent and Channel dropdowns on the Signal's results page, which narrow the set of sessions you're viewing after they've already been evaluated. Post-filters don't require a new run.
In general, run Signals on a representative sample of all your traffic so you can post-filter freely. If you pre-filter too narrowly upfront, you'll need to re-run the Signal to look at a different slice.
Best practices for running
Start small, then scale. Run a first test under 50 sessions, confirm the AI's classifications look right, then expand the date range or turn on a recurring schedule.
Widen your scope if results look thin. If a run keeps returning the same one or two labels and never touches a third, don't assume the criterion is broken first — try removing agent/guide/channel filters and extending the date range as far back as it goes. A narrow scope can easily miss whichever scenario type happened to run on a different day, especially in a lower-volume or test org.
Recurring schedules should match how often you'd actually act on the data. Daily is useful for high-volume, fast-moving signals; weekly is usually enough for slower-moving quality trends. There's no monthly option — for seasonal windows, use a manual run with a Custom date range instead so you control exactly when the window starts and ends.
Signals evaluates a sample today, not every session. Today's sample (about 20%) keeps results comparable across runs; full-volume evaluation is on the roadmap.
