Widget performance
One widget's own funnel: views, opens, starts, its engagement rate, messages per conversation, and a version-over-version comparison that is explicitly not a controlled test.
Every widget's own editor has a Performance tab, right after Versions, with the widget's preview kept beside it. Where the site-wide Engagement report answers "how is the site doing," this tab answers "how is this widget doing": its own funnel, on its own selectable date range.
Choosing a range
A date range picker sits at the top, with the same presets and custom-range calendar as the other reports. It remembers its own range independently of the Engagement report's, so changing one doesn't move the other. There's no compare-to-previous-period toggle here; the version comparison below covers that need in a different way.
What it shows
If the widget had no views and no opens in the selected range, the tab shows a single empty state instead of zeroed tiles.
KPIs
- Widget views: sessions where this widget was visible.
- Conversations opened: sessions where the chat panel opened from this widget. This can sit slightly above the numerator of the engagement rate: a session that opened from the widget with no recorded view counts here but not in the rate.
- Conversations started: sessions where the shopper sent a first message.
- Engagement rate: sessions that viewed this widget and opened the conversation from it, divided by sessions that viewed it. Same definition as everywhere else, see How the engagement rate is defined. Shown with one decimal, rounded up. A dash when the widget had no views in range.
- Messages per conversation: the average number of messages (shopper and assistant combined) across every conversation this widget started in range. A dash when nothing started.
A widget's figures roll up every version that was live during the range, the same way the Engagement report's per-widget table does. Republishing mid-range doesn't split the widget's history in two.
Version comparison
Not a controlled test
This card compares the engagement rate of the version that went live most recently against the version it replaced, over two windows of matching length on either side of the switch. It deliberately shows two numbers side by side and never a winner, a confidence figure, or significance language: seasonality, traffic mix, and catalog changes can all explain a difference on their own. Treat it as a smell test, not proof that the new version is better or worse.
The comparison reads the periods each version was live, not the order of the numbers. A version you put back live is compared with the version it replaced: if you put V2 back live after V5, the card compares V2 with V5, not with V1. The two sides are labelled with each version's name, or its V number when it has none.
The card is absent entirely for a widget that has never had two different versions live: there's nothing to compare yet. It's also independent of the date range picker above: it always measures around the most recent switch, not the range you've selected for the KPIs.
Where this comes from
The Performance tab and the Engagement report's per-widget table read the same underlying figures, computed from the events ledger rather than from any per-conversation field, so an erased conversation never changes a past figure retroactively, even though the conversation itself is gone.
The widget editor
A header with a status tag, Save and one primary action, Go live or Take offline, four editing tabs and two that inspect the widget, and a preview beside every tab.
A/B testing a widget
Run the version on screen against your live version with real traffic split between them, and get a verdict you can trust: the traffic floor, the detection limit, and the one-session-at-a-time limitation to know before you start.