Bot Performance is most useful as a recurring operational habit, not a one-off audit. Establish the cadence early and tune continuously.
- Open the Performance tab
Einstein Bot Builder, open the bot, click Performance. The page shows containment, intent recognition, transfer rate, and per-dialog metrics for the selected date range.
- Pair every metric with CSAT
Containment, intent recognition, and transfer rate are operational metrics. CSAT is the customer signal. Always read them together to avoid optimising for the wrong outcome.
- Investigate per-dialog drop-offs
Find the dialog with the highest drop-off rate. Open it in Bot Builder, walk the steps, identify why users abandon. Fixes here have the highest leverage.
- Improve intent training when recognition lags
For intents with low recognition, add utterances. Real customer phrasings come from transcripts of failed conversations; mine those for new training data.
- Hold a weekly review
Treat Bot Performance as a weekly operational meeting. Without cadence, the data accumulates without action.
- Containment without CSAT confuses deflection for resolution. Always read the two together.
- Aggregate metrics hide dialog-specific drop-offs. Drill into per-dialog data weekly.
- Intent recognition issues quietly destroy user trust. Sorry I didnt understand replies are the loudest signal of a training gap.
- Performance data is retained for a limited window. Snapshot to a custom history object for trend analysis beyond a few weeks.