Bot Performance Dashboard Metrics Discrepancy in Architect Flows

Trying to understand the divergence between the reported ‘Bot Success Rate’ in the standard Performance Dashboard and the actual conversation outcomes visible in the Interaction History view. We have configured a specific Architect flow utilizing the Natural Language Understanding (NLU) widget for intent classification, followed by a routing block to a human queue upon intent confidence scores below 0.75. The dashboard metric for ‘Bot Success’ appears to count any interaction that does not result in a transfer as a success, regardless of whether the user explicitly requested an agent or abandoned the session due to frustration. This definition conflicts with our internal Service Level Agreement (SLA) requirements, which mandate a successful resolution based on post-interaction survey data or explicit intent completion markers. The discrepancy is particularly pronounced during peak hours in the Europe/Paris timezone, where high volume correlates with an increase in ‘successful’ bot interactions that subsequently trigger high abandonment rates in the downstream human queue. The current metric definition does not seem to account for the semantic outcome of the conversation, only the technical routing path.

Is there a method to customize the ‘Bot Success’ definition within the default performance views, or must we construct a custom dashboard utilizing specific Architect flow events? We require a metric that reflects true resolution rather than mere transfer avoidance. The standard views lack the granularity to filter by specific intent completion nodes within the flow. Additionally, the latency in data propagation to the dashboard appears to introduce a significant delay, making real-time monitoring ineffective for immediate intervention. We need to align the reported metrics with operational reality to accurately assess the bot’s contribution to overall contact center efficiency. Any guidance on configuring custom metrics that leverage Architect flow event data to replace the generic ‘Bot Success’ indicator would be appreciated. The current reporting structure creates a misleading performance narrative that does not support strategic decision-making regarding bot deployment and training.

The simplest way to resolve this is to align your dashboard logic with the actual wfm.schedule.status. Often, the dashboard counts success based on initial intent, ignoring subsequent routing. Check if intent.confidence drops below 0.75 but the interaction still logs as successful due to a missing transfer event in the interaction.history.

Have you tried filtering the bulk export by interaction.outcome instead of relying on dashboard aggregates? The intent.confidence threshold often fails to trigger a proper transfer event in the metadata, so checking the raw interaction.history for missing routing steps is usually more reliable.