No idea why this is happening, the bot deflection rates reported in the Architect Analytics dashboard differ significantly from the conversation count metrics in the standard Queue Performance view.
Background
Our organization operates a hybrid support model using Genesys Cloud CX, with a primary inbound queue handled by a conversational AI bot before human agent handoff. The environment is configured for the Paris region (eu-west-1). We utilize the standard Architect flow structure where the bot attempts resolution; if unsuccessful or if the user requests an agent, the flow transitions to the human queue. The current Architect version is v2.1. We rely on the ‘Bot Performance’ and ‘Queue Performance’ dashboards to audit our automation efficiency.
Issue
Over the past 48 hours, the Architect Analytics report indicates a bot deflection rate of approximately 62%. However, when correlating this with the Queue Performance dashboard for the associated human queue, the ‘Handled by Bot’ metric implies a deflection rate closer to 45%. This 17% gap represents a significant variance in our operational reporting. The discrepancy appears most pronounced during peak call volumes between 09:00 and 11:00 CET. Specifically, conversations that are terminated by the user within 10 seconds of bot engagement are counted as ‘Deflected’ in Architect Analytics but are not reflected in the queue’s successful resolution metrics. This creates a conflict in our KPI reporting for the executive team.
Troubleshooting
- Verified that the Architect flow logic correctly tags conversations with the ‘Bot Handled’ disposition upon successful resolution.
- Checked for any lag in data synchronization between the Architect engine and the reporting services, but the delay seems consistent across all time blocks.
- Reviewed the conversation detail view for sample IDs during the peak window. The ‘Bot Score’ is populated, and the ‘Reason for Transfer’ is explicitly logged as ‘User Request’ for non-deflected calls.
- Confirmed that no custom scripts or API calls are interfering with the disposition mapping.
Is there a known latency or logic difference in how Architect Analytics calculates deflection versus how the Queue Performance module aggregates bot interactions? We need a unified definition for our monthly reporting.