Predictive routing analytics gaps during byoc trunk failover in apac

is it possible to reconcile the predictive routing queue metrics with the actual sip trunk failover events in our apac deployment?

we have fifteen byoc trunks configured with active-passive failover logic. when the primary carrier drops calls due to sip 408 request timeout, the system routes to the secondary trunk as expected. however, the predictive routing analytics dashboard still reports the initial queue wait time and service level breach based on the failed primary leg.

the api response from /api/v2/analytics/routing/queues shows a high abandon rate because the call is technically abandoned by the predictive engine before the failover succeeds. we are using the java sdk version 12.1.0 to pull these metrics.

we need to correlate the sip registration logs with the routing analytics to exclude these failover-related abandons from our sla calculations. the current setup treats the failover delay as user abandonment.

has anyone implemented a custom metric or filter in the architect flow to tag these specific failover events? we want to ensure the reporting accurately reflects the customer experience rather than the carrier instability. any advice on adjusting the outbound routing settings to better handle this latency would be appreciated.

1. Query the conversation_events endpoint for the specific date range where the 408 timeouts occurred. Filter by type: "conversation:updated" and look for routing_queue changes.
2. Isolate events where routing_queue transitions from the primary queue to the secondary queue. The timestamp of this transition marks the actual failover point.
3. Calculate the true wait time by subtracting the primary queue entry timestamp from the secondary queue entry timestamp, rather than using the final disposition time.
4. Use the /api/v2/analytics/conversations/queues/metrics endpoint with a custom date range that aligns with the secondary queue entry times. This bypasses the dashboard’s default aggregation which often ties metrics to the initial routing attempt.

The dashboard aggregates based on the initial routing decision, not the successful agent connection. For BYOC failovers, the initial leg failure creates a “ghost” wait time in the primary queue metrics. Pulling raw events allows for accurate reconciliation of service level breaches against the actual secondary trunk performance. This approach works consistently across APAC regions where latency spikes often trigger false 408s.