I am working on a deployment where agents need to perform attended transfers to external PSTN numbers. The transfer initiates correctly - the agent clicks Transfer, dials the external number, speaks to the third party, and then completes the transfer. However, approximately 25% of completed attended transfers fail with a SIP 481 “Call/Transaction Does Not Exist” error.
When the transfer fails, the customer hears dead air for about 5 seconds and then the call drops entirely. Neither the agent nor the external party remain connected.
The SIP trace from the Genesys Cloud diagnostic logs shows:
10:32:15.123 --> REFER sip:[email protected] SIP/2.0
10:32:15.234 <-- SIP/2.0 481 Call/Transaction Does Not Exist
10:32:15.345 --> BYE sip:[email protected] SIP/2.0
The REFER is sent to the external party’s carrier after the agent completes the transfer, but the carrier responds with 481, indicating the call leg to the external party no longer exists by the time the REFER arrives.
We are using Genesys Cloud Voice (not BYOC) in the US-East-1 region. The external numbers are on various carriers - AT&T, Verizon, and CenturyLink. The 481 failures seem slightly more common on CenturyLink numbers.
Is this a known interoperability issue with certain carriers, or is there a configuration we are missing?
The 481 on REFER is a classic carrier interoperability issue. Not all PSTN carriers support SIP REFER for call transfers. When Genesys Cloud sends a REFER to transfer the call, the carrier’s SBC needs to support the REFER method and maintain the dialog state long enough to process it.
CenturyLink (now Lumen) is notorious for having SBCs that aggressively clean up SIP dialog state. If the consult call to the external party has any pause or hold during the attended transfer process, the carrier SBC may time out the dialog before the REFER arrives.
The fix is to switch the transfer method from REFER to re-INVITE. In Genesys Cloud Voice, this is not directly configurable because you do not control the trunk settings. You need to open a support ticket and request that your GC Voice trunk be configured to use “re-INVITE based transfers” instead of “REFER based transfers.”
With re-INVITE transfers, Genesys Cloud acts as a B2BUA (Back-to-Back User Agent) and maintains both call legs throughout the transfer. The carrier never receives a REFER - it just sees the existing call continue with different media. This is universally compatible with all carriers.
The re-INVITE approach is the permanent fix. While waiting for Genesys support to process the trunk change, there is a workaround at the Architect flow level.
Instead of using the agent desktop’s built-in Transfer button, build a custom transfer flow that uses Blind Transfer instead of Attended Transfer. With blind transfer, Genesys Cloud sends a REFER to its own internal media server (which always supports it), and then the media server establishes a new outbound call to the external party.
The user experience is slightly different - the agent cannot speak to the external party before completing the transfer. But the transfer reliability goes to 100% because the REFER never reaches the external carrier.
If attended transfer is a business requirement, you can simulate it by having the agent place the customer on hold, initiate a separate outbound call to the external party via click-to-call, confirm the transfer verbally, then use the Merge feature to conference all three parties, and finally drop from the conference. This avoids REFER entirely.
From a supervisor perspective, the agent experience matters a lot here. The blind transfer workaround works technically but our agents absolutely need to speak to the external party before transferring a customer - especially for medical referral calls where context must be communicated.
What worked for us was the conference-and-drop approach. We trained our agents on a three-step process: Hold customer, Call external party, Conference all three, Then the agent disconnects. It takes about 10 seconds longer than a regular attended transfer but the success rate is 100%.
We added a quick-reference card to the agent Script panel that walks through the steps with screenshots. Agent adoption was high because they could immediately see that transfers stopped failing.