This one is tricky, and here is how I've accomplished this in couple of environments as this was the only way I could think of since AAR is not available on SIP trunks as oppose to SCCP ports:
configure your VM Route List to inlcude "Standard Local Route Group" as the last Route Group and configure
Called Party Transform Mask on this route group to point to a DID that is delivered to centralized site where UCXN is hosted.
This way when call arrives to site A and call needs to be redirected to VM per call forward no answer/busy the VM profile/VM pilot/Route Pattern, since the primary SIP trunk(s) are in different "location" and no bandwidth is available call will be rejected as there is no way to add AAR Group/destination to SIP trunk so the next available route group in the route list will be tried and since local route group is in the same "location" call will be send to it and then redirected to the masked DID and delivered to central site. In CUCM matching TP is defined to match the DID on ingress side and translated to VM pilot. In UCXN you will need forwarding routing rule that matches this DID and sends the call to "attempt to sign-in" conversation.
HTH, please rate all useful posts!
Chris
