Baggio AmouzouMontréal
loading fonts and photographs

CASE STUDY · AGENT TO AGENT TRANSFER

A call that changes agent

The memory follows the call

Consulting contract, voice agents in production · Two agents, one conversation

Result: 1 to 2 s both directions, verified on a real call

01 / THE PROBLEM

On a transfer, the caller starts over from the beginning.

They have already explained the problem once. A transfer hands over the line, not the conversation: the second agent opens as if the call had just started, and everything has to be said again. That is the moment in a call where people hang up.

02 / THE CONSTRAINT

The receiver has to have read before its first word.

A transfer is not a page load. On a phone call the loading state is silence, and silence is expensive. Whatever the second agent needs to know has to be there before it speaks, inside the length of the transfer itself. Second constraint: adding an agent to the fleet must not mean editing the ones already there.

03 / WHAT I CHOSE

One shared module, and a directory each agent declares itself in.

The memory could have been copied into every agent. It lives in one shared module instead: the logic exists in a single place, and a new agent registers in the directory rather than being wired to each of its neighbours. That is what makes adding one more agent cost the same as adding the second.

04 / WHAT IT DOES TODAY

What happens between one agent handing over and the next one speaking.

  1. 01

    The caller explains

    once

  2. 02

    First agent

    qualifies,

    decides to transfer

  3. 03

    Write

    call memory:

    what was said,

    what is left to do

  4. 04

    Read

    the receiver reads

    before its first word

    1 to 2 s

  5. 05

    Resume

    opens with a recap,

    asking nothing again

  6. 06

    Directory

    each agent registers,

    the others do not move

Verified on a real call, both directions. Nothing here describes the client's infrastructure.

05 / WHAT I WOULD CHANGE

One to two seconds, on a phone call, is still silence.

The memory is written at the moment of transfer, so the time it takes to write lands squarely in the gap the caller hears. Writing it as the conversation goes, rather than at the end, would leave nothing to compose at handover: the receiver would only have to read. The silence would fall to the read latency alone.