Configuring “AI Picks the Team” for Voice Handover
Voice handover used to mean picking one team and, optionally, one queue in advance — every call that needed a human went to that same destination, no matter what the caller actually needed. AI picks the team replaces that fixed assignment with a model decision: when a voice call needs to hand over to a person, the model looks at every team’s description and picks the team and the queue in a single choice, instead of a person having pre-decided it for every call ahead of time.
This is a per-stack setting on the voice handover configuration, stored as voiceHandoverAiTeam on the agent stack. It sits alongside — and overrides, for voice — the existing handoverTeamId used by fixed team assignment.
Where to enable it
- Open the agent stack’s Handover tab and find the Handover to a human (voice) card. A voice stack has no Voice tab — its voice settings sit inside General, Handover and Testing.
- Make sure Send caller to a human queue is the selected action; the Team dropdown only appears for that option.
- Open the Team dropdown. Alongside
(none)and your tenant’s actual teams, you’ll see an AI picks the team option. - Select it and click Save.
Selecting AI picks the team doesn’t touch handoverTeamId — that field is shared with the text-handover configuration and is left exactly as it was. The AI-picks flag simply takes priority over it for voice calls; if you later switch the dropdown back to a specific team, that fixed choice is what applies again.
Why every team needs a description
The model isn’t guessing from team names alone. What it is actually offered is a flat list of the enabled queues belonging to teams that have a description, each one labelled with its team’s name and description (and the queue’s own description, when it has one). Choosing the queue is choosing the team. A team with no description is left out of that list entirely — the model has nothing to judge whether it’s the right destination, so it won’t be considered as a candidate. A described team with no enabled queue contributes nothing either, because there is no queue to offer.
That’s why the card shows a red warning when no team has a description, and says plainly that the transfer-to-human tool will not be available. Before any of this reaches the model, team names and descriptions are run through the same sanitization used for text handover’s by-description routing, so a team description can’t be used to smuggle instructions into the routing prompt.
Practically, this means the quality of your team descriptions is the configuration. A one-line “Handles refund requests and chargebacks” gives the model something to route on; a blank field or a name alone does not.
How this differs from a fixed team or a fixed queue
- A named team in the Team dropdown, with the queue left on AI picks the queue — The person for the team; the model for the queue, from the queue descriptions inside it — Team ahead of time; queue per call
- A named team plus Always use this queue — Whoever configured the stack — Ahead of time, for every call
- AI picks the team — The model, from each described team’s name and description and the queue’s own description — At the moment of handover, per call
With a named team, every handover from this stack lands on that one team regardless of what the call was about — the model can at most choose between the described queues inside it. With AI picks the team, the model reads the caller’s situation and the available teams’ descriptions and picks both the team and its queue together, as a single decision — it isn’t choosing a team and then falling back to that team’s default queue separately.
You can’t combine it with a fixed queue
Always use this queue and AI picks the team are mutually exclusive in the UI: selecting AI picks the team clears the queue selection and greys the queue dropdown out, and saving in that state writes the stack’s fixed queue back to empty. So there is no configuration where a fixed queue overrides an AI-routed team — if you want every handover in one queue, pick the owning team explicitly and set the queue, and leave AI picks the team alone.
What happens when nothing routes
When the model names no queue, or names one it wasn’t offered, the transfer uses the plan’s fallback queue — under AI picks the team that’s the first eligible queue across the described teams, so the call still reaches a human rather than being dropped.
If nothing is eligible at all (no team has a description, or none of the described teams has an enabled queue), there is no fallback queue. The transfer-to-human tool is then never offered to the model in the first place; and if a transfer is somehow attempted anyway, the bridge tells the caller “I’m transferring you, but no agent queue is available right now. Please call back later.” and hangs up, ending the session as a failed handover. That is the cost of leaving team descriptions blank.
Relationship to text handover’s “by description” routing
If you’ve already configured text handover to route by description, this will feel familiar: it’s the same underlying idea — let the model choose a destination from team descriptions instead of a fixed assignment — applied to voice. The two never collide on one stack, because a stack is either a text stack or a voice stack: on a voice stack the Triggers and Delivery sub-tabs aren’t shown at all, and the text handover mode and its delivery toggles are not read on the call path. What the two do share is the same team descriptions and the same sanitization step, so a description you write with text handover’s by-description routing in mind will also be usable, unchanged, by the voice version of this feature.