AI voice agents

An agent that only says what you approved

Voice agents take the calls that do not need a person — qualification, reminders, renewals, the inbound at 2am. They run an approved playbook, they hand over the moment intent gets complex, and they do not improvise.

Approval required before any play runs · Handover with full context

Live call workspace: queue on the left, transcript and the WhatsApp chaser in the centre, CRM context, sentiment and call summary on the right
Rollout measured in the analytics: transcribed calls scored against the approved playbook version
Guardrails

No play, no sentence

The agent has no general licence to talk. Every line it can say came from a play a manager read and approved, and a question outside that set becomes a handover rather than a guess. That is a design decision, not a setting somebody can turn off on a busy afternoon.

  • Runs approved plays only — there is no improvisation mode to switch on.
  • Off-playbook intent triggers a handover instead of an answer.
  • Every call recorded, transcribed and scored like any other.
  • Roll back to the previous version and it takes effect on the next call.
How approval works How approval works →
A floor manager with the agent roster beside him, showing who is available, who is on a call and who is logged off, and the conditions routing each call to a queue
Where it earns its keep

The calls nobody wants at 2am

Overflow, out of hours, reminder runs and the third follow-up attempt: the volume that never justified headcount and quietly cost you pipeline every time it went unanswered.

  • Answers the overflow when the queue is deep, under the same routing rules.
  • Qualifies inbound and books the meeting straight into the calendar.
  • Runs reminder and renewal calls on a schedule you set.
  • Warm handover to a rep, transcript and all, the moment intent gets real.
See the support page See the support page →

How an agent goes live

The same loop the floor runs on. An agent is a playbook that answers the phone, so it goes through the same approval as one.

  1. Learn

    Plays are drawn from your own recordings rather than from a generic script.

  2. Approve

    A manager reads the diff and signs it off. Nothing runs before that happens.

  3. Run

    The agent takes the calls you point at it, on your numbers and your routing.

  4. Measure

    Every call scored against the version that was live, with rollback in one click.

Hear it on your own objections

Bring one recorded call. We will show you the same conversation run by an agent on an approved play, and exactly where it hands over.

Questions about voice agents

See how playbooks are learned →
Will the caller know they are talking to an agent?

Yes, and it says so. Pretending otherwise is a compliance problem in several of the regions we operate in, and a trust problem in all of them.

What stops it saying something we would never say?

It has no vocabulary outside the approved plays. Anything the playbook does not cover becomes a handover, so the failure mode is a transfer rather than an invention.

When does it hand over to a human?

On complex intent, on a request for a person, on anything off-playbook, and on the escalation triggers you set. The rep receives the transcript and the context, not a cold transfer.

Does it replace reps?

It takes the volume that never justified headcount — overflow, reminders, third attempts. Reps get the calls where a person changes the outcome, with live coaching on them.

Can it run on our existing numbers?

Yes. Agents sit behind the same numbers and the same routing flows as your team, so switching one on is a routing change rather than a migration.

Never let the phone ring out

SECURITY & COMPLIANCE
GDPR COMPLIANT
Encrypted in transit and at rest, stored in region