Hour 00 — kickoff & scoping.
Day one starts with a 60-minute working call. Three people on your side — ops lead, IT, one decision-maker. Three people on ours — solutions engineer, voice designer, integration engineer. Outcome of this hour: a written scoping document with the success criterion for the next 23 hours.
The single sentence that defines "success" for your first live call. We refuse to start without it. "Answers an inbound call" is not enough. "Books a service appointment in < 90s and writes to ServiceTitan" is enough.
- Channel — inbound or outbound (we recommend inbound for hour-one).
- Number — your existing number ported, or a new DID provisioned.
- Stack — one integration that the agent must read from or write to.
- Voice — off-the-shelf brand-adjacent or clone of a reference clip (8-second sample minimum).
Hour 02 — numbers + routing.
Telephony is the silent killer of week-long pilots. We provision a DID and front-route it through your existing IVR (or, if you don't have one, point your number forwards to ours during business hours). Two-way audio confirmed by hour-04.
| Provider | Twilio (default) · Vonage · Plivo · SignalWire · Bandwidth |
| Latency target | < 200ms first word |
| Caller-ID | Matched to receiving region · STIR/SHAKEN where applicable |
| Recording | Optional · stored in your tenant · encrypted at rest |
Hour 06 — voice + intent.
The voice designer tunes prosody and tone against your reference clip. We rebuild the intent playbook against the success criterion from hour 00. By hour 12 the agent runs an end-to-end mock call in staging that the ops lead approves verbatim.
Hour 14 — one integration.
One. Not seven. The first 24 hours ship exactly one integration — the one named in the success criterion. Everything else queues for day-2 onwards.
// example: ServiceTitan booking on hour-14 POST /v4/automations { "trigger": "call.intent.classified", "if": "intent === 'book_service'", "action": "servicetitan.appointment.create", "retry": 5, "on_fail": "queue_human" }
Hour 20 — dress rehearsal.
Four hours before go-live we run a dress rehearsal: ten real calls placed by named volunteers from your team, recorded, analysed, scored. Anything below 8/10 holds the launch.
Hour-20 readiness gate
- 10/10 calls handled end to end
- Zero PII leaks in transcripts
- Integration writes succeed on retry-1 or better
- Failover to human queue confirmed
- Ops lead listens to every recording
Hour 23 — go-live + on-call.
The number flips. The agent answers. The team sits in a war-room channel for the first four hours — anything anomalous halts the rollout and routes to humans. By hour-26, in our experience, the war room dissolves and the team moves to their normal calendar.
Day-2 onwards we add the second integration, layer the outbound campaign, and start the multi-language expansion. But day-1 is one channel, one integration, one criterion — measured.