Last Tuesday we watched a Melbourne digital agency spin up their first white-label AI receptionist for a physio clinic in Dandenong. Three people on the call: their ops lead, our support engineer, and the physio owner dialling in from between appointments. By end of day the agent was live, taking bookings, and the agency had invoiced their first month at a 68% margin.

This post is what actually happened in those first 24 hours. Not the polished case study version - the messy bits, the blockers, the five-minute fixes that unlocked ten hours of value.

Hour 0: kickoff and access

We started the call at 10am. First step was white-label DNS and branding. The agency had purchased the creator tier two weeks earlier, tested inbound on their own number, then upgraded to white-label the day before go-live. Their rebrand domain was already pointed, so we pushed their logo, colour hex, and custom email domain into the platform in about eight minutes.

The ops lead had done her homework. She'd already scraped the clinic's existing voicemail greeting, pulled FAQ copy from the website, and noted the booking system was Cliniko. We cloned our Cliniko template, swapped in her knowledge base, and ran a test call. The agent answered, knew the clinic hours, and offered to book. Eleven minutes elapsed.

What surprised us: she'd written the knowledge base in dot-points, not full sentences. We told her to leave it - the LLM handled it better than the over-formal paragraphs most people submit. Dot-points train faster.

Hour 4: live handoff and the SMS loop

By 2pm the number port from Telstra was done and the agent was fielding real calls. The physio owner joined a second call to listen to three back-to-back interactions. One booking, one reschedule, one "I'll call back" that triggered an outbound follow-up SMS two hours later.

The SMS response came back at 4.07pm: "Can you do Thursday 3pm instead?" The agent read it, checked Cliniko via API, confirmed the slot, wrote the appointment, and sent a calendar invite. Total human involvement: zero.

The clinic owner said one thing on that call that stuck with us: "I thought I'd need to babysit this for a week." She hung up after twenty minutes and never logged back in that day.

What blocked us for 90 minutes

Between hour 1 and hour 2.5 we hit a wall. The Cliniko API key the clinic provided had read-only permissions. The agent could see availability but couldn't write appointments. The physio's practice manager was in back-to-back sessions and unreachable.

The agency ops lead made a call: switch the agent to "confirmation mode" - take details, send the clinic an email summary, let them manually book in Cliniko for the afternoon. We pushed that change in four minutes. Three calls came in during the wait window. All three got handled, all three got emailed to the front desk, and all three were booked by 5pm when the practice manager finally generated the correct API token.

Lesson: plan for credential lag. A half-smart agent beats a silent phone every time.

Hour 24: margin check and next three clients

Wednesday morning the agency sent us their math. They'd invoiced the clinic $990/month on a 12-month contract. VoxReach white-label cost them $320/month base. The clinic took 47 minutes of inbound talk time on day one, so another $24 in usage. Total cost to the agency: $344. Margin: $646, or 65% before their own support load.

The ops lead was already on calendar invites for three more kickoffs that week: a conveyancing firm in Geelong, a vet in Frankston, a pest control outfit in Pakenham. Same playbook, different knowledge bases. She told us she'd cloned the physio setup as a template and swapped the Cliniko integration tag for ezyVet and Smokeball.

We asked what she'd do differently next time. Two things:

• Get API credentials 48 hours before go-live, not day-of.
• Record a sample call during onboarding and send it to the client as proof before handoff.

Both good. We've added a pre-flight checklist to the white-label onboarding doc that now lives in the partner portal.

What this tells you

Agencies win with AI voice when they treat it like hosting: fast setup, recurring revenue, low touch once it's running. The Melbourne agency didn't sell AI. They sold "your phone answered 24/7, bookings in your system, no extra staff". The AI was infrastructure.

If you're an agency testing this model, focus on three things: pick clients with online booking systems you can integrate, get correct API access before kickoff, and clone your first successful setup as a template. The second client onboarding takes 40% of the time of the first.

Try the same playbook

VoxReach white-label starts at $2,000 setup and 20% ongoing trail. You keep the client relationship, set your own pricing, and we handle platform uptime and voice infrastructure. If you want to test inbound on your own number first, get started at app.voxreach.com.au/signup - a free 90-second demo call, live in ten minutes.