How to Onboard a Voice AI Client: A Step-by-Step Playbook for Agencies

TL;DR: Onboarding is where voice AI retainers are won or lost. Close the deal, then move fast: collect the business details and call scripts you need, build and test the agent on Retell AI, port the client's phone number carefully, run a supervised go-live, and hand them a branded portal so day one feels like software, not a favor. I build production voice agents on Retell for US clients and run VoiceDash, a white-label client portal for voice AI agencies, so this is the exact sequence I use to take a client from signed contract to a live, visible agent without the messy first two weeks that get you cancelled.

Most agency owners obsess over selling and ignore onboarding. That is backwards. The sale gets you the first check; onboarding decides whether you get the next twelve. A client who has a rocky first two weeks, whose phone number went down for an afternoon, or who never got a login and had to email you for a call count, is already half gone. A client whose agent went live cleanly and who watches bookings pile up in a portal on their own domain renews without thinking about it.

Here is the onboarding sequence I run, in order.

Step 0: Set expectations before the contract is signed

Onboarding starts during the sale, not after it. The single biggest cause of a bad first two weeks is a client who expected the agent to be live tomorrow and perfect on the first call.

Before anyone signs, I tell them three things plainly. First, go-live takes a few days, not a few hours, because we test before we point real customers at it. Second, the agent gets better in the first week as we tune it on real calls, so early hiccups are expected and fixable. Third, porting a phone number, if they want to keep their existing line, is the slow part and depends on their current carrier, not on me.

Saying this up front turns a nervous client into a patient one. It also filters out the buyer who wanted magic, who was going to churn in month two anyway.

Step 1: Run a structured intake, not a vague "send me your info"

The fastest way to stall onboarding is to email "send me anything that might help" and wait. You will get nothing useful. Instead I run a tight intake form or a 30-minute call that collects exactly what I need to build the agent:

• The business basics: hours, location, services, what they do and do not offer, pricing they are willing to quote on a call.
• The top five reasons people call. This is the single most valuable input. Booking an appointment, checking hours, asking about pricing, rescheduling, and one or two niche questions cover the vast majority of a local business's inbound calls.
• What a good outcome looks like. For a dental office it is a booked appointment written into the calendar. For an HVAC company it might be a qualified lead with the address and the problem captured. Define the win before you build.
• The handoff rules. When should the agent transfer to a human, take a message, or send a text? Get the escalation path in writing.
• Their calendar and CRM. Which system, and who can grant access. If they run GoHighLevel or Google Calendar, I need booking access on day one, not day five.

I pull this into one document per client so the build is deterministic. The niche discipline I described in how to start a voice AI agency pays off here: if you only serve dental offices, your intake form barely changes between clients and onboarding gets faster every time.

Step 2: Build and test the agent before the client hears it

With the intake in hand, I build the agent on Retell AI, wired into the same stack I use for every client: Twilio for the number, GoHighLevel for the CRM and calendar, and n8n to automate the handoff from a completed call into a booking or a follow-up text. That stack, and why I standardized on it, is covered in Retell AI vs VAPI vs Bland.

The rule I never break: the client does not hear the agent until I have. I call it myself, a dozen times, from the caller's point of view. I try to book. I try to confuse it. I ask the awkward pricing question. I hang up mid-sentence and call back. Every one of those calls surfaces a prompt gap that would have been the client's first impression if I had skipped it.

Only when the agent handles the top five call reasons cleanly do I move to go-live. Testing on your own dime for a day is far cheaper than testing on the client's real customers.

Step 3: Handle the phone number carefully

Phone numbers are where onboarding goes wrong for people who rush. There are two paths, and I pick based on the client's risk tolerance.

The safe path is a new number. I provision a fresh Twilio number, point it at the agent, and have the client forward their existing line to it or use it as the published number. Nothing breaks, and if something is off, their old line still rings. This is how I onboard almost everyone in week one.

The full path is porting their existing number into the system so the agent answers their real, known line directly. Porting is worth it for the long term, but it is carrier-dependent and can take days, and there is a cutover moment where the number moves. I never port on day one. I go live on a forwarded or new number first, prove the agent works, and port later once the client trusts it. Doing it in that order means a porting delay never blocks go-live.

Step 4: Supervised go-live, not a silent handoff

Go-live is not "it is on now, good luck." For the first day or two the agent is live, I watch it. I check the first real calls, read the transcripts, and tune the prompt on anything the agent fumbled. Real callers do things your test calls never will, and the first 48 hours is where you catch them.

This is also the moment to catch integration failures: a booking that did not write to the calendar, a transfer that did not connect, a follow-up text that did not fire. Watching the first live calls turns those from a week-two complaint into a same-day fix the client never even sees.

Step 5: Hand over a branded portal, not a spreadsheet

This is the step that separates a professional onboarding from an amateur one, and it is the one most agencies get wrong.

You just delivered something impressive. Then, if you hand the client a raw Retell dashboard, a shared Google Sheet, or a promise to "send a report every Friday," you undo it. A developer dashboard makes your serious retainer look like a side project and quietly teaches the client they could have bought Retell themselves, which is exactly the objection I break down in how to sell AI receptionists.

Instead, day one ends with the client logging into a portal on their own domain, with their logo, showing their calls, recordings, transcripts, and outcomes updating in real time. That is what onboarding should culminate in: not a file, a product. It is also the difference that lets you charge and keep a software-sized retainer, which I covered in voice AI agency pricing.

With VoiceDash this step takes minutes instead of being a project of its own. I connect Retell once, create the client's workspace, drop in their branding and domain, and share the login. The portal is live in under 10 minutes, no code, and every client is isolated in their own workspace seeing only their own data. The client's first experience of the agency relationship is a polished dashboard with their name on it, which is precisely what a white-label Retell dashboard is for.

Step 6: The first-week check-in that locks in the renewal

I book a 15-minute check-in for the end of week one before go-live even happens. On that call I do not ask "how is it going." I show them the numbers already sitting in their portal: calls handled, appointments booked, minutes used, and I play them a recording of the agent nailing a real booking.

Hearing their own agent close a real customer, on a call they can see logged in their own branded dashboard, is what converts a nervous new client into a long-term one. The value is not abstract anymore. It is on the screen with their logo on it.

The onboarding checklist, condensed

• Before signing: set expectations on timeline, tuning, and number porting.
• Intake: hours, services, top five call reasons, win definition, escalation rules, calendar and CRM access.
• Build: agent on Retell plus Twilio, GoHighLevel, and n8n, tested by you first.
• Number: go live on a new or forwarded number; port the real line later.
• Go-live: supervised for 48 hours, tune on real calls, verify integrations fire.
• Handover: a branded client portal, not a spreadsheet.
• Week one: a check-in that shows the numbers and plays a winning call.

Get this sequence right and onboarding stops being the scary gap between the sale and the retainer. It becomes the part of your service that clients remember, and the reason they stay.

Want the portal handover to take minutes instead of being its own build? Start free on VoiceDash or book a demo and I will walk you through onboarding a client end to end on a live Retell setup.