A Voice AI callback should finish once the agent has confirmed the caller's name and number. Open a pre-filled web form for details that are harder to capture reliably by voice, then let the call end while the form stays available.
This keeps the voice agent focused on reception. The caller can correct their details, type an email address, add context, and make a separate optional marketing choice without extending the call.
Give voice and the form different jobs
The Miles walkthrough starts with a narrow callback task. The caller asks Patrick to call back, Miles learns that the incoming number is withheld, collects the callback number in segments, and confirms that the number is complete.
Only then does Miles open the Contact page. The page already contains the caller's name and phone number. The caller can add a message, email address, and business name without keeping the voice session open.
The division is deliberate. Voice handles the immediate conversational task. The form handles information that benefits from a keyboard, visible fields, and time to review.
Move fragile details off the call
Email addresses are difficult voice inputs because a single incorrect character can make the address unusable. Dots, symbols, unusual spellings, and transcription errors create more failure points than an ordinary callback number.
Patrick's operating conclusion from the walkthrough is simple: even a strong voice and post-call transcription process is not reliable enough when exact email capture matters. Let the caller type the address and inspect it before submitting.
The same form lets the caller correct a misheard name or number. That reduces the chance of a wasted callback without forcing the agent through another long confirmation sequence.
Keep callback intent separate from marketing consent
The caller asked for a callback. Marketing consent is a second decision, so the workflow should present it separately and keep it optional.
In the example, the form uses a distinct tick box while the caller is already engaged with the business. That creates a visible action and a written record without making consent a condition of receiving the callback.
Patrick also flags a South African constraint in the walkthrough: treat the chance to ask a non-customer for direct-marketing consent carefully because a refusal may prevent another request. His commercial point is to present the optional choice when the caller is already completing the form, not to interrupt the callback with a separate pitch.
This is an operational design principle, not legal advice. Consent requirements vary by jurisdiction and use case. Patrick's recommendation is to have the exact wording, timing, and process reviewed before relying on it, especially because spoken consent may be harder to evidence later.
Let the call end while the task continues
Once Miles opens the form, the caller can complete it in their own time. Privacy wording, optional fields, corrections, and extra context stay on the page instead of becoming more dialogue for the voice agent. This avoids making the agent read a long privacy explanation while the caller is waiting to end a simple callback request.
That keeps call duration and voice cost under control. It also preserves a submitted database record that the business can use when the transcript is unclear. If the eventual callback reaches voicemail, the typed email address can provide a fallback contact route.
Keep the voice agent as receptionist
The practical boundary is clear: use the voice agent to understand the request, confirm the callback details, and guide the caller to the next step. Do not turn the agent into a long data-entry or compliance script.
The best handoff is one the caller can see and correct. Confirm the minimum viable details on the call, open the right form with those details already filled, and leave the rest to a channel designed for exact input.