What this covers
A few years ago, "AI on the phone" meant a robotic voice tree that frustrated callers. That changed in 2024. Modern voice agents understand context, handle interruptions, ask clarifying questions, and sound like a person you'd actually want to talk to. They're not science fiction — they're working today.
For most growing businesses, three use cases cover 90% of the value:
- Inbound qualifier — picks up after hours or in overflow, qualifies the caller, books a follow-up if they're a fit, takes a message if they're not.
- Outbound follow-up — calls leads from a list with a specific reason (appointment confirmation, post-purchase check-in, lapsed-customer outreach).
- Triage and routing — receives the call, understands the request, and routes to the right human or handles it end-to-end.
How we build them
I'm vendor-neutral. The platform that fits depends on your call volume, voice quality requirements, and what tools the agent needs to call.
- Retell, Vapi, or Bland for the voice infrastructure (real-time transcription, low-latency turn-taking, function calling).
- Twilio or RingCentral for the phone number and PSTN routing.
- Claude, GPT, or Gemini as the underlying language model, picked for the use case.
- Your existing systems — Salesforce, Cal.com or Calendly, your CRM, your scheduling tool — wired into the agent's tool calls so it can actually do things.
How we typically engage
Voice agent projects are usually a four-week build:
- Week one: scope, conversation design, integration list.
- Week two: build and internal testing.
- Week three: closed beta with a small fraction of real traffic.
- Week four: ramp, monitoring, handoff.
Pricing has two parts: the build (fixed-fee), and the ongoing per-minute cost (usage-based, billed at cost from the underlying providers — typically eight to twenty cents per minute).
What "done" usually looks like
- An agent that handles the majority of calls in scope without escalation.
- Transparent transcripts and recordings of every call, with sentiment and outcome tagging.
- A clear escalation path when the agent should hand off to a human.
- Cost dashboards so you know what each call costs.