Most AI customer service posts make you pick: voice agent or chatbot. That framing has cost a lot of businesses real money.

In practice, voice and chat aren't competitors. They're different moments in the same funnel. A prospect on their lunch break, sitting at a desk, doesn't want to call you — they want to type a question and see an answer. The same prospect at 8pm on the couch with a question that needs context doesn't want to type a paragraph — they want to talk. The same prospect at 11pm thinking "I'll call them tomorrow" almost never calls back; they go to a competitor.

A properly built 24/7 customer service funnel uses both. Here's how I set this up for clients, where the handoffs are, and what most builds get wrong.

The two-tool funnel

Think of inbound prospect contact as a tree. The first branch is the channel they reach for, which is driven entirely by context — desk vs. couch, hands free vs. hands busy, complex question vs. simple one.

Chat is the right tool when the prospect is in a browsing context. They're on your website, on a product page, comparing options. They want a fast answer to a specific question. They don't want to commit to a call. Time-of-day skews business hours and evenings.
Voice is the right tool when the prospect has decided to take action — booking, complaining, closing. They've moved past "is this for me?" to "let's do this." Time-of-day skews late afternoons, evenings, and weekends — the hours a human team usually can't cover.

If you only have one tool, you lose the people in the other context. A standalone chatbot misses the prospects who'd commit if they could just talk to someone. A standalone voice agent misses the prospects who would never pick up the phone but would have closed via 4 chat exchanges.

The handoff points

The expensive failures happen at the handoffs. Two patterns matter most:

Chat → Voice escalation. A chat conversation reaches a question the bot can't answer well, or the prospect's intent shifts to "I want to book" or "I want to talk to a human." The right behavior is the chatbot offering an immediate phone call — connecting them to your AI voice agent in the same interface, with the chat transcript passed in as context so the voice agent already knows what they were asking about. The wrong behavior — what most setups do — is the chatbot saying "please call us during business hours."

Voice → Chat fallback. A voice call reaches an answer that's complex (think: pricing breakdown, document delivery, scheduling with options). The voice agent should offer to text the rest of the conversation as a follow-up, with the relevant link or document attached. The wrong behavior is reading a 12-item list aloud.

When I build a voice + chat system for a client, the integration layer between the two is what actually determines whether the funnel works. It's usually a shared CRM record (so both agents are reading and writing to the same conversation), plus webhooks that fire when one agent decides to hand off.

What this looks like in your stack

A typical RAS AI deployment for a service business:

Chatbot on the website built with our standard stack (OpenAI for the LLM, retrieval from your business knowledge base, with intent detection on every message). Embedded on every page. Average response time under 2 seconds.
Voice agent on your phone line (VAPI for the conversational layer, Twilio for the telephony). Same underlying knowledge base, so both agents give consistent answers.
Shared CRM (RAS Flow if you're using ours, or HubSpot / Salesforce / Zoho if not). Both agents write contact records and conversation history into the same database.
Cross-channel handoff webhooks. When the chatbot detects the user wants to call, it triggers a Twilio click-to-call flow with the conversation history pre-loaded. When the voice agent needs to send a document, it triggers an SMS or email from the same contact record.

The result is one funnel with two front doors. A prospect can drop in via chat at lunch, get their three questions answered, leave without booking, come back via phone that evening, and your voice agent picks up the conversation already knowing who they are and what they asked earlier.

What most builds get wrong

Three patterns I see again and again when I audit existing setups for prospects:

1. Two robots that don't know about each other. The chatbot is on a separate platform from the voice line. There's no shared state. The same prospect interacts with both in the same week and gets contradictory information. By the third interaction they're done.

2. The "we'll call you back" trap. The chatbot can't answer a question, so it asks for a phone number and promises a callback. The callback never happens, or happens 18 hours later when the prospect has moved on. Always-on voice agents exist specifically so this doesn't have to happen.

3. Over-scripting the voice agent. Builders try to make voice agents do everything chatbots do — long lists, document delivery, complex visual flows. Voice is bad at this. The right move is to keep voice agents focused on the moments where voice is genuinely better (qualifying, booking, transferring) and let the chatbot or SMS pick up everything text-shaped.

The result you can expect

For Canadian SMB clients running this integrated setup, the lift versus single-channel is consistent:

Inbound contact → qualified lead conversion goes up 30–60%, because the people who would have left at 11pm get captured.
Qualified lead → booked appointment goes up by 20–40%, because handoffs don't drop leads.
Coverage cost stays flat: both agents run on usage-based pricing, so a business with 100 inbound contacts a month and a business with 1,000 inbound contacts a month pay close to the same fixed cost.

What we charge for the combined setup is roughly $1,950 setup for the integrated voice + chat build, then ~$299/month for both agents bundled, plus usage (~$0.05–$0.12/min voice, near-zero for chat). Standard go-live is 72 hours.

When this is overkill

A few cases where you don't need both:

Pure-walk-in business with negligible online traffic: chatbot is wasted effort. Voice only.
B2B service with long sales cycles and high-touch deals: chatbot is fine, voice agent is unnecessary — your team is already on every call.
Sub-30 inbound contacts a month total: any AI setup is overkill. Email and a calendar link work.

If your business sees real volume across both channels and the goal is to stop losing prospects to "wrong time, wrong channel," it's worth building both. Book a call and I'll walk through your specific funnel, or read more about how the AI receptionist and chatbot work together in our stack.