How to set up voice support with AskVault
Early access. Early access Voice and phone support is built but not switched on for every account yet. Email [email protected] to join early access. This page describes how it works once it is on for your workspace.
What voice support gets you
Voice support is the right channel for three patterns: appointment booking and changes where customers prefer to call rather than type, basic account questions where keyboards are inconvenient (driving customers, older demographics), and overflow handling when your phone queue is long and you want a bot to triage before a human picks up.
Voice is on the Business plan and above. Voice minutes are sold separately as prepaid packs; no plan includes them. Business+
Voice-minute packs are offered only on plans that include the voice channel (Business and Enterprise). Starter and Growth won't see packs to buy, because those plans can't connect a voice number. Minutes you already bought stay on your balance.
Once a change now rolling out is enabled, the Voice Agent page shows your voice-minute balance next to your number. If you have no minutes, the page says Not answering calls instead of Active, because callers hear "Voice support is not active" and are hung up on until you add minutes. The page also warns you, and asks you to choose a supported voice, if your saved voice can't speak on the current voice engine. Call History shows what the caller experienced (for example Declined: no voice minutes or After hours; when calls are paid for in message credits it names the exact reason instead: Declined: out of message credits, Declined: monthly call minutes used up, or Declined: your plan does not include phone calls), shows the number you dialled for outbound calls, and loads older calls with Load more.
How a call actually flows
When a customer calls your AskVault voice number:
- Twilio answers the call and plays your custom greeting (configurable text).
- The customer speaks their question. Twilio transcribes the audio in real time.
- The transcript goes to AskVault. The AI agent retrieves relevant content from your knowledge base.
- AskVault returns the text answer. Twilio converts it to speech using a voice you pick.
- The customer hears the answer. They can ask a follow-up. Repeat.
Each turn takes a moment, because the caller's speech is transcribed, answered and spoken back. We have not published a latency figure.
Step 1: buy a voice-capable Twilio number
- Sign in to Twilio Console.
- Phone Numbers > Manage > Buy a Number.
- Pick the country, search for numbers with Voice capability enabled.
- Purchase the number. US voice numbers cost about $1 per month plus per-minute fees.
Step 2: connect to AskVault
- In Twilio Console, copy your Account SID and Auth Token.
- In AskVault, open Channels > Voice and go to the Phone Number step.
- Under Telephony Account, click Connect your own and paste the Account SID and Auth Token.
- Pick that account in the Telephony Account list. The Use Existing Number From This Account
field appears. Enter the number you bought in international format (for example
+14155551234). - Click Connect Existing Number.
A number can only be connected from the Twilio account that owns it. It also can't be connected while it is in use by another AskVault workspace. If AskVault bought a number for this workspace earlier, it shows under Existing numbers on this workspace. Click Reuse there instead.
AskVault configures Twilio's voice webhook so inbound calls route through our pipeline. Test it: call the Twilio number from your personal phone. The bot picks up and plays your default greeting.
Step 3: customize the greeting
Under Channels > Voice > Greeting, edit the welcome message. Examples:
Hi, you've reached Acme support. I'm an AI assistant. What can I help you with today?
Or for a more guided experience:
Hello and thanks for calling Acme. You can ask me about your account, your order, or our products. Go ahead, I'm listening.
Keep greetings under 8 seconds. Anything longer and callers start to disconnect.
Step 4: configure voice settings
Three settings matter most.
Voice. Pick the text-to-speech voice the bot uses. Press play next to any voice to hear a sample before you save it. If one voice's sample fails while others play, pick a different voice: calls use the same speech engine, so callers would not hear that voice either. If every sample fails, the problem is on our side, not your voice choice: email [email protected]. For US English, the default voice sounds natural and is well-tested.
Speech recognition language. Defaults to English. Set per-number for multilingual deployments.
Interrupt mode. Whether the caller can interrupt the bot's speech to ask the next question. We recommend On: more natural conversation, no waiting for the bot to finish.
Optional: IVR menu
For traffic high enough to need routing, configure an IVR (Interactive Voice Response) menu under Voice Settings > Menu:
Hi, you've reached Acme. Press 1 for account help, press 2 for order status, press 3 for anything else.
Press 1 or 2 routes the call to the matching menu branch. Press 3 sends the call to the general AI agent. All branches answer from the same workspace knowledge base: there is no per-branch knowledge subset.
For most use cases an IVR adds friction and the default "just ask" flow is better. Use IVR only if your inbound mix is bimodal (e.g., 60% order-status and 30% other) and you want explicit routing.
Escalating to a human
The escalate_to_human skill works on voice. When triggered, the bot says "Let me transfer you to someone who can help" and hands off the call to a human agent at a configured number. The customer hears hold music for a few seconds, then a human picks up.
Configure the handoff number under Voice Settings > Escalation > Forward to. Multiple numbers can rotate based on time of day, day of week, or skill type.
Limits and costs
Twilio voice costs about $0.013 per minute inbound and $0.014 per minute outbound for US numbers. Plus per-minute speech-recognition and text-to-speech charges (about $0.05 per minute combined). Total: roughly $0.07 per minute of conversation.
AskVault counts each AI-generated voice reply as 1 message against your monthly quota. A 5-minute conversation with 10 AI turns uses 10 messages.
For high-volume voice deployments (over 5,000 minutes per month), reach out to [email protected] for a volume agreement that flattens per-minute costs.
Compliance
Voice has more regulations than text. Three to know:
- Call recording. AskVault records every call by default. If you record customer conversations, you must inform them at the start (most jurisdictions). The default greeting includes "This call may be recorded for quality and training" automatically.
- PII handling. Voice transcripts often contain spoken account numbers, dates of birth, addresses. Transcripts are stored like chat messages, in the same encrypted database, and the transcript and keypad entries on a call log are cleared after 90 days.
- TCPA (US). Outbound automated calls and texts are heavily regulated. AskVault doesn't ship an outbound voice campaign feature for this reason.
Common pitfalls
Caller speaks but bot doesn't reply. Background noise is high enough that speech recognition fails. The bot asks them to repeat. If it keeps happening, the customer's signal is bad. Bot escalates to human after 2 failed attempts.
Bot misunderstands accents or industry jargon. Speech-to-text accuracy varies by accent and by jargon. AskVault has no custom vocabulary setting today. Add your product names and common terms to your knowledge base so the agent can match what it hears to your content.
A call ends after 30 minutes. Each voice call has a hard time limit of 30 minutes. If a conversation needs longer, the caller can call back.
Customer says "agent" but bot keeps talking. Make sure escalate_to_human is enabled and the trigger phrase "agent", "human", "representative" is in the skill's trigger list.
FAQ
Can the bot make outbound calls?
Not in the standard AskVault flow. Outbound automated calling is heavily regulated (TCPA, GDPR e-Privacy). For specific outbound use cases (appointment confirmations, callback campaigns), reach out to [email protected] for a custom integration.
Can I use a non-Twilio voice provider?
Twilio is the default. For Plivo, Vonage, or self-hosted Asterisk, the REST API channel gives you a programmatic interface.
Does the bot work with toll-free numbers?
Yes. US toll-free numbers require a separate Twilio Toll-Free Verification process. Takes 2 to 4 weeks.
How accurate is the speech recognition?
Transcription is most accurate on clear audio without heavy background noise; accents, jargon and poor audio reduce it, and we have not published a benchmark.
Can I get a transcript of every call?
Yes. Every call appears in Inbox > Chat History filtered by Voice channel, with the full transcript and audio recording attached. Export as CSV from Analytics > Export.