Inbound support on a live phone line

A support call is a conversation, not a form. The caller talks over the greeting, changes their mind mid-sentence, remembers a second question before you finish the first. Agent Vani is built for exactly that. It runs a two-way, realtime phone conversation: the caller can interrupt, correct, and steer, and the agent keeps up in the same breath instead of waiting for a beep and reading back a scripted turn. That is the difference between a phone line people trust and one they hang up on.

Most "AI on the phone" you have heard is turn-based. You speak, silence, a pause, then a paragraph plays. Callers hate it because it does not sound like talking to a person — it sounds like leaving a voicemail that answers back slowly. Agent Vani is designed to feel live. Callers can cut in, steer the conversation, and stay in flow without waiting for a robotic turn to finish. For inbound support, where frustration is already on the line, that responsiveness is the whole product.

What inbound support actually sounds like

Picture a caller who bought something last week and wants to know where it is. They dial your published number. The agent answers in the caller's language, confirms what they need, and handles the routine part of the exchange: order status, store hours, return windows, warranty basics, "do you have this in stock," "how do I reset it," "what are your timings today." When the caller interrupts with "no, the other order," the agent doesn't restart from the top — it adjusts, the way a good desk agent would.

The point of inbound support coverage is not to replace your people. It is to take the volume of predictable, repetitive questions off the queue so your humans spend their time on the calls that genuinely need a human. The agent finishes what it can finish cleanly, and it hands off the rest without making the caller repeat themselves.

What you control

Agent Vani is your agent, configured by your organisation. Nothing about it is hardcoded by us:

• Persona and voice. You write how it should sound — warm, brisk, formal — and choose a voice that fits your brand. You decide what it may say, what it may promise, and what it must never claim.
• Knowledge. You give it the facts it should use: policies, FAQs, product details, return windows, service areas. Answers stay grounded in what you approved, not invented on the spot.
• Hours. You set when it is on. It can cover the whole day, only after hours, only overflow when your team is busy, or any schedule you define.
• Numbers. You bring the numbers your customers already dial. Those lines route to your workspace's Agent Vani so the right org answers.
• Handoff rules. You decide the conditions under which it stops and gets a person involved.

You publish when you are ready, and not before. Before anything reaches a real caller, you dial your own attached number and test the conversation yourself.

Where it fits in a support operation

Inbound support coverage tends to land in a few shapes, and you can mix them:

1. Front-line triage. The agent answers first, resolves the easy questions, and routes the rest. Your team stops fielding "what are your hours" for the hundredth time.
2. Overflow. Your people answer first; the agent catches calls when everyone is busy so nobody hits a dead ring or an endless hold.
3. After-hours and weekends. The line stays open when the desk is closed. The agent answers what you allow and captures intent on everything else. See after-hours coverage for that pattern in depth.

In every shape, the calls are visible in your workspace. Operators see who called, how long the conversation ran, and how it ended — resolved, handed off, or captured for follow-up. That visibility is what makes the agent something you can actually manage rather than a black box on your main line.

A configuration checklist for support

Before you point real support traffic at the agent, walk this list:

• Write the greeting for interruption. Assume the caller will talk over it. Keep it short, name the business, and get to "how can I help" fast.
• List your top twenty questions. Pull them from your existing tickets or call notes. These are what the agent must answer flawlessly. Everything else can escalate.
• Draw the escalation line. Decide precisely what the agent should not attempt: refunds above a threshold, account changes, complaints, anything involving a promise you cannot automate. When the caller hits that line, hand off.
• Ground every claim. For each thing the agent can say — return window, delivery estimate, price — make sure the fact lives in the knowledge you gave it. Do not let it improvise policy.
• Set the "I don't know" behaviour. The agent should admit uncertainty and route the caller, not guess. A confident wrong answer costs you more than an honest handoff.
• Test the messy calls, not the clean ones. Call in and interrupt. Change your question halfway. Mumble. Switch languages. The clean script always works; the real calls are where quality shows.

Failure modes to design around

Honest coverage means planning for the calls that go sideways.

• The angry caller. Agent Vani is good at information, not at absorbing frustration or negotiating exceptions. Route anger to a person quickly. A caller who is already upset does not want to explain themselves to software.
• The out-of-scope request. Someone asks about something you never configured. The right outcome is a clean handoff or a captured callback request — never a made-up answer.
• The ambiguous account action. Anything that changes money, access, or personal records should escalate by default unless you have deliberately and carefully enabled it. Fail closed.
• The caller who just wants a human. Some people will not talk to an agent, and that is fine. Make "let me get you to a person" an easy, always-available path. Trapping callers is the fastest way to lose them.

Designing for these up front is what separates a support line that helps from one that annoys. The agent's job is to be genuinely useful on the calls it is good at and to get out of the way fast on the ones it is not.

Handoff is a feature, not an admission

The measure of a good support agent — human or otherwise — is not that it never needs help. It is that it hands off cleanly when it should. Agent Vani is built to transfer to a person when the call needs one, and to do it without making the caller start over. When you design escalation well, callers barely notice the seam; they just get helped. The full pattern lives in human handoff and escalation.

Getting the handoff right also protects your team. Instead of every call landing on a person, only the ones that need judgment, empathy, or authority do. Your agents get calls that use their skills, and the routine volume stops burning them out.

What this costs you

The control plane — where you configure persona, knowledge, hours, and numbers, and where you review calls — is free. You do not pay for seats. Agent Vani itself is billed by the minute of conversation, so what you pay maps to what actually happened on the line, and finance can forecast it from usage in the console. You can start on a trial with no credit card, run real test calls, and see the behaviour for yourself before you commit. The details are on the pricing page and in how billing works.

Because billing is per minute, the economics reward a well-scoped agent: tight answers, clean handoffs, no rambling. A concise, helpful call is cheaper and better at the same time, which is a rare alignment.

Going live without surprises

The path from configured to live is deliberately boring, which is what you want on your main support line:

1. Write the persona and the knowledge.
2. Attach your number.
3. Set hours and escalation rules.
4. Dial in and test the hard calls yourself.
5. Publish, then watch the first real calls in your workspace and tune.

You are not flipping a switch and hoping. You are reviewing calls as they complete and adjusting the script, the knowledge, and the escalation line until the line sounds the way you want. For a step-by-step version, see how to write a phone agent script and the product overview.

A walkthrough of one good call

It helps to picture the whole thing end to end. A caller dials your published support number at 11am. The agent answers in their language and, before it finishes the greeting, the caller cuts in: "yeah, hi, where's my order." The agent does not restart — it acknowledges, asks the one identifying detail it needs, and confirms the status from the facts you gave it. The caller then remembers a second thing: "and can I change the delivery address." That is outside what you allowed the agent to do, so it does not fumble a guess. It says it will get a person to help with the address change, captures the request with the detail your team needs, and either connects a human if one is free or logs a callback if not. The whole exchange took ninety seconds, resolved one thing cleanly, and handed off the other without the caller repeating themselves. Afterward it appears in your workspace: caller, duration, outcome, resolved-plus-handoff. That is what "the agent finishes what it can and hands off the rest" looks like on a real line.

What to measure once you are live

Coverage you cannot see is coverage you cannot trust, so watch a small, honest set of signals rather than vanity numbers:

• Containment. What share of calls the agent resolved end to end without a human. Rising containment on the questions you configured means your knowledge and script are landing.
• Handoff rate and reason. How often it escalates, and why. A spike in one reason usually means a gap in knowledge you can close, or a scope line you drew in the wrong place.
• Repeat callers. People calling back soon after a "resolved" call are a sign the resolution was not real. This is the metric that catches confident-but-wrong answers.
• Interruption recovery. Whether the agent keeps up when callers talk over it. This is the difference between a call that feels live and one that feels like a slow menu, and it is worth listening for directly.

None of these require a dashboard obsession. Skim a handful of real calls each week, watch these signals, and tune the knowledge, the script, and the escalation line from what you actually hear. Because every call is visible, tuning is grounded in reality rather than assumption. For a deeper treatment, see measuring phone agent quality.

FAQ

Does Agent Vani replace my support team?

No, and it should not. It takes the predictable, repetitive volume — status, hours, policies, common how-tos — off your queue so your people handle the calls that need a human. The goal is fewer wasted human minutes on questions software can answer, and cleaner handoffs on the ones it cannot. See when you still need humans.

Can callers interrupt it, or do they have to wait for it to finish?

They can interrupt. The conversation is two-way and realtime — the caller can talk over the agent, cut in, and change direction, and the agent yields and adapts instead of finishing a scripted turn. That is the core of why it feels like a real call rather than a slow menu.

Can it transfer to a person?

Yes. You define the conditions, and when a call meets them, the agent hands off to a human without making the caller repeat everything. Read human handoff and escalation for how to design it.

How does it answer questions specific to my business?

You give it your knowledge — policies, FAQs, product details, hours — and it answers from that. It does not invent policy. When it does not know, it says so and routes the caller rather than guessing. Details on the product page.

Is this inbound only right now?

Yes. At launch, Agent Vani answers inbound calls to the numbers you attach. Outbound calling using the same agent is a later release. If your use case depends on the agent calling people, plan around inbound for now. See pricing for what is available today.

What does it cost, and do I need a card to try it?

The control plane is free and there is no per-seat charge. Agent Vani is billed per minute of conversation. You can start a trial with no credit card, run real test calls, and only subscribe when you are ready for production volume. See pricing and how billing works.