Going-live checklist for Agent Vani

The gap between "we set up a phone agent" and "our phone agent is actually good" is entirely in the going-live process. Rush it and your real customers become your test group, on your real number, at scale. Do it deliberately and the switch to live is invisible to callers and boring for you — which is exactly what you want a launch to be. This is the practical, ordered checklist for taking Agent Vani live, from a blank workspace to answering real callers, without the mistakes that turn a launch into a fire drill.

The sequence matters. Each step depends on the one before it, and the single most common failure is skipping straight to "turn it on" before the earlier steps are solid. Work top to bottom.

Before you start: know what "good" looks like

Have a picture of success before you configure anything. What are the calls your line gets? What should the agent handle, and what must always reach a person? Pull a week of real calls if you can — that list of what callers actually ask is your specification. Everything below is easier when you know what you're building toward. If you're still choosing between options, run the buying guide checklist first.

Step 1: Persona and voice

Start with who's answering, because it's the through-line for everything the agent says.

• Tone set to match your brand and callers — warm, crisp, formal, casual.
• Language default chosen. Agent Vani handles Hindi, English, and Hinglish and follows the caller's lead; pick your default. See multilingual phone support.
• Voice selected to fit your brand.
• Identity and honesty decided — a name if you like, but no pretending to be a specific human or hiding that it's an AI. See privacy and caller trust.

Deep guidance in how to write a phone agent script. Don't over-script literal lines — set the persona and let it converse.

Step 2: Knowledge — the facts it answers from

This is the highest-leverage step and the one people under-invest in. The agent answers from what you give it and won't invent policy, so thin knowledge means thin calls.

• Hours loaded — including lunch breaks, holidays, and "are you open today" edge cases.
• Location and directions — landmarks, parking, which entrance.
• Policies — returns, cancellations, deposits, what's included and what's extra.
• Products and services in plain terms, with the details callers actually ask.
• Common questions from your real-call review, answered as facts (not scripted sentences).
• Do-not-promise list written — unapproved prices, guarantees, regulated advice, actions it can't take.

If you do one thing thoroughly, do this. A rich knowledge base is the difference between an agent that helps and one that says "let me have someone call you back" to everything.

Step 3: Hours and escalation

Decide when the agent is on and when it hands off — the two settings that most shape the caller experience.

• Business hours set for staffed times.
• After-hours behavior defined — the agent still answers, handles what it can, and captures details, rather than going dead. See after-hours coverage.
• Holidays set explicitly so it doesn't claim you're open when you're closed.
• Escalation triggers named — explicit request for a person, out-of-knowledge questions, upset callers, high-value or regulated requests.
• Handoff behavior set to be warm and to carry context, with an honest fallback when no one's available.

Full treatment in how to set business hours and escalation and human handoff and escalation. Avoid both failure modes: the dead line and the trap.

Step 4: Numbers

Connect the lines customers already dial so the switch is invisible to them.

• Numbers attached — the ones customers already call, routed to your workspace's agent via Integrations (Exotel, Plivo, or Twilio). See bring your own business number.
• Decide the scope — start with one line if you prefer, expand once proven.
• Confirm routing points to your configured agent, not a generic default.

Step 5: The test call — do not skip this

This is the step that separates a good launch from a bad one. A script that reads well on paper can still fail on the line, because paper doesn't interrupt you. Dial the attached number yourself and try to break the conversation the way a real caller would.

• Interrupt it mid-sentence. Does it stop and listen, or steamroll? This is the truest test — the two-way, realtime feel. See two-way realtime vs turn-based AI calls.
• Change your mind mid-call. Does it follow the correction?
• Talk naturally — trail off, mumble a number. Are you understood without slowing to computer-pace?
• Ask something off-script. Does it stay honest ("I don't have that, let me get someone") or invent?
• Ask for a human. Is the handoff quick, warm, and context-carrying?
• Test after-hours behavior by checking what happens outside your set hours.
• Check the languages your callers actually use.

Iterate on what you hear, not what you imagined. Run this loop until the call feels right.

Step 6: Go live — deliberately and reversibly

When the test call feels genuinely good, let real traffic flow.

• Switch on deliberately, knowing you can adjust or pull back.
• Watch the first real calls closely. Review completed calls — who called, how long, how it ended — in the early days.
• Tune fast. The first days of live traffic reveal more than any testing. Fix knowledge gaps and escalation issues as they surface.
• Keep it reversible. If something's off, adjust configuration or routing rather than being stuck.

Step 7: Operate and improve

Going live isn't the finish line; it's the start of the improvement loop.

• Review calls regularly — resolution, appropriate handoffs, and conversational naturalness (by listening). See measuring phone agent quality.
• Feed learnings back into knowledge, script, and escalation. Recurring out-of-knowledge handoffs mean a knowledge gap to fill.
• Watch usage. Agent Vani is billed by the minute and usage is visible, so finance can forecast. See pricing and how billing works.

A realistic timeline

People often ask how long going live takes. The honest answer: the mechanical steps are quick, but the steps that matter deserve real attention, and rushing them is the classic mistake. A useful way to think about the effort:

• Persona and voice — short. An hour of thought about tone, language, and identity, if you know your brand.
• Knowledge — the big one. Budget real time here. Pulling a week of real calls and writing down what people actually ask, then loading hours, policies, and FAQs as facts, is where most of the value is made. Under-investing here is the single most common cause of a disappointing launch.
• Hours and escalation — moderate. Decide your three time-states (staffed, after-hours, holidays) and name your escalation triggers deliberately.
• Numbers — short, once knowledge and behavior are set.
• Testing — as long as it takes to feel genuinely good. Iterate until the call survives interruptions, corrections, and off-script questions.
• Go live and tune — ongoing. The first days of real traffic are part of the launch, not after it.

The pattern to internalize: the fast steps are fast, but knowledge and testing are where you should spend your time. A line that's live but thin serves callers worse than one that took an extra day to get right.

Who should be involved

Going live well is rarely a solo task, and pulling in the right people early prevents rework:

• Whoever knows the answers. The person who actually knows your policies, hours, and common questions should shape the knowledge base. If those facts live only in someone's head, get them out and into the agent.
• Whoever handles escalations. The team that will receive handed-off calls should agree on the escalation triggers, so the right calls reach them with context.
• Whoever owns the brand voice. Persona and tone should match how you want to sound. A quick review by whoever owns that avoids a launch that feels off-brand.
• Whoever owns the numbers. Confirm which lines are being connected and that no line is attached by surprise.

You don't need a committee. You need the facts to be right, the escalation path to be agreed, and the voice to fit — which usually means a couple of the right people in the loop before you flip the switch.

Verifying before you flip the switch

Before production traffic, run a final honest pass against the earlier steps. The single most valuable verification is the test call from Step 5, done properly: not a scripted walk-through, but a genuine attempt to break the conversation. Interrupt it, correct yourself, go off-script, ask for a human, and check after-hours behavior. If any of those feel wrong, you're not ready — fix and re-test rather than launching and hoping. The whole reason this is a checklist and not a single button is that each step catches a different failure, and the test call catches the ones that matter most to real callers. See buying guide: AI phone agent for the specific tests to run.

The common launch mistakes

• Skipping the test call. The single biggest one. Real customers become your test group.
• Thin knowledge. Under-feeding facts means "call you back" to everything.
• Dead after-hours line. Turning the agent off outside hours wastes the highest-value coverage window.
• The trap. Not honoring requests for a human, or looping. Worse than no agent.
• Turning it on everywhere at once. Start on one line, prove it, expand.
• Launch-and-forget. The improvement is in the loop after go-live, not the switch itself.

An honest note on scope

Agent Vani handles inbound calls at launch — the calls coming to your connected numbers. Outbound dialing uses the same agent in a later release. So this checklist is about launching a great inbound line. Plan around that, and go live on the thing that's ready today.

Frequently asked questions

What's the most important step in going live? The test call. Dial the attached number and try to break the conversation — interrupt it, correct yourself, ask for a human. It's the only step that reveals what real callers will actually hit, and it's the one people skip.

How much time should I spend on the knowledge base? More than you think. The agent answers from what you give it and won't invent policy, so a thin knowledge base produces thin calls. Pull a week of real calls and write down what people actually ask.

Should the agent answer after hours? Yes. Turning it off wastes the highest-value coverage window. Have it still answer, handle what it can, and capture details for follow-up. See after-hours coverage.

Can I go live on just one number first? Yes, and you should. Prove it on one line, watch the real calls, then expand. Keep the change reversible.

What do I watch after going live? The first real calls closely, then a regular review of resolution, handoffs, and conversational naturalness. Feed what you learn back into the configuration. See measuring phone agent quality.

Does this cover outbound calls? No — inbound is available at launch. Outbound dialing comes on the same agent later. This checklist is for launching a great inbound line.

Who needs to be involved in going live? Whoever knows your answers (for the knowledge base), whoever handles escalations (to agree the triggers), whoever owns the brand voice (to check persona and tone), and whoever owns the numbers (to confirm which lines are connected). You don't need a committee — you need the facts right, the escalation path agreed, and the voice to fit.

How long does going live take? The mechanical steps are quick; the ones that matter — building a thorough knowledge base and testing on real calls until it feels genuinely good — deserve real time. A line that's live but thin serves callers worse than one that took an extra day to get right. Budget your time toward knowledge and testing.

What's the single biggest launch mistake? Skipping or rushing the test call, which makes your real customers the test group on your real number. Dial the attached line yourself and try to break the conversation before any production traffic flows.

Can I really pull it back if something's wrong? Yes — keep the change reversible. If early calls surface a problem, adjust the configuration or routing rather than being stuck. That reversibility is exactly why you go live deliberately and watch the first real calls.

Where to go next

• How to write a phone agent script — persona and knowledge in depth
• How to set business hours and escalation — the two make-or-break settings
• Bring your own business number — connect the lines callers already dial