AI voice bot for business: how to create a bot and get leads

An AI voice bot for business solves a familiar problem: your CRM holds thousands of contacts — people who once left a request and then went quiet. Managers have no time to call them, and the database earns nothing for years. And a website request that comes in at 9 p.m. waits until morning, while the client goes elsewhere.

A voice robot is an AI bot that talks to clients over the phone on its own, with no call center agents. It calls your old database, calls back new requests within seconds and answers incoming calls 24/7. The bot writes the outcome of each conversation to the CRM itself and transfers hot clients to a manager. A minute costs $0.12 on average — no salaries, sick days or weekends.

Watch the four-minute video, and below you will find step-by-step setup and a launch method that actually brings in leads.

What an AI bot for phone calls can do

• Answers in about 1.5 seconds in a lively voice with emotions. Many people do not realize they are talking to a bot.
• Keeps the whole conversation and large databases in memory — it does not forget what the client said at the start.
• Does not slow down at peak hours. With many services, replies drag on for 4–5 seconds in the middle of the day, and the pause gives the robot away. Not here.
• Works with large dynamic databases. During the call, the bot requests free time slots, an address or an employee from your system. That is why it books for an exact time and does not mix up dates and addresses.
• Runs the CRM itself and learns without prompts — it sets the step, writes a summary and makes up to 200 calls per minute.

Why your business needs an AI bot that calls customers

• An online school or a real estate agency calls its old database and invites people to a new course intake or to a viewing. The manager gets only the interested ones.
• A clinic, salon or car service stops losing calls at night and on weekends: the bot answers questions about prices and books the client.
• Any business with a website keeps its “we will call you back in 60 seconds” promise. The widget's “Callback request” button creates a lead, and the bot calls back even at night.
• Law firms and B2B services warm up a cold database: in a couple of questions the bot finds out the situation and the amount, politely says goodbye to those who don't fit and books the rest for a free review with a specialist. The specialist spends time only on large, clear cases.

AI bot vs. sales rep on cold calls

A comparison on the same 145 numbers a day: the rep had 92 unanswered calls, 53 conversations and 4 meetings; the bot had 51 unanswered calls, 94 conversations and 7 meetings. The difference is in redials: the bot calls the unreachable ones 5–6 more times by itself, while a rep has no time for that in a day.

The bot does not replace your sales team — it takes over the first-line routine of cold calling and qualification, so the rep gets only people who are ready to talk. On inbound requests the bot performs about 30% better than a person on average: it calls back within the first minute, talks to several clients at once and never strays from the script. Calling the base of refusals and stalled leads brings back up to 20% of clients, which is cheaper than a new client from ads.

How an AI voice bot works

The assistant bot asks about your business and writes the instruction (prompt) — the AI voice bot script: who the bot introduces itself as, how it speaks and what goal the conversation leads to. The bot takes leads (client contacts) from the CRM according to the “Launch conditions” and calls on a schedule. After the conversation, it sets the lead's funnel step — a stage of work with the client, such as “Interest” or “Refusal” — and fills in the card fields. The bot, phone numbers, telephony and CRM are all in one account, with no integrations or custom development. What exactly the bot writes to the card is covered in “AI voice bot CRM integration”.

The “Leads” column shows how many leads are waiting for a call. The “Status” tag is a button: “Running” means the bot is calling, “Stopped” means it only answers incoming calls, if a number is linked.

How to create an AI voice bot and start calls

1. 1
   Create the bot

   Open the “AI bot” → “AI voice bot” menu and click “Add AI bot”. Enter a name and click “Add”. The bot card opens together with the “Generate prompt” wizard.
2. 2
   Answer the assistant bot

   Paste a link to your website and click “Analyze the website” or “I have no website”. You can answer by voice.
3. 3
   Generate the instruction

   Click “Confirm and generate”. A couple of minutes later the bot saves itself, and you can already call it.
4. 4
   Go through the checklist

   The “Prompt generated 🎉” window shows what is left to do and takes you to the right place in the card.
5. 5
   Call the bot

   On the “Instruction” tab, click the “Test call” phone icon, pick a number and call it yourself.
6. 6
   Set up calling

   On the “Settings” tab, set the “Number carousel”, “Schedule”, “Launch conditions”, “Dialing speed per minute” and “Dialing settings”. Upload your database via “Lead import”. Each setting is covered in “AI voice bot for outbound calls”.
7. 7
   Launch

   Top up your balance, click the “Stopped” tag in the bot list, then “Got it, start”. The account checks the numbers and the checklist: “Accept the risk” starts calling without a completed checklist, but only this once.

New users get a trial period and $1 on the balance after activating the account — enough for the first test calls.

What to tell the assistant so your AI voice bot sells

The assistant writes the script from your answers. Vague words make a vague script, while specifics make a script that sells. Look for material in your managers' call recordings: what clients value and which three objections come up most often. Turn a frequent objection into an advantage. Here is what matters in your answers:

• Advantages — with proof and the client's pain. “Many years on the market” and “a personal approach” tell the client nothing. Write what you do differently from competitors and which problem it solves. For example: “We don't just file lawsuits — we trace where the debtor moved the money, so we recover the debt even when the bailiffs found nothing.”
• Strong but honest. “Increases the probability” sounds unsure, and “99% success” makes people ask about the remaining percent. Go as far as you can back up: “practically guarantees debt recovery”.
• Not sure the client has the problem? Write “probably”. “We call companies that probably have an unenforced court judgment”: the bot will ask rather than assert.
• The goal — together with qualification and an exit. “Find out the debt amount: from $10,000 — arrange a call with a lawyer; less — explain that we cannot help and say a polite goodbye.”
• Mark reference questions. “Find out whether the debtor is in bankruptcy: the lawyer needs this, and it is no reason to end the call.”
• Describe the action, not a ready-made phrase. Write “explain”, “clarify”, “act like this” rather than “say: …”. The bot quotes ready-made phrases, and its speech sounds wooden.

Before generating, read the “Please review the collected information” card. Forgot a criterion or a condition of your work? Click “Add more” and add it. This way the AI takes it into account in all instruction blocks at once, not as one rule on the side.

How an AI voice bot sells over the phone: conversation techniques

A script sells when every bot reply moves toward the goal instead of just answering the question. Techniques that bring meetings in cold calling:

• Start with the client's situation, not with your company. “You have a court judgment you can't collect on, right?” — and only then the name and the company. Then ask permission: “A couple of questions to see whether we can help, all right?”
• Offer an easy next step. A free review of the situation with a specialist: “It commits you to nothing, you lose nothing.” People agree to this step more often than to a purchase.
• A ladder of agreements. First a small “yes”, such as sending the details by message, then a bigger step: a short talk with a specialist “to tailor the offer rather than send you things you don't need.” Until the client agrees to the first step, the bot does not offer the second — so completely cold clients never reach a manager. How to send a message after the call is covered in “Follow-up message after an AI call”.
• Price and prepayment questions go to the specialist. The bot briefly says the specialist will discuss the terms and returns to booking. This way the conversation does not get stuck on price.
• “Are you a robot?” — answer honestly and move to the benefit. The bot admits it is an AI assistant and steers back to what the client gains: “Yes, I'm an AI assistant. But if I can help you get more leads, that's what matters, right? After all, we're having a normal conversation.” The objection turns into an argument, and the call continues with the plan.
• End every reply with a question that brings the conversation closer to the goal.
• Two options instead of an open question. “June 30 or July 1?” — it is easier for the client to choose, and the call is shorter.
• “It's not a good time” is no reason to give up. The bot finds out whether a callback makes sense and continues with the plan if the client is ready.
• Promise only what your team will deliver. If the specialist cannot call at an exact time, the bot agrees on a day, not an hour.

You add these techniques to the “Conversation plan” and “Important rules” — see “AI voice bot script” for how.

Test call to an AI voice bot: what to check and how to fix it

Call the bot yourself on speakerphone and talk like a client — as if the bot had called you. Ask awkward questions: “where did you get my number?”, “do I have to pay upfront?”, “are you a robot?”. In a few minutes you will see where the bot falters. To test without a call, chat with the bot in “Training AI”.

Check the conversation mechanics too: interrupt the bot mid-sentence, give your name and a few digits, and a minute later ask it to repeat them. If the bot keeps talking when interrupted, check “Minimum customer words before the AI can be interrupted” in “Settings” → “Speech recognition settings”: at 100 the bot can hardly be interrupted.

What to fix after the test:

• The bot does not know a fact about the company, such as whether you require prepayment — add an item to “Arguments and facts”.
• The bot acts the wrong way — click “Fix” under its reply in “Training AI”: the AI writes the rule in without contradicting the others. More in “How to train an AI voice bot”.
• You forgot something important in your answers to the assistant and have not edited the instruction yet — generate the prompt again. Do not do this after manual edits: the new instruction replaces the current one along with your edits.

Technical questions — starting calls, the schedule, uploading a database — are handled free of charge by support at a Zoom meeting: how to contact customer support.

How many phone numbers you need for outbound calls and why they get flagged as spam

A small business needs five numbers to start; for large volumes, take 15–25. The “Number carousel” is number rotation: each call goes out from a different number, so spam flags build up more slowly. If the client does not pick up, the second attempt goes out right away from another number.

A “spam” flag is a normal part of outbound calling: numbers get used up like rounds in a magazine. The “Check numbers against spam lists” checkbox removes flagged numbers, and new ones are bought in the “Phone numbers” section.

Before the start, the window “The best strategy for outgoing calls” recommends 200 calls per minute for no longer than 90 minutes: carriers' anti-bot systems do not have time to react. Then stop calling and replace the numbers.

How to launch the bot so it brings results

From generating the prompt to the first call takes about two minutes. But the goal is requests and meetings, not a bot that “sounds nice”. Results come from iterations.

The fastest way is to step out of your comfort zone:

1. Make 50 calls yourself. By the 51st, the script is polished: people do not hang up, and meetings get booked.
2. Write the script on paper and make 10 calls strictly by it.
3. If at least one meeting comes out of ten, move the script into the bot.

This way automation takes about 14 days, while trying entry points (reasons for the call and opening lines) at random can take more than a month with no result. The main rule: do not automate what has not been tested in practice — it saves money and nerves.

Cold database (they do not know you and are not expecting your call): run the main scenarios in “Automatic training” — training on synthetic data where the AI plays the clients. Come up with entry points together with the AI. Then launch 100 calls, listen to them, improve the script — and another 100, until the lead conversion rate and cost per lead become acceptable. How to go through these rounds is covered in detail in “How to train an AI voice bot”.

Launch timelines by scenario

• Calling back website requests — the fastest, within a couple of days.
• Calling people who left a request and went quiet — usually 5–10 days.
• Incoming calls — a couple of days of training, no more than seven. The bot learns like a child: in tests you will miss about 10% of the ways people actually talk. Switch calls over in portions: first the ones operators missed, then on a schedule, and when about 95% of cases are covered — all of them.

Don't want to figure it out yourself and need a guaranteed result? Order turnkey implementation: the button is at the end of the article.

Copying a bot or a project: what to set up again

A copy saves hours: you do not have to build a bot for a branch or a similar scenario from scratch. But not everything is carried over, and the copy needs finishing before launch.

• Copying a bot. The “⋮” menu in the bot row → “Copy” → “Select where to copy”: “Copy to the current project” or “Enter project number”. A copy to another account's project is sent as a request: its owner sees the “Accept an autofunnel copy?” window and clicks “Confirm”.
• Save the copy before launch. Open it and click “Save”: saving pulls the variables in for the new funnel. Without this, the launch is stopped by the “The voice bot has not been saved since the copy” window.
• Copying a project. In the project list in the header, click the “Rename project” pencil → “Copy”, enter the “Account number” and click “Copy” again. Only the settings are carried over: leads, phone numbers and integrations do not get into the copy, so you connect them again.
• Voice bot → chat bot. “Settings” → “General” → “Create a chat bot”: the instruction moves to a new chat bot without emotion tags and with channels turned off. How to connect channels and move a bot back is covered in “AI chatbot for website”.

Read also: AI outbound calling bot: call your contact list without wasting budget, How to train an AI voice bot to answer like your best sales rep, AI voice bot knowledge base: price list, catalog and webhooks