An AI agent for inbound calls is a voice assistant powered by a neural network that answers calls to your company's number on its own: it talks to callers in natural speech, answers from your knowledge base, books meetings, fills in the lead card in your CRM and transfers the call to a manager when the caller asks for a person or is ready to buy. Unlike a phone menu (IVR), callers don't have to listen to “press 1” — they simply say what they need. One agent handles hundreds of conversations at once and works around the clock, so no call has to wait for a free operator — not at peak hours and not at night.
- Answers calls on your numbers: telephony is connected over SIP, and customers call the number they already know.
- Speaks naturally and answers strictly from company materials: price list, FAQ, terms, documents.
- Takes the call to an outcome: a calendar booking, a completed lead card, a call summary in your CRM and Telegram.
- Transfers to a manager within the same call — on average about 20% of callers ask for a person (Sales Workshop data).
How an AI agent answers an inbound call
For the caller it's an ordinary call to a company. Behind the scenes there are eight steps:
- The call comes in to your number. The number stays yours: the agent connects to your telephony over SIP, and calls are recorded and saved to the lead card.
- The agent greets the caller and introduces itself. Honesty works best — “the company's AI assistant” — followed straight away by what it can help with.
- The caller speaks in their own words. “I'd like to book for tomorrow after six”, “how much is delivery to another city” — the agent understands natural speech, “the day after tomorrow” or “Friday at 3 pm”, and takes the caller's time zone into account.
- The agent answers from the knowledge base. Prices, terms, hours, addresses — only from company materials; if there's no answer, it doesn't make one up but offers to connect a specialist.
- It finds out what the caller needs. Name, phone number, budget, timing — the agent asks during the conversation and records them in the lead card.
- It drives to the target action. It books a meeting in Google Calendar, texts a link to the proposal at the end of the call and creates a lead card in the CRM.
- It transfers to a manager if the caller asks for a person or is ready to discuss a deal — within the same call, with no call back.
- It wraps up. After the call the lead card holds the recording and a summary: what was discussed, what objections came up, how ready the customer is to buy and what to do next. Hot leads trigger an instant Telegram notification.
What an AI agent does on an inbound line
| Task | Example request | Outcome |
|---|---|---|
| Information | “How much is it?”, “Are you open on Sunday?” | An accurate answer from the knowledge base in seconds |
| Booking | “Book me in for Friday evening” | A Google Calendar event with the caller's contact details |
| Qualification | “Tell me about your plans for teams” | A lead card with needs, budget and timing |
| Support | “Where's my order?”, “How do I return an item?” | An answer based on company rules, or a hand-off to a person |
| Transfer | “Put me through to a manager” | The call goes to a manager with the context of the conversation |
| Out of hours | A call at 11:40 pm or on a Sunday | The agent answers as it would by day; the manager gets the summary in the morning |
Night and weekend calls deserve a closer look: we calculated how many enquiries a business loses when nobody can pick up in our article on leads lost at night.
How an AI agent differs from IVR and voicemail
| Criterion | Phone menu (IVR) | AI agent |
|---|---|---|
| How the caller states the request | Presses menu buttons | Says it in their own words |
| Answering a question | No — only routes to a department | Answers from the knowledge base |
| Bookings and enquiries | The caller has to wait for an operator | Books and fills in the CRM itself |
| Peak load | “All our operators are busy” | Hundreds of conversations at once |
| Transfer to a person | After the menu, with no context | On the caller's request, with context |
| Nights and weekends | Voicemail | A full conversation |
An AI agent differs from robocalls with pre-recorded phrases in the same way: it holds a dialogue rather than playing a script. A detailed comparison is in Robocalls or an AI agent.
What an AI agent should never do
Trust in voice AI is fragile: in a Gartner survey, 64% of customers said they would prefer companies didn't use AI in customer service, and the top concern was that it would become harder to reach a person. That makes the agent's boundaries more important than its capabilities:
- Make up answers. If something isn't in the knowledge base, the agent says so and offers to connect a specialist.
- Hold on to a caller who asks for a person. One clarifying question is fine; persuasion is not.
- Give medical, legal or financial advice. Its job on such topics is to book a consultation or transfer to a specialist.
- Promise what hasn't been agreed. Discounts and special terms only within the rules set out in the knowledge base.
- Hide that it's AI. According to Five9, 71% of consumers consider it very or extremely important to know when they're interacting with an AI agent.
How many calls an AI agent handles on its own
On average across Sales Workshop clients, about 20% of callers on an inbound line ask to be connected to a manager; the remaining roughly 80% of conversations happen without such a request. The spread between industries is wide: requests are rarer on booking and information calls, and more common in real estate, B2B and on existing-customer lines. Why that is and how to work with it is covered in How often callers ask for a human.
That's why a transfer to a manager is a mandatory part of any rollout, not a “just in case” option. How to set it up so that calls aren't lost and the caller's number is kept is covered in our guide to transferring calls from AI to a human.
How to roll out an AI agent on an inbound line: 6 steps
- Step 1. Choose the line and scenarios. Start with one number and the most frequent calls — usually bookings, prices and terms.
- Step 2. Build the knowledge base: price list, terms, addresses, opening hours, answers to frequent questions. The best source is recordings of real calls: they show what people actually ask.
- Step 3. Connect telephony. The agent connects to your carrier or cloud PBX with SIP credentials — the number customers call doesn't change.
- Step 4. Set up actions: a calendar for bookings, CRM integration (HubSpot, Bitrix24 and others), hot-lead notifications in Telegram, a text message after the call.
- Step 5. Set up the transfer to a manager: a number or queue, the transfer method, the out-of-hours scenario.
- Step 6. Test and launch. Make test calls for typical scenarios, then put the agent on part of your calls and review the first days' recordings.
On the PRO plan you configure the agent yourself on the platform; on Business the Sales Workshop team runs a turnkey launch in 3–5 business days. What the launch looks like day by day is in our article on rolling out an AI sales rep in 5 days.
How much it costs
A voice agent for inbound calls is part of the PRO plan and above; PRO costs from $683/mo. The final price depends on the volume of minutes and messages you choose within the plan; see the pricing calculator for an exact figure.
Metrics to track after launch
- Share of missed calls — the main reason to put an agent on an inbound line.
- Share of calls that end in a target action: a booking, an enquiry, a CRM record.
- Share of requests for a manager and their reasons: the benchmark is about 20%, but compare yourself with yourself, not with the average.
- Share of completed transfers and the manager's answer time.
- Call-to-sale conversion before and after launch.
Bottom line
An AI agent on an inbound line takes routine conversations — information, bookings, qualification — off your team and keeps calls from slipping away at night and at peak hours. It works well under three conditions: a detailed knowledge base, honest boundaries (no making things up, no holding on, no hiding that it's AI) and a fast transfer to a person for those who need one. Start with one line and the most frequent scenarios, and expand based on the numbers from the first weeks.
Frequently asked questions
- Can an AI agent answer calls on our current number?
- Yes. The agent connects to your company's telephony over SIP, so customers call the number they already know. Calls are recorded and saved to the lead card together with a summary.
- How many calls can an AI agent take at once?
- One agent handles hundreds of conversations in parallel, so calls don't wait for a free operator even at peak hours.
- Should callers be told they're talking to AI?
- We recommend that the agent says so in its greeting: according to Five9, 71% of consumers consider it very or extremely important to know when they're interacting with an AI agent. Gartner also advises telling customers up front that they'll be connected to an employee if needed.
- What happens if the AI agent doesn't know the answer?
- It doesn't make one up: it answers only from company materials and hands an unusual question to a manager — by transferring the call or through a lead card with the recording and call summary.
- How much does an AI agent for inbound calls cost?
- Inbound calls come with the PRO plan, which costs from $683/mo. The price depends on the volume of minutes and messages; use the calculator on the website for an exact figure.