Chatbot and AI voice agent: when chat should become a call

Some conversations shouldn't end in chat

A customer has sent ten WhatsApp messages about a flat: budget, district, number of rooms, a mortgage question. The chatbot answered them all, but the sale is still far off — the remaining questions will be settled by talking, not typing. Elsewhere, another customer asked a price, left a number and went quiet. In both cases the right next step is a call. The question is: who calls, when, and with what information?

This article sets out how a chatbot works together with an AI voice agent and a call centre: when a chat should move to a call, how consent is obtained, what context must travel, whether an AI agent or a person makes the call, and how the result is measured. Handing a chat to an agent is a separate topic; here the subject is changing channel.

Channel change versus agent handover

1. 1Handover to an agent in chatThe conversation continues on the same channel; a person simply writes instead of the bot. Nothing changes for the customer except who is answering.
2. 2Moving from chat to a callThe channel changes: the customer moves from typing to talking. That needs their consent, a suitable time and the chat's context reaching the call.

The rules for handing a chat to a person are in handing over from chatbot to agent; handing over from AI to a person inside a call is covered in AI-to-human call transfer.

When to move to a call

• The conversation has got complex: many variables, many questions, a decision that needs explaining — property, construction, B2B
• The customer asks: «call me», «I'd like to talk»
• A high-value lead: a large budget, a corporate order
• The customer gave a number but went quiet in chat — a call is a chance to win back a warm lead
• Something needs confirming: a meeting, a delivery time, the details of an order

Consent

Having a phone number is not consent to call. The chatbot should offer the move openly:

1. 1Offer«We can sort these questions out in 5 minutes on the phone. Shall a manager call you?»
2. 2Time«When suits you — this evening or tomorrow morning?»
3. 3RecordThe customer's answer — agreed, the time, preferred language — is recorded.
4. 4RefusalIf the customer says «no, let's keep writing», that is recorded and no call is made.

The context has to travel

The worst start to a call: «Hi, how can I help?» — when the customer has just explained everything in ten messages. Whoever calls — AI agent or person — should know:

• A one-sentence summary of the conversation
• The fields collected: budget, district, product, date
• What the bot has already said and promised
• The customer's open questions
• The agreed call time and language

In Vexvon, fields extracted from the conversation and the summary are written to the customer card, and the caller starts from that card. The same knowledge base can serve both the chatbot and the call agent — so the customer hears the same answer in chat and on the phone.

Who calls: the AI agent or a person

1. 1AI voice agentA standard script: confirming a meeting, agreeing a time, a few clarifying questions, the first call to a warm lead. The script, question list and fields to extract are written in advance.
2. 2A personA complex negotiation, a large amount, agreeing a price during the call, an unhappy customer.
3. 3TogetherThe AI agent starts the call and, when needed — the customer asks, or the call leaves the script — transfers it to a person.

In Vexvon the voice agent works through calling campaigns: a filter is chosen in the CRM — leads at the «call needed» stage, for example — a target list is built, and the AI agent calls to the script. Call-time rules prevent calls outside set hours. A lead moves from chat to the call list through its CRM stage or a manager's decision.

A sample flow: real estate

1. 1Chat — WhatsAppThe customer asks about a two-room flat, gives a budget and district, asks about a mortgage.
2. 2BotShows matching projects, gives general information on the mortgage question and offers: «Shall our manager explain the details by phone? When suits you?»
3. 3ConsentThe customer writes «tomorrow 11:00». On the card: budget, district, rooms, mortgage interest, call time.
4. 4CRMThe lead moves to the «call needed» stage and is assigned to a manager.
5. 5CallTomorrow at 11:00 the manager or AI agent starts from the card: «Yesterday on WhatsApp you asked about a two-room flat and a mortgage…» — the customer repeats nothing.

The AI agent's first 30 seconds

1. 1Who and why«Hello Ms Leyla, this is Vexvon Estates calling — as we agreed on WhatsApp yesterday.»
2. 2Context«You were interested in a two-room flat in the Nasimi district, and a mortgage.»
3. 3Purpose«I'd like to book a viewing today and briefly explain the mortgage terms. Is now a good time?»
4. 4A way outIf the customer is busy — a new time; if they want a person — transfer to a manager.

Common mistakes

• Calling without consent just because there is a number
• Not carrying the chat's context into the call — the customer tells it all again
• Missing the agreed time
• Not returning the call's outcome to the chat history — the bot later repeats old questions
• Calling every lead — for a customer with a simple question, it's a nuisance

After the call

• The call's outcome is written to the same customer card — chat and call in one history
• The agreed next step — a meeting, documents, a quote — as a CRM stage
• If the customer writes in chat again, the bot should know the call's outcome
• A document or link promised on the call can be sent in chat

What to measure

• In how many conversations offered a call the customer agreed
• Share of calls made at the agreed time
• Call answer rate — separately for leads with and without consent
• Conversion from the call to a meeting or a sale
• Cases where the customer had to repeat on the call what they said in chat — context was lost

Limits

How far the move from chat to call is automated depends on the process set up: a lead reaches the call list through a CRM stage, a campaign or a manager's decision. The voice agent should leave negotiations that need a decision — a price reduction, contract terms — to a person. The company's rules for recording and storing calls, and for informing the customer, should be checked with a lawyer.

Conclusion: the channel changes, the context must not

When a chatbot and a voice agent work together, the customer moves from typing to talking without starting from scratch. It takes four things: offering the call at the right moment, clear consent and a time, writing the chat's context to the card, and returning the call's outcome to the same history. The channel changes; the customer feels they are continuing the same conversation.

The AI voice agent's capabilities are described on the call center page; the automatic steps after a conversation are in chatbot workflow automation. The rest of the section is in this category. To design your chat and call flow together, ask for a demo.

Frequently asked questions

1. 1Can the chatbot make the call itself?The chatbot writes; the AI voice agent or a manager calls. A lead moves to a call through its CRM stage, a calling campaign or a manager's decision.
2. 2Can we call without the customer's consent?We don't recommend it. Offering the call in chat and agreeing a time is both honest and more effective.
3. 3Does the voice agent know what was said in chat?Yes, if the context is written to the customer card and the script uses it. The same knowledge base can also serve both channels.