Your customer is not only messaging you on WhatsApp — they are calling you on WhatsApp. There is a call button inside the app and customers use it. The problem: in most businesses those calls are never answered. The number is managed through WhatsApp Business, a team or an AI assistant handles the messages, but when a call comes in there is no ringing phone on anyone's desk.
That gap is now closed. Your AI agent answers incoming WhatsApp calls, holds the conversation, collects what it needs and records the call. The same agent can also place outbound WhatsApp calls when needed.
What it does, briefly
Three capabilities arrived at once, and all three mirror how the agent already behaves on a phone line:
- It answers incoming calls. When a customer calls over WhatsApp the agent picks up, introduces itself and starts talking. No hold music, no keypad menu, no voicemail box.
- It places outbound calls. Appointment reminders, quote follow-ups, a callback to an unreachable lead — the agent initiates the call itself.
- It records. The recording and transcript land under the customer record, exactly as they do for phone calls.
In other words, WhatsApp is no longer a second messaging channel for you. It is a second phone line.
Why this is a continuation of one channel, not a new one
The real value here is not the calling capability itself. It is where the call connects to.
If a customer messaged you on WhatsApp two days ago asking about pricing and calls from the same number today, the agent knows that exchange. It does not open with a blank "how can I help you." Likewise, something shared during the call — a changed delivery address, an appointment preference — is still there when the conversation returns to text.
Businesses running a separate call center tool and a separate WhatsApp panel cannot make that connection. The two systems cannot see each other's history, so the customer explains themselves again in every channel. That is the hidden cost we run into most often in competitive analysis: not buying two tools, but trying to make the two tools talk.
Recording: everything the phone line has
What makes a voice conversation useful is that it is still in your hands after it ends. The same applies to WhatsApp calls:
- Audio recording and written transcript
- Call summary and outcome tag (purchase intent, complaint, appointment request, and so on)
- Duration, date and which line it was handled on
- Automatic write-back of any resulting task to your CRM
The legal side of recording is not a detail to skip. Callers must be told clearly at the start that the conversation is recorded, a retention period has to be set, and access to recordings must be limited by role. Later in this series we cover call permission and recording management for WhatsApp specifically, from both the Meta policy and the data protection angle.
What the conversation itself is like
The channel changed, but the model behind it works the same way: voice to voice. The agent does not run a chain that transcribes incoming audio, asks a language model, and converts the answer back to speech. The model processes audio directly and answers in audio.
Two practical consequences. First, intonation, emphasis and pauses are not lost in translation — when a customer hesitates or cuts their own sentence short, that signal reaches the model. Second, since every conversion step piles on delay, dropping the chain makes the conversation faster.
Measured on our phone line, the model composes its answer in 0.65 seconds (two independent measurements: 659 and 624 milliseconds). That figure is the model's composition time; audio travelling in both directions and detecting that the caller has finished add a little on top. WhatsApp calls run through the same architecture, but we are not publishing a separate figure for that channel until we have measured it.
In a chained architecture this step never happens alone: speech is converted to text, the text goes to a model, and the answer is converted back to speech. Each of the three conversions adds its own share, and in most setups speech generation cannot even start before the sentence is complete.
Where it pays off immediately
Answering a WhatsApp call catches a different customer profile than answering a phone call. Someone calling from WhatsApp usually already has the app open and has messaged you or was about to — their interest is fresh.
The fastest-returning scenarios:
- Appointment-driven businesses. Clinics, beauty centers, service points. When a customer abandons messaging for a call, the matter is usually urgent.
- E-commerce order problems. Where is my shipment, how do I return this. Moving from text to a call is the moment patience runs out.
- Customers abroad. A WhatsApp call carries no international calling charge. For businesses with overseas customers that alone opens a channel.
- After-hours volume. Evening and weekend calls are the segment that goes completely unanswered in most businesses.
How to start
Setup is shorter than provisioning a phone line: if your WhatsApp Business number is already configured in the system, enabling calling does not require a new line, a new PBX or a new number. Your agent's knowledge base, conversation rules and handoff thresholds carry over as well — the agent you built for the phone can answer WhatsApp calls too.
If you would rather hear it first, you can have the agent call you from the demo page, and see how the voice agent works on the AI call center page.
In the next post we go into what the WhatsApp Business Calling API actually is, which limits Meta sets, and what remains in your hands.