AI Answering

Ricardo Updated by Ricardo

Since we launched Calling last year, which represented a significant expansion of communication capabilities via WhatsApp beyond messaging, organisations all over the world have been addressing critical needs in healthcare, especially in regions with low literacy or where a phone call is faster and more convenient.

The main piece of feedback we get is: we get so many missed calls. We just can't handle the volume. Unlike text messages, it's hard to scale.

AI Answering is the capability for you automate a voice call with AI, enabling people to interact with your Journeys entirely in voice.

What this enables

The core unlock is turning a WhatsApp number into a 24/7 clinical front door — powered by voice. Here are a few examples of what you can do with it:

  • After-hours triage: A patient calls outside clinic hours, describes their symptoms by speaking naturally, and the AI agent triages urgency — advising them to come in the morning, go to the ER, or providing self-care guidance. No staff needed.
  • Appointment booking by voice: Instead of navigating menus or typing, a patient just says "I need to see Dr. Molefe on Thursday afternoon" and the AI handles scheduling. This is especially important for elderly patients or those with limited literacy.
  • Pre-consultation intake: Before a scheduled visit, the patient calls and the AI walks them through medical history questions, current symptoms, and medication lists — all by voice. The clinician gets a structured summary before the patient even walks in.
  • Medication adherence: The AI calls patients (or patients call in) to confirm they're taking medications correctly, flag side effects, and escalate to a nurse if something sounds wrong.
  • Community health worker support: CHWs in the field call in to report patient visits, get clinical guidance, or check protocols — hands-free, while they're on the move.
  • Language accessibility: In multilingual markets like South Africa or India, patients can speak in their preferred language instead of trying to type in English. Voice + AI removes the literacy and language barrier simultaneously.

AI Answering uses Journeys to know how to behave. You can use any journey you already have. However, because voice calls are more fluid than messaging, we strongly recommend you use Agents — which naturally adapt to zig-zags in the conversation.

⚙️ Setting up

1. Enable calling

The first thing you need to do is to enable and get acquainted with our Calls feature. Here’s an in-depth guide ».

You probably want to enable the display the call buttons setting, so that your customers can call you.

2. Integrate with an AI provider

AI Answering is currently compatible with:

  • OpenAI, which works standalone as it both combines voice and the brain
  • ElevenLabs, which requires you to integrate with another provider as it only provides the voice — we recommend Google Gemini, as it's the fastest brain model we've found

Here are instructions on how to integrate with an AI provider: Integrate with an AI provider.

Note: make sure you have credits on the provider you integrate with — otherwise your calls will fail.

3. Setup an assistant

You also need an assistant. If you already have one, you can probably use that one — and add any voice-related instructions you’d like.

If you don’t, on the Settings > AI page hit “New assistant”.

4. Configure AI Answering

In the Settings > Calls page, simply enable AI Answering. Here's what you can configure:

  • When should AI Answer: you can choose if it should always answer, or when no operators are available
  • Model & voice: select the model you integrated with earlier. If you select ElevenLabs, which only provides the actual voice, you'll get a few extra options to select the brain model
  • Assistant: select the assistant you created earlier
  • Starting Journey: which journey you want the AI model to follow when it picks up the call.

And that’s it! Really. And if you already have the journey you want to offer over voice, things should already be working. Call your number! (if you’re switching on calling for the first time, sometimes the WhatsApp app takes a while until the call button appears in the interface).

💡 Tips & tricks

We’ve been trialling this feature internally for a while. Here are a few tips & tricks:

Agents work best

Any kind of journey works well. However, user’s expectations of a voice call are different from texting. They expect the ability to say anything, and the system to handle that well. Non-AI journeys tend not to work great here.

We strongly suggest you use Agents, which naturally solve this problem.

Escalate!

If you have folks providing support already, and want to provide escalation pathways, the Route to operator card routes the call to an actual person (make sure they mark themselves as Available on the inbox to receive the call!).

Testing

Testing a voice call works the same as testing a text journey. We respect your !preview on/off settings, so you can keep working on your draft journey and calling it and, when you’re ready to share your changes to your users, Publish it.

Additionally, you can still preview your journey on the built-in text simulator. Just bear in mind that the Realtime model might behave differently to your text generation model.

Debugging, Logs & Evals

You can debug issues with:

  • The AI Logs page, which has every single trace for how a response got constructed
  • You can run Evals & Logs via our integration with Maxim

Was this article helpful?

Agents

Evals & Logs

Contact