Configuring the babelforce AI Agent module with Zendesk

Katrin Geske
Katrin Geske
  • Updated

This article explains how to connect the babelforce AI Agent module with a Zendesk AI agent (AI agents – Advanced, formerly Ultimate). babelforce gives your Zendesk AI agent a voice: the caller’s speech is transcribed, sent to your AI agent as a chat message, and the AI agent’s reply is read out to the caller.

Before you start

  • The AI Agent module is a paid feature. Please contact the babelforce Sales team regarding pricing first.
  • You need Zendesk AI agents – Advanced. Essential (legacy) agents are not supported.
  • The connection uses a Custom CRM integration in Zendesk, not a messaging channel.
  • The babelforce Success team will provide you with:
    • the Webhook URL for your babelforce environment
    • a Bearer token

Step 1 – Create a Custom CRM in Zendesk

  1. In Zendesk, open the product switcher (top left) and select AI agents.
  2. Go to Admin → CRM integrations and click Create custom CRM.
  3. Enter a name, e.g. babelforce, and click Create.
  4. In the Webhook URL field, enter the URL provided by the Success team. It follows this pattern:
    https://<your-environment>.babelforce.com/ivr/api/v1/webhooks/ultimate/chat
  5. Under Headers, click Add header:
    • Key: Authorization
    • Value: Bearer <token provided by babelforce>
  6. The field Webhook secret is optional and can be left as it is.
  7. Click Save.

Note: A Custom CRM can only belong to one organization. If saving fails with “organizationIds must contain no more than 1 elements”, the CRM was created for several organizations and cannot be edited anymore. Create a new Custom CRM instead.

Step 2 – Connect your AI agent to the Custom CRM

  1. Select your AI agent in the top right corner of the AI agents dashboard.
  2. Go to Settings (gear icon) → CRM integration.
  3. Select the Custom CRM you created in Step 1.

Important: If your AI agents workspace is linked to a Zendesk account, newly created AI agents are connected to Zendesk Messaging automatically, and the CRM integration page shows no option to select a Custom CRM. In this case, please contact your Zendesk account manager or Zendesk Support and ask them to connect your AI agent to your Custom CRM. When you create a new AI agent, choose the channel Messaging (not Voice – this is Zendesk’s own telephony product).

Step 3 – Generate the access token and share the details with babelforce

  1. On the CRM integration page of your AI agent, click Regenerate next to Access token.
  2. Copy the token right away – it is displayed only once.
  3. Send the following three details to the babelforce Success team via a secure channel:
    • the Access token
    • the Bot ID – you find it in the URL of your AI agent, e.g. …/bot/67e2c8a982d7fc37005695c3/…
    • your Zendesk subdomain – e.g. mycompany if your account is mycompany.zendesk.com

Note: Regenerating the token invalidates the previous one immediately. If you regenerate it later, please send the new token to babelforce, otherwise calls will fail.

Do not click Remove integration.

Step 4 – Prepare your AI agent for voice

Please check the following settings in your AI agent before testing. Missing settings are the most common reason why the AI agent stays silent on the phone.

  1. Welcome reply – go to Content → Use cases → Welcome reply, open the reply for each language, click Edit dialogue, enter a greeting and click Publish. babelforce waits about 10 seconds for this first message. Without it, the call ends

 

2. Identity – go to Content → Personalization → Identity and fill in the AI agent name, company name and business profile. Without these details, generative replies (e.g. answers from your knowledge sources) fail with a technical error.

3. Technical error reply – go to Content → Use cases → Technical error response and add a short text, e.g. “Sorry, something went wrong. Could you rephrase your question?”. Otherwise the caller hears nothing if an error occurs.

4. Languages – Zendesk does not detect the caller’s language on voice. Set the welcome and escalation replies for every language you use, and make sure the language in the babelforce module (Step 5) matches.

 

Step 5 – Create the AI Agent module in babelforce

  1. In the babelforce Manager, go to IVR / Voicebot → Call flows and click Add module → AI agent.
  2. Settings tab:
    • Provider name: ultimate
    • Bot Id: the Bot ID of your Zendesk AI agent. Please make sure the Bot ID is entered in the Bot Id field, not in Provider name.
    • Only play last message from the bot: if enabled, babelforce plays only the last message of each reply and skips messages such as “Please bear with me while I find the answer.”
    • Transcribe conversation: enable it and add a trigger with transcription conditions if you want to transcribe the conversation.
    • Language settings: select the input language (speech-to-text) and the voice of the bot (text-to-speech). Use the same language for both.
    • Request body: add the metadata you want to send to Zendesk, e.g. sessionId = {session.id}. See Data communication between babelforce and Zendesk AI Agent.

3. Audio tab:

  • Waiting audio: played while the AI agent prepares its answer. Generative answers can take 20–30 seconds, so a waiting audio is strongly recommended.
  • Failure audio: played if the AI agent takes too long or does not answer.

4. Exit condition tab:

  • Max. duration: the maximum time a caller can stay with the AI agent. The default is 60 seconds, which is usually too short for more than one question. We recommend at least 300 seconds.
  • Max. iterations: the maximum number of questions and answers.
  • Trigger (optional): ends the AI agent conversation when a condition is met, e.g. an escalation.
  • If exit condition applies, send to: the module the call continues with afterwards, e.g. a queue.

 

5. Click Save and connect the module to a number or call flow.

 

Step 6 – Test the setup

  1. Call the number connected to the AI Agent module.
  2. You should hear the welcome reply of your AI agent.
  3. Ask a short question in the configured language and wait for the answer.
  4. Check the conversation in Zendesk under Conversations. If an AI agent message is empty, open it to see which reply was used (e.g. Technical error reply).

Troubleshooting

What you notice Possible cause What to do
The call ends after 1–2 seconds Bot ID, access token or Zendesk subdomain missing or wrong on the babelforce side Check the Bot ID in the module and contact the babelforce Success team
Silence for about 10 seconds, then the call ends No published welcome reply Publish a welcome reply (Step 4)
Waiting audio for a long time, then the failure audio The AI agent replied with an empty message, e.g. a technical error Check Identity and Technical error reply (Step 4) and the conversation log in Zendesk
The call leaves the AI agent in the middle of a conversation Max. duration reached Increase Max. duration (Step 5)
The AI agent answers in the wrong language Language in the module does not match the AI agent Align the input language and voice with the AI agent’s language

If you need support, please contact support@babelforce.com.

 

Related to

Was this article helpful?

/

Comments

0 comments

Please sign in to leave a comment.