AXL Knowledge Base

AI Calls

~7 min read Section: Telephony

AI Calls let you call contacts automatically with a voice AI agent. Add an AI Call to an AXL automation, and the agent handles the conversation. The automation then determines the call outcome, extracts the data you need from the transcript, and continues along the matching branch.

For example, an AI agent can qualify a new lead by asking about the company’s number of employees, industry, and revenue. After the call, AXL can save the answers to the contact record and determine whether the customer is ready to buy.

Adding an AI Call to an Automation

Go to Settings → Automations. Create a new automation or open an existing one. To learn more about creating and using automations, see the Automations section.

In AXL, an automation is a sequence of triggers and actions that runs automatically once it’s triggered.

Add the AI Call action to the automation and connect it to the previous step. Then open the settings of the new step.

Automation palette on the Actions tab with AI Call found by search, and the AI Call step connected to Automation Started on the canvas.

AI Call Step Settings

The following fields are available in the step settings:

  1. Step Name. Enter a name for the step. The name is used only within the automation and helps you tell one AI Call from another.

  2. Voice Agent. Select one of your voice agents from Settings → AI Agents.

  3. Outbound Number. Select the number the agent calls from, or leave the automatic choice to use a number routed to that agent.

  4. Wait for Call to End (Minutes). If AXL gets no response within this time, the automation continues through the No Answer branch.

  5. Call back if there is no answer. Turn this on to retry unanswered or busy calls, and set the delay before each additional attempt. Voicemail also triggers these retries; after the last attempt, the call follows the Voicemail branch. Calls answered by a person are not redialed by this setting.

  6. Analyze call outcome. Turn this on to have AI read an answered call’s transcript and choose a matching outcome. This uses paid AI credits when there is a transcript to analyze and a configured outcome to evaluate. The provider’s answered, unanswered, voicemail and failed statuses work without this transcript analysis; that does not make the call itself free. Calling charges are separate. Add custom outcomes such as Ready to Buy with the condition “The customer confirmed that they want to purchase or asked how to pay,” or Not Ready to Buy with a condition describing a declined purchase. Connect each outcome to its next action.

  7. Call Branches. In addition to custom outcomes, the step can include technical call branches:

    - Call Connected: The person answered and talked with the agent. This branch runs for every answered call, alongside the outcome branch. Connect anything here that should happen regardless of the outcome, such as saving call variables.

    - Call Placed: Runs the moment the call is placed, while the step waits for the call to end. Use this branch for actions that should run in parallel, such as letting the contact know a call is coming.

    No Answer. Nobody answered, the line was busy, or the wait time expired. If retries are configured, this branch runs after they are exhausted. The provider status does not require paid transcript analysis.

    Voicemail. The call reached an answering machine. This branch runs after any configured retries. The provider status does not require paid transcript analysis.

  8. Outcome Unclear. With call analysis on, the person was reached but the conversation matched none of your outcomes.

  9. Fill variables from the call. AI reads an answered call’s transcript and extracts the facts you specify, such as budget or preferred appointment time. Give each variable a key and a description. Each extracted value becomes a {CallVar.key} tag that later actions can save to a contact field or insert into an email. Extraction uses paid AI credits. If outcome analysis and extraction are both enabled, they share one transcript-analysis request.

  10. Form the Agent Fills Out. A form whose questions the agent asks during the call, in conversation. Questions the contact card already answers are not asked. After the call, the answers are saved to the form on the contact card, and the step also leaves through Form filled in or Form not filled in, beside the outcome. Leave it empty to use the agent's own form, if it has one.

  11. Meetings the Agent Can Book. Booking pages the agent may check and book for the contact during the call. If none are chosen, the agent's own booking pages apply.

  12. Selling on Calls. Let the agent sell products sets what the agent may sell on this call and how the payment link reaches the person once the order exists. It replaces the agent's own selling settings for this step, and the step also leaves through Order created when an order was made.

  13. Testing. To check a voice agent before you activate the automation, place a test call from the agent's page (Settings → AI Agents). After you change the step, save it to apply your changes.

If an answering machine picks up lets you choose to hang up or leave a message. If you choose Leave a message, enter the message and review the notice in the form. When retries are enabled, the message is left only on the last attempt.

AI Call step settings: step name Qualify the lead, the Northstar Front Desk voice agent, the outbound number, a 30-minute wait and a second attempt after 30 minutes.
Analyze call outcome turned on, the Call Connected, Call Placed, No Answer and Voicemail branches, a custom Ready to buy outcome with its condition, Add Outcome and Outcome Unclear.
Fill variables from the call with a team_size variable and its CallVar tag, Form the Agent Fills Out, Meetings the Agent Can Book and Selling on Calls.

Running the Automation

Connect all the required actions and set the automation to Active. When the automation runs for a contact, the AI Call action automatically calls the phone number on that contact’s record.

Important: with an AXL voice agent, the call is not placed when the contact is marked Do Not Call in their card, when the account's telephony balance is used up, or when the AI spending limit for agents is reached. Check these before testing, and make sure you have the contacts' consent to be called.

After the call ends, the automation:

  1. determines the technical call result;

  2. analyzes the conversation if call analysis is on;

  3. extracts the variables you configured;

  4. selects the matching branch;

  5. runs the actions connected to that branch.

Automation with Automation Started, the Qualify the lead AI Call step and its branches, and an Add Task step connected to the Ready to buy branch.

As a result, AI Calls work as a complete step in CRM automation, from the first customer interaction to lead qualification and saving the results of the conversation.

Check the call flow before activation

  1. Choose the voice agent and outbound number in the AI Call step. Give the step a name that describes its purpose, such as “Confirm consultation requirements”.
  2. Set the wait time and any retry intervals, then connect each outcome you use. Keep the Call Placed branch distinct from the branches that run after the conversation.
  3. If you extract call variables, connect the actions that save or use those values after the call. Defining a variable alone does not explain where the information will be used.
  4. Place a test call from the voice agent’s page to a phone number you control; it is a real call, not a preview. Review the conversation, then save the step.
  5. Activate the automation only after checking the trigger and every connected outcome. Review the resulting call and contact information after your controlled test.