Note: Fin Tasks are currently in closed beta. We’re focused on ensuring that customers get maximum value from the capability and are set up for success. During this phase, we’re prioritizing access and support for customers with higher support volumes and greater technical readiness. We’re actively working on a version of the product that will be suitable for all customers in the full release.

What to consider when setting up Fin Tasks

Setting up Fin Tasks and using Data connectors takes a blend of technical awareness, process thinking, and collaboration. While the experience is becoming easier over time, here are the key areas to focus on:

Choose high-impact processes to automate: Start by identifying 1–2 workflows where automation could meaningfully reduce support volume. Look for repeatable processes that currently rely on human intervention. Your Intercom Overview report is a great place to spot trends and uncover common issues being routed to your team.
Understand your API requirements: Many Tasks depend on connecting to APIs to fetch or update data (i.e Data connectors). Make sure you know which APIs are needed, what they return, and how they’ll be used within the Task.
Work cross-functionally: Setting up Tasks often requires input from product, engineering, or operations teams—especially when APIs need to be built or updated. Clear communication around the desired workflow ensures the right systems are in place.
Test and iterate: Building great Tasks is rarely a one-shot effort. Expect to test different approaches with Task instructions, debug edge cases, and refine as you go to improve reliability and end-user experience.

Getting these fundamentals right will help you launch faster and build more effective Fin-powered Tasks.

Create a Fin Task

Visit Fin AI Agent > Train > Tasks and click on New task.

1. Give your task trigger a title and description

Every Fin task should be given a title telling Fin when to trigger the task. This is very important, so that Fin triggers it at the right time.

Similarly to the title, the description also tells Fin when it should trigger the task, in greater detail. This describes the situation Fin should use the task, provide example customer queries, and could also call out situations where Fin should not use this task.

Read our best practices for writing a task title and description.

2. Set channels and audience rules

The task trigger is also where you define the channels and audience rules which need to match before Fin performs the task.

3. Add instructions

Use the instructions block to give Fin custom instructions to perform the task. You can template the instructions with data, get Fin to use Data connectors, store response data and reply to users.

Provide as much information as possible in the instructions, any information not provided may be incorrectly inferred by Fin.

Start each instruction with a verb. Be clear, concise, and as complete as possible. Long instructions are less reliable so use multiple steps for each instruction if needed.

Fin can send a one-time password to the customer’s email and require them to enter it before performing the task.

Read our best practices for writing effective instructions.

AI-generated task instructions

With AI-generated task instructions, you can simply describe what you want Fin to do and it will automatically generate a structured task prompt.

This allows for:

Faster setup: Skip the blank page and let AI do the heavy lifting!
Better structure: Generated prompts follow best practices for Fin Tasks —complete with steps, end conditions, and optional guidance.
Easy to edit: Once generated, prompts are fully editable, so you can tweak and customize as needed.

Add Data connectors `[optional]`

Data connectors can be used in the instruction block to retrieve or update data in an external system. Add a Data connector by typing "@" in line with your text instructions and click Connect to external, then select one of your pre-configured Data connectors.

Note: You'll need to configure Data connectors before you can use them in Fin Tasks. For more info on how to design and use your APIs with Data connectors, go to this article.

The response data from the Data connector will become available as "temporary attributes" within the instructions block (see below).

Use attributes `[optional]`

Add attributes to the instructions block by typing "@" in line with your text instructions. This allows Fin to read and update attributes based on your specific business rules, at clear moments, with precise values. You can use any people, company, conversation, or temporary attribute.

Use temporary attributes `[optional]`

Temporary attributes are useful when you need data only for the current task and don't want to save it permanently in Intercom for reporting or other purposes. You can create these attributes to hold information that's relevant to the specific task being executed.

For example you can create a temporary attribute to:

Capture a customer's selected booking ID
Customer’s preferred delivery date
Ask Fin to classify user query or collect some data

Learn more about attributes and how they can power Fin Tasks and Data connectors.

Tell Fin when a task is complete

In the "End this task when" section, specify the conditions that indicate the task is complete. This helps Fin avoid ending the task too early, repeating steps, or getting stuck. Define clear criteria for completion, regardless of whether it's a positive or negative outcome for the customer.

Guide how Fin behaves during a task `[optional]`

Guide how Fin interacts, responds, and behaves during a task by providing clear guidance. Simply describe any specific behaviors you’d like Fin to follow while performing the steps.

Note: Fin will combine any global Fin Guidance you've set with any custom guidance you add here.

Use "Wait for webhook" `[optional]`

Use Wait for webhook to pause the task until it receives a webhook request. This is ideal for processes that require external input before moving forward.

Imagine your task needs to integrate with an external system—such as a third-party application for identity verification. With Wait for webhook, the task stops at this step until the external system sends a webhook response. Once received, the task resumes and takes action based on the provided data.

Examples

Identity Verification: Pause the task until a third-party service confirms a user's identity.
Payment Confirmation: Wait for a webhook from your payment provider to confirm a transaction before proceeding.
Order Fulfillment: Pause until your inventory system sends an update about stock availability.
Approval Processes: Hold the task until a third-party tool sends an approval or rejection status.

This feature ensures your tasks are seamless and synchronized with external systems, reducing manual intervention and keeping things efficient! To add Wait for webhook to your task, simply click Add step and search for “Wait for webhook”.

Best practices when using "Wait for webhook"

Add a clear name to the step to make it easy to identify temporary variables created for the step
Configure the example request properly by including all the fields you expect to receive in the webhook callback within the data object.
Ensure the webhook returns a JSON body that matches the structure in the step. If the JSON structure is updated, no data will be received. Missing fields in the configuration will also make those fields inaccessible in the task.
Click the "Copy" button to copy the webhook URL to your clipboard. On the third-party site where the webhook is triggered, ensure the conversationId is passed as expected.

Note: The webhook endpoint supports idempotency to safely retry requests and prevent the same operation from being performed more than once. When sending a request, include an 'Idempotency-Key' header with a unique value. We recommend using UUIDs. Idempotency keys expire after 24 hours.