Use this article to deploy Fin AI Agent over live chat (Web, iOS, Android, WhatsApp, Slack, Facebook, Instagram, and SMS), test your configuration before going live, and troubleshoot common issues — including unwanted conversation closures and Fin re-engaging after escalation. Fin is available on trial and paid plans.

You'll need to identify the audience or support topics you want Fin to handle over chat and set a success metric (e.g. resolution rate, CSAT). For example, this could include specific high-volume topics or a sub-set of customers.

Go to Fin AI Agent > Train > Content and ensure you've enabled content which Fin can use to support chat queries. Update your content to cover your frequently asked chat queries. You can add public articles, documents, snippets, or public URLs.

Go to Fin AI Agent > Train > Guidance and set up guidance to train Fin how to speak in your brand’s voice, follow your policies, and handle conversations the way you want—using simple, natural-language instructions. You can set clear rules for what Fin should say and do, from using the right terminology to handing sensitive topics over to your human support team.

Go to Fin AI Agent > Train > Attributes and set up Fin Attributes to automatically classify conversations by topic, sentiment, urgency, or any custom attribute you define. Fin sets these attributes during a conversation, and you can use them as conditions in workflows to automatically route or escalate conversations to the right team.

Go to Fin AI Agent > Train > Escalation and configure Escalation guidance to define when and how Fin should hand conversations over to your human support team. Use natural-language or rules to specify escalation conditions—for example, escalating when a customer asks the same question twice, mentions a specific keyword, or expresses frustration.

Go to Fin AI Agent > Train > Procedures and build Procedures to automate more complex processes. Often times, these processes may involve actions in your external systems that need to reliably follow your specific business rules (e.g. cancel an order, refund a subscription). Fin will trigger the procedure and be actively involved each step to resolve customer queries.

From Fin AI Agent > Test you can see if you're ready to deploy Fin by testing how Fin responds to actual customer queries.

Simply generate questions from previous conversations, bulk upload questions via a CSV, or manually add specific questions you want to test Fin on. Then review Fin's responses and mark them as "Good" or "Poor".

When marking a response as "Poor" you can leave a reason and dig into where Fin's answer came from (i.e. which content source, guidance, or personality settings were used). You'll also get recommendations on how to improve Fin's answer. Then you can export this as a report to share with your teammates.

It's a good idea to set Fin live over chat to a small test or internal audience (e.g. yourself and teammates) and review the full customer experience from Fin's greeting to it's response and handover.

Expand the audience Fin is replying to, so that Fin is replying to some real customer conversations. Review the results from Fin AI Agent > Analyze then expand Fin to handle more of your volume or topics.

Go to Fin AI Agent > Deploy > Chat. At the top of this page, you'll see "Simple deploy" where you can:

First, decide which of your customers can interact with Fin AI Agent. You can allow Fin to provide answers to your users, leads, and/or visitors.

Click + Add audience rule and use data attributes in Intercom to target your audience further. Then click Save.

You can also use channel attributes in the audience rules to configure channel-specific actions and content. Then click Save and move to the next step.

Now select which chat channels you'd like Fin to be available on:

Then click Save.

Now configure Fin's introduction message to customers who reach out through chat. This helps your customers better understand what Fin can do and how to interact with it. You only need to write Fin's introductory messages in your workspace's default language. When you click Save, these messages will be automatically translated into all of the supported languages you have enabled for Fin. Fin will then greet customers in their detected language.

Review which content sources Fin uses to answer questions. Go to Fin AI Agent > Train > Content and confirm the articles, snippets, and documents you want Fin to use are enabled.

Review the guidance rules that shape how Fin responds. Go to Fin AI Agent > Train > Guidance to check your active rules — these control Fin’s tone, terminology, escalation behavior, and any topic-specific instructions.

Specify what action should be taken if Fin cannot resolve the conversation. You can choose how Fin hands over or escalates to a particular team or teammate.

To gather more information from a customer before Fin hands over to your team, simply toggle on "Collect more information when a customer asks to speak to the team".

Fin will encourage the user to share more information. The exact message it sends will be something along the lines of "Sure! While I connect you, could you provide us with more details about your issue to help us find an answer faster?" This will be adapted to the context of the conversation and your workspace’s Fin personality settings.

By prompting the customer for more context before Fin hands over to a teammate, there is another opportunity to get more and/or better information than presented the first time around. The benefit to this is two-fold:

Or, instead of escalating conversations straight to your team, choose to close conversation and offer other ways for the customer to get help. For example, provide a contact number or link to another resource.

Then click Save and move to the next step.

Choose when to auto-close conversations by specifying a length of time Fin should wait before closing conversations.

Select if you want Fin to automatically close the conversation after customer becomes inactive:

You can also Customize what Fin says when closing the conversation to make sure the customer got what they needed and remind them they can still speak to your support team, even if the conversation is being closed. The closing messages will also be automatically translated into all your supported languages when you click Save.

At the end of a customer's interaction with Fin AI Agent you can ask them to give a conversation rating by toggling these settings on:

To preview Fin over chat before setting it live:

To preview Fin over chat as a specific user or brand, select the Preview user dropdown at the top of the preview and then select User or lead. This let's you impersonate real users/leads and see exactly how Fin will respond. You can simulate real scenarios using live or dummy data, test data connectors, and validate every answer.

Once you've finished configuring Fin's chat experience, click Set Fin Live 🎉 Fin will immediately begin handling conversations for the audience you’ve configured. To pause Fin at any time, return to Fin AI Agent > Deploy > Chat and toggle Fin off.

If you want to add Fin to a workflow, or customize Fin in an existing workflow, go to Fin AI Agent > Deploy > Chat and open Advanced setup through Workflows.

If you have an existing workflow set up specifically for inbound conversations through the Messenger or other chat channels:

If you don't have a workflow set up specifically for inbound conversations through the Messenger or other chat channels:

To customize Fin's behavior in a workflow, open your workflow and click on the Let Fin handle step. Here, you can:

By using branching, you can create different Fin experiences for different audiences.

For example:

Add as many branches as you need to create different experiences for your different customers.

You can also use branching to create different Fin experiences and paths based on your different conversation topics.

For example, you can use a "Message Content" filter to escalate conversations which contain the keyword "billing" directly to your Billing team.

If you want to test a Fin over chat workflow, select the Preview button at the top of the Workflows builder. This will give you an interactive preview of the Messenger experience when talking to Fin.

After you've deployed Fin over chat, go to Fin AI Agent > Analyze to get real-time metrics like Customer Experience (CX) score, automated conversation analysis at scale, and AI-powered suggestions for optimizing Fin.

Here, you'll find the following dashboards:

As you analyze Fin’s performance and identify opportunities for improvement, you’ll need to continuously train, test, and deploy Fin to achieve an increasing number of Outcomes.

If Fin is closing conversations prematurely:

If Fin is interfering with handovers when customers request a human agent:

When Fin escalates a conversation, it ends its active session and stops responding. However, Fin can be re-engaged on a subsequent customer message if a workflow trigger fires again after the escalation. The most common cause is using the “Customer sends any message” trigger in a workflow that contains Fin — this fires on every customer message, including those sent after escalation, routing the conversation back to Fin.

To avoid this:

Monitor conversations escalated to humans by creating a custom view with filters:

This view allows detailed tracking of escalated items.

Control routing through the bot inbox configuration. Define handover rules that: