This feature is currently in beta:
The feature scope described here is expected to change frequently.
We may run A/B testing of this feature.
What are personalized answers?
Personalized answers allow you to connect Fin AI Agent to your external systems like admin tools, project management and payment processing platforms, CRMs and more. This means that Fin can answer questions that are very specific to each customer, for example:
What's the status of my order?
What subscription do I have?
Is there an ongoing incident?
Each personalized answer consists of an API call which you can configure behind the scenes. Fin AI Agent will automatically choose when to use it and deliver an answer specific to the customer.
You can connect any system which has an API. That might be your unique internal backend tool you have built for your business or a third party tool you're using like Shopify, Salesforce, Stripe, Jira, and more.
Setting up personalized answers
Navigate to Fin AI Agent > Overview > Add content and select Use Actions.
To create a new personalized answer, select Manage or navigate directly to Custom Actions in Settings > Integrations > Custom Actions.
Select Create action and choose the setup option:
Create custom action (manual setup)
Create action from template (currently only Shopify and Statuspage is supported)
Create custom action (manual setup)
We’ll use the following use case as an example:
An electricity utility company that has customer electricity usage information in their own backend system.
About section
Give your personalized answer a unique name that explains exactly what it will be doing (e.g. “Get electricity usage”).
Select "Fin AI Agent" in the "Usage" section.
Audience rules can be used to define the customers for who Fin can call the action during a conversation. If you wish to test the personalized answer before enabling it for customers, you can use audience rules to enable it just for test accounts initially.
‘When should Fin AI Chabot use this action ’ input specifies when this personalized answer should be triggered to read the external data.
Note:
Identity verification must be configured in order to set a personalized answer live for users. Identity verification stops anyone impersonating your logged-in users or seeing their conversations. Learn how to set it up.
Personalized answers only support use cases to read data in external systems. Updating data in external systems is not yet supported (coming soon).
Request section
Next, fill in the request details. This is where you’ll need to enter the API Endpoint to the 3rd party system you’re connecting to (e.g. Shopify, Salesforce, etc.). You’ll typically find this information in the API docs of the system you’re connecting to.
API Endpoint: Enter the API endpoint method and URL. You can insert attributes from User, Conversation, Company, and events as templated values in the URL.
HTTP Headers: You can choose to add any additional parameters to this request by selecting + New key value pair under HTTP Headers.
Authentication: Personalized answers support both fixed and dynamic tokens for authentication. Under the Authentication tab, you can select the authentication token you want to use in the request, which will then be added to the header.
Testing section
Next, you’ll need to test this personalized answer to make sure it’s fetching the right data from the 3rd party system you’re connecting to.
There are two options for testing:
1. Test Response - Select when you want to test with an API endpoint
Enter a test value (e.g. User ID) to check the personalized answer has been configured correctly.
You should get a green tick and see the details of that request if the connection is validated with the API.
2. Mock Response - Select when you do not have an API available
With this option you can add mock data in JSON format. This enables you to test how Fin would work with certain data responses without you needing to have the APIs built out first.
Click Save and set live in order to use the preview.
Response section
When using personalized answers with Fin, there is no need to map response data to Intercom attributes or objects. Instead, Fin directly interprets the JSON response and can then use it to resolve questions. The ‘Response’ section is generated based on the response in the ‘Testing’ section. Each line item corresponds to a data point in the JSON response.
By default, Fin can access the full response data to generate replies.
Select Fin can access selected data if you wish to limit the data Fin can read. Then check the data fields you want to give Fin access to.
Field name and field values can be transformed using the edit button.
When you’re done setting up your action select Set live.
Create custom action from a template
Personalized answer templates are available when the following apps are installed in your workspace:
If the app is not already installed, got to the App Store to install. Once the app is installed, action templates for each app will appear as an option when creating a new personalized answer.
On selecting a template, this will automatically complete all sections. No further configuration is required.
If you’re happy with the action behavior, just select Set live.
Preview
The 'Preview Answer' feature is not enabled unless either 'Test Response' or a 'Mock Response' has been generated (see more in the Test Section).
Once enabled, the Preview will have access to:
All live personalized answers in that workspace.
The specific personalized answer that is being built.
Setting live / roll-out
We recommend using the audience targeting to roll out personalized answers in phases to your customer base. This enables you to validate the performance of the action and make tweaks/changes where necessary.
Monitor and manage
Navigate to Fin AI Agent > Overview > Add content and select Use Actions.
The number of actions that are currently live for Fin to use is highlighted in blue.
Viewing in the inbox
Go to the Help Desk conversation and select Show conversation events. Events will show if Fin has access to the action and if the action is successfully triggered.
Note:
If there’s an error with the personalized answer triggering, select See logs to find out why.
Fin may not always choose to use the personalized answer despite making an API request if other content is determined to be more relevant.
Inbox views can be created with the attribute “Fin AI Agent: Action used in reply”. This attribute is set if Fin calls a personalized answer and uses some/all of the response in an answer.
Data privacy and security
How do I ensure personalized answers don't share personal data with someone who isn't the user?
Before setting a personalized answer live for your logged in users, you are required to set up Identify Verification. Personalized answers are only exposed to users who's identity has been verified.
With Identity Verification, you generate a unique user hash for each of your users based on their email address or user_id and your workspaces’s identity verification secret (available from your Intercom security settings). Your integration will generate and send these hashes along with every Messenger request, which avoids bad actors from impersonating your users and Fin delivering a personalized answer using someone else's data.
How do I ensure that personalized answers don't accidentally share information from another user?
When setting up a personalized answer, you can pass a parameter into the request which ensures that the answer that is shared will be specific to the user Fin is talking to. These parameters can be templated as one of the values on the user record.
For example, here you’re passing a “User ID” parameter in the request, meaning that Fin will only return data from the specific user who matches this “User ID”.
Additionally, we recommend that you prevent the ability for users to make updates via the Messenger to parameters used in a personalized answer. You can do so from Settings > Data > People, go into the relevant attribute and toggle on "Prevent updates via the Messenger".
Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts