Fin actions offer a powerful way to deliver personalized answers by connecting to data in your third-party systems. Instead of customers waiting for teammates to manually look up the information, or being directed to find information themselves, Fin can instantly access the relevant data and craft a personalized reply.

To ensure these actions work correctly and are used in the right scenarios, here are a few best practices to consider…

One of the most important steps is to write an effective description in the When to use section of your action.

You should write 3-5 sentences to describe when Fin AI Agent should trigger this action. Be as specific as possible and make sure to include one or more of the following:

You can also check out the AI recommended templates as an example.

Clearly explain when this action should be used and the types of details it should provide.

Replace pronouns such as “we”, “us”, and “their” with the proper noun you’re referring to such as your company name.

Use common scenarios in which triggering this action would be appropriate.

Look at the way customers have phrased their queries in previous conversations and include the keywords and phrases they use.

In order to get customers the correct data, it’s important to choose the right API endpoint, whether it’s from a third-party app or an external system that is owned by your company.

If you’re not sure where to start, you may want to find an engineer to help you figure out which API will return the data that is required.

For example, if you have an internal booking system that stores a customer’s booking details, you will want to use an API for that system that can access booking details directly.

You will also want to use an API that can find a customer’s details using a unique identifier as a parameter, like an ID or an email address.

If you decided to use a customer’s name or email address as a request parameter, and they had multiple bookings, you would probably need to specify the request further by using the date of the reservation. Rather than requesting multiple pieces of data, it would be optimal to just ask for one — the Booking ID.

This way you can collect the Booking ID as a data input, which will prompt Fin to get this information from the customer in the conversation. Then the ID can be inserted into the endpoint by hovering over the menu at the end of the URL input form box.

You will also want to ensure you are using the correct HTTP headers for the request. If these will be used across many actions, you can add them as a stored authentication token, which will securely save them in your workspace.

Using our same booking example, let’s say that a customer wants to rebook their trip for another day. If there is an action available for checking booking rates and availability, Fin 2 will automatically choose it within the same conversation.

When you are choosing API endpoints to use for actions, you may want to think about cases where the data your customers need would be across multiple endpoints.

To ensure the customer can see the rates and availability, you can create another action to an endpoint that contains that information.

If you are just starting out with actions, you may want to experiment with which actions configurations lead to more resolutions. One way to do this is to first activate the action for a subset of users to see how it performs. You can do this by using audience rules.

Audience rules can be selected from a list of company or contact attributes that are already available in your workspace. If you have Shopify, Statuspage, or Stripe installed, you can also access attributes specific to the app.

If an attribute that you want to use does not exist, you can set up custom data attributes to identify users who meet the criteria that you want to target for the action.

Once you have created the rule, you can see a preview of the users who may receive a response powered by the action.

While you can use the logs to check on the performance of individual actions calls, another way to measure performance is to take a look at conversations where an action was used and see what worked well or what missed the mark.

If you are using custom data attributes to narrow the audience, you can then utilize the attribute to filter through the inbox and take a look to see how it performed, and if any adjustments need to be made.

Some questions to consider include: